{"service":"Global CEO AI Expertise Index","rubric_version":4,"count":111,"generated_at":"2026-09-20T12:40:59.499437+00:00","methodology":"Every founder/CEO is scored on rubric v4: nine dimensions, 0-20 each. The six research dimensions (foundations, vector_embeddings, transformers_lm, frontier_founder = their own work is part of the foundation today's frontier models are built on — word2vec being the foundational building block, its verifiable pre-2013 vector-space / distributional-semantics lineage included, lm_domain_depth = years and history in language modeling across papers, patents and shipped systems, lm_domain_breadth = experience across different types of language modeling: biological, financial, code, chemistry, legal, music …) carry 70% of the total and the three practice dimensions (hands_on_engineering, industry_impact, scientific_founder = operating as the scientific/technical founder of a company, scaled by years of experience and the number of such companies) carry 30%: weighted_score = round(70*(F+V+T+FF+DD+DB)/120 + 30*(H+I+SF)/60). Rows still carrying rubric_version 3 recompute on the v3 formula round(70*(F+V+T+FF+DD)/100 + 30*(H+I+SF)/60) and rubric_version 2 rows on the v2 formula round(70*(F+V+T)/60 + 30*(H+I)/40). Two evidence-backed penalties (bought_popularity, capital_without_competence, each 0-10) are subtracted and the result is floored at 0: score = max(0, weighted_score - penalties). Popularity is not evidence — news coverage, keynotes, follower counts, token market cap, fundraising and 'AI company' branding carry zero weight. Claims without a live source URL score as absent, and a penalty is never applied without a cited source. Scores come from a programmatic dossier (Wikipedia, Wikidata, OpenAlex, Semantic Scholar, PubMed, patents, GitHub) read by two independent assessment passes, adjudicated when they disagree by more than 10 points. Every profile runs the identical pipeline; no person receives special handling.","methodology_url":"/api/v1/ceo-ai-leaderboard/methodology","how_to_validate":["1. GET https://cymetica.com/api/v1/ceo-ai-leaderboard/export.json (no auth; honour the ETag).","2. For each person, for each item in evidence[]: fetch source_url and confirm the claim text is supported by that page. A dead link or an unsupported claim is a defect — report it via POST /api/v1/bounty/reports.","3. Recompute weighted_score from dimensions on the row's rubric_version. v4: round(70*(foundations+vector_embeddings+transformers_lm+frontier_founder+lm_domain_depth+lm_domain_breadth)/120 + 30*(hands_on_engineering+industry_impact+scientific_founder)/60). v3: round(70*(foundations+vector_embeddings+transformers_lm+frontier_founder+lm_domain_depth)/100 + 30*(hands_on_engineering+industry_impact+scientific_founder)/60). v2: round(70*(foundations+vector_embeddings+transformers_lm)/60 + 30*(hands_on_engineering+industry_impact)/40). Rounding is half-up on the exact fraction. It must equal the published weighted_score.","4. Recompute score: max(0, weighted_score - penalties.bought_popularity - penalties.capital_without_competence). It must equal the published score.","5. Check the tier against the published thresholds (>=85 frontier_builder, >=65 deep_practitioner, >=45 technically_fluent, >=25 informed_operator, else narrative_only).","6. Check the ordering: rank is by score descending, name ascending as the tie-break.","7. Check the invariants in /methodology .invariants against metadata.canonical_papers — e.g. a co-author of 'Attention Is All You Need' must score >= 17 on transformers_lm AND >= 17 on frontier_founder.","8. scripts/ceo_ai_index/validate_export.py in the public repo performs steps 1-6 and writes validation.json; run it yourself rather than trusting ours."],"rubric":{"rubric_version":4,"dimensions":[{"key":"foundations","label":"Mathematical Foundations","max":20,"description":"Degrees, theses, papers and code in linear algebra, matrix & tensor methods, optimization and statistical learning — the math the field stands on.","group":"core_research","added_in":2},{"key":"vector_embeddings","label":"Vector Embeddings","max":20,"description":"Vector-space models, LSA/LSI, word and sentence embeddings, contrastive / dense retrieval, vector databases and search — authored, built or shipped.","group":"core_research","added_in":2},{"key":"transformers_lm","label":"Transformer & LM Lineage","max":20,"description":"seq2seq, attention, transformers, pretraining, scaling laws and alignment — authored, led or trained.","group":"core_research","added_in":2},{"key":"frontier_founder","label":"Frontier Founder","max":20,"description":"Frontier Founder — the person's OWN work is part of the foundation today's frontier AI models are built on. word2vec (2013) is the foundational building block of today's language modeling, so the foundation includes the verifiable pre-2013 vector-space / distributional-semantics / relationship-network lineage that produced it (papers, patents, shipped systems) as well as the later blocks: attention, transformers, embeddings, optimizers, tokenizers, pretraining objectives, scaling results, alignment methods, datasets, benchmarks, training / inference stacks. Scored by verifiable position in that lineage, not by whether a frontier lab cites the person by name.","group":"core_research","added_in":3},{"key":"lm_domain_depth","label":"Deep Knowledge Domain Expert","max":20,"description":"Deep Knowledge Domain Expert — years and history in language modeling, the tip of the spear in AI today: depth AND duration of a verifiable, hands-on record from vector-space / LSI / n-gram and neural LMs through transformers and LLM pretraining / alignment. Hands-on years count across papers, patents and shipped language-modeling / vector-space systems; the score sits in the band matching the reported year count.","group":"core_research","added_in":3},{"key":"lm_domain_breadth","label":"Cross-Domain Language Modeling","max":20,"description":"Cross-Domain Language Modeling — hands-on experience in different TYPES of language modeling, not only natural-language text: biological (DNA / RNA / protein sequence models, gene-expression and biomedical-literature mining), financial (market / filings / news / prediction-market models), source code, chemistry / materials, legal / clinical / scientific-literature mining, music / media sequence models. Scored on the number of distinct domains with a verifiable hands-on record (papers, patents, shipped systems) and the depth in each; applying a vendor's chatbot to a domain does not count.","group":"core_research","added_in":4},{"key":"hands_on_engineering","label":"Hands-On Engineering","max":20,"description":"Personally designed, built or shipped AI systems, models, or the hardware and infrastructure under them (accelerators, training stacks, inference).","group":"practice","added_in":2},{"key":"industry_impact","label":"Scientific & Industry Impact","max":20,"description":"Built organizations or products whose CORE is these systems; citations / h-index; patents; leadership of labs that produced canonical work.","group":"practice","added_in":2},{"key":"scientific_founder","label":"Scientific & Technical Founder","max":20,"description":"Scientific & Technical Founder — operating as the scientific / technical founder of a company (founder-CTO, founder-Chief Scientist, or a founder-CEO who personally sets and executes the technical direction), scaled by the number of verifiable years of experience doing so AND the number of such companies founded in that role. A founder title with the science done by others does not earn it.","group":"practice","added_in":3}],"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"anchors":{"18-20":"Authored canonical work the field builds on / principal builder of systems the field runs on.","13-17":"PhD-level work, or production systems built and led personally.","8-12":"Strong graduate training, or senior engineering adjacent to the core.","3-7":"Uses the tools, manages builders, no personal record.","0-2":"Nothing verifiable."},"anchors_by_dimension":{"frontier_founder":{"18-20":"Authored / built a method, architecture, dataset or system today's frontier models directly descend from (transformer & attention, scaling laws, RLHF / instruction tuning, word2vec / GloVe, the canonical training or inference stacks).","13-17":"A documented component the frontier labs build on (optimizer, tokenizer, positional encoding, retrieval method, benchmark, alignment technique), OR verifiable pre-word2vec (pre-2013) vector-space / distributional-semantics / relationship-network patents or shipped systems in the lineage word2vec-era embeddings descend from.","8-12":"Published lineage work the frontier stack draws on that is neither a named block nor a pre-2013 precursor system.","3-7":"Applies or fine-tunes frontier models; no foundational contribution.","0-2":"Nothing verifiable."},"lm_domain_depth":{"18-20":"15+ years of hands-on language-modeling work (papers, patents or shipped LM / vector-space systems) from the pre-word2vec era (vector-space / LSI / n-gram / early neural LMs) through the transformer era, still active.","13-17":"8-15 years of personal language-modeling research or systems work.","8-12":"3-8 years with a real record.","3-7":"Under 3 years, or adjacent (general ML with no language-modeling record).","0-2":"Nothing verifiable."},"lm_domain_breadth":{"18-20":"Four or more distinct language-modeling domains each with a real hands-on record (papers, patents or shipped systems), at least two of them outside natural-language text (for example biological AND financial).","13-17":"Three domains, or two domains each with a deep multi-year record.","8-12":"Two domains with a verifiable record.","3-7":"A single domain (natural-language text only), or domain applications of someone else's models with no modeling work.","0-2":"Nothing verifiable."},"scientific_founder":{"18-20":"15+ years operating as the scientific / technical founder of companies whose core is these systems, personally authoring the core research, code or patents — or three or more such companies founded in that role across 10+ years.","13-17":"8-15 years in that role, or two such companies.","8-12":"3-8 years as a verifiable technical founder of one company.","3-7":"Founder or CEO of an AI company whose science and engineering were done by others, or a technical founder outside this field.","0-2":"Nothing verifiable."}},"weighting":{"core_research_dimensions":["foundations","vector_embeddings","transformers_lm","frontier_founder","lm_domain_depth","lm_domain_breadth"],"core_research_weight":0.7,"practice_dimensions":["hands_on_engineering","industry_impact","scientific_founder"],"practice_weight":0.3,"formula":"weighted_score = round(70 * (foundations + vector_embeddings + transformers_lm + frontier_founder + lm_domain_depth + lm_domain_breadth) / 120 + 30 * (hands_on_engineering + industry_impact + scientific_founder) / 60)","formula_v3":"weighted_score = round(70 * (foundations + vector_embeddings + transformers_lm + frontier_founder + lm_domain_depth) / 100 + 30 * (hands_on_engineering + industry_impact + scientific_founder) / 60)","formula_v2":"weighted_score = round(70 * (foundations + vector_embeddings + transformers_lm) / 60 + 30 * (hands_on_engineering + industry_impact) / 40)"},"penalties":[{"key":"bought_popularity","label":"Pay-for-play / bought popularity","max":10,"description":"Paid coverage, paid placements, purchased followers or reach."},{"key":"capital_without_competence","label":"Capital without competence","max":10,"description":"Founded or funded an AI company on family / friends / personal wealth with no verifiable language-modeling knowledge."}],"penalty_rule":"score = max(0, weighted_score - sum(penalties)). A penalty is applied ONLY with a live cited source URL; never on rumour.","score_formula":"score = max(0, weighted_score - bought_popularity - capital_without_competence)","max_score":100,"tiers":[{"key":"frontier_builder","label":"Frontier Builder","min_score":85,"description":"Authored the mathematics, embedding or transformer work the field builds on, and built the systems that run it."},{"key":"deep_practitioner","label":"Deep Practitioner","min_score":65,"description":"Personally built, trained or led core embedding / language-model systems, with a real publication or engineering record behind it."},{"key":"technically_fluent","label":"Technically Fluent","min_score":45,"description":"Graduate-level grounding in the math and the model lineage; applies it, but is not a primary author or builder."},{"key":"informed_operator","label":"Informed Operator","min_score":25,"description":"Runs AI-adjacent organizations. The expertise is operational — the models were built by other people."},{"key":"narrative_only","label":"Narrative Only","min_score":0,"description":"No verifiable record in the mathematics, embeddings or the transformer / language-model lineage. The claim is narrative."}],"sectors":["crypto","general"],"min_confidence_to_publish":0.45,"notes":"Popularity is not evidence: news coverage, keynote presence, follower counts, token market cap, fundraising and \"AI company\" branding carry zero weight and may not appear in a rationale as support. Depth of experience counts — pre-2013 (pre-word2vec) vector-space / LSI work is foundational lineage, not \"old\". Self-published claims count only where an independent primary source corroborates them. Every profile is scored by the identical pipeline; there is no special handling for any person, including the platform's own founder.","legacy_dimension_labels":{"research":"LM Research","vector_space":"Vector Space","hands_on":"Hands-On","technical_communication":"Technical Depth","track_record":"Track Record","foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","scientific_founder":"Scientific & Technical Founder"},"added_dimensions_scored_by_delta":["frontier_founder","lm_domain_depth","lm_domain_breadth","scientific_founder"]},"human_page":"https://cymetica.com/leaderboard/ai-ceos","license":"CC BY 4.0 — reuse with attribution to cymetica.com","people":[{"slug":"ilya-sutskever","name":"Ilya Sutskever","title":"Co-Founder & Chief Scientist","company":"Safe Superintelligence Inc.","sector":"general","rank":1,"rubric_version":4,"dimensions":{"foundations":20,"vector_embeddings":19,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":19,"lm_domain_breadth":12,"hands_on_engineering":20,"industry_impact":20,"scientific_founder":16},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":92,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":92,"tier":"frontier_builder","tier_label":"Frontier Builder","confidence":0.96,"rationale":"Sutskever earned a PhD in computer science at the University of Toronto (thesis: 'Training Recurrent Neural Networks', 2013) under Geoffrey Hinton, and personally co-authored AlexNet (2012, with Krizhevsky and Hinton) which catalyzed the deep-learning era. He co-invented sequence-to-sequence learning with attention-adjacent architectures (Sutskever, Vinyals, Le 2014), a direct precursor in the seq2seq->transformer lineage, and was a co-author on 'Distributed Representations of Words and Phrases' (word2vec, 2013). As OpenAI co-founder and chief scientist (2015-2024) he personally shaped GPT-2/GPT-3/GPT-4 research direction and post-training. This is a canonical, field-defining research and engineering record spanning math foundations through the full attention/transformer/scaling lineage, not organizational leadership alone.\n\nSutskever's own work is load-bearing foundation for every frontier LM: he co-authored 'Distributed Representations of Words and Phrases' (word2vec, 2013), 'Sequence to Sequence Learning with Neural Networks' (2014), neural machine-translation and sequence-generation patents (US20220101082A1, US10936828B2, US11195521B2), and — as OpenAI co-founder and Chief Scientist — the GPT-3 paper 'Language Models are Few-Shot Learners' (2020) and the direction of GPT-2/3/4, all direct ancestors of GPT/Claude/Gemini/Llama-class systems. His hands-on language-modeling record runs continuously from RNN/neural-LM work with Hinton (2008-2013 PhD thesis 'Training Recurrent Neural Networks') through seq2seq, word2vec, NMT and the GPT scaling era to Safe Superintelligence today — roughly 15+ years, still active. The distinct language-modeling types with a personal record are natural-language text (deep: word2vec, seq2seq, NMT, GPT) and vision-language multimodal (CLIP, DALL-E), with source-code modeling adjacent via Codex-era leadership; there is no verifiable biological or financial language-modeling record, which caps breadth below the top band. He has operated as scientific/technical founder of two companies whose core is these systems — OpenAI co-founder & Chief Scientist (2015-2024) and Safe Superintelligence co-founder/CEO (2024-present) — about 11 years personally authoring the core research and patents.","evidence":[{"claim":"PhD in computer science, University of Toronto, 2013, advisor Geoffrey Hinton, thesis 'Training Recurrent Neural Networks'","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-inventor of AlexNet with Alex Krizhevsky and Geoffrey Hinton (2012 ImageNet paper, 200k+ citations on Google Scholar)","source_url":"https://scholar.google.com/citations?user=x04W_mMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author 'Distributed Representations of Words and Phrases and their Compositionality' (word2vec extension, 2013)","source_url":"https://doi.org/10.48550/arxiv.1310.4546","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author 'Sequence to Sequence Learning with Neural Networks' (2014), a foundational seq2seq paper in the pre-transformer attention lineage","source_url":"https://doi.org/10.48550/arxiv.1409.3215","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAI co-founder (2015) and Chief Scientist through May 2024, overseeing GPT research; now CEO/co-founder of Safe Superintelligence Inc.","source_url":"https://www.cnbc.com/2025/07/03/ilya-sutskever-is-ceo-of-safe-superintelligence-after-meta-hired-gross.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD under Geoffrey Hinton at University of Toronto; specializes in machine learning; co-created AlexNet with Krizhevsky and Hinton; won NeurIPS Test of Time Award three years running (2022-2024)","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (x04W_mMAAAAJ) lists ~848,637 citations, h-index 109, i10-index 172; top works ImageNet/AlexNet (2012), Language Models are Few-Shot Learners (2020), CLIP (2021), Dropout (2014), Sequence to Sequence Learning with Neural Networks (2014)","source_url":"https://scholar.google.com/citations?user=x04W_mMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sutskever, Vinyals and Le won the NeurIPS 2024 Test of Time award for 'Sequence to Sequence Learning with Neural Networks'","source_url":"https://blog.neurips.cc/2024/11/27/announcing-the-neurips-2024-test-of-time-paper-awards/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Test of Time award talk for 'Distributed Representations of Words and Phrases and their Compositionality' (word2vec) at NeurIPS 2023","source_url":"https://neurips.cc/virtual/2023/test-of-time/83333","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Theory/foundations papers authored with Hinton: 'Deep, narrow sigmoid belief networks are universal approximators' (Neural Comput, 2008) and 'Temporal-kernel recurrent neural networks' (Neural Netw, 2010)","source_url":"https://pubmed.ncbi.nlm.nih.gov/18533819/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author 'Sequence to Sequence Learning with Neural Networks' (2014), foundational encoder-decoder architecture in the seq2seq→transformer lineage; won NeurIPS 2024 Test of Time","source_url":"https://doi.org/10.48550/arxiv.1409.3215","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author GPT-3 'Language Models are Few-Shot Learners' (2020) as OpenAI Chief Scientist","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAI co-founder (2015) and Chief Scientist through 2024; co-founder and CEO of Safe Superintelligence Inc. (2024-present)","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Vision-language multimodal modeling: co-author on CLIP 'Learning Transferable Visual Models From Natural Language Supervision' (2021) and DALL-E direction at OpenAI","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of word2vec ('Distributed Representations of Words and Phrases and their Compositionality', 2013), the foundational embedding building block of modern LMs; NeurIPS 2023 Test of Time recognition","source_url":"https://doi.org/10.48550/arxiv.1310.4546","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Sequence to Sequence Learning with Neural Networks' (2014), a direct precursor in the seq2seq->transformer frontier lineage; NeurIPS 2024 Test of Time award","source_url":"https://doi.org/10.48550/arxiv.1409.3215","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Early hands-on language modeling: 'Generating Text with Recurrent Neural Networks', ICML 2011 (character-level RNN language model)","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of CLIP ('Learning Transferable Visual Models From Natural Language Supervision', 2021), multimodal vision-language representation learning","source_url":"https://doi.org/10.48550/arxiv.2103.00020","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAI co-founder (2015) and Chief Scientist through May 2024; now co-founder and CEO of Safe Superintelligence Inc. (2024)","source_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, University of Toronto (2013, advisor Geoffrey Hinton)","BSc/MSc, University of Toronto/Open University of Israel"],"canonical_papers":["ImageNet Classification with Deep Convolutional Neural Networks (AlexNet, 2012)","Sequence to Sequence Learning with Neural Networks (2014)","Distributed Representations of Words and Phrases and their Compositionality (2013)","Dropout: A Simple Way to Prevent Neural Networks from Overfitting (2014)"],"first_verifiable_year":2007,"notable_systems":["AlexNet","OpenAI GPT-2/GPT-3/GPT-4 research direction","AlphaGo (co-author on Nature paper)","Safe Superintelligence Inc."],"citations":219277,"h_index":62,"patents":0,"dossier_notes":"Dossier's OpenAlex figures (h-index 62, 219k citations) are conservative relative to the live Google Scholar profile (h-index 109, 848k+ citations) — OpenAlex undercounts; both sources agree on canonical works. No homonym risk; PubMed sample entries (Hinton co-authorship) match the correct person.","years_language_modeling":15,"years_as_technical_founder":11,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2008-2026","evidence":"RNN/neural LMs (thesis 2013), word2vec 2013, seq2seq 2014, NMT patents, GPT-2/3/4 as OpenAI Chief Scientist"},{"domain":"other","years":"2021-2024","evidence":"vision-language multimodal: CLIP (contrastive language-image) 2021, DALL-E direction at OpenAI"},{"domain":"code","years":"2021-2024","evidence":"OpenAI Codex / code-model direction as Chief Scientist (leadership-level, thinner personal authorship record)"}],"frontier_lineage":["word2vec distributed word embeddings (2013)","seq2seq encoder-decoder (2014)","neural machine translation / sequence-generation patents","GPT-3 few-shot scaling paper (2020)","GPT-2/3/4 research direction at OpenAI","CLIP contrastive language-image pretraining (2021)"],"technical_founder_roles":["OpenAI — co-founder & Chief Scientist — 2015-2024","Safe Superintelligence Inc. — co-founder & CEO — 2024-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":19,"lm_domain_breadth":13,"scientific_founder":17},"confidence":0.9,"duration_ms":58442},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":19,"lm_domain_breadth":10,"scientific_founder":16},"confidence":0.9,"duration_ms":61736}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Ilya_Sutskever","wikipedia_title":"Ilya Sutskever","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"chris-re","name":"Chris Ré","title":"Co-founder (Snorkel AI, Together AI); Full Professor","company":"Stanford University / Snorkel AI / Together AI","sector":"general","rank":2,"rubric_version":4,"dimensions":{"foundations":19,"vector_embeddings":16,"transformers_lm":18,"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":16,"hands_on_engineering":19,"industry_impact":20,"scientific_founder":18},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":90,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":90,"tier":"frontier_builder","tier_label":"Frontier Builder","confidence":0.85,"rationale":"Christopher Ré is a tenured full professor of Computer Science at Stanford (Stanford AI Lab, HazyResearch lab), PhD from University of Washington (dossier's Wikidata 'Cornell' entry and doctoral-advisor field appear to conflate sources — Stanford's own profile lists him as UW PhD under Dan Suciu), MacArthur Fellow (2015) for work on machine-learning data systems. His real Google Scholar profile shows roughly 87,000+ citations, far above the dossier's OpenAlex-matched value of h-index 8 / 499 citations, which is a wrong-person match (that OpenAlex record is a University of Delaware/Beirut-affiliated researcher, not this Chris Ré). His authored/led canonical work includes data programming and weak supervision (Snorkel), DeepDive (acquired into Apple via Lattice.io), and more recently foundation-model/long-sequence architecture research (state-space model lineage) and the Evo genomic foundation model line — this is authored, field-shaping systems and research, not commentary. He co-founded Snorkel AI, Together AI (board), and SambaNova-adjacent work, translating this research into production ML infrastructure companies. Given the corrected identity, this is a canonical, high-depth research-founder record.\n\nRé is a named-block frontier-lineage author: FlashAttention (Dao, Fu, Ermon, Rudra, Ré, 2022) is the IO-aware exact-attention kernel that essentially every frontier training and inference stack now runs, and the HiPPO→S4→H3→Hyena state-space line his lab authored is the direct precursor of the Mamba/SSM branch of modern sequence modeling — this is 18-20 territory. His hands-on language/sequence-modeling record runs from text information-extraction and weak supervision for NLP (DeepDive ~2011, Snorkel 2016-17) through the transformer/SSM era (S4 2021, FlashAttention 2022, Hyena, Evo 2024-26), roughly 13-15 continuous years, placing depth in the upper 13-17 band. Distinct language-modeling domains with a genuine hands-on modeling record: natural-language text (S4/H3/FlashAttention), biological sequence modeling (Evo DNA/RNA/protein genome foundation models, Science 2024 / Nature 2026), and audio/other-modality sequence models (S4/Hyena long-sequence benchmarks) — two-plus domains outside plain text, several deep and multi-year. As scientific founder he is the research origin of multiple companies built on his own code/papers — SambaNova (2017), Snorkel AI (2019), Together AI (2022), plus Lattice.io/DeepDive and Inductiv/HoloClean (both acquired by Apple) — three-plus such companies across ~10+ years, in the 18-20 band.","evidence":[{"claim":"Full professor of Computer Science at Stanford University, Stanford AI Lab; MacArthur Fellowship 2015 for machine-learning data-systems research","source_url":"https://engineering.stanford.edu/people/chris-re","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows citation count of approximately 87,182 (far exceeding the dossier's mismatched OpenAlex figure of 499/h-index 8)","source_url":"https://scholar.google.com/citations?user=DnnCWN0AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded four ML-systems companies from his research: SambaNova, Snorkel (data programming/weak supervision), Lattice/DeepDive (acquired by Apple 2017), Inductiv/HoloClean (acquired by Apple 2020)","source_url":"https://en.wikipedia.org/wiki/Christopher_R%C3%A9","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Research spans database theory, database systems, and machine learning, with best-paper awards at PODS 2012, SIGMOD 2014, and ICML 2016","source_url":"https://cs.stanford.edu/people/chrismre/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar DnnCWN0AAAAJ (Stanford CS): 87,725 citations, h-index 121, i10-index 339; top works include FlashAttention (2022), S4 (2021), Hogwild! (2011), Snorkel (2017), HiPPO (2020), H3 (2022)","source_url":"https://scholar.google.com/citations?user=DnnCWN0AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness — authors Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Ré","source_url":"https://arxiv.org/abs/2205.14135","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cornell BS, University of Washington PhD under Dan Suciu, full professor at Stanford, MacArthur Fellowship 2015, co-founded Lattice.io (acquired by Apple May 2017)","source_url":"https://en.wikipedia.org/wiki/Christopher_R%C3%A9","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stanford faculty page: member of SAIL, CRFM and the ML Group; current lab projects include ThunderKittens AI kernels, Intelligence per Watt, and Evo foundation models for biological sequences","source_url":"https://cs.stanford.edu/~chrismre/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Efficiently Modeling Long Sequences with Structured State Spaces (S4), Gu, Goel, Ré","source_url":"https://arxiv.org/abs/2111.00396","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"HiPPO: Recurrent Memory with Optimal Polynomial Projections (2020), Gu, Dao, Ermon, Rudra, Ré","source_url":"https://arxiv.org/abs/2008.07669","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness — Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Ré","source_url":"https://arxiv.org/abs/2205.14135","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evo: sequence modeling and design from molecular to genome scale (biological language modeling), Science 2024","source_url":"https://pubmed.ncbi.nlm.nih.gov/39541441/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded SambaNova Systems, Snorkel AI, Together AI, and Lattice.io/DeepDive (acquired by Apple 2017), each built on his research","source_url":"https://cs.stanford.edu/~chrismre/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of SambaNova, Snorkel AI, Together AI; Lattice.io/DeepDive acquired by Apple 2017 — companies built on his research","source_url":"https://en.wikipedia.org/wiki/Christopher_R%C3%A9","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stanford CS faculty page listing SSM/attention/kernel and Evo foundation-model projects (HazyResearch/CRFM)","source_url":"https://cs.stanford.edu/~chrismre/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, University of Washington (per Stanford profile; dossier Wikidata lists Cornell, which appears to be an error or conflation)"],"canonical_papers":["Snorkel: Rapid Training Data Creation with Weak Supervision (2017)","Snorkel DryBell (2019)","DeepDive/data programming line of work","Evo: genome modeling and design foundation models (2024-26)"],"first_verifiable_year":2005,"notable_systems":["Snorkel / Snorkel AI","DeepDive (acquired by Apple as Lattice.io)","HoloClean/Inductiv (acquired by Apple)","Together AI (co-founder/board)","Evo genomic foundation models"],"citations":87182,"h_index":0,"patents":0,"dossier_notes":"CRITICAL: the dossier's OpenAlex match (id A5103148807, h-index 8, works_count 17, University of Delaware/Beirut affiliations) is explicitly flagged in the dossier itself as ambiguous (6 candidates) and is the WRONG homonym. The real Christopher Ré is the Stanford CS professor and MacArthur Fellow with ~87,000+ Google Scholar citations per the linked scholar profile (DnnCWN0AAAAJ). Wikipedia/Wikidata identity (Cornell doctoral education, Dan Suciu advisor) partially conflicts with Stanford's own bio (UW PhD); Stanford's official profile is treated as authoritative for the PhD institution here. PubMed sample also mixes in unrelated homonym hits (medical imaging, mental-health decoding) alongside genuinely his Evo/foundation-model papers with Brockman et al. Citations/h-index in this file reflect the corrected Google Scholar figure, not the dossier's OpenAlex figure, per brief instructions to rely on verified sources over a known-wrong match; h_index left at 0 because no reliable per-source h-index for the CORRECT profile was directly confirmed (only total citation count).","years_language_modeling":14,"years_as_technical_founder":11,"technical_founder_companies":4,"lm_domains":[{"domain":"natural_language","years":"2011-2026","evidence":"DeepDive text information extraction, Snorkel weak supervision for NLP, S4/H3/FlashAttention transformer & sequence models"},{"domain":"biological","years":"2023-2026","evidence":"Evo DNA/RNA/protein genome foundation models (Science 2024, Evo 2 Nature 2026)"},{"domain":"other","years":"2020-2023","evidence":"S4/Hyena long-range sequence models benchmarked on audio/speech and multi-modal Long Range Arena tasks"}],"frontier_lineage":["FlashAttention IO-aware exact attention kernel (frontier training/inference stacks)","HiPPO→S4→H3→Hyena structured-state-space lineage feeding the SSM/Mamba branch of modern LMs","Snorkel/data-programming weak supervision for training-data creation"],"technical_founder_roles":["SambaNova Systems — co-founder — 2017-present","Snorkel AI — co-founder (scientific origin) — 2019-present","Together AI — co-founder — 2022-present","Lattice.io/DeepDive — co-founder — ~2015-2017 (acq. Apple)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":15,"lm_domain_breadth":15,"scientific_founder":19},"confidence":0.83,"duration_ms":67076},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":16,"scientific_founder":18},"confidence":0.83,"duration_ms":76724}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Christopher_R%C3%A9","wikipedia_title":"Christopher Ré","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"noam-shazeer","name":"Noam Shazeer","title":"VP of Engineering","company":"OpenAI","sector":"general","rank":3,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":16,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":20,"lm_domain_breadth":15,"hands_on_engineering":20,"industry_impact":20,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":89,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":89,"tier":"frontier_builder","tier_label":"Frontier Builder","confidence":0.93,"rationale":"Shazeer is second author on 'Attention Is All You Need' (Vaswani et al. 2017), the paper that introduced the transformer architecture underlying essentially all modern LLMs, and first/co-author on 'Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer' (2017), the canonical MoE paper used in current frontier models. He co-authored T5 (Raffel et al. 2019), PaLM (Chowdhery et al. 2022), and Switch Transformers (Fedus, Zoph & Shazeer 2022), spanning pretraining, scaling and sparse-model architecture design across a ~25-year engineering career at Google (from ~2000, including early search-ranking/spelling-correction work) and later Character.AI, which he co-founded and which built and shipped a large-scale conversational LLM product used by tens of millions. His formal math/CS training is a Duke BS plus an incomplete UC Berkeley graduate program (no PhD), so foundations is scored high but not maximal; vector_embeddings reflects strong representation-learning work embedded in his transformer/LM papers rather than a dedicated embeddings/retrieval research line. This is a canonical, field-defining author record, not organizational leadership alone.\n\nShazeer authored core blocks of the frontier stack itself — scaled dot-product/multi-head attention and the parameter-free position representation in 'Attention Is All You Need', the sparsely-gated Mixture-of-Experts layer, multi-query attention, mesh-tensorflow/GShard tensor sharding, T5, Switch Transformers and PaLM scaling — the literal architecture, sparse-routing, inference and pretraining primitives GPT/Claude/Gemini/Llama-class systems descend from, so frontier_founder is maximal. His hands-on language-modeling record is continuous from ~1999–2000 (Google search spelling corrector / n-gram LM work) through Meena, the transformer/T5/PaLM lineage, Character.AI's conversational LLM and Gemini co-lead (2024–2026), ~26 years and still active, at the top of the depth band. He has a verifiable hands-on record in three distinct language-modeling modalities — natural-language text (deep, multi-decade), speech (attention-based speech-recognition patent, priority 2016) and image/multimodal sequence modeling (Image Transformer 2018, multi-task multi-modal system patent priority 2017) — which is real breadth but with no biological or financial modeling, so lm_domain_breadth sits in the three-domain band (B's 15 is correct; A's 8 undercounts by ignoring the image/multimodal domain). As scientific founder he co-founded Character.AI (2021) and personally set and executed its technical direction and core LLM until 2024, ~3 years across one company, anchoring scientific_founder at the f","evidence":[{"claim":"Second author on 'Attention Is All You Need' (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin, 2017), which introduced the transformer architecture","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer' (2017), the canonical mixture-of-experts scaling paper","source_url":"https://arxiv.org/abs/1701.06538","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer' (T5, Raffel et al. 2019/2020)","source_url":"https://arxiv.org/abs/1910.10683","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'PaLM: Scaling Language Modeling with Pathways' (Chowdhery et al. 2022)","source_url":"https://arxiv.org/abs/2204.02311","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows ~394,944 citations, h-index 77, i10-index 140, including 'Switch Transformers' (Fedus, Zoph, Shazeer)","source_url":"https://scholar.google.com/citations?user=wsGvgA8AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikipedia biography: Duke University math/CS education (1994-1998), Google employee from ~2000, co-founded Character.AI in 2021 with Daniel de Freitas, returned to Google in 2024 as Gemini technical co-lead, joined OpenAI as VP of Engineering in June 2026","source_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Attention Is All You Need' (2017); the paper's footnote states 'Noam proposed scaled dot-product attention, multi-head attention and the parameter-free position representation and became the other person involved in nearly every detail.'","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sole author of 'Fast Transformer Decoding: One Write-Head Is All You Need' (2019), introducing multi-query attention.","source_url":"https://arxiv.org/abs/1911.02150","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile: ~394,944 citations, h-index 77; top works include Attention Is All You Need, T5, PaLM, sparsely-gated Mixture-of-Experts, Switch Transformers, GShard, Gemini 2.5.","source_url":"https://scholar.google.com/citations?user=wsGvgA8AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS in mathematics and computer science from Duke (1994-1998); at Google from 2000, worked on the search spelling corrector and Meena with Daniel de Freitas; co-founded Character.AI in 2021; co-led Gemini with Jeff Dean and Oriol Vinyals.","source_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Shazeer proposed scaled dot-product attention, multi-head attention and the parameter-free position representation in 'Attention Is All You Need' — foundational transformer architecture the frontier stack descends from","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of the Sparsely-Gated Mixture-of-Experts layer and Switch Transformers — sparse-scaling methods used in current frontier LLMs","source_url":"https://arxiv.org/abs/1701.06538","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sole author of multi-query attention ('Fast Transformer Decoding: One Write-Head Is All You Need'), an inference-stack primitive in current LLM serving","source_url":"https://arxiv.org/abs/1911.02150","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"At Google from ~2000, worked on the search spelling corrector and Meena before the transformer lineage; co-founded Character.AI in 2021 and returned to Google to co-lead Gemini in 2024 — a continuous ~26-year language-modeling history","source_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Image Transformer' (2018), applying self-attention to image generation — a distinct sequence-modeling modality","source_url":"https://arxiv.org/abs/1802.05751","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored the sparsely-gated Mixture-of-Experts layer and introduced multi-query attention, both used in current frontier LLMs","source_url":"https://arxiv.org/abs/1911.02150","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"At Google from ~2000, worked on the search spelling corrector and the Meena chatbot before the transformer lineage — continuous language-modeling history","source_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Character.AI in 2021 as its technical founder, building and shipping a large-scale conversational LLM, before returning to Google to co-lead Gemini in 2024","source_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Shazeer proposed scaled dot-product attention, multi-head attention and the position representation in 'Attention Is All You Need' — foundational transformer architecture","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Math/Computer Science, Duke University (1994-1998)","Began but did not complete a graduate program at UC Berkeley"],"canonical_papers":["Attention Is All You Need (2017)","Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer (2017)","Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer / T5 (2019)","PaLM: Scaling Language Modeling with Pathways (2022)","Switch Transformers (2022)"],"first_verifiable_year":1999,"notable_systems":["Transformer architecture (co-inventor)","Mixture-of-Experts layer","T5","PaLM","Meena chatbot","Character.AI","Google Gemini (technical co-lead, 2024-2026)"],"citations":394944,"h_index":77,"patents":0,"dossier_notes":"Dossier's OpenAlex entry for 'Attention Is All You Need' shows year 2025 with a broken/placeholder DOI (10.65215/2q58a426) — this is a metadata artifact; the paper is actually NeurIPS/arXiv 2017 (arXiv:1706.03762), confirmed directly from the arXiv abstract page listing Shazeer as second author. Dossier citation counts (OpenAlex 20,783 cited-by, Semantic Scholar 249,110) are both far below the live Google Scholar total (~394,944) — Google Scholar is the more current figure and was used for h-index/citations in this report. No homonym risk identified. Note: dossier's 'company' field (Character.AI/Google) is now stale — as of June 2026 Shazeer is VP of Engineering at OpenAI per Wikipedia; used current title/company in this report.","years_language_modeling":26,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2000-2026","evidence":"Google search spelling corrector/n-gram LM, Meena, transformer/T5/PaLM lineage, Character.AI conversational LLM, Gemini co-lead"},{"domain":"other","years":"2016-2021","evidence":"speech: 'Speech recognition with attention-based recurrent neural networks' patent (priority 2016), speech-recognition topics in OpenAlex"},{"domain":"other","years":"2017-2018","evidence":"image/multimodal: Image Transformer (2018) and multi-task multi-modal machine learning system patent (US11494561B2, priority 2017)"}],"frontier_lineage":["transformer / scaled-dot-product & multi-head attention & positional representation","sparsely-gated Mixture-of-Experts + Switch Transformers sparse routing","multi-query attention (inference stack)","T5 text-to-text pretraining","mesh-tensorflow / GShard tensor sharding","PaLM scaling"],"technical_founder_roles":["Character.AI — co-founder & CEO / technical founder (core LLM & architecture) — 2021-2024"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":20,"lm_domain_breadth":8,"scientific_founder":10},"confidence":0.86,"duration_ms":49493},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":19,"lm_domain_breadth":15,"scientific_founder":11},"confidence":0.9,"duration_ms":58713}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Noam_Shazeer","wikipedia_title":"Noam Shazeer","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"richard-socher","name":"Richard Socher","title":"Co-founder & CEO","company":"You.com","sector":"general","rank":4,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":20,"transformers_lm":16,"frontier_founder":19,"lm_domain_depth":18,"lm_domain_breadth":14,"hands_on_engineering":17,"industry_impact":18,"scientific_founder":17},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":87,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":87,"tier":"frontier_builder","tier_label":"Frontier Builder","confidence":0.89,"rationale":"Socher earned a Stanford CS PhD (2014) under Christopher D. Manning and co-authored GloVe (2014), one of the two canonical word-embedding algorithms the field's vector-space lineage is built on, alongside foundational pre-transformer deep-learning NLP work (Recursive Deep Models for Semantic Compositionality / Recursive Neural Tensor Networks, 2013; Tree-LSTM, 2015) that directly extended the matrix/tensor and compositional-representation-learning tradition. His verified Google Scholar profile (stanford.edu-verified) shows 256,483 citations and an h-index of 114, with GloVe alone carrying ~51,700 citations. He personally built and led MetaMind (founded 2014, acquired by Salesforce 2016, becoming Salesforce Chief Scientist through 2020) and founded You.com (2020), an AI search company, giving him a rare combination of canonical authored research plus personally-led production AI systems. His transformers_lm score reflects strong pre-transformer language-modeling-lineage authorship (recursive/compositional neural nets, embeddings feeding into later LM pretraining) and applied leadership of transformer-era products at You.com, rather than being a co-author of the Transformer paper itself. This is a researcher-founder profile the rubric explicitly identifies as scoring high — not fame-driven.\n\nSocher co-authored GloVe (EMNLP 2014), which the rubric names alongside word2vec as a canonical, named building block of the word-embedding lineage that today's frontier language models descend from — placing him squarely in the 18-20 frontier band — and extended the lineage with CoVe contextualized vectors (2017) and CTRL, a 1.63B-parameter conditional transformer LM (2019). His hands-on language-modeling record is continuous from the recursive-neural-net / compositional-representation work of ~2011-2013 (Recursive Deep Models 2013) through GloVe (2014), seq2seq/CoVe (2017), CTRL (2019) and You.com's LLM-based search (2020+) — ~15 years spanning the pre/post-word2vec eras, still active. Across distinct LM domains he has a verifiable hands-on record in natural-language text (deep, ~15 yrs), biological language modeling (ProGen protein-sequence LM, Salesforce Research, Nat Biotechnol 2023) and clinical/biomedical-literature mining (COVID-19 deep semantic search/QA/summarization, npj Digital Medicine 2021) — three domains, two outside NL text, matching the 13-17 band. As a scientific/technical founder he founded MetaMind (2014, deep-learning NLP platform, acquired by Salesforce 2016 where he was Chief Scientist), co-founded and set the technical direction of You.com (2020), and co-founded Recursive (2025) — three companies whose core is these systems across ~12 years (10+), meeting the 18-20 company-count anchor.","evidence":[{"claim":"Richard Socher received his PhD in Computer Science from Stanford University in 2014.","source_url":"https://en.wikipedia.org/wiki/Richard_Socher","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher co-authored 'GloVe: Global Vectors for Word Representation' (2014) with Jeffrey Pennington and Christopher D. Manning, a canonical word-embedding paper with ~51,700 citations on Google Scholar.","source_url":"https://scholar.google.com/citations?user=FaOcyfMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher's verified Google Scholar profile (stanford.edu email) shows 256,483 total citations, h-index 114, i10-index 244, including 'Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank' (2013, ~12,280 citations) and 'Improved Semantic Representations from Tree-Structured LST","source_url":"https://scholar.google.com/citations?user=FaOcyfMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher founded MetaMind in 2014, which was acquired by Salesforce in 2016, after which he served as Salesforce's Chief Scientist; he later founded You.com in 2020.","source_url":"https://en.wikipedia.org/wiki/Richard_Socher","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GloVe: Global Vectors for Word Representation — Jeffrey Pennington, Richard Socher, Christopher Manning, EMNLP 2014","source_url":"https://aclanthology.org/D14-1162/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (Recursive, you.com, AIX): 256,483 citations, h-index 114; top works ImageNet 2009, GloVe 2014, Recursive Deep Models 2013","source_url":"https://scholar.google.com/citations?user=FaOcyfMAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Learned in Translation: Contextualized Word Vectors (CoVe) — McCann, Bradbury, Xiong, Socher, 2017; contextual vectors from a deep LSTM attentional seq2seq encoder","source_url":"https://arxiv.org/abs/1708.00107","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CTRL: A Conditional Transformer Language Model (1.63B parameters) — Keskar, McCann, Varshney, Xiong, Socher, 2019","source_url":"https://arxiv.org/abs/1909.05858","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stanford CS PhD 2014 under Christopher Manning; founded MetaMind 2014 (acquired by Salesforce 2016); Salesforce Chief Scientist 2016-2020; co-founded You.com 2020","source_url":"https://en.wikipedia.org/wiki/Richard_Socher","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GloVe: Global Vectors for Word Representation — Pennington, Socher, Manning, EMNLP 2014, a canonical word-embedding algorithm in the vector-space lineage frontier LMs build on.","source_url":"https://aclanthology.org/D14-1162/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CTRL: A Conditional Transformer Language Model (1.63B params) — Keskar, McCann, Varshney, Xiong, Socher, 2019.","source_url":"https://arxiv.org/abs/1909.05858","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Large language models generate functional protein sequences across diverse families (ProGen) — Salesforce Research, Nat Biotechnol 2023, biological language modeling with Socher.","source_url":"https://pubmed.ncbi.nlm.nih.gov/36702895/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"COVID-19 information retrieval with deep-learning based semantic search, question answering, and abstractive summarization (Esteva, ..., Socher R), npj Digital Medicine 2021 — biomedical/clinical literature mining.","source_url":"https://pubmed.ncbi.nlm.nih.gov/33846532/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher founded MetaMind (2014, acquired by Salesforce 2016; Chief Scientist 2016-2020), co-founded You.com (2020) and Recursive (2025) as scientific/technical founder.","source_url":"https://en.wikipedia.org/wiki/Richard_Socher","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher co-authored 'GloVe: Global Vectors for Word Representation' (EMNLP 2014), a canonical word-embedding algorithm in the vector-space lineage frontier LMs build on.","source_url":"https://aclanthology.org/D14-1162/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher is a co-author of 'CTRL: A Conditional Transformer Language Model' (1.63B parameters, 2019), an early large transformer LM.","source_url":"https://arxiv.org/abs/1909.05858","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Large language models generate functional protein sequences across diverse families' (Nat Biotechnol 2023, Salesforce Research / ProGen) — biological language modeling from the lab Socher led as Chief Scientist.","source_url":"https://pubmed.ncbi.nlm.nih.gov/36702895/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'COVID-19 information retrieval with deep-learning based semantic search, question answering, and abstractive summarization' (NPJ Digit Med 2021), Socher R co-author — clinical/biomedical language work.","source_url":"https://pubmed.ncbi.nlm.nih.gov/33846532/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Socher founded MetaMind (2014, acquired by Salesforce 2016), was Salesforce Chief Scientist (2016-2020), and co-founded and is CEO of You.com (2020) and Recursive (2025).","source_url":"https://en.wikipedia.org/wiki/Richard_Socher","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, Stanford University (2014), advisor Christopher D. Manning","Undergraduate study, University of Leipzig / Saarland University"],"canonical_papers":["GloVe: Global Vectors for Word Representation (2014)","Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank (2013)","Improved Semantic Representations from Tree-Structured LSTM Networks (2015)","ImageNet: A Large-Scale Hierarchical Image Database (2009, co-author)"],"first_verifiable_year":2009,"notable_systems":["MetaMind (founded 2014, acquired by Salesforce 2016)","GloVe embeddings library","You.com (AI search, founded 2020)","Recursive (2025)"],"citations":256483,"h_index":114,"patents":0,"dossier_notes":"The dossier's OpenAlex block (works_count 225, cited_by_count 140330, affiliations: Seoul National University, National University of Singapore, University of Colorado Boulder, Princeton University, Cornell University, Prostate Cancer Foundation) is a HOMONYM MISMATCH — none of these affiliations or the medical-adjacent 'Prostate Cancer Foundation' entry belong to the real Richard Socher (Stanford NLP PhD, GloVe co-author, You.com/MetaMind founder). The Wikipedia and Wikidata blocks (Stanford PhD 2014, advisor Christopher D. Manning, google_scholar_id FaOcyfMAAAAJ) are correctly matched and were independently verified via his actual Google Scholar profile (256,483 citations, h-index 114), which was used instead of the OpenAlex numbers for this scoring.","years_language_modeling":15,"years_as_technical_founder":12,"technical_founder_companies":3,"lm_domains":[{"domain":"natural_language","years":"2011-2026","evidence":"Recursive Deep Models (2013), GloVe (2014), CoVe (2017), CTRL (2019), You.com LLM search (2020+)"},{"domain":"biological","years":"2020-2023","evidence":"ProGen — LLMs generate functional protein sequences, Salesforce Research, Nat Biotechnol 2023"},{"domain":"clinical","years":"2021","evidence":"COVID-19 deep-learning semantic search/QA/summarization over biomedical literature, npj Digital Medicine 2021"}],"frontier_lineage":["GloVe global word-vector embeddings — canonical word-embedding block underlying LM input representations","CoVe contextualized word vectors (2017)","CTRL conditional transformer language model (2019)"],"technical_founder_roles":["MetaMind — founder / Chief Scientist — 2014-2016","Salesforce — Chief Scientist (research leadership) — 2016-2020","You.com — co-founder & CEO, technical direction — 2020-present","Recursive — co-founder & CEO — 2025-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":19,"lm_domain_depth":18,"lm_domain_breadth":15,"scientific_founder":18},"confidence":0.82,"duration_ms":73959},{"pass":"pass_2","dimensions":{"frontier_founder":19,"lm_domain_depth":18,"lm_domain_breadth":14,"scientific_founder":16},"confidence":0.8,"duration_ms":89323}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Richard_Socher","wikipedia_title":"Richard Socher","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"demis-hassabis","name":"Demis Hassabis","title":"Co-founder & CEO","company":"Google DeepMind","sector":"general","rank":5,"rubric_version":4,"dimensions":{"foundations":17,"vector_embeddings":14,"transformers_lm":18,"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":18,"hands_on_engineering":18,"industry_impact":20,"scientific_founder":18},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":85,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":85,"tier":"frontier_builder","tier_label":"Frontier Builder","confidence":0.95,"rationale":"Hassabis holds a PhD from University College London (2009, advisor Eleanor Maguire, thesis on the neural basis of episodic memory) and an undergraduate degree in Computer Science from Cambridge, giving him genuine formal training bridging neuroscience and computer science. He co-founded DeepMind in 2010 and personally led the research programs behind DQN ('Human-level control through deep reinforcement learning,' Nature 2015, 31,045 citations), AlphaGo ('Mastering the game of Go with deep neural networks and tree search,' Nature 2016), and AlphaFold ('Highly accurate protein structure prediction with AlphaFold,' Nature 2021, 47,191 citations) -- for which he and John Jumper won the 2024 Nobel Prize in Chemistry, a rare case of a tech CEO with a Nobel for the underlying science itself. OpenAlex shows an unambiguous match (176 works, 197,926 citations, h-index 92, i10-index 129), and he is listed as a contributor to 'Gemini: A Family of Highly Capable Multimodal Models' (2023), placing him directly in the transformer/scaling-era LM lineage as DeepMind (merged with Google Brain in 2023) became the org building Google's frontier LLMs. This is as close to the rubric's top anchor as any CEO in this batch: principal builder of systems the field runs on (AlphaFold, AlphaGo, DQN), not a business-only executive.\n\nHassabis is a Gemini co-author and, as CEO/co-founder of Google DeepMind and Chief Scientist of Alphabet, directs the lab that produced load-bearing frontier-LM foundations — the Chinchilla compute-optimal scaling laws (Hoffmann et al. 2022), RETRO retrieval, Gopher, and the Gemini family itself — placing his own org's work directly in the stack today's frontier models descend from, though his personal authorship centers on deep RL (DQN) rather than the attention/transformer primitives. His hands-on language-modeling record is recent and org-led (DeepMind's LM push, ~2021→2026, roughly five years), which caps lm_domain_depth in the 8-12 band despite a two-decade AI career, because pre-2021 his personal work was neuroscience, deep RL, games and protein folding, not language modeling. Cross-domain, however, is exceptional: biological sequence modeling (AlphaFold/AlphaFold3 proteins, AlphaGenome DNA regulatory variants), natural-language text (Gemini), source code (AlphaCode), and chemistry/materials + drug design (GNoME, Isomorphic Labs) — four-plus distinct language-modeling domains, several outside natural text, each with a verifiable authored/led record. He is a genuine scientific/technical founder: DeepMind (2010→present, ~16 years) and Isomorphic Labs (2021→present), personally co-authoring the core papers (a 2024 Nobel Prize in Chemistry for the AlphaFold science), satisfying the 15+-year / multi-company top band.","evidence":[{"claim":"PhD, University College London (2009), advisor Eleanor Maguire, thesis on episodic memory","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded DeepMind in 2010 with Shane Legg and Mustafa Suleyman; led AlphaGo (2016) and AlphaFold research programs","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"2024 Nobel Prize in Chemistry (with John Jumper) for AI research contributions to protein structure prediction","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cambridge CS double first 1997; PhD 2009 UCL under Eleanor Maguire on episodic memory; Bullfrog Theme Park lead programmer 1994; Lionhead lead AI programmer on Black & White 2001; DeepMind founded 2010; Nobel Prize in Chemistry 2024","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (dYpPMQEAAAAJ): ~310,736 citations, h-index 109; top works AlphaFold (2021), DQN Nature (2015), AlphaGo (2016), AlphaFold 3 (2024), EWC (2017), Gemini (2023)","source_url":"https://scholar.google.com/citations?user=dYpPMQEAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Human-level control through deep reinforcement learning', Nature 518, 2015","source_url":"https://doi.org/10.1038/nature14236","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Highly accurate protein structure prediction with AlphaFold', Nature 596, 2021","source_url":"https://doi.org/10.1038/s41586-021-03819-2","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on 'Gemini: A Family of Highly Capable Multimodal Models' (2023), DeepMind's frontier LLM family","source_url":"https://arxiv.org/abs/2312.11805","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepMind's Chinchilla compute-optimal scaling laws (Hoffmann et al., 2022) are a foundational input to frontier-LLM training","source_url":"https://arxiv.org/abs/2203.15556","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"AlphaGenome (Nature 2026) applies sequence modeling to DNA/regulatory-variant prediction — biological language modeling","source_url":"https://doi.org/10.1038/s41586-025-09099-4","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"AlphaFold protein-structure prediction (Nature 2021), for which Hassabis shared the 2024 Nobel Prize in Chemistry","source_url":"https://doi.org/10.1038/s41586-021-03819-2","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder & CEO of DeepMind (2010) and co-founder & CEO of Isomorphic Labs (2021)","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder, CEO and chief scientist of DeepMind/Google DeepMind; listed contributor to 'Gemini: A Family of Highly Capable Multimodal Models' (2023)","source_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepMind authored Chinchilla compute-optimal scaling laws, a foundational component frontier LM training builds on","source_url":"https://arxiv.org/abs/2203.15556","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"AlphaFold protein structure prediction (biological sequence modeling) — 2024 Nobel Prize in Chemistry with John Jumper","source_url":"https://doi.org/10.1038/s41586-021-03819-2","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder and CEO of Isomorphic Labs (2021), AI drug-discovery / molecular modeling company","source_url":"https://en.wikipedia.org/wiki/Isomorphic_Labs","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BA Computer Science, University of Cambridge","PhD Neuroscience, University College London (2009, advisor Eleanor Maguire)"],"canonical_papers":["Human-level control through deep reinforcement learning (DQN, Nature 2015)","Mastering the game of Go with deep neural networks and tree search (AlphaGo, Nature 2016)","Mastering the game of Go without human knowledge (AlphaGo Zero, Nature 2017)","Highly accurate protein structure prediction with AlphaFold (Nature 2021)","Gemini: A Family of Highly Capable Multimodal Models (2023)"],"first_verifiable_year":2007,"notable_systems":["DQN","AlphaGo / AlphaGo Zero / AlphaZero","AlphaFold / AlphaFold 3","Gemini"],"citations":197926,"h_index":92,"patents":0,"dossier_notes":"Dossier's OpenAlex match is clean and unambiguous (1 candidate); no homonym issues found. This is one of the strongest, most unambiguous records in the batch.","years_language_modeling":5,"years_as_technical_founder":16,"technical_founder_companies":2,"lm_domains":[{"domain":"biological","years":"2018-2026","evidence":"AlphaFold/AlphaFold3 protein sequences; AlphaGenome DNA regulatory-variant modeling; author/leader, Nature"},{"domain":"natural_language","years":"2021-2026","evidence":"Gemini family, Gopher, RETRO — DeepMind LLMs; Gemini co-author"},{"domain":"code","years":"2022-2026","evidence":"AlphaCode competitive-programming code generation, DeepMind"},{"domain":"chemistry","years":"2021-2026","evidence":"GNoME materials discovery; Isomorphic Labs molecular/drug-design models (CEO, co-founder)"}],"frontier_lineage":["Gemini frontier LLM family (co-author, leads the lab)","Chinchilla compute-optimal scaling laws (DeepMind, 2022)","RETRO retrieval-augmented LM","deep RL / DQN lineage underpinning RLHF-style methods"],"technical_founder_roles":["Google DeepMind — co-founder & CEO / Chief Scientist of Alphabet — 2010-present","Isomorphic Labs — co-founder & CEO — 2021-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":17,"lm_domain_depth":11,"lm_domain_breadth":18,"scientific_founder":18},"confidence":0.82,"duration_ms":69895},{"pass":"pass_2","dimensions":{"frontier_founder":18,"lm_domain_depth":13,"lm_domain_breadth":19,"scientific_founder":18},"confidence":0.82,"duration_ms":68599}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Demis_Hassabis","wikipedia_title":"Demis Hassabis","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jakob-uszkoreit","name":"Jakob Uszkoreit","title":"Co-founder & CEO","company":"Inceptive","sector":"general","rank":6,"rubric_version":4,"dimensions":{"foundations":15,"vector_embeddings":15,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":18,"lm_domain_breadth":15,"hands_on_engineering":18,"industry_impact":18,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":84,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":84,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.86,"rationale":"Jakob Uszkoreit is a confirmed co-author of 'Attention Is All You Need' (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin, arXiv 1706.03762, 2017) — the canonical paper that introduced the Transformer architecture, the direct origin point of the entire modern language-model lineage the rubric describes. This alone places him at the top anchor for transformers_lm (authored canonical work the field builds on). He led NLP research and engineering teams at Google Research/Google Brain prior to co-authoring the Transformer, work reflected in the dossier's OpenAlex-linked record (Google, 'Brain (Germany)', University of Washington affiliations; 53 works, ~39,325 citations), giving strong, verifiable foundations and vector_embeddings credit for pre-transformer attention/sequence-modeling and representation-learning research (attention mechanisms operate directly over learned embeddings). His hands_on_engineering is rated highly given his personal role building NLP infrastructure at Google (e.g., Tensor2Tensor-era tooling) prior to founding Inceptive, an AI-for-biology company applying transformer-style modeling to mRNA/molecule design — a genuine technical-founder profile, not a business-only leader. industry_impact is high both for the field-defining influence of the Transformer paper itself and for founding a company whose technical core is applying this lineage to a new domain.\n\nUszkoreit is a confirmed co-author of 'Attention Is All You Need' (arXiv 1706.03762, 2017), the Transformer architecture that every frontier model — GPT, Claude, Gemini, Llama — directly descends from; he also co-authored the 2016 Decomposable Attention model (early attention over embeddings) and the 2020 Vision Transformer (ViT), so his own work is squarely the foundation, earning the top frontier_founder anchor. His verifiable language-modeling record runs from a 2006 first publication through NLP/machine-translation research and engineering leadership at Google Research/Brain, the Transformer, the Natural Questions QA benchmark (2019), and the mRNA/RNA sequence models at Inceptive — roughly 18-20 continuous hands-on years in language modeling, placing lm_domain_depth in the 15+ band. He has a verifiable hands-on record across natural-language text (deep, ~15+ yr), biological language modeling (Inceptive foundation models of mRNA/siRNA/ASO/peptide sequences, 2021-present), and vision/multimodal sequence modeling (ViT, MLP-Mixer) — three distinct domains, two outside natural-language text, matching the 13-17 breadth band. As co-founder & CEO of Inceptive (2021) he personally sets and executes the technical direction, applying research he himself authored — one company across ~5 years, which sits in the 3-8-year single-company scientific_founder band.","evidence":[{"claim":"Jakob Uszkoreit is a co-author of 'Attention Is All You Need' (arXiv 1706.03762, 2017), alongside Ashish Vaswani, Noam Shazeer, Niki Parmar, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin — the paper that introduced the Transformer architecture.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's OpenAlex-linked profile for Uszkoreit shows affiliations at Google (United States), 'Brain (Germany)' (i.e. Google Brain), University of Washington, and UC Berkeley, with 53 works and approximately 39,325 citations — consistent with a senior NLP/ML research career at Google Research/Go","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uszkoreit's education is recorded as Technische Universität Berlin (TU Berlin), and his career trajectory (Google Research/Brain NLP team lead, then Attention Is All You Need co-author, then founder of Inceptive applying AI to mRNA/biological molecule design) is consistent across the dossier's wikid","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attention Is All You Need — Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin; arXiv 1706.03762, 12 June 2017; Uszkoreit is fourth author","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ViT), 2020 — Uszkoreit is eleventh of twelve authors","source_url":"https://arxiv.org/abs/2010.11929","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Inceptive founded 2021, offices in Palo Alto, Berlin and Zurich; builds foundation models of life for mRNA, siRNA, ASO and peptide therapeutics","source_url":"https://inceptive.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata: machine learning researcher, educated at Technische Universität Berlin, employer Google, ORCID 0000-0001-5066-7530","source_url":"https://www.wikidata.org/wiki/Q98891246","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uszkoreit is a co-author of 'Attention Is All You Need' (arXiv 1706.03762, 2017), which introduced the Transformer architecture underpinning all frontier LLMs.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uszkoreit co-authored 'An Image is Worth 16x16 Words' (ViT, 2020) and a long Google NLP research record (Decomposable Attention 2016, Natural Questions 2019) documenting continuous language/sequence-modeling work.","source_url":"https://arxiv.org/abs/2010.11929","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uszkoreit co-founded Inceptive (2021), which builds transformer-style 'foundation models of life' for mRNA, siRNA, ASO and peptide therapeutics — biological language modeling under his technical direction.","source_url":"https://inceptive.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uszkoreit co-authored 'An Image is Worth 16x16 Words' (ViT, 2020) and the 2016 Decomposable Attention model, extending the attention/transformer lineage across NLP and vision.","source_url":"https://arxiv.org/abs/2010.11929","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Inceptive, founded 2021 with Uszkoreit as co-founder & CEO, builds transformer-style foundation models of biological molecules (mRNA, siRNA, ASO, peptides) — biological language modeling.","source_url":"https://inceptive.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Technische Universität Berlin (per dossier wikidata)"],"canonical_papers":["Attention Is All You Need (2017, co-author)"],"first_verifiable_year":2006,"notable_systems":["Transformer architecture (co-author)","Google Research/Google Brain NLP systems","Inceptive (AI-designed mRNA/biological molecules, co-founder & CEO)"],"citations":39325,"h_index":0,"patents":0,"dossier_notes":"No Wikipedia page was found for this person in this session (404) and Google Scholar / Semantic Scholar direct-profile lookups failed to return usable content (404/empty), so h_index could not be independently confirmed and is left at 0 despite the strong citation count from the dossier's OpenAlex block, which was used since it is well-corroborated by the confirmed Attention Is All You Need co-authorship and plausible Google/Google-Brain/TU-Berlin affiliation trail — no homonym mismatch signal was found here, unlike richard-socher/thomas-wolf/humayun-sheikh.","years_language_modeling":18,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2006-2020","evidence":"Google NLP/machine translation, Attention Is All You Need (2017), Decomposable Attention (2016), Natural Questions QA benchmark (2019)"},{"domain":"biological","years":"2021-2026","evidence":"Inceptive foundation models of mRNA/siRNA/ASO/peptide sequences (co-founder & CEO)"},{"domain":"other","years":"2020-2021","evidence":"vision/multimodal sequence modeling — Vision Transformer (ViT, 2020) and MLP-Mixer (2021) co-author"}],"frontier_lineage":["Transformer architecture (Attention Is All You Need, 2017) — the direct architectural basis of GPT/Claude/Gemini/Llama","Vision Transformer (ViT, 2020) — transformer applied to images, basis of multimodal frontier models","Decomposable Attention (2016) — early attention-over-embeddings NLP work"],"technical_founder_roles":["Inceptive — co-founder & CEO / technical founder — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":18,"lm_domain_breadth":15,"scientific_founder":12},"confidence":0.83,"duration_ms":56038},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":18,"lm_domain_breadth":15,"scientific_founder":11},"confidence":0.83,"duration_ms":64309}],"source":"seeded","profile_url":null,"wikipedia_title":"Jakob Uszkoreit","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ashish-vaswani","name":"Ashish Vaswani","title":"Co-founder & CEO (Essential AI); joined Nvidia via 2026 acqui-hire","company":"Essential AI","sector":"general","rank":7,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":15,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":17,"lm_domain_breadth":8,"hands_on_engineering":18,"industry_impact":20,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":82,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":82,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Ashish Vaswani is the lead/first author of 'Attention Is All You Need' (NeurIPS 2017), the paper that introduced the Transformer architecture and is arguably the single most consequential paper in the modern language-model lineage — an unambiguous 18-20 anchor case ('authored canonical work the field builds on'). He holds a PhD in Computer Science from USC (2014, advisors David Chiang and Liang Huang), thesis 'Smaller, Faster, and Accurate Models for Statistical Machine Translation', giving him genuine graduate-level foundations in statistical/optimization methods for sequence modeling predating the Transformer. He worked at Google Brain, co-founded Adept AI (2022) and Essential AI (2023) as CEO, building applied AI-agent/foundation-model companies, and as of mid-2026 he and the Essential AI team were acqui-hired by Nvidia to work on the Nemotron model family — continued hands-on model-building at a frontier lab. vector_embeddings is scored below transformers_lm/foundations since his direct authored work is concentrated in attention/seq2seq/MT rather than embeddings/retrieval specifically, though attention mechanisms are adjacent.\n\nVaswani is the lead/first author of 'Attention Is All You Need' (NeurIPS 2017), the Transformer architecture from which every current frontier model — GPT, Claude, Gemini, Llama — directly descends; this is the single clearest 18-20 frontier-founder anchor ('transformer & attention authors'). His hands-on language-modeling record is deep and continuous: statistical machine translation in his USC PhD (thesis 'Smaller, Faster, and Accurate Models for Statistical Machine Translation', 2014, with dissertation research spanning the preceding pre-word2vec SMT years) → self-attention/Transformer at Google Brain → Adept and Essential AI foundation models → Nvidia Nemotron post-acquihire, ~15 years still active. Distinct LM domains with a verifiable hands-on record are natural-language text (deep, SMT→LLMs) and image/media sequence modeling (Image Transformer 2018, Stand-Alone Self-Attention in Vision Models 2019); the DeepConsensus 'sequence transformer' PubMed hit is a likely homonym per the prior pass and is not counted, so breadth is two domains. As scientific/technical founder he co-founded two such companies whose core is these systems — Adept AI (2022) and Essential AI (co-founder & CEO, 2023) — over ~4 years, personally setting technical direction; two companies places him in the 13-17 band despite the short tenure.","evidence":[{"claim":"Lead author, 'Attention Is All You Need', Google Brain, NeurIPS 2017 — introduced the Transformer architecture","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Adept AI (2022) then Essential AI (2023) as CEO after Google Brain","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"As of June 2026, Vaswani and Essential AI were acqui-hired by Nvidia, joining to work on the Nemotron open-source model family","source_url":"https://www.groundlevel-ai.com/p/nvidia-quietly-acquihires-essential","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar record (name-exact match): 57 papers, 195,978 citations, h-index 26","source_url":"https://www.semanticscholar.org/author/Ashish-Vaswani/40348417","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 40348417: 195,978 citations, h-index 26, with 'Attention is All you Need' (2017) at ~191,942 citations","source_url":"https://api.semanticscholar.org/graph/v1/author/40348417?fields=name,citationCount,hIndex,paperCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attention Is All You Need (2017) — Ashish Vaswani listed as first author","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from University of Southern California (2014) under David Chiang and Liang Huang, thesis 'Smaller, Faster, and Accurate Models for Statistical Machine Translation'; Google Brain 2016-2021; co-founded Adept AI then Essential AI; acqui-hired by Nvidia June 2026","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Image Transformer (ICML 2018) — Vaswani co-author, extends self-attention to image generation","source_url":"https://arxiv.org/abs/1802.05751","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stand-Alone Self-Attention in Vision Models (NeurIPS 2019) — Vaswani co-author","source_url":"https://arxiv.org/abs/1906.05909","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ashish Vaswani is first author of 'Attention Is All You Need' (2017), which introduced the Transformer architecture underlying today's frontier language models","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD thesis 'Smaller, Faster, and Accurate Models for Statistical Machine Translation', USC 2014 — statistical/neural machine-translation language modeling","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Adept AI (2022) then Essential AI (2023) as co-founder and CEO, both foundation-model companies; acqui-hired by Nvidia June 2026 to work on the Nemotron model family","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ashish Vaswani is lead/first author of 'Attention Is All You Need' (2017), introducing the Transformer architecture that today's frontier models build on","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD, USC (2014), thesis 'Smaller, Faster, and Accurate Models for Statistical Machine Translation' — SMT language-modeling work predating the Transformer","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Adept AI, then co-founder & CEO of Essential AI; acqui-hired by Nvidia June 2026 to work on the Nemotron model family","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, University of Southern California (2014)"],"canonical_papers":["Attention Is All You Need (2017)"],"first_verifiable_year":2014,"notable_systems":["Transformer architecture","Adept AI (ACT-1 agent)","Essential AI","Nvidia Nemotron (post-acquihire)"],"citations":195978,"h_index":26,"patents":0,"dossier_notes":"Dossier's OpenAlex found:false but Semantic Scholar (name-exact, 3 candidates) and PubMed refined matches (DeepConsensus gap-aware sequence transformer, Nat Biotechnol 2023, plausibly co-authored given transformer/genomics crossover work) are consistent with this person; the earlier 2014 PLoS One Alzheimer's paper and 2017 Hum Brain Mapp paper in the PubMed sample are likely unrelated homonyms and were not used as evidence. The 195,978 citation / h-index 26 Semantic Scholar figure is almost certainly dominated by the single Attention Is All You Need paper's massive citation count and should not be read as broad h-index depth across many independent works.","years_language_modeling":15,"years_as_technical_founder":4,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2011-2026","evidence":"SMT PhD thesis (USC 2014), Transformer/Attention (Google Brain 2017), Adept & Essential AI foundation models, Nvidia Nemotron"},{"domain":"other","years":"2018-2019","evidence":"Image Transformer (ICML 2018) and Stand-Alone Self-Attention in Vision Models (NeurIPS 2019) — self-attention sequence modeling of image/media"}],"frontier_lineage":["Transformer architecture (self-attention, multi-head attention, positional encoding) — the foundational architecture of all GPT/Claude/Gemini/Llama-class frontier models"],"technical_founder_roles":["Adept AI — co-founder — 2022","Essential AI — co-founder & CEO — 2023-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":6,"scientific_founder":15},"confidence":0.85,"duration_ms":57240},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":18,"lm_domain_breadth":11,"scientific_founder":14},"confidence":0.82,"duration_ms":76980}],"source":"seeded","profile_url":null,"wikipedia_title":"Ashish Vaswani","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"dario-amodei","name":"Dario Amodei","title":"Co-founder & CEO","company":"Anthropic","sector":"general","rank":8,"rubric_version":4,"dimensions":{"foundations":17,"vector_embeddings":11,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":14,"hands_on_engineering":18,"industry_impact":20,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":82,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":82,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.93,"rationale":"Amodei's mathematical foundations are real rather than nominal: a Princeton PhD in biophysics under William Bialek and Michael J. Berry produced maximum-entropy and statistical-physics models of neural populations published in PNAS ('Thermodynamics and signatures of criticality in a network of neurons', 2015) and PLoS Computational Biology — statistical learning and optimization applied to high-dimensional data, though not a CS/ML degree, which is why foundations sits at 17 rather than at the canonical anchor. His transformer/language-model record is canonical by any reading: he is the final author of 'Language Models are Few-Shot Learners' (GPT-3, arXiv:2005.14165, verified author list: Brown … Sutskever, Amodei), an author of 'Scaling Laws for Neural Language Models' (2020), of 'Deep Reinforcement Learning from Human Preferences' (2017, the origin of RLHF) and of Constitutional AI (2022) — he authored work in both the scaling and the alignment halves of the lineage. His hands-on engineering predates the LLM era: he was a core author of Baidu's Deep Speech 2 (2015), a large-scale GPU-trained end-to-end sequence model. His verified Google Scholar profile (the ID supplied by the dossier's own Wikidata block) shows 196,269 citations and h-index 67, of which GPT-3 alone accounts for 82,085 — several times the dossier's OpenAlex figures. Vector embeddings is his one thin dimension: representation learning is implicit in his speech and LM work but he has authored no embedding, contrastive or dense-retrieval paper, so that dimension is scored on implicit representation-learning content only and lands well below the rest.\n\nAmodei's own work is load-bearing foundation for today's frontier models: he is the final author of GPT-3 'Language Models are Few-Shot Learners' (few-shot prompting), an author of 'Scaling Laws for Neural Language Models', of 'Deep RL from Human Preferences' (the origin of RLHF), and of Constitutional AI — scaling and alignment methods that GPT/Claude/Gemini/Llama-class systems directly descend from, so frontier_founder sits at the top anchor. His hands-on language/sequence-modeling record runs from Baidu's Deep Speech 2 (2015) through GPT-2/GPT-3/Codex at OpenAI to the Claude family at Anthropic — roughly 11 years continuously hands-on, placing lm_domain_depth in the 8-15-year band. Across those years he has a verifiable hands-on record in three distinct sequence/language-modeling domains: natural-language text (deep, multi-year), source code (Codex, 'Evaluating Large Language Models Trained on Code', 2021), and speech/audio sequence modeling (Deep Speech 2); his pre-2015 biophysics/proteomics/neural-population work is statistical modeling but not sequence language modeling, so no biological-LM domain is credited. As co-founder and CEO of Anthropic since 2021 (~5 years) he is a genuine scientific/technical founder who personally authored the core research the company runs on (Constitutional AI, Claude), but that is one company over ~5 years, placing scientific_founder in the 3-8-year single-company band.","evidence":[{"claim":"Final author of 'Language Models are Few-Shot Learners' (GPT-3); verified author list begins Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan and ends Ilya Sutskever, Dario Amodei; submitted 28 May 2020","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile 6-e-ZBEAAAAJ (CEO and Co-Founder at Anthropic): 196,269 citations, h-index 67, i10-index 101; top works 'Language models are few-shot learners' (82,085), 'Language models are unsupervised multitask learners' (24,702), 'Scaling laws for neural language models' (8,966), 'Deep re","source_url":"https://scholar.google.com/citations?user=6-e-ZBEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Scaling Laws for Neural Language Models' (Kaplan et al., 2020)","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Statistical-physics neuroscience record under Bialek and Berry: 'Thermodynamics and signatures of criticality in a network of neurons', PNAS 2015","source_url":"https://pubmed.ncbi.nlm.nih.gov/26330611/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Deep Speech 2: End-to-End Speech Recognition in English and Mandarin' (2015), built at Baidu","source_url":"https://arxiv.org/abs/1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD biophysics Princeton (advisors William Bialek, Michael J. Berry); OpenAI VP of Research; co-founded Anthropic 2021","source_url":"https://en.wikipedia.org/wiki/Dario_Amodei","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in biophysics from Princeton University, Hertz Thesis Prize 2011","source_url":"https://en.wikipedia.org/wiki/Dario_Amodei","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Worked at Baidu under Andrew Ng (Nov 2014-Oct 2015) on Deep Speech end-to-end speech recognition","source_url":"https://ai.miraheze.org/wiki/Dario_Amodei","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Language Models are Few-Shot Learners' (GPT-3, 2020) and 'Scaling Laws for Neural Language Models' (2020) as VP of Research at OpenAI","source_url":"https://scholar.google.com/citations?user=6-e-ZBEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Anthropic in 2021 and leads development of the Claude model family","source_url":"https://en.wikipedia.org/wiki/Dario_Amodei","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final author of GPT-3 'Language Models are Few-Shot Learners' (2020) and author of 'Scaling Laws for Neural Language Models' — foundational building blocks of frontier LMs","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Core author of 'Deep Speech 2: End-to-End Speech Recognition' (2015, Baidu) — speech/audio sequence-modeling domain","source_url":"https://arxiv.org/abs/1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Evaluating Large Language Models Trained on Code' (Codex, 2021) — source-code language-modeling domain","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Anthropic in 2021 as CEO and leads development of the Claude model family; PhD biophysics Princeton under Bialek/Berry, ex-OpenAI VP Research","source_url":"https://en.wikipedia.org/wiki/Dario_Amodei","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final author of 'Language Models are Few-Shot Learners' (GPT-3) and author of 'Scaling Laws for Neural Language Models' — the scaling results frontier LMs build on","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Scaling Laws for Neural Language Models' (Kaplan, McCandlish, Amodei et al., 2020)","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Core author of 'Deep Speech 2: End-to-End Speech Recognition in English and Mandarin' (2015), a large-scale GPU-trained sequence model built at Baidu; granted speech/RNN patents EP3245597B1 and EP3179415B1","source_url":"https://arxiv.org/abs/1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Evaluating Large Language Models Trained on Code' (Codex, 2021) among his top works — source-code language modeling","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Physics, Stanford University","PhD Biophysics, Princeton University (advisors William Bialek, Michael J. Berry)"],"canonical_papers":["Language Models are Few-Shot Learners (GPT-3, 2020)","Scaling Laws for Neural Language Models (2020)","Deep Reinforcement Learning from Human Preferences (2017)","Concrete Problems in AI Safety (2016)","Constitutional AI: Harmlessness from AI Feedback (2022)","Deep Speech 2: End-to-End Speech Recognition in English and Mandarin (2015)"],"first_verifiable_year":2003,"notable_systems":["Baidu Deep Speech 2","GPT-2","GPT-3","Codex","Claude model family"],"citations":196269,"h_index":67,"patents":0,"dossier_notes":"The dossier's OpenAlex and PubMed matches are the correct person — the biophysics co-authors Bialek, Berry, Marre and Tkacik match his Princeton thesis lineage, and the mass-spectrometry papers come from his Stanford postdoc, not a homonym. But its bibliometrics badly undercount him (OpenAlex 18,991 citations / h-index 35; Semantic Scholar 136,088 / h-index 30) against the verified Google Scholar record of 196,269 / 67, and its top-works list omits GPT-2, the 2017 RLHF paper and Constitutional AI. OpenAlex affiliations 'Palo Alto University' and 'Institute of Molecular Biology' are parsing artifacts. Patents were skipped, not searched.","adjudication_notes":["foundations (pass_1 15, pass_2 18, final 17): pass_2 was substantially right. Pass_1 discounted the PhD as 'not CS/ML per se'; the verified PNAS 2015 maximum-entropy/criticality paper with Bialek and Berry is genuine statistical-learning mathematics on high-dimensional data, not adjacent biology. Scored 17 rather than 18 because the degree and thesis are in biophysics and there is no authored work in optimization or learning theory as such.","vector_embeddings (pass_1 9, pass_2 14, final 11): pass_1 was closer. Pass_2 offered no embedding paper and conceded the record is 'indirect'; under the rubric an uncorroborated implicit contribution cannot carry a 14. No authored embedding, contrastive or retrieval work was found on the verified Scholar profile's top 12.","transformers_lm (pass_1 19, pass_2 20, final 20): pass_2 right. Verified arXiv author list confirms he is the final author of GPT-3 and an author of Scaling Laws, RLHF (2017) and Constitutional AI — authorship in both the scaling and alignment halves of the lineage meets the canonical anchor.","industry_impact (pass_1 19, pass_2 20, final 20): pass_2 right. Verified Google Scholar 196,269 citations / h-index 67, with GPT-3 alone at 82,085, is canonical-tier impact whose core is language models."],"years_language_modeling":11,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2019-2026","evidence":"GPT-2/GPT-3 at OpenAI, Claude family at Anthropic; final author of GPT-3 and Scaling Laws"},{"domain":"code","years":"2021","evidence":"'Evaluating Large Language Models Trained on Code' (Codex), OpenAI"},{"domain":"other","years":"2015","evidence":"'Deep Speech 2' end-to-end speech recognition sequence model, Baidu"}],"frontier_lineage":["few-shot in-context learning (GPT-3, 2020)","neural scaling laws (2020)","RLHF / learning from human preferences (2017)","Constitutional AI alignment (2022)"],"technical_founder_roles":["Anthropic — co-founder & CEO, sets technical direction and authored core research (Claude, Constitutional AI) — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":14,"scientific_founder":12},"confidence":0.8,"duration_ms":57672},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":14,"scientific_founder":12},"confidence":0.83,"duration_ms":64482}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Dario_Amodei","wikipedia_title":"Dario Amodei","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"niki-parmar","name":"Niki Parmar","title":"Co-founder","company":"Essential AI","sector":"general","rank":9,"rubric_version":4,"dimensions":{"foundations":15,"vector_embeddings":16,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":15,"lm_domain_breadth":14,"hands_on_engineering":16,"industry_impact":18,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":82,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":82,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.79,"rationale":"Parmar is one of the eight equal-contributor co-authors of 'Attention Is All You Need' (NeurIPS 2017), the paper that introduced the Transformer architecture and is the direct founding text of the entire modern language-model lineage — this alone places her at the top of the transformers_lm anchor as a personal, canonical, field-defining contribution, not an adjacent or managerial one. She was a research engineer/researcher at Google Brain during this period, giving her hands-on architecture/implementation credit on the systems (attention mechanisms, encoder-decoder Transformer stack) that underlie vector-embedding-based retrieval and every subsequent LLM. She later co-founded Adept AI and, in December 2022, co-founded Essential AI with fellow Transformer co-author Ashish Vaswani (acqui-hired by Nvidia as of June 2026), extending her personal research record into founder-level industry impact building foundation-model companies. Her Semantic Scholar profile (exact name match, single unambiguous candidate) shows 80 papers and 192,205 citations with an h-index of 20, consistent with authorship of one of the most-cited papers in computer science history. Foundations is scored high (deep neural-architecture/optimization work) but not maximal since her record is concentrated in this lineage rather than broader mathematical foundations work.\n\nParmar is a co-author of 'Attention Is All You Need' (NeurIPS 2017) — the Transformer architecture is the direct, load-bearing foundation of every GPT/Claude/Gemini/Llama-class frontier model, so her position in that lineage is maximal (frontier_founder 20). Her hands-on language-modeling record runs from Google Brain research (~2016) through the Transformer, Image Transformer (2018), Stand-Alone Self-Attention (2019), Conformer (2020) and Bottleneck Transformers (2021) into Adept and Essential AI, roughly a decade of continuous work in the transformer/attention core (lm_domain_depth 15, ~10 years, 8-15 band). She has applied self-attention/transformer sequence modeling across at least three distinct modalities — natural-language text, image generation/vision, and speech recognition (Conformer) — two of them outside natural-language text (lm_domain_breadth 15, three domains). She co-founded two AI research companies, Adept AI and Essential AI (both December 2022), operating as a technical/scientific founder authoring the core research, roughly four years in that role across two companies (scientific_founder 14, two-company band).","evidence":[{"claim":"Niki Parmar is listed as one of eight equal-contributor authors of 'Attention Is All You Need' (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin), the paper introducing the Transformer architecture.","source_url":"https://en.wikipedia.org/wiki/Attention_Is_All_You_Need","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ashish Vaswani (fellow Transformer-paper co-author) co-founded Essential AI with Niki Parmar in December 2022; Vaswani and Essential AI were acqui-hired by Nvidia as of June 2026.","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar record for Niki Parmar (id 3877127): exact name match, single candidate (unambiguous), 80 papers, 192,205 citations, h-index 20.","source_url":"https://www.semanticscholar.org/author/3877127","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Third author of 'Attention Is All You Need' (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin, 2017), the paper introducing the transformer architecture","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Image Transformer' (Parmar, Vaswani, Uszkoreit, Kaiser, Shazeer, Ku, Tran, 2018), generalising the transformer to autoregressive image generation with locally restricted self-attention, improving ImageNet NLL from 3.83 to 3.77","source_url":"https://arxiv.org/abs/1802.05751","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Second author of 'Stand-Alone Self-Attention in Vision Models' (Ramachandran, Parmar, Vaswani, Bello, Levskaya, Shlens, 2019), demonstrating self-attention as an effective stand-alone replacement for spatial convolutions in ResNet","source_url":"https://arxiv.org/abs/1906.05909","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fourth author of 'Conformer: Convolution-augmented Transformer for Speech Recognition' (2020), achieving 1.9%/3.9% WER on LibriSpeech with language models","source_url":"https://arxiv.org/abs/2005.08100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 3877127 'Niki Parmar': 80 papers, 192,205 citations, h-index 20; Attention is All you Need at 191,946 citations, Conformer 4,295, Image Transformer 1,945, Stand-Alone Self-Attention 1,383, Bottleneck Transformers 1,217","source_url":"https://api.semanticscholar.org/graph/v1/author/3877127?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ashish Vaswani co-founded Essential AI with Niki Parmar in December 2022","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Niki Parmar is one of eight equal-contributor authors of 'Attention Is All You Need' (2017), introducing the Transformer architecture that frontier LMs descend from.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Image Transformer' (2018), applying self-attention to autoregressive image generation.","source_url":"https://arxiv.org/abs/1802.05751","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Conformer: Convolution-augmented Transformer for Speech Recognition' (2020) — a speech-domain sequence model.","source_url":"https://arxiv.org/abs/2005.08100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Stand-Alone Self-Attention in Vision Models' (2019), self-attention as a vision-model primitive.","source_url":"https://arxiv.org/abs/1906.05909","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Niki Parmar co-founded Essential AI with Ashish Vaswani in December 2022 (after co-founding Adept AI).","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Niki Parmar is an equal-contributor author of 'Attention Is All You Need' (Vaswani, Shazeer, Parmar, et al., 2017), the paper introducing the Transformer architecture that modern frontier LLMs descend from.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Image Transformer' (Parmar et al., 2018), applying self-attention to autoregressive image generation — a distinct modeling domain.","source_url":"https://arxiv.org/abs/1802.05751","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ashish Vaswani co-founded Essential AI with Niki Parmar in December 2022; she had earlier co-founded Adept AI.","source_url":"https://en.wikipedia.org/wiki/Ashish_Vaswani","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["Attention Is All You Need (NeurIPS 2017)"],"first_verifiable_year":2017,"notable_systems":["Transformer architecture","Essential AI foundation models"],"citations":192205,"h_index":20,"patents":0,"dossier_notes":"MAJOR DOSSIER ERROR: the dossier's Wikipedia and Wikidata blocks are entirely about Ashish Vaswani (a different, though closely related, Attention Is All You Need co-author) — wrong person, not Niki Parmar. There is no standalone Wikipedia page for Niki Parmar (confirmed via direct fetch, 404). Her identity and co-authorship were instead confirmed via the Attention Is All You Need Wikipedia page (author list) and the Ashish Vaswani Wikipedia page's explicit statement that he co-founded Essential AI with her. The dossier's Semantic Scholar block does appear correctly matched (name_exact:true, 1 candidate) and was relied on for citation metrics. The dossier's PubMed block (368 raw/47 refined, all titles/years null) is unusable and was disregarded. This session's WebSearch budget was exhausted; identity confirmation relied on WebFetch of Wikipedia pages only — a fuller pass (Google Scholar, Essential AI team page, Adept AI records) could not be completed and would strengthen confidence further.","years_language_modeling":10,"years_as_technical_founder":4,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2016-2026","evidence":"Attention Is All You Need (2017), transformer/pretraining research at Google Brain, Adept, Essential AI"},{"domain":"other","years":"2018-2021","evidence":"Image Transformer (2018), Stand-Alone Self-Attention (2019), Bottleneck Transformers (2021) — vision/image sequence models"},{"domain":"other","years":"2020","evidence":"Conformer (2020) — speech-recognition sequence model"}],"frontier_lineage":["Transformer architecture / multi-head self-attention (Attention Is All You Need, 2017) — the core block of all frontier language models","self-attention generalizations across vision and speech modalities"],"technical_founder_roles":["Adept AI — co-founder / research scientist — 2022","Essential AI — co-founder — 2022-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":15,"lm_domain_breadth":15,"scientific_founder":14},"confidence":0.72,"duration_ms":56130},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":15,"lm_domain_breadth":14,"scientific_founder":14},"confidence":0.8,"duration_ms":55293}],"source":"seeded","profile_url":null,"wikipedia_title":"Ashish Vaswani","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"tengyu-ma","name":"Tengyu Ma","title":"Co-founder & former CEO (now MongoDB Chief AI Scientist); Assistant Professor","company":"Voyage AI (acquired by MongoDB, Feb 2025)","sector":"general","rank":10,"rubric_version":4,"dimensions":{"foundations":19,"vector_embeddings":20,"transformers_lm":16,"frontier_founder":15,"lm_domain_depth":15,"lm_domain_breadth":16,"hands_on_engineering":16,"industry_impact":16,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":80,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":80,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.85,"rationale":"Ma holds a PhD from Princeton under Sanjeev Arora on the mathematics of non-convex optimization (sparse coding, topic models, word embeddings, dynamical systems), giving deep, authored foundations in optimization/statistical learning theory. He is a Stanford assistant professor and co-author of 'A Simple but Tough-to-Beat Baseline for Sentence Embeddings' (ICLR 2017, the SIF method, ~1,051 citations) and 'A Latent Variable Model Approach to PMI-based Word Embeddings' (TACL, ~299 citations) — both are canonical, authored contributions to the vector-embeddings lineage, not adjacent work. He personally founded and led Voyage AI (2023) as CEO, a company built specifically around embedding models and retrieval, later acquired by MongoDB where he is now Chief AI Scientist — this is founder-level industry impact whose core is exactly the embeddings/vector-search space the rubric targets. His broader research spans deep learning, representation learning, and foundation models (co-author of the widely-cited 'On the Opportunities and Risks of Foundation Models' survey, ~2,279 citations), giving solid but not first-author-canonical standing in the transformers_lm dimension. OpenAlex shows 173 works, 9,501 citations, h-index 38; Semantic Scholar for the Stanford-affiliated profile shows 678 papers, 26,116 citations, h-index 73.\n\nMa sits verifiably in the frontier lineage through named, reusable building blocks rather than fame: the PMI latent-variable theory of word embeddings (TACL 2016) that mathematically justifies word2vec/GloVe, the SIF sentence-embedding baseline (ICLR 2017), the Sophia second-order optimizer for LM pre-training (2023, demonstrated on 125M–1.5B GPT models with ~2x speedup over Adam), and Voyage's production embedding/reranker models now inside MongoDB Atlas Vector Search — an optimizer plus a retrieval stack are exactly the 13–17 'documented component the frontier labs build on' band. His hands-on language-modeling record runs continuously from Arora-group topic-model/word-embedding theory (~2013–2016) through sentence embeddings, foundation-model work, Sophia and Voyage retrieval models to today (~13 years). Voyage shipped domain-specialized embedding models across natural-language text, source code (voyage-code), finance (voyage-finance), and legal (voyage-law) — four distinct language-modeling domains with a shipped record, though all text-retrieval-based rather than reaching biological/chemistry sequence modeling. He is the genuine scientific/technical founder of ONE company (Voyage AI, founded Sept 2023, CEO who personally authored the core research and models), acquired by MongoDB Feb 2025 where he continues as Chief AI Scientist — roughly 2–3 years in that role, placing him at the bottom of the one-company band.","evidence":[{"claim":"PhD Princeton under Sanjeev Arora, work on non-convex optimization theory applied to sparse coding, topic models, word embeddings","source_url":"https://ai.engineer/speakers/tengyu-ma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A Simple but Tough-to-Beat Baseline for Sentence Embeddings, ICLR 2017 (Arora, Liang, Ma) — SIF sentence embedding method","source_url":"https://dblp.org/rec/conf/iclr/AroraLM17.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Voyage AI (Sept 2023) as CEO; acquired by MongoDB Feb 2025; now MongoDB Chief AI Scientist and Stanford CS professor","source_url":"https://www.voyageai.com/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stanford CS faculty profile","source_url":"https://www.cs.stanford.edu/people/tengyu-ma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'A Latent Variable Model Approach to PMI-based Word Embeddings' (Arora, Li, Liang, Ma, Risteski; TACL 2016) — theoretical justification for PMI, word2vec and GloVe and for the linear-algebraic structure of low-dimensional semantic embeddings","source_url":"https://arxiv.org/abs/1502.03520","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Assistant Professor of Computer Science at Stanford; research areas include deep learning, pre-training / foundation models, non-convex optimization, distributed optimization and high-dimensional statistics; awards include ACM Doctoral Dissertation Award Honorable Mention (2018), COLT Best Paper (20","source_url":"https://ai.stanford.edu/~tengyuma/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of 'Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training' (Liu, Li, Hall, Liang, Ma, 2023), demonstrated on GPT models 125M-1.5B with a 2x step/compute/wall-clock speed-up over Adam","source_url":"https://arxiv.org/abs/2305.14342","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (ID i38QlUwAAAAJ, Stanford): 50,334 citations, h-index 82; recent work includes 'Chain of thought empowers transformers to solve inherently serial problems' and 'One step of gradient descent is provably the optimal in-context learner with one layer of linear self-attention'","source_url":"https://scholar.google.com/citations?user=i38QlUwAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Voyage AI builds embedding models and rerankers for retrieval-augmented generation, including general-purpose, domain-specific (finance, legal, code) and company-specific models; acquired by MongoDB, whose announcement notes the team has 'roots at Stanford, MIT, UC Berkeley, and Princeton'","source_url":"https://www.mongodb.com/company/blog/news/redefining-database-ai-why-mongodb-acquired-voyage-ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A Latent Variable Model Approach to PMI-based Word Embeddings (Arora, Li, Liang, Ma, Risteski; TACL 2016) — theoretical justification for PMI, word2vec and GloVe embedding structure","source_url":"https://arxiv.org/abs/1502.03520","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training (Liu, Li, Hall, Liang, Ma, 2023), demonstrated on GPT 125M–1.5B with ~2x speedup over Adam","source_url":"https://arxiv.org/abs/2305.14342","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Voyage AI ships domain-specialized embedding/reranker models for retrieval — general, code, finance, and legal — founded by Ma as CEO, acquired by MongoDB into Atlas Vector Search","source_url":"https://www.mongodb.com/company/blog/news/redefining-database-ai-why-mongodb-acquired-voyage-ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A Simple but Tough-to-Beat Baseline for Sentence Embeddings (Arora, Liang, Ma; ICLR 2017) — the SIF sentence-embedding method","source_url":"https://dblp.org/rec/conf/iclr/AroraLM17.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Voyage AI ships general-purpose plus domain-specific embedding models for finance, legal and code, and rerankers for RAG; founded by Ma as CEO (Sept 2023), acquired by MongoDB Feb 2025 where he is Chief AI Scientist","source_url":"https://www.mongodb.com/company/blog/news/redefining-database-ai-why-mongodb-acquired-voyage-ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, Princeton University, advisor Sanjeev Arora","BS Tsinghua University (per public bio, not independently verified in this pass)"],"canonical_papers":["A Simple but Tough-to-Beat Baseline for Sentence Embeddings (ICLR 2017)","A Latent Variable Model Approach to PMI-based Word Embeddings (TACL 2016)","On the Opportunities and Risks of Foundation Models (2021, co-author)"],"first_verifiable_year":2011,"notable_systems":["Voyage AI embedding models (now part of MongoDB Atlas Vector Search)"],"citations":26116,"h_index":73,"patents":0,"dossier_notes":"Dossier's Wikipedia block is correctly absent (no Wikipedia page found). OpenAlex flags the match ambiguous (10 candidates) with some likely-mismatched affiliated works mixed in (e.g. a low-light image enhancement CVPR paper) — but the core canonical works (SIF sentence embeddings, PMI word embeddings, foundation models survey) are verifiably this Tengyu Ma via DBLP/Princeton/Stanford cross-reference. The dossier's PubMed block (radiology, oncology, RNA structure papers under 'Ma T') is entirely homonym noise and was disregarded — no legitimate PubMed record exists for this researcher's CS/ML work. Semantic Scholar h-index (73) reported over OpenAlex (38) as it appears to better capture the full Stanford-affiliated record.","years_language_modeling":13,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2015-2026","evidence":"PMI/word-embedding theory (TACL 2016), SIF sentence embeddings (ICLR 2017), Voyage general-purpose text embedding models"},{"domain":"code","years":"2023-2026","evidence":"Voyage voyage-code embedding models for code retrieval"},{"domain":"financial","years":"2024-2026","evidence":"Voyage voyage-finance domain-specific embedding models"},{"domain":"legal","years":"2024-2026","evidence":"Voyage voyage-law domain-specific embedding models"}],"frontier_lineage":["PMI latent-variable theory justifying word2vec/GloVe embeddings (TACL 2016)","SIF sentence-embedding baseline (ICLR 2017)","Sophia optimizer for LM pre-training (2023)","Voyage embedding/reranker models powering RAG in MongoDB Atlas Vector Search"],"technical_founder_roles":["Voyage AI — founder & CEO / core research author — 2023-2025 (then MongoDB Chief AI Scientist 2025-2026)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":15,"lm_domain_breadth":14,"scientific_founder":11},"confidence":0.78,"duration_ms":79216},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":15,"lm_domain_breadth":17,"scientific_founder":10},"confidence":0.82,"duration_ms":62271}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"aidan-gomez","name":"Aidan Gomez","title":"Co-founder & CEO","company":"Cohere","sector":"general","rank":11,"rubric_version":4,"dimensions":{"foundations":15,"vector_embeddings":16,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":7,"hands_on_engineering":18,"industry_impact":18,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":79,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":79,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Gomez is a co-author of \"Attention Is All You Need\" (Vaswani et al., 2017), the paper that introduced the Transformer architecture underlying essentially all modern LLMs — canonical, field-defining work, done as a 20-year-old Google Brain intern out of the University of Toronto, later formalized into an Oxford DPhil (advisors Yarin Gal, Yee Whye Teh) completed 2024. He is also lead/co-author of \"The Reversible Residual Network\" (RevNet, NeurIPS 2017) and an original author of Tensor2Tensor, the reference Transformer implementation — both hands-on engineering contributions to the training-efficiency side of the transformer lineage, not just the attention paper. His Google Scholar record (h-index 49, ~306k citations, dominated by the Transformer paper) shows a narrower but extremely deep footprint concentrated almost entirely in the transformer/attention area rather than broad classical ML foundations or embeddings work (no LSA/word2vec-era contributions; vector_embeddings credit here is mostly transfer from representation-learning work adjacent to the Transformer and RevNet, plus later Cohere embedding-model shipping). Post-2017 he founded Cohere, one of the few labs that trains and ships frontier-scale LLMs and production text-embedding models, giving him personal, technical leadership over systems the field runs on rather than purely business leadership. Foundations score reflects strong applied deep-learning mathematics (backprop, residual/reversible network theory) demonstrated in his own papers rather than a classical linear-algebra/optimization theory record.\n\nGomez is a named co-author of 'Attention Is All You Need' (2017), the transformer paper that today's GPT/Claude/Gemini/Llama-class frontier models directly descend from, and he is an inventor on the assigned transformer patent (ES2934313T3, priority 2017-05-23) plus a cluster of training-efficiency patents (sliceout for transformers, depthwise-separable NMT, low-rank training) — this is a top-of-the-lineage foundational position, not a downstream component. His verifiable language-modeling record runs from the 2017 transformer/Tensor2Tensor NMT work through founding Cohere in 2019 to shipping Command and Embed models still in 2026, ~9 continuous hands-on years, placing lm_domain_depth in the 8-15-year band. His hands-on modeling is concentrated in natural-language text (English NLP through multilingual Aya/Tiny Aya and dense text embeddings/retrieval) — a single deep domain rather than distinct biological/financial/chemistry domains, so breadth stays low. As co-founder and CEO who personally authored the core research and patents Cohere is built on, he has ~7 years as the technical founder of one company, matching the 3-8-year single-company band.","evidence":[{"claim":"Co-author of \"Attention Is All You Need\" (2017), introducing the Transformer architecture, written as a Google Brain intern during undergrad at University of Toronto.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows 306,588 total citations and h-index 49, led by the Transformer paper (284,792 citations on the 2023-updated entry).","source_url":"https://scholar.google.com/citations?user=2oq9614AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of \"The Reversible Residual Network: Backpropagation Without Storing Activations\" (NeurIPS 2017), a memory-efficient deep network training method.","source_url":"https://arxiv.org/abs/1707.04585","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Doctoral studies at University of Oxford under Yarin Gal and Yee Whye Teh (OATML group), DPhil awarded 2024.","source_url":"https://oatml.cs.ox.ac.uk/members/aidan_gomez/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Cohere, an enterprise LLM company that trains and ships foundation and embedding models.","source_url":"https://en.wikipedia.org/wiki/Aidan_Gomez","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Attention Is All You Need' (2017), the transformer paper; interned at Google Brain at age 20; founded Cohere in 2019; Oxford PhD completed 2024; BSc CS+math Toronto","source_url":"https://en.wikipedia.org/wiki/Aidan_Gomez","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'The Reversible Residual Network: Backpropagation Without Storing Activations' (Gomez, Ren, Urtasun, Grosse, arXiv:1707.04585, July 2017)","source_url":"https://arxiv.org/abs/1707.04585","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (Cohere affiliation): ~306,838 citations, h-index 49; top works include Attention Is All You Need, RevNet, Tensor2Tensor for Neural Machine Translation, One Model to Learn Them All, Depthwise Separable Convolutions for NMT","source_url":"https://scholar.google.com/citations?user=2oq9614AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education at Oxford and Toronto with doctoral advisors Yarin Gal and Yee Whye Teh, and Google Scholar id 2oq9614AAAAJ","source_url":"https://www.wikidata.org/wiki/Q110864219","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Attention Is All You Need' (2017), introducing the Transformer.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Cohere, an enterprise LLM company training and shipping Command generation models and Embed text-embedding models.","source_url":"https://en.wikipedia.org/wiki/Aidan_Gomez","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Attention Is All You Need' (2017), introducing the Transformer — the architecture underlying modern frontier LLMs.","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Cohere, an enterprise LLM company training and shipping foundation (Command), multilingual (Aya) and text-embedding (Embed) models; founded 2019.","source_url":"https://en.wikipedia.org/wiki/Aidan_Gomez","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'The Reversible Residual Network' (2017) and original author of Tensor2Tensor, hands-on transformer-lineage engineering.","source_url":"https://arxiv.org/abs/1707.04585","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BSc Computer Science & Mathematics, University of Toronto","DPhil Computer Science, University of Oxford (advisors Yarin Gal, Yee Whye Teh), completed 2024"],"canonical_papers":["Attention Is All You Need (Vaswani et al., 2017)","The Reversible Residual Network: Backpropagation Without Storing Activations (2017)","Tensor2Tensor for Neural Machine Translation (2018)"],"first_verifiable_year":2017,"notable_systems":["Transformer / Tensor2Tensor reference implementation","Cohere Command and Embed model families"],"citations":306588,"h_index":49,"patents":0,"dossier_notes":"Dossier's OpenAlex block is severely undercounting/likely mismatched to the wrong author record: it shows only 3 works, h-index 1, cited_by_count 257, earliest_year 2026 — this omits the 2017 Attention Is All You Need and RevNet papers entirely. Verified directly via Google Scholar (ID 2oq9614AAAAJ, matches Wikidata) that the true record is h-index 49, ~306,588 citations, with the earliest major work in 2017, not 2026. Scored using the verified Scholar data, not the dossier's OpenAlex numbers. PubMed block in dossier is an unrelated homonym (medical imaging/coral reef authors) and was disregarded entirely.","years_language_modeling":9,"years_as_technical_founder":7,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2017-2026","evidence":"Transformer/Tensor2Tensor NMT (2017-18); Cohere Command generative LLMs and Embed text-embedding/retrieval models; multilingual Aya/Tiny Aya (2026)"}],"frontier_lineage":["Transformer architecture ('Attention Is All You Need', 2017) — the direct foundation of all GPT/Claude/Gemini/Llama-class models","Assigned transformer patent ES2934313T3 (2017)","Transformer training-efficiency methods (sliceout, low-rank training, depthwise-separable NMT)"],"technical_founder_roles":["Cohere — co-founder & CEO / technical direction, core transformer research & patents — 2019-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":15,"lm_domain_breadth":7,"scientific_founder":12},"confidence":0.86,"duration_ms":60567},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":7,"scientific_founder":12},"confidence":0.85,"duration_ms":68133}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Aidan_Gomez","wikipedia_title":"Aidan Gomez","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"karen-simonyan","name":"Karén Simonyan","title":"Chief Scientist, Microsoft AI (formerly Co-founder & Chief Scientist, Inflection AI)","company":"Microsoft AI","sector":"general","rank":12,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":16,"transformers_lm":18,"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":14,"hands_on_engineering":19,"industry_impact":18,"scientific_founder":9},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":79,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":79,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.82,"rationale":"Karén Simonyan holds a PhD in computer vision from Oxford (2013, thesis 'Large-Scale Learning of Discriminative Image Representations', advisors Andrew Zisserman and Antonio Criminisi) and, with Zisserman, co-authored VGGNet ('Very Deep Convolutional Networks for Large-Scale Image Recognition'), a canonical deep-representation-learning paper that is one of the most-cited works in computer vision. He then became a Principal Research Scientist at DeepMind, where he was a key contributor to WaveNet, AlphaZero, AlphaFold, BigGAN and Flamingo — systems central to representation learning, generative modeling and reinforcement learning, several published in Nature/Science. He co-founded Inflection AI in 2022 as Chief Scientist and moved to Microsoft AI as Chief Scientist in 2024. This is a strong researcher-builder profile: canonical authored work (VGG), principal engineering role on systems the field runs on (WaveNet, AlphaZero, AlphaFold), and leadership of frontier labs — squarely a high scorer on core dimensions, with vector_embeddings scored slightly below transformers_lm/hands_on since his direct authored contributions to the seq2seq/attention/transformer lineage specifically (vs. representation learning broadly) are less documented in the sources found.\n\nSimonyan co-authored Chinchilla ('Training Compute-Optimal Large Language Models', 2022), a named scaling-law result that today's frontier LMs directly build on, and Flamingo (2022), a foundational few-shot visual-language model — placing his own work squarely in the frontier LM lineage (frontier_founder 17). His language-modeling-specific record is recent and concentrated: Chinchilla and Flamingo (2022), Chief Scientist of Inflection AI building the Pi conversational LLM (2022-2024), then Chief Scientist of Microsoft AI (2024-present) — roughly 4 hands-on LM years, so lm_domain_depth sits in the 8-12 band; his earlier canonical work (VGGNet, AlphaZero) is representation learning and RL, not language modeling. Across sequence-modeling domains he has verifiable hands-on contributions in natural-language text (Chinchilla, Flamingo, Pi), biological sequence modeling (AlphaFold protein structure) and audio/music generation (WaveNet raw-audio autoregressive model) — three distinct domains with two outside natural-language text, matching the lm_domain_breadth 13-17 band. He was co-founder and Chief Scientist of one company, Inflection AI, for about two years (2022-2024), personally setting and authoring the core science, which places scientific_founder at the top of the 'one company, under 3 years' range.","evidence":[{"claim":"Oxford PhD 2013 in computer vision, thesis 'Large-Scale Learning of Discriminative Image Representations', advisors Zisserman/Criminisi","source_url":"https://www.robots.ox.ac.uk/~karen/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (id L7lMQkQAAAAJ) lists him as Chief Scientist, Microsoft AI with ~315,795 citations and h-index 61; top works include VGG (165,619), WaveNet (12,082), Flamingo (9,673) and Training Compute-Optimal Large Language Models (5,421)","source_url":"https://scholar.google.com/citations?user=L7lMQkQAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Very Deep Convolutional Networks for Large-Scale Image Recognition (VGG) — Karen Simonyan first author with Andrew Zisserman","source_url":"https://arxiv.org/abs/1409.1556","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Training Compute-Optimal Large Language Models (Chinchilla) — Simonyan co-author","source_url":"https://arxiv.org/abs/2203.15556","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Flamingo: a Visual Language Model for Few-Shot Learning — Simonyan co-author","source_url":"https://arxiv.org/abs/2204.14198","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q58492861: DPhil University of Oxford 2013, doctoral advisors Andrew Zisserman and Antonio Criminisi, employer Google DeepMind","source_url":"https://www.wikidata.org/wiki/Q58492861","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Mastering Atari, Go, chess and shogi by planning with a learned model (MuZero, Nature 2020) — Simonyan K listed among authors","source_url":"https://pubmed.ncbi.nlm.nih.gov/33361790/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"WaveNet: A Generative Model for Raw Audio — autoregressive audio sequence modeling (Simonyan contributor)","source_url":"https://arxiv.org/abs/1609.03499","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Karén Simonyan co-founded Inflection AI in 2022 as Chief Scientist, then joined Microsoft AI as Chief Scientist in 2024","source_url":"https://en.wikipedia.org/wiki/Inflection_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Flamingo: a Visual Language Model for Few-Shot Learning — Simonyan co-author (visual language modeling)","source_url":"https://arxiv.org/abs/2204.14198","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"WaveNet: A Generative Model for Raw Audio — autoregressive audio/music sequence model, DeepMind (Simonyan contributor)","source_url":"https://arxiv.org/abs/1609.03499","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Improved protein structure prediction using potentials from deep learning (AlphaFold, Nature 2020) — DeepMind team incl. Simonyan; biological sequence modeling","source_url":"https://pubmed.ncbi.nlm.nih.gov/31942072/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Inflection AI co-founded 2022 with Simonyan as Chief Scientist; he and much of the team moved to Microsoft AI in March 2024","source_url":"https://en.wikipedia.org/wiki/Inflection_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Vision, University of Oxford (2013)"],"canonical_papers":["Very Deep Convolutional Networks for Large-Scale Image Recognition (VGGNet, 2014)","WaveNet: A Generative Model for Raw Audio (contributor)","AlphaZero / AlphaFold (contributor, DeepMind)"],"first_verifiable_year":2011,"notable_systems":["VGGNet","WaveNet","AlphaZero","AlphaFold","BigGAN","Flamingo","Inflection Pi"],"citations":180000,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex match (A5042688487, Armenian-script display name, neuroscience/Parkinson's/thyroid topics) is a clear homonym, not this Karen Simonyan; disregarded. Dossier's PubMed refined sample correctly surfaces two genuine matches (Mastering Atari/Go/chess/shogi with a learned model, Nature 2020; Skilful precipitation nowcasting, Nature 2021) among mostly-unrelated homonym rows — the scorer should not use the full refined_count of 10 as his output. Google Scholar ID L7lMQkQAAAAJ in Wikidata was not independently queried this pass for exact citation/h-index; 180,000+ citation figure sourced from a secondary article rather than Scholar directly.","years_language_modeling":4,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2022-2026","evidence":"Chinchilla scaling laws, Flamingo VLM, Inflection Pi LLM, Microsoft AI"},{"domain":"biological","years":"2020","evidence":"AlphaFold protein structure prediction, DeepMind Nature paper"},{"domain":"music","years":"2016","evidence":"WaveNet raw-audio autoregressive generative model, DeepMind"}],"frontier_lineage":["Chinchilla compute-optimal scaling laws — a training-recipe foundation for frontier LLMs","Flamingo few-shot visual language modeling — lineage block for multimodal LMs"],"technical_founder_roles":["Inflection AI — Co-founder & Chief Scientist — 2022-2024"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":18,"lm_domain_depth":14,"lm_domain_breadth":14,"scientific_founder":10},"confidence":0.8,"duration_ms":58769},{"pass":"pass_2","dimensions":{"frontier_founder":17,"lm_domain_depth":11,"lm_domain_breadth":15,"scientific_founder":8},"confidence":0.8,"duration_ms":59446}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"llion-jones","name":"Llion Jones","title":"Co-founder & CTO","company":"Sakana AI","sector":"general","rank":13,"rubric_version":4,"dimensions":{"foundations":13,"vector_embeddings":14,"transformers_lm":20,"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":15,"hands_on_engineering":18,"industry_impact":16,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":79,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":79,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.88,"rationale":"Llion Jones is a co-author (5th of 8) of 'Attention Is All You Need' (NeurIPS 2017), the paper that introduced the Transformer architecture underlying essentially all modern LLMs — he reportedly proposed the paper's title, which is squarely canonical, field-defining work and anchors transformers_lm at the top of the scale. He holds a BSc in Artificial Intelligence and Computer Science and an MSc in Advanced Computer Science from the University of Birmingham (completed 2009), then worked as a software engineer at YouTube/Google before moving into Google Research/Google Brain in 2015 doing machine intelligence and NLP research, personally building the sequence-to-sequence and attention-based systems that led to the Transformer. In 2023 he co-founded Sakana AI (with David Ha and Ren Ito) as CTO, where he leads a research organization personally building novel model architectures (nature-inspired/evolutionary methods) rather than merely directing others. His formal ML/math training is master's-level (not a research PhD), and while his attention/seq2seq work is deeply tied to representation learning, I found no dedicated vector-embeddings/retrieval-system publication under his name, so vector_embeddings is scored as strong-adjacent-lineage rather than top-tier authored work.\n\nJones is a co-author of 'Attention Is All You Need' (NeurIPS 2017) and, per the paper's footnote, wrote the initial Transformer codebase, experimented with novel model variants, and built efficient inference and visualizations — the Transformer is the literal architectural foundation that every GPT/Claude/Gemini/Llama-class frontier model directly descends from, anchoring frontier_founder at the top. His hands-on language-modeling record runs continuously from Google Research/Brain NLP work (~2015) through WikiReading (2016), One Model To Learn Them All / Tensor2Tensor, character-level self-attention LMs (2018), and on to Sakana AI's Transformer-squared and Continuous Thought Machines (2024-2025) — roughly 11 verifiable years, placing lm_domain_depth in the 8-15-year band. He has a hands-on record across three distinct language-modeling domains: natural-language text (transformers, NMT, WikiReading), source code (CodeTrans, 2021), and biological sequence modeling (ProtTrans protein language models, 2020) — two of them outside natural-language text — supporting the 13-17 breadth band. He has operated as a scientific/technical founder of one company, Sakana AI (co-founder & CTO, 2023-present, ~3 years), personally authoring core research, which sits at the low end of the single-company technical-founder band.","evidence":[{"claim":"Llion Jones is listed as an author (5th of 8) on 'Attention Is All You Need', arXiv:1706.03762 / NeurIPS 2017, which introduced the Transformer architecture","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jones holds a BSc in Artificial Intelligence and Computer Science and an MSc in Advanced Computer Science from the University of Birmingham, completed 2009","source_url":"https://en.wikipedia.org/wiki/Llion_Jones","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jones worked at YouTube/Google as a software engineer starting ~2011-2012, then moved into Google Research doing machine intelligence and NLP work from 2015 before co-founding Sakana AI in 2023 as CTO","source_url":"https://en.wikipedia.org/wiki/Llion_Jones","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile for Llion Jones (Sakana AI, verified sakana.ai email) shows very high citation counts driven substantially by the Transformer paper","source_url":"https://scholar.google.com/citations?user=_3_P5VwAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DBLP lists a Llion Jones publication record including recent (2024-2025) papers on transformer/model-architecture and evaluation topics","source_url":"https://dblp.org/pers/j/Jones:Llion","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Attention Is All You Need' (2017); footnote: 'Llion also experimented with novel model variants, was responsible for our initial codebase, and efficient inference and visualizations.'","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"arXiv author listing shows a sustained LM/transformer record: WikiReading (2016), One Model To Learn Them All (2017), Tensor2Tensor for NMT (2018), Character-Level Language Modeling with Deeper Self-Attention (2018), Lingvo (2019), ProtTrans (2020), CodeTrans (2021), Transformer Layers as Painters (","source_url":"http://export.arxiv.org/api/query?search_query=au:%22Llion+Jones%22&start=0&max_results=30","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Welsh ML researcher; BSc in AI and computer science and MSc in advanced computer science from the University of Birmingham; at Google Research in machine intelligence/NLP from 2015; co-founded Sakana AI in 2023 as CTO with David Ha and Ren Ito.","source_url":"https://en.wikipedia.org/wiki/Llion_Jones","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sakana AI's published research line includes Evolutionary Model Merge, The AI Scientist, Transformer-squared self-adaptive LLMs and Continuous Thought Machines.","source_url":"https://sakana.ai/blog/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jones is an author of 'Attention Is All You Need' (2017), which introduced the Transformer; the paper's footnote credits him with the initial codebase, novel model variants, efficient inference and visualizations","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jones co-founded Sakana AI in 2023 as CTO with David Ha and Ren Ito, leading research on novel model architectures (Evolutionary Model Merge, Transformer² self-adaptive LLMs, Continuous Thought Machines)","source_url":"https://en.wikipedia.org/wiki/Llion_Jones","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jones is a co-author of 'Attention Is All You Need' (arXiv:1706.03762); footnote credits him with the initial Transformer codebase, novel model variants, and efficient inference/visualizations","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ProtTrans applies transformer/protein language models to biological sequences (protein language modeling)","source_url":"https://arxiv.org/abs/2007.06225","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CodeTrans applies transformer language models to source code","source_url":"https://arxiv.org/abs/2104.02443","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Llion Jones co-founded Sakana AI in 2023 as CTO, leading research including Transformer-squared and Continuous Thought Machines","source_url":"https://en.wikipedia.org/wiki/Llion_Jones","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["University of Birmingham — BSc Artificial Intelligence and Computer Science","University of Birmingham — MSc Advanced Computer Science, completed 2009"],"canonical_papers":["Attention Is All You Need (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin; NeurIPS 2017, arXiv:1706.03762)"],"first_verifiable_year":2015,"notable_systems":["Transformer architecture (Google Brain)","Sakana AI research systems (evolutionary/nature-inspired model architectures)"],"citations":0,"h_index":15,"patents":0,"dossier_notes":"The dossier's Wikipedia/Wikidata entries correctly identify the right Llion Jones (Sakana AI CTO, co-author of Attention Is All You Need) — no homonym issue found. OpenAlex returned no match (a gap, since Google Scholar/DBLP confirm a real publication record exists — OpenAlex simply appears not to index him well). Semantic Scholar's reported citation_count of 188,954 for 19 papers is consistent with the outsized citation count of the Transformer paper alone and looks plausible rather than a homonym artifact, but I did not independently verify every one of the 19 papers belongs to him, so I did not rely on the exact h-index figure it reports (h_index above is from Semantic Scholar and should be treated as approximate). Citations figure left at 0 rather than repeating an unverified six-figure number as a precise claim.","years_language_modeling":11,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2015-2025","evidence":"Transformer, WikiReading, Tensor2Tensor NMT, character-level self-attention LMs; Sakana AI LLM research"},{"domain":"code","years":"2021","evidence":"CodeTrans — transformer LMs for source code"},{"domain":"biological","years":"2020","evidence":"ProtTrans — protein sequence (language) models"}],"frontier_lineage":["Transformer architecture (Attention Is All You Need) — the base architecture of all frontier LLMs","initial Transformer reference codebase / Tensor2Tensor lineage"],"technical_founder_roles":["Sakana AI — co-founder & CTO — 2023-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":15,"scientific_founder":10},"confidence":0.83,"duration_ms":43390},{"pass":"pass_2","dimensions":{"frontier_founder":20,"lm_domain_depth":16,"lm_domain_breadth":15,"scientific_founder":10},"confidence":0.82,"duration_ms":48161}],"source":"seeded","profile_url":null,"wikipedia_title":"Llion Jones","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"percy-liang","name":"Percy Liang","title":"Professor of Computer Science; Director, Center for Research on Foundation Models; Co-founder","company":"Together AI","sector":"general","rank":14,"rubric_version":4,"dimensions":{"foundations":19,"vector_embeddings":16,"transformers_lm":18,"frontier_founder":16,"lm_domain_depth":18,"lm_domain_breadth":10,"hands_on_engineering":16,"industry_impact":19,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":79,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":79,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Percy Liang has a canonical statistical-learning and NLP research record: PhD from UC Berkeley (2011, advisors Dan Klein and Michael I. Jordan) in structured prediction and semantic parsing, followed by 15 years as a Stanford CS professor producing foundational work across probabilistic modeling, optimization, and representation learning. He created SQuAD (2016) and SQuAD 2.0 (2018), the benchmark datasets that shaped a generation of reading-comprehension and embedding/retrieval-adjacent NLP research, and authored \"Prefix-Tuning\" (2021), a widely used parameter-efficient adaptation method for transformer LMs. He is lead author of \"On the Opportunities and Risks of Foundation Models\" (2021), which named and framed the foundation-model paradigm, and founded/directs Stanford's Center for Research on Foundation Models, building HELM, the standard holistic evaluation framework for LLMs — direct, personally-led technical leadership of the transformers/scaling/alignment lineage rather than commentary about it. His verified Google Scholar record (h-index 143, ~167,000 citations) substantially exceeds the figures in the dossier's OpenAlex/Semantic Scholar blocks, confirming an exceptionally deep and continuous 20+ year research record. His industry role is co-founder of Together AI (an open LLM infrastructure/training company) alongside his Stanford professorship, not a from-scratch CEO-only business role, which is why hands_on_engineering is scored as senior-research-leadership rather than the top anchor reserved for principal builders of shipped consumer/enterprise production infra.\n\nLiang authored several documented components that today's frontier stacks build on: Prefix-Tuning (2021), a widely-used parameter-efficient adaptation method for transformer LMs; the SQuAD / SQuAD 2.0 benchmarks that shaped a generation of reading-comprehension and retrieval evaluation; HELM, the standard holistic LLM evaluation framework; and Stanford Alpaca, a canonical instruction-tuning recipe — named benchmark/alignment/adaptation blocks (13-17 band) rather than the core architecture itself. His language-modeling record is continuous and deep: first verifiable NLP work in 2004 (MEng under Michael Collins), a 2011 Berkeley PhD in semantic parsing/structured prediction, and ~22 years of hands-on work through the transformer era, still active as CRFM director and CS336 (LMs from Scratch) instructor — a 15+-year record placing him in the 18-20 band. His verifiable hands-on domains are natural-language text (deep, 2004-present) and code/program synthesis via executable semantic parsing (~2011-2016), i.e. two domains (8-12 band); there is no verifiable biological/financial/chemistry LM record (the PubMed hits are homonyms). As a technical founder he is one of several scientific co-founders of Together AI (2022, ~4 years, one company), with much of the core systems science done by co-founders (Ré, Dao) — a verifiable technical co-founder of one company in the 3-8-year band.","evidence":[{"claim":"PhD in Computer Science, UC Berkeley (2011), advisors Dan Klein and Michael I. Jordan.","source_url":"https://en.wikipedia.org/wiki/Percy_Liang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows 167,157 total citations and h-index 143.","source_url":"https://scholar.google.com/citations?user=pouyVyUAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of \"On the Opportunities and Risks of Foundation Models\" (2021), which coined/framed the term \"foundation model\"; Director of Stanford's Center for Research on Foundation Models (CRFM), which built HELM.","source_url":"https://arxiv.org/abs/2108.07258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Together AI, an open-source/open-weight LLM training and inference infrastructure company.","source_url":"https://cs.stanford.edu/~pliang/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD UC Berkeley 2011 under Michael I. Jordan and Dan Klein; MEng MIT 2005 under Michael Collins; BS MIT 2004; Professor of CS (courtesy Statistics) at Stanford; projects include Marin and CodaLab Worksheets; teaches CS336 Language Models from Scratch","source_url":"https://cs.stanford.edu/~pliang/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar: ~167,157 citations, h-index 143, i10-index 357; top works include On the Opportunities and Risks of Foundation Models, SQuAD, Prefix-Tuning, Emergent Abilities of LLMs, Lost in the Middle, Understanding Black-box Predictions via Influence Functions, HELM, Stanford Alpaca","source_url":"https://scholar.google.com/citations?user=pouyVyUAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Stanford's Center for Research on Foundation Models; NSF CAREER, PECASE, IJCAI Computers and Thought Award, Sloan Fellowship","source_url":"https://en.wikipedia.org/wiki/Percy_Liang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as a Founder of Together AI (founded 2022) alongside Vipul Ved Prakash, Ce Zhang, Chris Re and Tri Dao","source_url":"https://www.together.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prefix-Tuning: Optimizing Continuous Prompts for Generation (2021), a parameter-efficient adaptation method for transformer LMs (2331 citations).","source_url":"https://doi.org/10.18653/v1/2021.acl-long.353","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"SQuAD (2016) and SQuAD 2.0 (2018) benchmark datasets, and On the Opportunities and Risks of Foundation Models (2021); Director of Stanford CRFM which built HELM.","source_url":"https://arxiv.org/abs/2108.07258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in NLP/semantic parsing, UC Berkeley 2011 (advisors Dan Klein, Michael I. Jordan); MEng MIT under Michael Collins; earliest publication 2004; ~22 years active in language modeling.","source_url":"https://en.wikipedia.org/wiki/Percy_Liang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Together AI (founded 2022) alongside Vipul Ved Prakash, Ce Zhang, Chris Ré and Tri Dao.","source_url":"https://www.together.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prefix-Tuning: Optimizing Continuous Prompts for Generation (2021), a widely used parameter-efficient adaptation method for transformer LMs.","source_url":"https://doi.org/10.18653/v1/2021.acl-long.353","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of On the Opportunities and Risks of Foundation Models (2021) and director of Stanford CRFM, which built HELM.","source_url":"https://arxiv.org/abs/2108.07258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD UC Berkeley 2011 in NLP/semantic parsing under Dan Klein and Michael I. Jordan; MEng MIT under Michael Collins; earliest publications ~2004.","source_url":"https://cs.stanford.edu/~pliang/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as a founder of Together AI (founded 2022) alongside Vipul Ved Prakash, Ce Zhang, Chris Ré and Tri Dao.","source_url":"https://www.together.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science and Mathematics, MIT","PhD Computer Science, UC Berkeley (2011), advisors Dan Klein and Michael I. Jordan"],"canonical_papers":["SQuAD: 100,000+ Questions for Machine Comprehension of Text (2016)","Know What You Don't Know: Unanswerable Questions for SQuAD (2018)","Prefix-Tuning: Optimizing Continuous Prompts for Generation (2021)","On the Opportunities and Risks of Foundation Models (2021)"],"first_verifiable_year":2004,"notable_systems":["SQuAD benchmark","HELM (Holistic Evaluation of Language Models)","Together AI training/inference infrastructure"],"citations":167157,"h_index":143,"patents":0,"dossier_notes":"Dossier's OpenAlex (h-index 87, 45,675 citations) and Semantic Scholar (h-index 104, 87,393 citations) blocks both meaningfully undercount his true record versus verified Google Scholar (h-index 143, 167,157 citations) — likely a coverage gap in OpenAlex/S2 rather than a homonym, since name, Stanford affiliation, Dan Klein advisor link, and SQuAD/foundation-models works all match cleanly. PubMed block is an unrelated set of homonyms (medical/biology authors named Liang P) and was disregarded. Dossier lists company as \"Together AI\"; his primary institutional role is Stanford professor with Together AI as a co-founded venture — both are reflected in the title field.","years_language_modeling":22,"years_as_technical_founder":4,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2004-2026","evidence":"semantic parsing, SQuAD, prefix-tuning, foundation models, HELM — Stanford/Berkeley/MIT research"},{"domain":"code","years":"2011-2016","evidence":"semantic parsing / program synthesis mapping language to executable logical forms & queries (Berkeley PhD and Stanford NLP group)"}],"frontier_lineage":["Prefix-Tuning parameter-efficient adaptation","SQuAD / SQuAD 2.0 benchmarks","HELM holistic LLM evaluation","Stanford Alpaca instruction-tuning recipe","foundation-model paradigm framing"],"technical_founder_roles":["Together AI — scientific co-founder — 2022-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":16,"lm_domain_depth":19,"lm_domain_breadth":10,"scientific_founder":9},"confidence":0.8,"duration_ms":57606},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":18,"lm_domain_breadth":10,"scientific_founder":10},"confidence":0.82,"duration_ms":59520}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Percy_Liang","wikipedia_title":"Percy Liang","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"andrew-ng","name":"Andrew Ng","title":"Founder","company":"DeepLearning.AI / Landing AI","sector":"general","rank":15,"rubric_version":4,"dimensions":{"foundations":19,"vector_embeddings":16,"transformers_lm":10,"frontier_founder":16,"lm_domain_depth":14,"lm_domain_breadth":13,"hands_on_engineering":18,"industry_impact":20,"scientific_founder":16},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":78,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":78,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.84,"rationale":"Ng earned a PhD from UC Berkeley (2002, advisor Michael I. Jordan; dossier's Wikidata '1993' date is wrong/mismatched) with a thesis on shaping and policy search in reinforcement learning, producing canonical statistical-learning and optimization work (the Stanford autonomous helicopter RL papers, and co-authorship of Latent Dirichlet Allocation, a foundational topic-modeling/representation paper). His representation-learning record is real but pre-transformer: sparse autoencoders and unsupervised feature learning at Stanford/Google Brain (the 'cat neuron' unsupervised-learning paper) sit squarely in the vector-embeddings/representation lineage, but he has no seq2seq/attention/transformer-architecture or LLM-pretraining/scaling-law authorship — transformers_lm is scored as adjacent-senior-leadership, not authorship, consistent with the brief's instruction not to inflate this dimension for him. He personally co-founded and led Google Brain (2011-2012, distributed deep learning at scale on commodity CPU clusters), then led a 1,300-person AI organization as Chief Scientist at Baidu (2014-2017) including the Deep Speech 2 end-to-end speech system (2181 citations per the dossier's own OpenAlex data). Industry impact is very high: founder of Coursera, deeplearning.ai and Landing AI, director of Stanford AI Lab, and an author credited with 200+ papers per his own institutional bio, corroborating that OpenAlex's 12-work/h-index-5 match here is a severe undercount of his true record.\n\nNg sits verifiably in the pre-word2vec distributional-semantics lineage frontier language models descend from: he co-authored Latent Dirichlet Allocation (2003, foundational topic modeling) and 'Learning word vectors for sentiment analysis' (2011, a documented pre-word2vec word-embedding paper), and co-founded Google Brain (2011-12), whose large-scale distributed deep-learning / COTS-HPC training work (2013) is a training-stack precursor — a documented component the frontier stack builds on rather than a named architecture like attention or scaling laws, which he did not author (frontier_founder 15). His hands-on language-modeling record runs ~2003→2018 across LDA, word vectors, Deep Speech / Deep Speech 2 end-to-end recognition (2015), and diverse-backtranslation grammar correction (2018), ~15 years but as one thread within broader ML rather than a continuous LM specialization (lm_domain_depth 13). Distinct language-modeling domains with a real hands-on record: natural-language text (LDA, word vectors, grammar correction), speech (Deep Speech 2), and clinical/biomedical (deep learning for palliative care 2017, CheXNet-line medical NLP/RAG referral routing) — two-to-three domains (lm_domain_breadth 13). As scientific/technical founder he set and executed the technical direction of Coursera (2012), DeepLearning.AI (2017) and Landing AI (2017), personally authoring the core ML curricula/research — ~14 years across three such companies (scientific_founder 16).","evidence":[{"claim":"PhD from UC Berkeley in 2002, advisor Michael I. Jordan, thesis on shaping and policy search in reinforcement learning; the Wikidata P582 '1993' date in the dossier does not match and is not corroborated by Wikipedia.","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded and led the Google Brain Deep Learning Project (2011-2012) with Jeff Dean and Greg Corrado, including the large-scale unsupervised feature-learning ('cat neuron') work on 16,000 CPU cores.","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Served as VP & Chief Scientist at Baidu (2014-2017), overseeing a 1,300-person AI team; authored/co-authored over 200 papers in AI and related fields per his own institutional biography.","source_url":"https://www.andrewng.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Deep Speech 2: End-to-End Speech Recognition in English and Mandarin (2015) and Deep learning with COTS HPC systems (2013) are real, high-impact papers with 2181 and 606 citations respectively.","source_url":"https://doi.org/10.48550/arxiv.1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Directed the Stanford Artificial Intelligence Laboratory (SAIL) as assistant professor (2002) and associate professor (2009); remains adjunct professor.","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile: ~320,529 citations, h-index 159; top works include 'Latent dirichlet allocation' (2003, 65,185), 'On spectral clustering' (2001, 14,602), 'Learning word vectors for sentiment analysis' (2011, 8,311)","source_url":"https://scholar.google.com/citations?hl=en&user=mG4imMEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS Carnegie Mellon 1997, MS MIT 1998, PhD UC Berkeley 2002 under Michael I. Jordan (thesis: shaping and policy search in reinforcement learning); founded and directed Google Brain 2011-2012; Baidu Chief Scientist 2014-2017; co-founded Coursera (2012), DeepLearning.AI (2017), Landing AI, AI Fund","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records doctorate with advisor Michael I. Jordan and Google Scholar ID JgDKULMAAAAJ, employers Stanford, Google, Coursera, Baidu","source_url":"https://www.wikidata.org/wiki/Q2846695","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on 'Deep Speech 2: End-to-End Speech Recognition in English and Mandarin' (2015) and 'Deep learning with COTS HPC systems' (2013)","source_url":"https://arxiv.org/abs/1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ng co-authored 'Learning word vectors for sentiment analysis' (ACL 2011), a pre-word2vec word-embedding / distributional-semantics paper (8,311 citations per his Google Scholar), placing his work in the vector-space lineage frontier LMs descend from.","source_url":"https://scholar.google.com/citations?hl=en&user=mG4imMEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded and led the Google Brain deep-learning project (2011-2012) and authored 'Deep learning with COTS HPC systems' (2013) and Deep Speech 2 (2015) — distributed-training and speech-recognition lineage work.","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-authored Latent Dirichlet Allocation (Blei, Ng, Jordan, 2003) — foundational topic modeling and a natural-language representation method (~65k citations).","source_url":"https://scholar.google.com/citations?hl=en&user=mG4imMEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded/led Coursera (2012), DeepLearning.AI (2017) and Landing AI (2017) as the technical founder who authored the core ML courses/research, plus AI Fund.","source_url":"https://en.wikipedia.org/wiki/Andrew_Ng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Clinical/biomedical language-modeling record: 'Improving palliative care with deep learning' (2017) and CheXNet-line medical NLP work.","source_url":"https://doi.org/10.1109/bibm.2017.8217669","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored Latent Dirichlet Allocation (Blei, Ng, Jordan, 2003) and 'Learning word vectors for sentiment analysis' (2011), a pre-word2vec word-vector paper in the distributional-semantics lineage.","source_url":"https://scholar.google.com/citations?hl=en&user=mG4imMEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"As VP & Chief Scientist at Baidu (2014-2017) led the Deep Speech / Deep Speech 2 end-to-end neural speech-recognition systems (2015, 2181 citations).","source_url":"https://doi.org/10.48550/arxiv.1512.02595","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder of DeepLearning.AI (2017), Landing AI (founder & CEO, 2017) and AI Fund (2018), and co-founder of Coursera (2012) — serving as the scientific/technical founder who personally authors the technical direction and curriculum.","source_url":"https://www.andrewng.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["B.Sc. Carnegie Mellon University","M.Sc. MIT","Ph.D. UC Berkeley (2002, advisor Michael I. Jordan)"],"canonical_papers":["Shaping and policy search in Reinforcement learning (PhD thesis)","Latent Dirichlet Allocation (Blei, Ng, Jordan)","Autonomous helicopter control via reinforcement learning (Stanford)","Deep learning with COTS HPC systems (2013)","Large-scale unsupervised feature learning / sparse autoencoders (Google Brain 'cat neuron' work)","Deep Speech 2: End-to-End Speech Recognition in English and Mandarin (2015)"],"first_verifiable_year":2002,"notable_systems":["Google Brain distributed deep learning infrastructure","Baidu Deep Speech / Deep Speech 2","Coursera","deeplearning.ai","Landing AI"],"citations":3171,"h_index":5,"patents":0,"dossier_notes":"OpenAlex match is thin and clearly incomplete for this person (only 12 works, h-index 5, earliest_year 2013) despite the dossier's own top_works listing a 2181-citation and a 606-citation paper -- those two citation counts alone are inconsistent with an h-index of 5, so the OpenAlex profile undercounts his true output; his own institutional bio (andrewng.org) states 200+ papers. Google Scholar profile page (id JgDKULMAAAAJ) could not be fetched directly (404) to pull an exact current h-index, so h_index/citations above are left as the dossier's (known-undercounted) OpenAlex figures rather than an unverified estimate. Wikidata's educated_at P582 end-date of 1993 is not corroborated by Wikipedia or andrewng.org, both of which give 2002 for the Berkeley PhD; 1993 appears to be a data error, possibly conflated with an earlier CMU/MIT degree date. PubMed data (3227/73 refined) is heavily homonym-contaminated with unrelated medical-oncology authors named 'Ng A' and was not used for scoring.","years_language_modeling":15,"years_as_technical_founder":14,"technical_founder_companies":3,"lm_domains":[{"domain":"natural_language","years":"2003-2018","evidence":"LDA topic modeling (2003), word vectors for sentiment (2011), diverse-backtranslation grammar correction (2018)"},{"domain":"clinical","years":"2017-2024","evidence":"deep learning for palliative care (2017), CheXNet-line medical NLP and RAG specialty-referral routing"},{"domain":"other","years":"2014-2015","evidence":"Deep Speech / Deep Speech 2 end-to-end speech recognition (speech language modeling)"}],"frontier_lineage":["Pre-word2vec word-vector embeddings (Learning word vectors for sentiment analysis, 2011)","Latent Dirichlet Allocation topic-model representation (2003)","Large-scale distributed deep-learning / COTS-HPC training-stack precursors (Google Brain, 2011-2013)"],"technical_founder_roles":["Coursera — co-founder (technical direction) — 2012","DeepLearning.AI — founder — 2017-present","Landing AI — founder-CEO — 2017-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":13,"scientific_founder":16},"confidence":0.7,"duration_ms":75548},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":16,"lm_domain_breadth":13,"scientific_founder":15},"confidence":0.72,"duration_ms":91999}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Andrew_Ng","wikipedia_title":"Andrew Ng","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jared-kaplan","name":"Jared Kaplan","title":"Co-founder & Chief Science Officer","company":"Anthropic","sector":"general","rank":16,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":13,"transformers_lm":20,"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":10,"hands_on_engineering":16,"industry_impact":19,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":77,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":77,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Kaplan holds a Harvard PhD in theoretical physics (advisor Nima Arkani-Hamed) and spent 15 years as an academic physicist before moving into AI, giving him deep, verifiable mathematical/statistical training even though it predates a formal ML degree. As lead author of 'Scaling Laws for Neural Language Models' (2020) he authored one of the canonical papers underlying the modern LM-scaling paradigm, and he is a listed co-author on GPT-3 ('Language Models are Few-Shot Learners') and Constitutional AI, placing him squarely in the transformer/scaling/alignment lineage. His Google Scholar profile (real citation count 169k+, h-index 82) is far higher than the dossier's Semantic Scholar figure, indicating the dossier undercounts him. He co-founded Anthropic and leads its research/alignment agenda, which is industry impact rooted directly in core LM research rather than pure business leadership. Vector-embeddings work specifically is thinner in the visible record (his focus is scaling laws and RLHF/Constitutional AI, not embeddings/retrieval per se), so that dimension is scored moderately on adjacency rather than direct authorship.\n\nKaplan is a first-order frontier_founder: as lead author of 'Scaling Laws for Neural Language Models' (2020) he authored a method that today's frontier training runs (GPT/Claude/Gemini/Llama-class) directly descend from, and he co-authored GPT-3, Codex, RLHF and Constitutional AI — named blocks in the frontier stack, not mere applications. His verifiable hands-on language-modeling record runs from ~2019 (OpenAI/scaling-laws work) to the present, about 7 years, placing lm_domain_depth in the 8-12 band; his prior ~15 years were theoretical physics, not language modeling, so they do not count toward LM years. Across those years he has a real hands-on record in two distinct LM domains — natural-language text (scaling laws, GPT-3, Claude, Constitutional AI) and source code (Codex co-author) — which sits in the two-domain 8-12 breadth band. As co-founder and Chief Science Officer of Anthropic since 2021 (~5 years) he personally sets and executes the core research direction of one company whose core is these systems, matching the 3-8-year / one-company scientific_founder band.","evidence":[{"claim":"Harvard PhD in physics (2009), advisor Nima Arkani-Hamed, thesis on holography","source_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of 'Scaling Laws for Neural Language Models' (arXiv 2001.08361), the canonical LM scaling-laws paper","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows 169,316 citations, h-index 82, including co-authorship of 'Language Models are Few-Shot Learners' (GPT-3) and 'Constitutional AI: Harmlessness from AI Feedback'","source_url":"https://scholar.google.com/citations?user=KNr3vb4AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and Chief Science Officer of Anthropic","source_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Scaling Laws for Neural Language Models' (2020), establishing power-law scaling of loss with model size, dataset size and compute over seven orders of magnitude","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (Johns Hopkins University & Anthropic): ~169,316 citations, h-index 82, i10-index 125; top papers include GPT-3, Codex, scaling laws, Constitutional AI and RLHF","source_url":"https://scholar.google.com/citations?user=KNr3vb4AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in Physics from Harvard (2009) advised by Nima Arkani-Hamed, thesis 'Aspects of holography'; professor at Johns Hopkins since 2012; joined OpenAI 2019; co-founder and Chief Science Officer of Anthropic","source_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'A Mathematical Framework for Transformer Circuits' (Anthropic, 2021), which decomposes attention into QK and OV circuits and identifies induction heads","source_url":"https://transformer-circuits.pub/2021/framework/index.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of 'Scaling Laws for Neural Language Models' (2020), the canonical scaling-law paper underlying frontier LM training","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of GPT-3, Codex, RLHF and Constitutional AI (Google Scholar KNr3vb4AAAAJ) — natural-language and code LM domains","source_url":"https://scholar.google.com/citations?user=KNr3vb4AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and Chief Science Officer of Anthropic (founded 2021), setting and executing its research direction","source_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of 'Scaling Laws for Neural Language Models' (arXiv 2001.08361), foundational to frontier-model training","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of GPT-3 'Language Models are Few-Shot Learners' and Codex 'Evaluating Large Language Models Trained on Code' — natural-language and code LM domains","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and Chief Science Officer of Anthropic; joined OpenAI in 2019 before founding Anthropic in 2021","source_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Constitutional AI: Harmlessness from AI Feedback', an alignment method the frontier stack draws on","source_url":"https://arxiv.org/abs/2212.08073","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Stanford University, BA physics and mathematics","Harvard University, PhD physics (2009, advisor Nima Arkani-Hamed)"],"canonical_papers":["Scaling Laws for Neural Language Models (2020)","Language Models are Few-Shot Learners (GPT-3, 2020)","Constitutional AI: Harmlessness from AI Feedback (2022)"],"first_verifiable_year":2009,"notable_systems":["Anthropic's Claude model family (scaling/alignment research)"],"citations":169316,"h_index":82,"patents":0,"dossier_notes":"The dossier's openalex block is missing (API budget error) and Semantic Scholar undercounts him (30,844 citations, h-index 33) vs. his real Google Scholar profile (169,316 citations, h-index 82, id KNr3vb4AAAAJ matching wikidata). The PubMed sample in the dossier is entirely homonym noise (pediatric gastroenterology, neuroimaging, etc.) from unrelated 'Kaplan J' authors and should be disregarded. Identity confirmed via Wikipedia/Wikidata (Q102649624) cross-checked against Google Scholar and the Harvard physics PhD / Arkani-Hamed advisor record.","years_language_modeling":7,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2019-2026","evidence":"Scaling laws, GPT-3, Claude family, Constitutional AI"},{"domain":"code","years":"2021-2022","evidence":"Co-author of Codex, 'Evaluating LLMs Trained on Code' (arXiv 2107.03374)"}],"frontier_lineage":["Scaling Laws for Neural Language Models (2020) — power-law scaling used to plan frontier training","GPT-3 few-shot in-context learning","Constitutional AI / RLHF alignment methods","Codex code-LM pretraining"],"technical_founder_roles":["Anthropic — co-founder & Chief Science Officer — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":19,"lm_domain_depth":12,"lm_domain_breadth":10,"scientific_founder":12},"confidence":0.82,"duration_ms":48676},{"pass":"pass_2","dimensions":{"frontier_founder":18,"lm_domain_depth":11,"lm_domain_breadth":9,"scientific_founder":12},"confidence":0.86,"duration_ms":56271}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Jared_Kaplan","wikipedia_title":"Jared Kaplan","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"yang-zhilin","name":"Yang Zhilin","title":"Co-founder & CEO","company":"Moonshot AI (Kimi)","sector":"general","rank":17,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":14,"transformers_lm":20,"frontier_founder":18,"lm_domain_depth":14,"lm_domain_breadth":8,"hands_on_engineering":18,"industry_impact":17,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":77,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":77,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.83,"rationale":"Yang Zhilin is a canonical contributor to the transformer/language-model lineage: first author of Transformer-XL (2019) and co-first-author of XLNet (2019), both widely-cited pretraining/architecture papers that predate GPT-3 and directly address long-context and permutation-based autoregressive language modeling. He holds a CMU machine-learning PhD (advisors Ruslan Salakhutdinov and William Cohen) after a Tsinghua CS undergraduate degree, and interned at Google Brain and FAIR before founding Moonshot AI in 2023, where he personally leads the Kimi model line known for very long context windows. This is a researcher-founder profile: canonical authored work plus hands-on leadership of a frontier lab producing shipped LLMs, not a fame-only executive. The dossier's OpenAlex match (id A5101639237, 'Z. Yang', wireless-communications/DTMB topics, works from 1989-2010) is clearly a different person and must be disregarded. citations note: XLNet has 10,000+ citations per secondary sources; exact h-index not independently confirmed beyond dossier's mismatched OpenAlex record.\n\nYang Zhilin authored named building blocks the frontier stack directly descends from: Transformer-XL (2019) introduced segment-level recurrence and relative positional encodings — a positional-encoding scheme reused across later transformer LMs — and XLNet (2019) introduced permutation-based autoregressive pretraining, both canonical pretraining/architecture contributions predating GPT-3. His verifiable hands-on language-modeling record runs from roughly 2017-2018 (PhD-era NLP/LM work, HotpotQA, Transformer-XL, XLNet) through Huawei PanGu and BAAI Wu Dao contributions to leading the Kimi model line at Moonshot AI in 2026 — about 8-9 years, placing him in the 13-17 depth band. The verifiable modeling record is concentrated in natural-language text (Transformer-XL, XLNet, Kimi), with code capability present only as part of a general LLM rather than a separately-attributable domain model (the CodeGeeX and PubMed hits in the dossier are homonym mismatches), so cross-domain breadth is a single well-established domain. He is a scientific/technical founder in the strict sense — personally authoring the core research — across two companies: Recurrent AI (co-founded 2016) and Moonshot AI (co-founded 2023, where he sets and executes the model direction), consistent with the two-company 13-17 band.","evidence":[{"claim":"First author of Transformer-XL and XLNet, published before GPT-3","source_url":"https://x.com/Michaelzsguo/status/2078154407611416935","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Studied CS at Tsinghua, PhD at Carnegie Mellon, interned at Google Brain and Meta AI (FAIR)","source_url":"https://en.wikipedia.org/wiki/Yang_Zhilin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"XLNet uses permutation language modeling, cited 10,000+ times; co-founded Moonshot AI in 2023 leading Kimi","source_url":"https://daily.dev/posts/yang-zhilin-and-moonshot-ai-the-researcher-behind-kimi-ufwiaatfo","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata doctoral advisors Ruslan Salakhutdinov and William W. Cohen at CMU","source_url":"https://www.wikidata.org/wiki/Q130865273","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (id 7qXxyJkAAAAJ, Carnegie Mellon) shows ~52,873 citations, h-index 42, with XLNet (18,229) and Transformer-XL (6,361) as top works","source_url":"https://scholar.google.com/citations?user=7qXxyJkAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context — Zhilin Yang is a lead author","source_url":"https://arxiv.org/abs/1901.02860","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"XLNet: Generalized Autoregressive Pretraining for Language Understanding — Zhilin Yang first author","source_url":"https://arxiv.org/abs/1906.08237","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD at Carnegie Mellon (2019) under Ruslan Salakhutdinov and William Cohen; co-founded Recurrent AI 2016; worked on Huawei PanGu and BAAI Wu Dao; co-founded Moonshot AI 2023","source_url":"https://en.wikipedia.org/wiki/Yang_Zhilin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kimi k1.5: Scaling Reinforcement Learning with LLMs — Moonshot AI technical report","source_url":"https://arxiv.org/abs/2501.12599","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Transformer-XL introduced segment-level recurrence and relative positional encoding, reused in later transformer language models; Zhilin Yang is lead author","source_url":"https://arxiv.org/abs/1901.02860","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"XLNet: generalized autoregressive (permutation) pretraining, Zhilin Yang first author","source_url":"https://arxiv.org/abs/1906.08237","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CMU ML PhD (advisors Salakhutdinov, Cohen), co-founded Recurrent AI 2016, worked on Huawei PanGu and BAAI Wu Dao, co-founded Moonshot AI 2023 leading Kimi","source_url":"https://en.wikipedia.org/wiki/Yang_Zhilin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kimi k1.5 technical report — Moonshot AI language-model scaling with RL, his current LM work","source_url":"https://arxiv.org/abs/2501.12599","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Recurrent AI (2016) and Moonshot AI (2023), leading the Kimi model line; researcher-founder profile","source_url":"https://en.wikipedia.org/wiki/Yang_Zhilin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, Tsinghua University","PhD Machine Learning, Carnegie Mellon University (advisors: Ruslan Salakhutdinov, William W. Cohen)"],"canonical_papers":["Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context (2019)","XLNet: Generalized Autoregressive Pretraining for Language Understanding (2019)"],"first_verifiable_year":2018,"notable_systems":["Transformer-XL","XLNet","Kimi (Moonshot AI)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex match (A5101639237, 'Z. Yang', 6 works, wireless/telecom topics, earliest 1989) is a clear homonym mismatch, not the AI researcher. The dossier's PubMed matches (surface engineering, glioma, fungal endophytes) are also unrelated homonyms. Real citation count/h-index for Yang Zhilin (AI) not independently verified via Semantic Scholar (dossier reports semantic_scholar not found); Google Scholar ID 7qXxyJkAAAAJ and GitHub kimiyoung in Wikidata corroborate identity but citation metrics were not pulled from Scholar directly in this pass.","years_language_modeling":9,"years_as_technical_founder":7,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2017-2026","evidence":"Transformer-XL, XLNet, PanGu/Wu Dao contributions, Kimi model line at Moonshot AI"},{"domain":"code","years":"2023-2026","evidence":"code generation/reasoning as a capability within the general-purpose Kimi LLM; no separately-attributable code-only model (CodeGeeX attribution in dossier is a homonym)"}],"frontier_lineage":["Transformer-XL relative positional encoding and segment-level recurrence","XLNet permutation language-modeling pretraining objective"],"technical_founder_roles":["Recurrent AI — co-founder (NLP) — 2016","Moonshot AI — co-founder & CEO setting model/technical direction — 2023-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":18,"lm_domain_depth":14,"lm_domain_breadth":6,"scientific_founder":14},"confidence":0.8,"duration_ms":57884},{"pass":"pass_2","dimensions":{"frontier_founder":18,"lm_domain_depth":14,"lm_domain_breadth":9,"scientific_founder":15},"confidence":0.8,"duration_ms":50284}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Yang_Zhilin","wikipedia_title":"Yang Zhilin","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"john-schulman","name":"John Schulman","title":"Co-founder & Chief Scientist","company":"Thinking Machines Lab","sector":"general","rank":18,"rubric_version":4,"dimensions":{"foundations":19,"vector_embeddings":9,"transformers_lm":19,"frontier_founder":19,"lm_domain_depth":12,"lm_domain_breadth":6,"hands_on_engineering":18,"industry_impact":19,"scientific_founder":15},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":75,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":75,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.89,"rationale":"Schulman authored the canonical reinforcement-learning optimization lineage the entire RLHF stack is built on: Trust Region Policy Optimization (ICML 2015) and Proximal Policy Optimization Algorithms (2017), plus Generalized Advantage Estimation (2016) and OpenAI Gym — these are foundational, field-defining contributions in optimization/statistical learning applied to policy learning, and PPO is the direct mechanism used to RLHF-train modern LLMs, placing him squarely in the 2020+ scaling/alignment lineage. PhD in Computer Science from UC Berkeley under Pieter Abbeel, undergraduate work at Caltech, gives strong foundations training. As an OpenAI co-founder and later head of the post-training/RLHF effort behind ChatGPT, and now chief scientist of Thinking Machines Lab, his industry impact and hands-on engineering are personally led, not managerial-only. Vector embeddings is not his direct research area (scored lower, adjacent competence via broader ML training), which is the one dimension without direct authored work found. Semantic Scholar shows 69 papers and 137,700 citations under his profile (h-index 45), consistent with a top RL researcher, though this count was not independently reconciled paper-by-paper.\n\nSchulman authored Proximal Policy Optimization (2017), the RL optimizer that is the literal mechanism used to RLHF-align every frontier chat model (GPT/Claude/Gemini/Llama-class), and he led OpenAI's post-training/RLHF effort behind InstructGPT (2022) and ChatGPT — instruction-tuning and RLHF are named blocks the frontier stack directly descends from, placing him firmly at 18-20 for frontier_founder. His hands-on language-modeling record is roughly 2020-2026 (InstructGPT, GSM8K verifiers 2021, ChatGPT, GPT-4 report 2023) — about 6 years of deep LM work built on his earlier RL foundations (PPO/TRPO/GAE were continuous-control/robotics work, not LM), which sits in the 3-8-year (8-12) band. His verifiable LM work is essentially all natural-language text plus math/reasoning within it, a single domain, so breadth is low. As an OpenAI co-founder (2015-2024) who personally authored the core research the company's alignment ran on, and now chief scientist setting technical direction at Thinking Machines Lab (2025-), he has ~10 years operating as a scientific/technical founder-scientist, placing scientific_founder in the 8-15-year band.","evidence":[{"claim":"Schulman is first author of Trust Region Policy Optimization (TRPO), ICML 2015, with Levine, Moritz, Jordan, Abbeel.","source_url":"https://arxiv.org/abs/1502.05477","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Schulman is first author of Proximal Policy Optimization Algorithms (2017), the RL optimization method used across modern RLHF pipelines.","source_url":"https://arxiv.org/abs/1707.06347","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from UC Berkeley advised by Pieter Abbeel; co-founder of OpenAI; later chief scientist at Thinking Machines Lab (after a stint at Anthropic).","source_url":"https://en.wikipedia.org/wiki/John_Schulman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar profile: 69 papers, 137,700 citations, h-index 45.","source_url":"https://www.semanticscholar.org/author/John-Schulman/47971768","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD thesis 'Optimizing Expectations: From Deep Reinforcement Learning to Stochastic Computation Graphs', UC Berkeley 2016, advisor Pieter Abbeel; introduced TRPO with monotonic improvement guarantee and a general gradient-estimator calculus","source_url":"https://www2.eecs.berkeley.edu/Pubs/TechRpts/2016/EECS-2016-217.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored Proximal Policy Optimization (2017, 30,865 citations), Trust Region Policy Optimization (2015, 8,273), High-Dimensional Continuous Control Using GAE (2015, 4,828), InstructGPT (2022, 23,966), GPT-4 Technical Report (2023, 27,061), Training Verifiers to Solve Math Word Problems (2021, 10,518","source_url":"https://api.semanticscholar.org/graph/v1/author/47971768?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of OpenAI; joined Anthropic August 2024; joined Thinking Machines Lab February 2025 as chief scientist","source_url":"https://en.wikipedia.org/wiki/John_Schulman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records UC Berkeley doctorate with doctoral advisor Pieter Abbeel and Caltech undergraduate education","source_url":"https://www.wikidata.org/wiki/Q103236782","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Proximal Policy Optimization (Schulman first author, 2017) is the RL optimization method used across modern RLHF pipelines that align frontier LLMs.","source_url":"https://arxiv.org/abs/1707.06347","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Schulman co-authored InstructGPT (Training language models to follow instructions with human feedback, 2022), the canonical RLHF instruction-tuning paper behind ChatGPT.","source_url":"https://arxiv.org/abs/2203.02155","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of OpenAI (2015-2024), then chief scientist at Thinking Machines Lab from 2025.","source_url":"https://en.wikipedia.org/wiki/John_Schulman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Training Verifiers to Solve Math Word Problems (GSM8K, 2021) — LM reasoning work under his authorship.","source_url":"https://arxiv.org/abs/2110.14168","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Schulman co-authored InstructGPT / 'Training language models to follow instructions with human feedback' (2022), the canonical RLHF instruction-tuning recipe underlying ChatGPT.","source_url":"https://arxiv.org/abs/2203.02155","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of OpenAI (2015-2024) where he led post-training/RLHF; joined Anthropic Aug 2024; chief scientist at Thinking Machines Lab from Feb 2025.","source_url":"https://en.wikipedia.org/wiki/John_Schulman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-authored 'Training Verifiers to Solve Math Word Problems' (GSM8K, 2021) — natural-language math reasoning.","source_url":"https://arxiv.org/abs/2110.14168","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, UC Berkeley (advisor Pieter Abbeel)","Caltech (undergraduate, physics)"],"canonical_papers":["Trust Region Policy Optimization (2015)","High-Dimensional Continuous Control Using Generalized Advantage Estimation (2016)","Proximal Policy Optimization Algorithms (2017)"],"first_verifiable_year":2015,"notable_systems":["OpenAI Gym","OpenAI RLHF/InstructGPT post-training","ChatGPT post-training","Thinking Machines Lab"],"citations":137700,"h_index":45,"patents":0,"dossier_notes":"Dossier's openalex block is marked found:false, confirmed to be an API-budget gap rather than an absence of record — Semantic Scholar (69 papers, h-index 45) and direct arXiv/ICML lookups confirm a substantial, high-impact, canonical publication record (TRPO, GAE, PPO) that the dossier under-represents. The pubmed block in the dossier is entirely homonym noise (dermatology/oncology papers by other Schulmans) and should be disregarded.","years_language_modeling":6,"years_as_technical_founder":10,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2020-2026","evidence":"RLHF post-training of InstructGPT/ChatGPT, GPT-4 technical report, GSM8K math-reasoning verifiers — natural-language text and reasoning"}],"frontier_lineage":["PPO — the RL optimizer used to RLHF-align frontier chat models","InstructGPT/RLHF instruction-tuning methodology behind ChatGPT-class post-training","GAE/TRPO policy-optimization lineage underpinning RLHF"],"technical_founder_roles":["OpenAI — co-founder / RLHF & post-training lead — 2015-2024","Thinking Machines Lab — chief scientist — 2025-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":19,"lm_domain_depth":12,"lm_domain_breadth":6,"scientific_founder":15},"confidence":0.83,"duration_ms":65373},{"pass":"pass_2","dimensions":{"frontier_founder":19,"lm_domain_depth":13,"lm_domain_breadth":7,"scientific_founder":15},"confidence":0.82,"duration_ms":62542}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/John_Schulman","wikipedia_title":"John Schulman","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"wojciech-zaremba","name":"Wojciech Zaremba","title":"Co-founder; Head of AI Resilience","company":"OpenAI","sector":"general","rank":19,"rubric_version":4,"dimensions":{"foundations":17,"vector_embeddings":9,"transformers_lm":17,"frontier_founder":16,"lm_domain_depth":14,"lm_domain_breadth":12,"hands_on_engineering":18,"industry_impact":18,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":75,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":75,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Zaremba holds a 2016 NYU PhD under Yann LeCun and Rob Fergus ('Learning Algorithms from Data') after mathematics and computer science degrees at the University of Warsaw and study at Ecole Polytechnique, and his authored record is squarely in the sequence-modeling half of the lineage that precedes transformers. The fact that decides the transformers_lm dispute is 'Recurrent Neural Network Regularization' (arXiv:1409.2329, 8 September 2014), which the verified arXiv record shows he FIRST-authored with Ilya Sutskever and Oriol Vinyals — the standard reference for applying dropout to LSTMs, evaluated on language modeling and machine translation. That is authored language-modeling research, not the applied-engineering-only record pass_1 described. He also first-authored 'Learning to Execute' (2014) and co-authored 'Intriguing Properties of Neural Networks' (2013, 23,150 citations), 'Spectral Networks and Locally Connected Networks on Graphs' (2013), OpenAI Gym, Codex (arXiv:2107.03374) and the GPT-4 Technical Report. His verified Google Scholar profile shows 146,695 citations, h-index 48, i10 66. He co-founded OpenAI in 2015, personally led the robotics work (Dactyl, domain randomization, Hindsight Experience Replay) and after 2020 led the GPT/Codex/Copilot teams — a genuine built-systems record. Vector embeddings remains his weakest dimension: the graph-spectral-networks work touches representation learning but he has authored no embedding, contrastive or retrieval contribution.\n\nZaremba's own work sits in the language-modeling lineage frontier models descend from: he FIRST-authored 'Recurrent Neural Network Regularization' (2014, with Sutskever and Vinyals — the standard reference for dropout in LSTM language models and MT) and 'Learning to Execute' (2014), and as an OpenAI co-founder he then led the GPT model, Codex and GitHub Copilot teams — Codex being a foundational code LLM the frontier code stack builds on — placing him in the built/led-a-frontier-component band rather than the transformer/word2vec-author top. His verifiable hands-on language-modeling record runs from 2014 (RNN LM regularization, Learning to Execute) through the GPT/Codex era and is still active in 2026, ~12 continuous years. Distinct language-modeling domains with a real hands-on record are two: natural-language text (GPT line) and source code (Learning to Execute 2014 → Codex 2021), both text-based, with no biological/financial/chemistry LM record, so breadth is bounded. He operated as a scientific/technical co-founder of ONE company (OpenAI, 2015–present, ~11 years), personally authoring core research and leading its language-modeling teams, which lands in the 8–15-year single-role band.","evidence":[{"claim":"First author of 'Recurrent Neural Network Regularization' (submitted 8 September 2014) with Ilya Sutskever and Oriol Vinyals, applying dropout to LSTMs for language modeling and machine translation","source_url":"https://arxiv.org/abs/1409.2329","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar XCZpOcAAAAAJ (Head of AI Resilience, OpenAI): 146,695 citations, h-index 48, i10-index 66; top works GPT-4 Technical Report (27,861), Intriguing Properties of Neural Networks (23,150), Improved Techniques for Training GANs (14,463), Codex (11,520), OpenAI Gym (10,957), Spectral Networ","source_url":"https://scholar.google.com/citations?user=XCZpOcAAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD NYU 2016 under Yann LeCun and Rob Fergus, dissertation 'Learning Algorithms from Data'; OpenAI co-founder 2015; led robotics (Rubik's-cube hand) then GPT/Codex/GitHub Copilot teams","source_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Learning to Execute' (2014) on neural networks learning to execute programs","source_url":"https://arxiv.org/abs/1410.4615","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author on 'Evaluating Large Language Models Trained on Code' (Codex, 2021)","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in deep learning, New York University, 2016, dissertation 'Learning Algorithms from Data', advised by Yann LeCun and Rob Fergus","source_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of OpenAI (2015); led robotics research (2015-2020) including a robotic arm/hand solving a Rubik's Cube; led GPT model, GitHub Copilot and Codex teams from 2020","source_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Learning to Execute' (2014) on neural networks learning to execute simple programs","source_url":"https://arxiv.org/abs/1410.4615","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Reinforcement Learning Neural Turing Machines' (2015)","source_url":"https://arxiv.org/abs/1505.00521","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Intriguing Properties of Neural Networks' (Szegedy et al. 2013), foundational adversarial-examples paper; Google Scholar profile shows 146,695 citations, h-index 48, i10-index 66","source_url":"https://scholar.google.com/citations?user=XCZpOcAAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Recurrent Neural Network Regularization' (2014) with Sutskever and Vinyals — dropout for LSTM language models, evaluated on language modeling and machine translation","source_url":"https://arxiv.org/abs/1409.2329","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of OpenAI (2015–present); led robotics (2015–2020) then the GPT model, Codex and GitHub Copilot teams","source_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Learning to Execute' (2014) — neural networks reading/executing source-code programs (code-domain sequence modeling)","source_url":"https://arxiv.org/abs/1410.4615","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Led OpenAI's Codex team; 'Evaluating Large Language Models Trained on Code' (2021), the code-LM foundation of GitHub Copilot","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded OpenAI (2015/2016), led robotics 2015-2020, then led GPT models, Codex and GitHub Copilot teams from 2020 — technical/scientific founder of one company","source_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Learning to Execute' (2014) — neural networks learning to execute programs, a code/sequence language-modeling domain","source_url":"https://arxiv.org/abs/1410.4615","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Mathematics and computer science, University of Warsaw","Ecole Polytechnique","PhD, New York University, 2016 (advisors Yann LeCun, Rob Fergus)"],"canonical_papers":["Intriguing Properties of Neural Networks (2013)","Recurrent Neural Network Regularization (2014) — first author","Learning to Execute (2014) — first author","An Empirical Exploration of Recurrent Network Architectures (2015)","OpenAI Gym (2016)","Evaluating Large Language Models Trained on Code / Codex (2021)","GPT-4 Technical Report (2023)"],"first_verifiable_year":2013,"notable_systems":["OpenAI Gym","Dactyl robotic hand (Rubik's cube)","Codex / GitHub Copilot","GPT model line (team lead)"],"citations":146695,"h_index":48,"patents":0,"dossier_notes":"OpenAlex found no record for him at all and timeline.first_verifiable_year is null; his real first verifiable publication year is 2013. The PubMed block is almost entirely wrong people — 'Zaremba W[Author]' with earliest_year 1959 returns veterinary reproduction papers in swine and cattle (Kauffold, Bostedt, Grunert), pediatric ophthalmology and peptide self-assembly; one genuine hit is buried in it ('Learning from M/EEG data with variable brain activation delays', 2013, with Blaschko and Gramfort). Semantic Scholar id 2563432 (41 papers, 98,837 citations, h-index 31) is the right person but undercounts the verified Google Scholar record. Patents were skipped, not searched.","adjudication_notes":["transformers_lm (pass_1 13, pass_2 18, final 17): pass_2 was right that pass_1 missed the evidence, and the deciding verified fact is arXiv:1409.2329 — he is the FIRST author of 'Recurrent Neural Network Regularization' with Sutskever and Vinyals, authored LSTM language-modeling research, which pass_1 never cited while concluding he had 'no first-author transformer-lineage paper'. Set at 17 rather than 18 because his contribution sits in the pre-transformer RNN/seq2seq segment and his transformer-era credits (Codex, GPT-4 report) are large-team author listings.","foundations (pass_1 15, pass_2 17, final 17): pass_2 right. A Warsaw mathematics degree plus an NYU PhD under LeCun and Fergus whose thesis subject is learning algorithms from data is PhD-level foundations work in the rubric's 13-17 band, at its top.","vector_embeddings (pass_1 8, pass_2 13, final 9): pass_1 was right. Pass_2 conceded he has 'no canonical embedding or dense-retrieval contribution' yet scored 13; the spectral-graph-networks paper is representation learning in a different sense and cannot carry a PhD-level embeddings score.","industry_impact (pass_1 17, pass_2 18, final 18): pass_2 marginally right — verified 146,695 citations plus leadership of the Codex/Copilot product line."],"years_language_modeling":12,"years_as_technical_founder":11,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2014-2026","evidence":"RNN LM regularization (2014); GPT model team lead at OpenAI (2020-present)"},{"domain":"code","years":"2014-2021","evidence":"Learning to Execute (2014); Codex / GitHub Copilot (2021), team lead"}],"frontier_lineage":["RNN/LSTM language-modeling regularization (dropout for recurrent LMs, 2014)","Codex code LLM behind GitHub Copilot (led team)","GPT model line (team lead)","OpenAI Gym as RL infrastructure"],"technical_founder_roles":["OpenAI — co-founder, robotics lead then GPT/Codex/Copilot team lead — 2015-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":16,"lm_domain_depth":15,"lm_domain_breadth":12,"scientific_founder":15},"confidence":0.78,"duration_ms":56380},{"pass":"pass_2","dimensions":{"frontier_founder":17,"lm_domain_depth":14,"lm_domain_breadth":13,"scientific_founder":14},"confidence":0.8,"duration_ms":55397}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Wojciech_Zaremba","wikipedia_title":"Wojciech Zaremba","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"andrej-karpathy","name":"Andrej Karpathy","title":"Founder","company":"Eureka Labs","sector":"general","rank":20,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":14,"transformers_lm":18,"frontier_founder":14,"lm_domain_depth":15,"lm_domain_breadth":6,"hands_on_engineering":20,"industry_impact":18,"scientific_founder":13},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":74,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":74,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"Karpathy earned a PhD in computer science at Stanford (2011-2015) advised by Fei-Fei Li on the intersection of computer vision and NLP (dissertation: Connecting Images and Natural Language), with widely cited work (ImageNet Large Scale Visual Recognition Challenge, Deep Visual-Semantic Alignments for Generating Image Descriptions, Large-Scale Video Classification with CNNs) that is directly in the representation-learning / vector-embedding lineage (joint image-text embedding spaces) though centered on vision rather than pure text embeddings. He is an OpenAI founding member, led Tesla Autopilot/AI, wrote and taught Stanford's CS231n, and personally authored influential minimal/from-scratch implementations of language models (char-rnn, minGPT, nanoGPT, llm.c) that are widely used pedagogical and practical references in the transformer/LM lineage — this is genuine, personal, hands-on engineering of the systems the field runs on, not managerial credit. OpenAlex shows 25 works, 59,127 citations, h-index 20; other aggregators (ResearchGate/SciSpace) suggest wider citation counts in the tens of thousands to 80k+ depending on source, consistent with major impact. Foundations (core math/optimization theory) is strong graduate-level but not his primary authored contribution, so scored just below the top anchor.\n\nKarpathy is a genuine builder inside the frontier lineage rather than the author of a named foundational block: he was an OpenAI founding member (2015) and joined Anthropic's pretraining team in 2026, and his 2015 'Deep Visual-Semantic Alignments' work sits in the joint image-text representation lineage that today's multimodal models descend from, but he did not author the transformer, word2vec, RLHF or the scaling laws, so he lands in the mid-high band, not the top. His hands-on language-modeling record runs continuously from char-rnn and neural image captioning (2015) through minGPT (2020), nanoGPT (2022), llm.c (2024) and nanochat (2025) into Anthropic pretraining (2026) — roughly 11 years, firmly in the 8-15-year depth band. On breadth his verifiable modeling work is essentially one domain — natural-language text (with vision-language captioning), no biological, financial, chemistry, legal or music sequence modeling — so it scores as a single domain. As a technical founder he co-founded OpenAI as a founding research member and founded Eureka Labs (2024), where he personally writes the core code/curriculum (nanochat, LLM101n); two founder-technical roles across roughly four years places him at the lower edge of the two-company band.","evidence":[{"claim":"PhD Stanford (2011-2015), advisor Fei-Fei Li, dissertation Connecting Images and Natural Language.","source_url":"https://cs.stanford.edu/people/karpathy/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of char-rnn, a minimal character-level RNN language model, and later minGPT/nanoGPT, minimal from-scratch GPT implementations.","source_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding member of OpenAI, Director of AI at Tesla (Autopilot), founder of Eureka Labs (2024), joined Anthropic pretraining team in 2026 per Wikipedia.","source_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Stanford 2015 under Fei-Fei Li, dissertation 'Connecting Images and Natural Language'; BSc Toronto 2009, MSc UBC 2011; OpenAI founding member 2015-2017 and 2023-2024; Tesla Director of AI and Autopilot Vision 2017-2022; founded Eureka Labs July 2024; joined Anthropic May 2026 to lead pretraining","source_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar: 80,896 citations; first-author 'Deep visual-semantic alignments for generating image descriptions' (2015, 8,314), 'Large-scale video classification with convolutional neural networks' (2014, 9,447), 'Visualizing and Understanding Recurrent Networks' (2015, 1,645); co-author ImageNet","source_url":"https://scholar.google.com/citations?user=l8WuQJgAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Personally authored open-source LLM training implementations: nanoGPT (62,995 stars), nanochat (57,949), llm.c 'LLM training in simple, raw C/CUDA' (30,975), minGPT (24,879), nn-zero-to-hero (24,343), LLM101n (37,499)","source_url":"https://github.com/karpathy?tab=repositories&sort=stargazers","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex records 25 works, 59,127 citations, h-index 20, earliest publication year 2011, Stanford affiliation","source_url":"https://api.openalex.org/authors/A5009290031","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAI founding member (2015-2017), returned 2023-2024, joined Anthropic pretraining team 2026; founded Eureka Labs 2024.","source_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored char-rnn, minGPT, nanoGPT, llm.c and nanochat — personal from-scratch language-model training implementations spanning 2015-2025.","source_url":"https://github.com/karpathy?tab=repositories&sort=stargazers","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First-author 'Deep Visual-Semantic Alignments for Generating Image Descriptions' (CVPR 2015) — joint image-text embedding / language-generation work in the representation-learning lineage.","source_url":"https://cs.stanford.edu/people/karpathy/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded OpenAI, specialized in deep learning; joined Anthropic pretraining team in 2026; founded Eureka Labs (2024), an AI education platform","source_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored char-rnn and 'The Unreasonable Effectiveness of Recurrent Neural Networks' (2015), and later minGPT/nanoGPT/llm.c/nanochat — minimal from-scratch language-model training implementations widely used as references","source_url":"https://github.com/karpathy?tab=repositories&sort=stargazers","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Stanford (2015, advisor Fei-Fei Li) on connecting images and natural language; canonical multimodal/vision-language works (Deep Visual-Semantic Alignments, DenseCap)","source_url":"https://cs.stanford.edu/people/karpathy/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, Stanford University (advisor Fei-Fei Li, 2015)","MSc University of British Columbia","BSc University of Toronto"],"canonical_papers":["ImageNet Large Scale Visual Recognition Challenge (2015)","Deep Visual-Semantic Alignments for Generating Image Descriptions (CVPR 2015)","Large-Scale Video Classification with Convolutional Neural Networks (CVPR 2014)"],"first_verifiable_year":2011,"notable_systems":["CS231n (Stanford)","char-rnn","minGPT / nanoGPT","llm.c","Tesla Autopilot AI","Eureka Labs"],"citations":59127,"h_index":20,"patents":0,"dossier_notes":"Dossier accurate; OpenAlex flags 'ambiguous' with 4 candidates but the matched profile (Stanford, ImageNet/CVPR papers, Fei-Fei Li lineage) is clearly the correct person, not a homonym. Semantic Scholar's separate 'A. Karpathy' entry (66,031 citations, h-index 16, name_exact:false) is a partially-overlapping alternate profile, not a different scholar.","years_language_modeling":11,"years_as_technical_founder":4,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2015-2026","evidence":"neural image captioning / char-rnn (2015), minGPT/nanoGPT/llm.c/nanochat, OpenAI + Anthropic pretraining"}],"frontier_lineage":["OpenAI founding member contributing to GPT-era pretraining that frontier models descend from","joint image-text embedding work (visual-semantic alignments, 2015) in the multimodal representation-learning lineage","widely-used from-scratch LM training reference implementations (nanoGPT, llm.c) used as practical/pedagogical building blocks"],"technical_founder_roles":["OpenAI — co-founding research member — 2015-2017","Eureka Labs — founder, sets/executes technical direction (writes core code & curriculum) — 2024-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":15,"lm_domain_breadth":5,"scientific_founder":14},"confidence":0.78,"duration_ms":65305},{"pass":"pass_2","dimensions":{"frontier_founder":14,"lm_domain_depth":15,"lm_domain_breadth":8,"scientific_founder":12},"confidence":0.7,"duration_ms":81465}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Andrej_Karpathy","wikipedia_title":"Andrej Karpathy","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"matei-zaharia","name":"Matei Zaharia","title":"Co-founder & CTO, Databricks; Associate Professor, UC Berkeley","company":"Databricks","sector":"general","rank":21,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":18,"transformers_lm":15,"frontier_founder":16,"lm_domain_depth":11,"lm_domain_breadth":5,"hands_on_engineering":20,"industry_impact":19,"scientific_founder":15},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":74,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":74,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.93,"rationale":"Zaharia's vector-embeddings record is personally authored and central, not supervisory: ColBERT (arXiv:2004.12832, SIGIR 2020) is a two-author paper — Omar Khattab and Matei Zaharia — introducing late-interaction contextualized retrieval, one of the reference architectures for dense passage search, verified at 3,143 citations on his Scholar profile and followed by ColBERTv2 and PLAID. That single verified fact is what separates the two passes, and it places him near the canonical anchor on the dimension the rubric names explicitly (dense retrieval, vector search). His transformer/LM record is training-systems rather than architecture, but it is real and authored: he is the final author of 'Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM' (arXiv:2104.04473, 1,875 citations) and the 12th of 13 authors on DSPy (arXiv:2310.03714) for compiling declarative LM pipelines — built LM training and programming infrastructure, not attention or pretraining research of his own, so that dimension stays mid-high. He created Apache Spark at Berkeley's AMPLab, with a 2013 PhD under Ion Stoica and Scott Shenker that won the 2014 ACM Doctoral Dissertation Award and the 2025 ACM Prize in Computing; his verified Google Scholar record is 116,578 citations, h-index 102, i10-index 257. Foundations is the one dimension both passes agreed on and is correctly mid-high: his mathematics is applied within distributed-systems and retrieval work rather than being the contribution itself.\n\nZaharia has authored documented components of the frontier stack rather than the core language-modeling architecture: ColBERT (SIGIR 2020, with Omar Khattab) is a reference late-interaction dense-retrieval method that RAG and neural-search systems build on, and he is the final author of 'Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM' (2021) plus PipeDream — pipeline-parallel training-stack work that large-LM training pipelines descend from — which places him in the 'documented building block' band, not the word2vec/transformer-author band. His hands-on language-modeling record is specific and real but recent: ColBERT (2020), the foundation-models report (2021), Megatron-LM training (2021) and DSPy (2023), i.e. roughly six years (2020-present), all within natural-language text (retrieval, LM training infrastructure, LM programming) — a single language-modeling domain, since the PubMed biomedical hits are homonyms and Spark/RDD/Mesos are distributed-systems work, not language modeling. As a technical founder he is the co-founder and CTO of Databricks since 2013 (~13 years), the one company whose core technology — Apache Spark, MLflow, Delta Lake — he personally authored, which sits in the 8-15-year single-company band. He is a canonical distributed-systems builder and a genuine dense-retrieval/LM-infra contributor, but not an author of attention, pretraining, scaling-law or pre-2013 vector-space precursor work.","evidence":[{"claim":"One of only two authors — Omar Khattab and Matei Zaharia — of 'ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT', SIGIR 2020, introducing late interaction for dense passage retrieval at ~100x lower query cost than prior BERT rankers","source_url":"https://arxiv.org/abs/2004.12832","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final author of 'Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM' (2021) with Narayanan, Shoeybi, Casper, Catanzaro and Phanishayee","source_url":"https://arxiv.org/abs/2104.04473","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar I1EvjZsAAAAJ (UC Berkeley and Databricks): 116,578 citations, h-index 102, i10-index 257; top works include Spark (13,296), foundation models report (12,333), ColBERT (3,143), MLlib (2,654), Megatron-LM LM training (1,875), PipeDream (1,586)","source_url":"https://scholar.google.com/citations?user=I1EvjZsAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"12th of 13 authors on 'DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines' (2023)","source_url":"https://arxiv.org/abs/2310.03714","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Resilient Distributed Datasets' (NSDI 2012, Best Paper) and 'Spark: Cluster Computing with Working Sets' (2010); PhD Berkeley under Ion Stoica and Scott Shenker, 2014 ACM Doctoral Dissertation Award","source_url":"https://www.usenix.org/conference/nsdi12/technical-sessions/presentation/zaharia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Computer Science, UC Berkeley (2007-2013), advisors Ion Stoica and Scott Shenker; dissertation on Spark/RDDs won the 2014 ACM Doctoral Dissertation Award","source_url":"https://www.csail.mit.edu/news/matei-zaharia-receives-acm-doctoral-dissertation-award","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing' (NSDI 2012, Best Paper Award) and 'Spark: Cluster Computing with Working Sets' (2010) — the foundational Apache Spark papers","source_url":"https://www.usenix.org/conference/nsdi12/technical-sessions/presentation/zaharia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author (with Omar Khattab) of 'ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT' (SIGIR 2020), a dense/late-interaction retrieval architecture","source_url":"https://arxiv.org/abs/2004.12832","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior co-author of 'DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines' (NeurIPS 2023)","source_url":"https://arxiv.org/abs/2310.03714","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT (SIGIR 2020), a two-author dense/late-interaction retrieval architecture with Omar Khattab — a retrieval method the frontier RAG/search stack builds on","source_url":"https://arxiv.org/abs/2004.12832","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior/last author of 'DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines' (2023), LM-programming infrastructure","source_url":"https://arxiv.org/abs/2310.03714","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CTO of Databricks (founded 2013), the company built on Apache Spark, which he personally created at UC Berkeley's AMPLab","source_url":"https://en.wikipedia.org/wiki/Matei_Zaharia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final/senior author of 'Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM' (2021), a canonical tensor/pipeline-parallel training-stack paper used to train frontier-scale LMs","source_url":"https://arxiv.org/abs/2104.04473","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Two-author (Khattab & Zaharia) 'ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT' (SIGIR 2020), a reference dense/late-interaction retrieval method extended by ColBERTv2 and PLAID","source_url":"https://arxiv.org/abs/2004.12832","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CTO of Databricks (founded 2013) and creator of Apache Spark, the data/ML core the company runs on; also professor at MIT/Stanford/Berkeley","source_url":"https://en.wikipedia.org/wiki/Matei_Zaharia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BMath, University of Waterloo (2003-2007)","PhD Computer Science, UC Berkeley (2013), advisors Ion Stoica and Scott Shenker"],"canonical_papers":["Spark: Cluster Computing with Working Sets (2010)","Resilient Distributed Datasets (NSDI 2012, Best Paper)","ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT (SIGIR 2020)","ColBERTv2 (2022)","Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM (2021)","DSPy (2023)"],"first_verifiable_year":2001,"notable_systems":["Apache Spark","Apache Mesos (co-author)","MLflow","MLlib","ColBERT / PLAID retrieval engine","DSPy","Delta Lake"],"citations":116578,"h_index":102,"patents":0,"dossier_notes":"The dossier's OpenAlex works list is his but its affiliation list is garbage ('Moscow Institute of Thermal Technology', 'Palo Alto University', 'Berkeley College', 'Georgia Institute of Technology'); its metrics (51,480 citations / h-index 66) badly undercount the verified Google Scholar record (116,578 / 102), and it flags only spark_2010 as canonical while missing ColBERT entirely — the single most relevant paper for the vector-embeddings dimension. The Semantic Scholar match (id 2253469012, 19 papers, h-index 12) is a fragment. The PubMed block is largely homonyms (a Romanian radiologist Zaharia MC, a Greek psycho-oncology trial); only the Stanford cardiology deep-learning papers with Rogers AJ and Bailis P plausibly involve him and none were used. Wikidata omits his Databricks CTO role and his current Berkeley post.","adjudication_notes":["vector_embeddings (pass_1 14, pass_2 19, final 18): pass_2 was right. Both passes cited ColBERT, but pass_1 treated it as one contribution among many while the verified arXiv record shows it is a TWO-author paper — Khattab and Zaharia — i.e. the senior author of a reference dense-retrieval architecture, with 3,143 verified citations and the ColBERTv2/PLAID line following it. That is the rubric's own named dimension at near-canonical level. Set at 18 not 19 because the architectural idea is Khattab-led doctoral work he supervised and co-authored.","transformers_lm (pass_1 12, pass_2 15, final 15): pass_2 right. Pass_1 missed Megatron-LM; verification confirms he is the FINAL author of 'Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM' (1,875 citations), which is authored large-scale LM training work, not tool use. Capped at 15 because he has authored no attention, pretraining or scaling-law research.","hands_on_engineering (pass_1 19, pass_2 20, final 20): pass_2 right. He personally wrote Apache Spark at AMPLab and built MLflow and the ColBERT/PLAID engine — systems the field runs on, which is the stated maximum anchor.","foundations (16) and industry_impact (19): both passes agreed; no dispute to resolve."],"years_language_modeling":6,"years_as_technical_founder":13,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2020-2026","evidence":"ColBERT/ColBERTv2/PLAID dense retrieval, Megatron-LM large-scale LM training, DSPy LM pipelines"}],"frontier_lineage":["ColBERT late-interaction dense retrieval (RAG/neural search)","Megatron-LM GPU-cluster training + PipeDream pipeline parallelism (LM training stack)","DSPy declarative LM pipeline compilation"],"technical_founder_roles":["Databricks — co-founder & CTO — 2013-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":11,"lm_domain_breadth":5,"scientific_founder":15},"confidence":0.82,"duration_ms":60247},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":11,"lm_domain_breadth":5,"scientific_founder":15},"confidence":0.82,"duration_ms":70232}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Matei_Zaharia","wikipedia_title":"Matei Zaharia","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"thomas-wolf","name":"Thomas Wolf","title":"Co-founder & Chief Science Officer","company":"Hugging Face","sector":"general","rank":22,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":14,"transformers_lm":18,"frontier_founder":16,"lm_domain_depth":12,"lm_domain_breadth":10,"hands_on_engineering":19,"industry_impact":19,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":73,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":73,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.85,"rationale":"Wolf is the verified first author of 'HuggingFace's Transformers: State-of-the-Art Natural Language Processing' (arXiv:1910.03771, EMNLP 2020 demos, 25,160 citations), the library that is the de facto implementation layer of the transformer lineage — principal-builder status rather than adjacency. The fact that separates the two passes is DistilBERT (arXiv:1910.01108), where the verified author order is Sanh, Debut, Chaumond, Wolf: he is the senior author of a knowledge-distillation result that is itself language-modeling research (14,772 citations), which pass_1 missed entirely and which lifts his transformers_lm above a pure-tooling reading. He is also an author of BLOOM (176B open multilingual LM), StarCoder, T0 multitask prompted training and Zephyr alignment distillation, and his teams shipped transformers, datasets, tokenizers and the Hub. His verified Google Scholar profile (Co-founder at Hugging Face) shows 65,616 citations and h-index 52, essentially all in the 2019+ transformer window, so depth of experience in this lineage is about seven years rather than the dossier's spurious 58. Foundations is his weakest dimension and neither pass could verify it: no degree, thesis or authored work in linear algebra, optimization or statistical learning could be confirmed from a primary source — his Scholar profile lists a polytechnique.edu contact and his own site returned HTTP 403 — so under the rubric's 'if unsure, score lower' instruction this is scored on the published record alone, which contains no foundations paper.\n\nWolf's own work is a load-bearing part of the foundation today's frontier and open models are built and distributed on: he is first author of the transformers library and an author of the tokenizers and datasets libraries — the canonical open training/inference/data stack — plus DistilBERT (a named knowledge-distillation LM method) and BLOOM (176B open LLM), which places him firmly in the frontier lineage even though he did not author attention, the transformer architecture, word2vec or scaling laws himself. His verifiable hands-on language-modeling record runs ~2018/2019→2026 (~7 years), continuously active and unusually deep for its duration, which sits at the top of the 3–8-year band. He has a hands-on record in two distinct language-modeling domains: natural-language text (transformers, DistilBERT, T0, BLOOM, Zephyr) and source code (StarCoder / BigCode), with no verifiable personal record in biological, financial or other LM domains. He co-founded Hugging Face in 2016 and operates as its Chief Science Officer, personally authoring/leading the core research and code the company runs on — one company, ~10 years in that technical-founder role, which lands in the two-companies-or-8-15-years band on the strength of the years even at a single company.","evidence":[{"claim":"First author of 'HuggingFace's Transformers: State-of-the-art Natural Language Processing' (2019/EMNLP 2020 demos), with Lysandre Debut, Victor Sanh, Julien Chaumond and Clement Delangue","source_url":"https://arxiv.org/abs/1910.03771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior (final) author of 'DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter' (2019); verified author order Victor Sanh, Lysandre Debut, Julien Chaumond, Thomas Wolf","source_url":"https://arxiv.org/abs/1910.01108","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar D2H5EFEAAAAJ (Co-founder at Hugging Face, contact polytechnique.edu): 65,616 citations, h-index 52, i10-index 76; top works Transformers (25,160), DistilBERT (14,772), T0 multitask prompted training (2,717), BLOOM (2,580), StarCoder (2,250), Zephyr (1,095), Datasets (997)","source_url":"https://scholar.google.com/citations?user=D2H5EFEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on the BLOOM 176B-parameter open multilingual language model (BigScience Workshop, 2022)","source_url":"https://arxiv.org/abs/2211.05100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hugging Face profile thomwolf lists NLP and open source, with membership of BigScience, BigCode, LeRobot and Open LLM Leaderboard organizations; it states no degrees, so education remains unverified","source_url":"https://huggingface.co/thomwolf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thomas Wolf is first author of the Transformers library paper, EMNLP 2020 demo track, ~8,300+ citations","source_url":"https://arxiv.org/abs/1910.03771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thomas Wolf co-founded Hugging Face in 2016 with Clément Delangue and Julien Chaumond, and serves as Chief Science Officer / Chief Strategy Officer","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thomas Wolf is a listed author on the BLOOM 176B-parameter open multilingual language model paper (BigScience Workshop, 2022)","source_url":"https://arxiv.org/abs/2211.05100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thomas Wolf's Hugging Face profile confirms his role in NLP/open-source and involvement in BigScience and core Hugging Face teams","source_url":"https://huggingface.co/thomwolf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (Co-founder at HuggingFace): 65,616 citations, h-index 52; top works Transformers (2020, 25,160 cites), DistilBERT (2019, 14,772), BLOOM (2022), StarCoder (2023), Zephyr (2023), Datasets (2021)","source_url":"https://scholar.google.com/citations?user=D2H5EFEAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'HuggingFace's Transformers: State-of-the-Art Natural Language Processing' (EMNLP 2020 demos), the de facto open implementation/distribution layer for the transformer lineage","source_url":"https://arxiv.org/abs/1910.03771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of DistilBERT, a distilled version of BERT — a named knowledge-distillation language-modeling method (author order Sanh, Debut, Chaumond, Wolf)","source_url":"https://arxiv.org/abs/1910.01108","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on BLOOM, the 176B-parameter open multilingual language model (BigScience, 2022)","source_url":"https://arxiv.org/abs/2211.05100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Involved as an author/leader in BigCode / StarCoder, a source-code language model — a second, distinct LM domain","source_url":"https://arxiv.org/abs/2305.06161","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior (final) author of DistilBERT (2019), a knowledge-distillation language-modeling result the frontier open ecosystem draws on","source_url":"https://arxiv.org/abs/1910.01108","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on StarCoder (2023), a source-code language model (BigCode) — a second, non-natural-language LM domain","source_url":"https://arxiv.org/abs/2305.06161","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Hugging Face in 2016 with Clément Delangue and Julien Chaumond and serves as Chief Science Officer — technical/scientific founder who authored the core libraries","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["HuggingFace's Transformers: State-of-the-Art Natural Language Processing (EMNLP 2020 demos) — first author","DistilBERT, a distilled version of BERT (2019) — senior author","BLOOM: A 176B-Parameter Open-Access Multilingual Language Model (2022)","Multitask Prompted Training Enables Zero-Shot Task Generalization / T0 (2021)","StarCoder (2023)","Zephyr: Direct Distillation of LM Alignment (2023)","Datasets: A Community Library for NLP (EMNLP 2021)"],"first_verifiable_year":2019,"notable_systems":["transformers library","datasets library","tokenizers","Hugging Face Hub","BLOOM (BigScience)","StarCoder (BigCode)"],"citations":65616,"h_index":52,"patents":0,"dossier_notes":"The dossier is badly contaminated. Wikidata Q112483880 ('researcher, sport, University of Mainz', born 1971) is a different Thomas Wolf. OpenAlex A5078865608 (flagged ambiguous, 10 candidates) merges at least two people: the Transformers 2020 and BLOOM entries are his, but the diamond-NV-centre, superconductivity and crystallography works and the affiliations (Stuttgart, KIT, Tulane, Augsburg, LSU, Washburn) belong to a physicist of the same name, and its earliest_year 1968 is that person's — so timeline.first_verifiable_year 1968 and years_active 58 are wrong; his real lineage record begins ~2019 (~7 years). Semantic Scholar 'T. Wolf' (7 papers, h-index 6) is also not him. Correct metrics come from his own Google Scholar profile. No primary source confirming his degrees could be retrieved (thomwolf.io returns HTTP 403; the HF profile states none), so education is left empty rather than asserted.","adjudication_notes":["transformers_lm (pass_1 17, pass_2 19, final 18): pass_2 was right that pass_1 under-scored him, and the deciding fact is DistilBERT — the verified arXiv author order (Sanh, Debut, Chaumond, Wolf) puts him in the senior-author position on a 14,772-citation language-modeling research result, which pass_1 never mentioned and which refutes its 'library builder, not LM researcher' framing. Set at 18 rather than 19 because he authored no architecture, pretraining or scaling result of his own.","foundations (pass_1 10, pass_2 14, final 11): pass_1 was right. Pass_2 scored 14 while itself recording education_verified=false and an empty education list; the rubric requires scoring lower when unsure. No degree, thesis or foundations publication is verifiable from any primary source, and his Scholar record contains no mathematics paper.","vector_embeddings (pass_1 13, pass_2 16, final 14): split toward pass_1. Neither pass cited an authored embeddings or retrieval paper; the credit rests on shipping the tokenizers/embedding-distribution infrastructure of the Hub, which supports a solid but not high-teens score.","hands_on_engineering (pass_1 18, pass_2 19, final 19) and industry_impact (pass_1 18, pass_2 19, final 19): pass_2 right on both — he personally shipped the transformers, datasets and tokenizers libraries, and Hugging Face's core product IS the transformer/LM distribution layer."],"years_language_modeling":7,"years_as_technical_founder":10,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2018-2026","evidence":"transformers library, DistilBERT, T0, BLOOM, Zephyr — authored/led"},{"domain":"code","years":"2023-2026","evidence":"StarCoder / BigCode source-code language model — author/leader"}],"frontier_lineage":["transformers library (canonical open training/inference stack)","tokenizers library","datasets library","DistilBERT knowledge distillation","BLOOM 176B open LLM"],"technical_founder_roles":["Hugging Face — Co-founder & Chief Science Officer — 2016-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":16,"lm_domain_depth":12,"lm_domain_breadth":10,"scientific_founder":14},"confidence":0.8,"duration_ms":67944},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":12,"lm_domain_breadth":10,"scientific_founder":14},"confidence":0.8,"duration_ms":70286}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"arthur-mensch","name":"Arthur Mensch","title":"Co-founder & CEO","company":"Mistral AI","sector":"general","rank":23,"rubric_version":4,"dimensions":{"foundations":17,"vector_embeddings":14,"transformers_lm":18,"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":8,"hands_on_engineering":16,"industry_impact":16,"scientific_founder":11},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":72,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":72,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.88,"rationale":"Mensch completed a PhD (2015-2018, Inria/NeuroSpin CEA-Saclay, advisors Bertrand Thirion, Gael Varoquaux, Julien Mairal) on stochastic optimization and dictionary learning for large-scale matrix factorization, publishing 'Dictionary Learning for Massive Matrix Factorization' at ICML 2016 and a 2018 IEEE Trans. Signal Processing follow-up -- genuine applied-math/optimization foundations work, applied to fMRI representation learning rather than text embeddings specifically. He then did a postdoc on optimal transport/stochastic optimization at ENS Paris (2018-2020) before joining Google DeepMind Paris (2020-2023) as a Senior Research Scientist contributing to Flamingo, Gemini, LM scaling and retrieval-augmented generation -- direct hands-on transformer/LM-lineage work at a top lab. He co-founded Mistral AI in 2023 and is a named co-author on both the Mistral 7B and Mixtral-of-Experts technical reports, i.e. personally involved in shipping widely-used open-weight transformer models, not merely a business-side founder. Semantic Scholar record (32 papers, ~19k citations, h-index 14, id 1697879) is consistent with this profile; a PubMed 'Mensch A' hit-set is a clinical homonym and was excluded. Overall: strong PhD-level optimization/math foundations, direct pre-founding LM/transformer research experience at DeepMind, and continued technical authorship post-founding -- a researcher-founder profile, not a pure business CEO.\n\nMensch is a genuine frontier-lineage contributor: he is a named author of 'Training Compute-Optimal Large Language Models' (Chinchilla scaling laws, 2022), a result virtually every frontier lab now trains against, and of the Mistral 7B and Mixtral-of-Experts reports (grouped-query + sliding-window attention, sparse MoE) — architectures and open weights that GPT/Claude/Gemini/Llama-class ecosystems build on and around — plus DeepMind-era work on RETRO and Flamingo. His hands-on language-modeling record runs from a 2018 attention paper ('Differentiable dynamic programming for structured prediction and attention') through DeepMind LM/scaling work (2020-2023) into Mistral (2023-), roughly 6-8 years — deep but not spanning the pre-word2vec era. Distinct LM domains with a real personal record are natural-language text (deep) and vision-language/multimodal (Flamingo, Gemini contributions); code (Codestral) exists at Mistral but his personal modeling record there is thin, so I count two domains. As scientific founder he is co-founder and CEO of exactly one company, Mistral AI (~3 years), where he personally co-authored the core technical reports the product runs on — a real technical-founder role of one company within the 3-8 year band.","evidence":[{"claim":"PhD 2015-2018 at Inria/NeuroSpin on stochastic optimization and representation learning for fMRI, advisors Thirion/Varoquaux/Mairal","source_url":"https://team.inria.fr/parietal/team-members/arthur-mensch/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Dictionary Learning for Massive Matrix Factorization' (ICML 2016), scalable optimization/matrix-factorization work","source_url":"https://arxiv.org/abs/1605.00937","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior Research Scientist at Google DeepMind Paris 2020-2023, contributed to Flamingo/Gemini, LM scaling, retrieval-augmented generation","source_url":"https://en.wikipedia.org/wiki/Arthur_Mensch","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author on the Mistral 7B technical report","source_url":"https://arxiv.org/abs/2310.06825","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author on the Mixtral of Experts technical report","source_url":"https://arxiv.org/abs/2401.04088","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD 2015-2018 Inria/NeuroSpin on stochastic optimization for large-scale fMRI; advisors Thirion, Varoquaux, Mairal; ENS postdoc on optimal transport; DeepMind Paris late 2020-May 2023; Mistral AI co-founder/CEO May 2023-","source_url":"https://en.wikipedia.org/wiki/Arthur_Mensch","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Third author of 'Training Compute-Optimal Large Language Models' (Chinchilla scaling laws), 29 March 2022","source_url":"https://arxiv.org/abs/2203.15556","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Third author of 'Mistral 7B' (grouped-query + sliding-window attention), 10 Oct 2023","source_url":"https://arxiv.org/abs/2310.06825","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (ID F8riAN8AAAAJ): ~53,164 citations, h-index 33; top works Mistral 7B, Gemini, Flamingo, Chinchilla, Mixtral, RETRO, Gopher, 'Differentiable dynamic programming for structured prediction and attention'","source_url":"https://scholar.google.com/citations?user=F8riAN8AAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author of 'Mistral 7B' (grouped-query + sliding-window attention)","source_url":"https://arxiv.org/abs/2310.06825","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author of 'Mixtral of Experts' (sparse mixture-of-experts LM)","source_url":"https://arxiv.org/abs/2401.04088","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Mistral AI (2023-), DeepMind Paris 2020-2023 contributing to Flamingo/Gemini/RAG","source_url":"https://en.wikipedia.org/wiki/Arthur_Mensch","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author of 'Mistral 7B' (grouped-query + sliding-window attention), 10 Oct 2023","source_url":"https://arxiv.org/abs/2310.06825","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author of 'Mixtral of Experts' (sparse MoE transformer), 2024","source_url":"https://arxiv.org/abs/2401.04088","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior Research Scientist at Google DeepMind Paris 2020-2023 (Flamingo/Gemini/RETRO, LM scaling); co-founder & CEO of Mistral AI since May 2023","source_url":"https://en.wikipedia.org/wiki/Arthur_Mensch","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Ecole Polytechnique","Telecom Paris","PhD, Inria / Paris-Saclay (2015-2018)"],"canonical_papers":["Dictionary Learning for Massive Matrix Factorization (ICML 2016)","Stochastic Subsampling for Factorizing Huge Matrices (IEEE TSP 2018)","Mistral 7B (2023)","Mixtral of Experts (2024)"],"first_verifiable_year":2015,"notable_systems":["Mistral 7B","Mixtral 8x7B","Google Gemini (contributor)","Flamingo (contributor)"],"citations":19356,"h_index":14,"patents":0,"dossier_notes":"Dossier's OpenAlex block is missing/empty (API budget error per task instructions), so Semantic Scholar (32 papers, ~19356 citations, h-index 14) was used instead; this understates true output since it likely does not fully capture DeepMind-era and Mistral technical reports. Independently confirmed via web search: PhD topic/years, DeepMind role (Flamingo/Gemini/RAG), and co-authorship on Mistral 7B and Mixtral technical reports, none of which were in the dossier's programmatic blocks.","years_language_modeling":8,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2020-2026","evidence":"Chinchilla scaling laws, RETRO, Gopher, Gemini at DeepMind; Mistral 7B/Mixtral at Mistral AI"},{"domain":"other","years":"2022-2026","evidence":"vision-language / multimodal modeling — Flamingo and Gemini multimodal contributions at DeepMind"}],"frontier_lineage":["Chinchilla compute-optimal scaling laws (used across frontier LM training)","Mistral 7B sliding-window + grouped-query attention","Mixtral sparse mixture-of-experts open weights","RETRO retrieval-augmented LM (contributor)"],"technical_founder_roles":["Mistral AI — co-founder & CEO, co-author of core Mistral 7B/Mixtral technical reports — 2023-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":18,"lm_domain_depth":13,"lm_domain_breadth":8,"scientific_founder":11},"confidence":0.82,"duration_ms":61332},{"pass":"pass_2","dimensions":{"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":9,"scientific_founder":11},"confidence":0.82,"duration_ms":65030}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Arthur_Mensch","wikipedia_title":"Arthur Mensch","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"yann-lecun","name":"Yann LeCun","title":"Executive Chairman & Co-founder","company":"AMI Labs (Advanced Machine Intelligence Labs)","sector":"general","rank":24,"rubric_version":4,"dimensions":{"foundations":20,"vector_embeddings":18,"transformers_lm":14,"frontier_founder":15,"lm_domain_depth":11,"lm_domain_breadth":5,"hands_on_engineering":20,"industry_impact":20,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":72,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":72,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.94,"rationale":"LeCun is a principal architect of the mathematical/engineering core of modern AI: he co-invented convolutional neural networks and pioneered practical backpropagation applications (LeNet, 'Gradient-based learning applied to document recognition' 1998; 'Backpropagation Applied to Handwritten Zip Code Recognition' 1989), work the field still runs on, earning the 2018 ACM Turing Award jointly with Hinton and Bengio. His hands-on record spans PhD-level optimization/statistical-learning training under Maurice Milgram, a Hinton postdoc, and a decade at Bell Labs building and shipping CNN-based systems for real-world document/check recognition. His embeddings/representation-learning work is deep and personally authored but sits in the contrastive/metric-learning and self-supervised lineage (siamese/contrastive embeddings 'Dimensionality Reduction by Learning an Invariant Mapping' 2006; energy-based models; later VICReg/JEPA self-supervised architectures) rather than the text vector-space/word-embedding line, so it scores high but not maximal on vector_embeddings. He is not an author of the seq2seq/attention/original Transformer papers and is publicly a skeptic of autoregressive LLM scaling as a path to intelligence; his transformer-era personal authorship is concentrated in self-supervised and JEPA/world-model architectures (I-JEPA, V-JEPA, 'A Path Towards Autonomous Machine Intelligence' 2022) rather than canonical transformer/LM/RLHF work itself, so transformers_lm is scored as strong-adjacent leadership/authorship rather than top-tier canonical — precise per the brief's instruction to distinguish CNN-era from transformer-era personal authorship. industry_impact is maximal: founding director of Facebook/Meta AI Research (FAIR) and Chief AI Scientist 2013-2025, h-index in the 120-175 range (OpenAlex 120 / Google Scholar 175) with 250k-500k citations, now Executive Chairman of AMI Labs (founded Dec 2025/Nov 2025 per sources, raised $1.03B per Google Scholar-linked reporting).\n\nLeCun's personally-authored foundation — practical backpropagation, convolutional networks (LeNet, 'Gradient-based learning applied to document recognition' 1998; 'Backpropagation Applied to Handwritten Zip Code Recognition' 1989), contrastive/metric embeddings (DrLIM 2006) and the self-supervised paradigm — is genuine deep-learning substrate that frontier multimodal and representation stacks build on, and the FAIR lab he led shipped the PyTorch training stack and the Llama family; but he authored NONE of the transformer/attention, word2vec, scaling-law or RLHF blocks the frontier LANGUAGE line directly descends from and holds no verifiable pre-2013 vector-space/LSI text patents, so he sits mid-band (15), not as a named LM-block author. His hands-on language-modeling record is real but adjacent and comparatively recent — character-level ConvNets for text (Zhang & LeCun, 2015) plus directional leadership of RoBERTa/BART/fastText/Llama pretraining as Chief AI Scientist — roughly 8 years of hands-on/directional LM work rather than a career-length dedicated LM record from the vector-space era (depth 11), and it is confined almost entirely to natural-language text, a single domain (breadth 5). On scientific_founder the record supports only ONE actual company founded in that role — AMI Labs (Dec 2025), co-founded as Executive Chairman, under one year — because 'founding director' of Facebook/Meta AI Research is leadership of a CORPORATE lab, not founding a company, and Chief AI Sci","evidence":[{"claim":"PhD from Université Pierre et Marie Curie (1987), advisor Maurice Milgram, proposed an early form of backpropagation in his thesis; postdoc under Geoffrey Hinton at University of Toronto starting 1987.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Joined AT&T Bell Laboratories in 1988, developed LeNet convolutional neural networks for handwriting/check recognition, commercially deployed.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"2018 ACM A.M. Turing Award, awarded jointly to Yann LeCun, Yoshua Bengio, and Geoffrey Hinton for conceptual and engineering breakthroughs making deep neural networks a critical component of computing.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jacob T. Schwartz Professor of Computer Science, Courant Institute of Mathematical Sciences, New York University, since 2003.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Facebook/Meta AI Research (FAIR), served as Chief AI Scientist 2013-2025; departed Meta to found Advanced Machine Intelligence Labs (AMI Labs) as Executive Chairman.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile shows h-index 175, i10-index 505, 495,810 total citations; top papers Deep Learning (Nature 2015, 122,796 cites) and Gradient-based learning applied to document recognition (1998, 89,891 cites).","source_url":"https://scholar.google.com/citations?user=WLN3QrAAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD 1987 Universite Pierre et Marie Curie, thesis 'Modeles connexionnistes de l'apprentissage'; Bell Labs 1988-1996 developing convolutional networks (LeNet), Optimal Brain Damage, Graph Transformer Networks, deployed bank-check reading, co-created DjVu; 2018 Turing Award with Bengio and Hinton; Met","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile: ~495,810 citations, h-index 175; top works 'Deep learning' (2015, ~122,796), 'Gradient-based learning applied to document recognition' (1998, ~89,891), 'Backpropagation applied to handwritten zip code recognition' (1989, ~20,709)","source_url":"https://scholar.google.com/citations?user=WLN3QrAAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Facebook/Meta AI Research (FAIR, a corporate lab) and Chief AI Scientist 2013-2025; co-founded Advanced Machine Intelligence Labs (AMI Labs) in December 2025 as Executive Chairman — his only company founded in that role.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"2018 ACM Turing Award (with Bengio and Hinton) for practical backpropagation and convolutional networks — deep-learning substrate frontier models build on — while personally not authoring transformer/attention/word2vec/scaling/RLHF work.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"LeCun's personal language-modeling contribution centers on character-level convolutional networks for text classification (Zhang & LeCun, NeurIPS 2015) and later self-supervised/JEPA work; he is publicly a skeptic of autoregressive LLM scaling.","source_url":"https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Dimensionality Reduction by Learning an Invariant Mapping' (Hadsell, Chopra, LeCun, CVPR 2006) established contrastive/siamese embedding learning, a documented precursor in the representation-learning lineage.","source_url":"https://doi.org/10.1109/cvpr.2006.100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Facebook/Meta AI Research (FAIR) and Chief AI Scientist 2013-2025; co-founded Advanced Machine Intelligence Labs (AMI Labs) in December 2025.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"2018 ACM Turing Award (with Bengio and Hinton) for conceptual and engineering breakthroughs — practical backpropagation and convolutional networks — that made deep neural networks a critical component of computing, the substrate frontier models build on.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Championed self-supervised learning and introduced the JEPA world-model direction ('A Path Towards Autonomous Machine Intelligence', 2022; I-JEPA / V-JEPA), the pretraining paradigm underpinning modern LM pretraining, while personally not authoring transformer/attention/LLM-scaling work.","source_url":"https://openreview.net/forum?id=BZ5a1r-kVsf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"LeCun co-invented convolutional networks and practical backprop applications (LeNet, 'Gradient-based learning applied to document recognition' 1998; 'Backpropagation Applied to Handwritten Zip Code Recognition' 1989), foundational deep-learning substrate; 2018 Turing Award.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"LeCun's personal language-modeling contribution is centered on character-level convolutional networks for text classification (Zhang & LeCun, 2015) and later self-supervised/JEPA work; he is publicly a skeptic of autoregressive LLM scaling and did not author transformer/word2vec/scaling/RLHF work.","source_url":"https://en.wikipedia.org/wiki/Yann_LeCun","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Diplôme d'ingénieur, ESIEE Paris","Diplôme d'études approfondies / Doctorate, Pierre and Marie Curie University (1987), advisor Maurice Milgram","Postdoctoral researcher, University of Toronto, under Geoffrey Hinton (1987-1988)"],"canonical_papers":["Gradient-based learning applied to document recognition (LeCun et al., Proceedings of the IEEE, 1998)","Backpropagation Applied to Handwritten Zip Code Recognition (LeCun et al., Neural Computation, 1989)","Deep learning (LeCun, Bengio, Hinton, Nature, 2015)","Dimensionality Reduction by Learning an Invariant Mapping (Hadsell, Chopra, LeCun, CVPR 2006)","A Path Towards Autonomous Machine Intelligence (LeCun, 2022 position paper introducing JEPA)"],"first_verifiable_year":1985,"notable_systems":["LeNet (convolutional neural network for handwritten/check digit recognition, deployed commercially at Bell Labs / NCR)","Facebook/Meta AI Research (FAIR) - founding director","I-JEPA / V-JEPA self-supervised world-model architectures","Torch7 / early deep learning tooling contributions at NYU"],"citations":255413,"h_index":120,"patents":0,"dossier_notes":"Dossier is accurate and well-matched (Wikipedia, Wikidata QID Q3571662 both confirm identity cleanly, no homonym issue despite the flag - PubMed sample co-authors, e.g. Bengio and Precup, corroborate correct person). One correction: dossier's OpenAlex figures (h-index 120, 255k citations) are conservative relative to Google Scholar (h-index 175, ~496k citations, ORCID/Wikidata-linked profile WLN3QrAAAAAJ) - both are cited in evidence since OpenAlex and Google Scholar commonly diverge on venue/preprint coverage; neither is wrong, Scholar is simply more inclusive. Dossier's patents field is empty/skipped, not investigated further here. Dossier correctly captures the Dec 2025 AMI Labs move; Wikipedia dates the Meta departure to November 2025 with AMI Labs founding shortly after - a minor date discrepancy across sources, not material to scoring.","years_language_modeling":8,"years_as_technical_founder":1,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2015-2025","evidence":"Character-level ConvNets for text (Zhang & LeCun 2015); directional leadership of RoBERTa/BART/fastText/Llama pretraining as Meta Chief AI Scientist; self-supervised text and JEPA lineage"}],"frontier_lineage":["Convolutional networks / practical backpropagation (deep-learning substrate)","Contrastive/metric-learning embeddings (DrLIM 2006, representation-learning lineage)","Self-supervised pretraining paradigm (JEPA direction)","PyTorch/Torch training stack (produced by FAIR under his leadership, not personally authored)"],"technical_founder_roles":["Meta AI Research (FAIR) — founding director / Chief AI Scientist (corporate lab leadership, not company founder) — 2013-2025","AMI Labs — co-founder & Executive Chairman — Dec 2025-present (<1 year)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":16,"lm_domain_depth":12,"lm_domain_breadth":5,"scientific_founder":14},"confidence":0.72,"duration_ms":113149},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":11,"lm_domain_breadth":6,"scientific_founder":8},"confidence":0.82,"duration_ms":93000}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Yann_LeCun","wikipedia_title":"Yann LeCun","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ali-farhadi","name":"Ali Farhadi","title":"Professor, Allen School of CSE, University of Washington; CEO of the Allen Institute for AI (Ai2) 2023-2026","company":"Allen Institute for AI (Ai2)","sector":"general","rank":25,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":15,"transformers_lm":14,"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":6,"hands_on_engineering":19,"industry_impact":18,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":70,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":70,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.9,"rationale":"The two passes diverged because pass_1 scored Farhadi from the dossier's OpenAlex record (276 works, 52,784 citations, h-index 58) and concluded his work was vision-only with no embedding or LM record. His verified Google Scholar profile refutes both halves: 230,069 citations, h-index 100, i10-index 221, and the top-15 list contains 'Unsupervised Deep Embedding for Clustering Analysis' (ICML 2016, 5,065 citations — authored representation/embedding-space research), 'Bidirectional Attention Flow for Machine Comprehension' (ICLR 2017, 2,570 — genuine pre-transformer attention-architecture work for language), 'HellaSwag' (ACL 2019, 5,175), 'Defending Against Neural Fake News' / Grover (1,827), 'Model soups' (ICML 2022, 2,127) and 'Editing models with task arithmetic' (1,756), the last two operating directly in model weight space. His hands-on record is the strongest dimension: YOLO (81,998 citations), YOLOv3 (42,656), YOLO9000 (29,619) and XNOR-Net (6,826), the last being first-principles numerical work binarising weights and activations so convolution reduces to XNOR/popcount — the low-precision arithmetic modern inference stacks depend on. A UIUC PhD under David Forsyth and this body of quantization, clustering-objective and weight-space work support strong foundations. On the LM axis he is a listed author of '2 OLMo 2 Furious' (arXiv:2501.00656, verified) but not of the original OLMo paper, so his open-LLM credit is partly organizational as Ai2 CEO; that caps transformers_lm in the mid-teens rather than higher.\n\nFarhadi's own work sits in several documented building blocks that today's frontier stack draws on: XNOR-Net's reduction of convolution to XNOR/popcount is foundational low-precision/binary-quantization lineage for modern quantized inference, HellaSwag (ACL 2019) is a canonical benchmark used to evaluate GPT/Claude/Gemini-class LLMs, 'Model soups' and 'Editing models with task arithmetic' are weight-space merging methods now used in model post-training, and as Ai2 CEO he leads the fully-open OLMo models and the Dolma pretraining corpus — placing him in the 'documented component the frontier labs build on' band rather than the named-architecture band. His personal language-modeling record is genuine but bounded: BiDAF (pre-transformer attention for reading comprehension, 2016/2017), Grover and HellaSwag (2019), and OLMo/OLMo 2 authorship (2024-2025), roughly a decade of hands-on LM work atop a career whose core was computer vision. That LM work is essentially all in natural-language text — a single verifiable language-modeling domain, with no biological, financial, code or other LM modeling record of his own (the PubMed hits are a Rush-University gastroenterologist homonym). As scientific/technical founder he co-founded and led Xnor.ai (~2017-2020), the edge-AI company built directly on his XNOR-Net research and acquired by Apple in 2020 — one company, roughly three years in that role, squarely the 3-8-year single-company band.","evidence":[{"claim":"Google Scholar jeOFRDsAAAAJ (Professor, CSE, University of Washington): 230,069 citations, h-index 100, i10-index 221; top works YOLO (81,998), YOLOv3 (42,656), YOLO9000 (29,619), XNOR-Net (6,826), HellaSwag (5,175), Unsupervised Deep Embedding for Clustering Analysis (5,065), Describing objects by","source_url":"https://scholar.google.com/citations?user=jeOFRDsAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author of '2 OLMo 2 Furious', Ai2's open language model report (43 authors, first listed as Team OLMo)","source_url":"https://arxiv.org/abs/2501.00656","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD, University of Illinois Urbana-Champaign, doctoral advisor David Forsyth; Google Scholar ID jeOFRDsAAAAJ","source_url":"https://www.wikidata.org/wiki/Q80873822","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Professor of computer science and CEO of the Allen Institute for Artificial Intelligence","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of YOLO9000: Better, Faster, Stronger with Joseph Redmon","source_url":"https://arxiv.org/abs/1612.08242","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD, University of Illinois Urbana-Champaign, advisor David Forsyth","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Professor at University of Washington, CEO of Allen Institute for Artificial Intelligence (Ai2)","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (jeOFRDsAAAAJ), UW professor: ~230,069 citations, h-index 100, i10 221; top works YOLO (2016), YOLOv3, YOLO9000, XNOR-Net, HellaSwag, Unsupervised Deep Embedding for Clustering Analysis, Bidirectional Attention Flow, Model soups, Editing models with task arithmetic","source_url":"https://scholar.google.com/citations?user=jeOFRDsAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from University of Illinois Urbana-Champaign, doctoral advisor David Forsyth; Google Scholar ID jeOFRDsAAAAJ","source_url":"https://www.wikidata.org/wiki/Q80873822","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bidirectional Attention Flow for Machine Comprehension (ICLR 2017) — authored attention architecture for language; and HellaSwag (ACL 2019), a benchmark used to evaluate frontier LLMs, on his verified Google Scholar top works","source_url":"https://scholar.google.com/citations?user=jeOFRDsAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author of '2 OLMo 2 Furious', Ai2's fully-open language-model report","source_url":"https://arxiv.org/abs/2501.00656","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Professor of computer science and CEO of the Allen Institute for AI; PhD UIUC under David Forsyth (vision, not LM, pre-2017)","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Xnor.ai, built on his XNOR-Net binary-CNN research; company acquired by Apple, ~2017–2020 as technical founder","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks — reduces convolution to XNOR/popcount, foundational binary/low-precision quantization work","source_url":"https://doi.org/10.1007/978-3-319-46493-0_32","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"HellaSwag: Can a Machine Really Finish Your Sentence? (ACL 2019) — a canonical commonsense-NLI benchmark now standard in frontier LLM evaluation","source_url":"https://aclanthology.org/P19-1472/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bidirectional Attention Flow for Machine Comprehension (ICLR 2017) — authored pre-transformer attention architecture for language","source_url":"https://arxiv.org/abs/1611.01603","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author of '2 OLMo 2 Furious', Ai2's fully-open language-model report; leads OLMo/Dolma as Ai2 CEO","source_url":"https://arxiv.org/abs/2501.00656","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ali Farhadi co-founded Xnor.ai (spun out of Ai2 on XNOR-Net research) and was CEO; acquired by Apple in 2020","source_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, University of Illinois Urbana-Champaign (advisor David Forsyth)"],"canonical_papers":["You Only Look Once: Unified, Real-Time Object Detection (CVPR 2016)","YOLO9000 (CVPR 2017)","YOLOv3 (2018)","XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks (ECCV 2016)","Unsupervised Deep Embedding for Clustering Analysis (ICML 2016)","Bidirectional Attention Flow for Machine Comprehension (ICLR 2017)","HellaSwag (ACL 2019)","Model soups (ICML 2022)","Editing models with task arithmetic (2022)","2 OLMo 2 Furious (2025)"],"first_verifiable_year":2009,"notable_systems":["YOLO detector family","XNOR-Net binary CNNs","Xnor.ai (acquired by Apple)","OLMo / OLMo 2 open LLMs (as Ai2 CEO and OLMo 2 co-author)","AI2-THOR"],"citations":230069,"h_index":100,"patents":0,"dossier_notes":"Two dossier blocks are wrong. The PubMed block is a complete homonym — 'Farhadi A' on the inflammatory-bowel-disease and alcohol-gut-leakiness papers with Keshavarzian and Banan is a Rush University gastroenterologist, and the dossier's first_verifiable_year of 2002 derives from that person. The Semantic Scholar block (61 papers, 427 citations, h=12) is a fragment off by orders of magnitude. OpenAlex A5101576595 is the right person but flagged ambiguous and undercounts him (h=58, 52,784 citations against the verified Scholar record of h=100, 230,069), and its affiliations include spurious entries ('Lorestan University of Medical Sciences', 'Nvidia (United Kingdom)'). Patents were skipped, not searched; XNOR-Net-related patents are plausible but unverified, so recorded as 0/unverified.","adjudication_notes":["vector_embeddings (pass_1 9, pass_2 16, final 15) and transformers_lm (pass_1 8, pass_2 15, final 14): pass_2 was right and pass_1 simply missed the evidence by scoring from the truncated OpenAlex top-works list. The deciding verified facts are on his Google Scholar top-15: 'Unsupervised Deep Embedding for Clustering Analysis' (ICML 2016, 5,065 citations) is authored embedding-space research, and 'Bidirectional Attention Flow for Machine Comprehension' (ICLR 2017, 2,570) is authored attention-architecture work for language — neither appears anywhere in pass_1's evidence. transformers_lm held at 14 because he is on OLMo 2 but not the original OLMo, so part of that credit is organizational.","foundations (pass_1 11, pass_2 17, final 16): pass_2 right. XNOR-Net's reduction of convolution to XNOR/popcount is first-principles numerical work, and the clustering-objective and weight-space papers are authored mathematics; a UIUC PhD under Forsyth plus that record is squarely in the PhD-level band, not the 8-12 'adjacent' band pass_1 used.","industry_impact (pass_1 15, pass_2 18, final 18): pass_2 right. Verified 230,069 citations, the YOLO family the field runs on, Xnor.ai's acquisition, and leadership of the org producing the fully open OLMo models."],"years_language_modeling":10,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2016-2026","evidence":"BiDAF (2016/17), Grover + HellaSwag (2019), OLMo/OLMo 2 authorship & Dolma corpus (2024-25) as Ai2 CEO"}],"frontier_lineage":["XNOR-Net binary/low-precision quantization → quantized inference stacks","HellaSwag benchmark used to evaluate frontier LLMs","Model soups + task arithmetic weight-space merging methods","OLMo open LLMs and Dolma open pretraining corpus (organizational, as Ai2 CEO)"],"technical_founder_roles":["Xnor.ai — co-founder & CEO (built on his XNOR-Net research; acquired by Apple) — ~2017-2020"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":5,"scientific_founder":10},"confidence":0.7,"duration_ms":86014},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":6,"scientific_founder":10},"confidence":0.72,"duration_ms":71321}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Ali_Farhadi","wikipedia_title":"Ali Farhadi","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"chris-olah","name":"Chris Olah","title":"Co-founder, Interpretability Lead","company":"Anthropic","sector":"general","rank":26,"rubric_version":4,"dimensions":{"foundations":14,"vector_embeddings":16,"transformers_lm":19,"frontier_founder":14,"lm_domain_depth":11,"lm_domain_breadth":5,"hands_on_engineering":18,"industry_impact":18,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":70,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":70,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.86,"rationale":"Olah has no formal PhD (he is largely self-taught, coming up through Google Brain and OpenAI as an engineer-researcher), so foundations is scored on strong demonstrated mathematical/statistical work rather than credentials. He personally coined the term 'mechanistic interpretability' and is the driving author behind the Circuits research program ('Zoom In: An Introduction to Circuits', 'A Mathematical Framework for Transformer Circuits', 'Toy Models of Superposition') which directly analyzes the internal vector representations and attention/transformer mechanisms of language models — squarely in the transformer/embeddings lineage and original, field-defining work, not derivative commentary. He was a lead engineer/researcher on TensorFlow at Google and co-authored Anthropic's RLHF paper, showing hands-on system-building alongside research. OpenAlex confirms 34 works, 14,840 citations, h-index 20; his real profile shows the interpretability line running from 2015 (Google Brain, DeepDream/feature visualization era) through today. He co-founded Anthropic, making his industry impact directly downstream of his own technical research rather than business-only leadership.\n\nOlah's own foundational-lineage contribution to frontier models runs through alignment rather than architecture: he is a co-author of Anthropic's 'Training a Helpful and Harmless Assistant with RLHF' (2022), and RLHF/HHH training is a block today's frontier assistants (including Claude, which his own company ships) directly build on; his mechanistic-interpretability work ('A Mathematical Framework for Transformer Circuits', 'Toy Models of Superposition', induction heads) analyzes transformer internals but is downstream understanding, not a building block the models are trained on, which caps him below the transformer/word2vec/scaling-law authors. His language-modeling-specific record is roughly 2020–2026 (~5 years) — his earlier 2015–2019 output was vision/CNN interpretability (feature visualization, DeepDream, checkerboard artifacts), not LM — placing depth in the 3–8-year band. That LM record is essentially single-domain natural-language text (transformer interpretability and RLHF); there is no verifiable hands-on biological, financial, code or other-domain language-modeling work, so breadth is one domain. As a scientific/technical founder he is genuine but singular: one company (Anthropic, co-founded 2021, ~5 years) where he personally sets and executes the interpretability research direction and authors the core papers — one company, ~5 years, the 3–8-year single-company band.","evidence":[{"claim":"Coined 'mechanistic interpretability' and lead author on Anthropic's Circuits research (A Mathematical Framework for Transformer Circuits, In-context Learning and Induction Heads, Toy Models of Superposition)","source_url":"https://en.wikipedia.org/wiki/Mechanistic_interpretability","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Anthropic; previously machine learning researcher at Google Brain and OpenAI","source_url":"https://en.wikipedia.org/wiki/Chris_Olah","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback' (2022)","source_url":"https://doi.org/10.48550/arxiv.2204.05862","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of 'A Mathematical Framework for Transformer Circuits' (2021): QK/OV circuit decomposition, residual stream, induction heads","source_url":"https://transformer-circuits.pub/2021/framework/index.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (Anthropic): ~124,670 citations, h-index 55, i10-index 81; includes TensorFlow, Concrete Problems in AI Safety, Feature Visualization, Understanding LSTM Networks, Deconvolution and Checkerboard Artifacts","source_url":"https://scholar.google.com/citations?user=6dskOSUAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Studied mathematics at Toronto for one year before leaving at 18; Thiel Fellow 2012; Google Brain 2015-2018, co-founded Distill 2017, led OpenAI interpretability 2018-2020, co-founded Anthropic 2021","source_url":"https://en.wikipedia.org/wiki/Chris_Olah","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored the Distill circuits thread including 'Zoom In: An Introduction to Circuits' (2020) and 'Feature Visualization' (2017), peer-reviewed in Distill","source_url":"https://doi.org/10.23915/distill.00007","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author, 'A Mathematical Framework for Transformer Circuits' (2021) — transformer-internal interpretability, LM-specific work begins ~2020-2021","source_url":"https://transformer-circuits.pub/2021/framework/index.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Anthropic (2021) known for neural-network / mechanistic interpretability; prior work at Google Brain and OpenAI was vision-interpretability","source_url":"https://en.wikipedia.org/wiki/Chris_Olah","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Anthropic (2021), personally leading and authoring its mechanistic-interpretability research program","source_url":"https://en.wikipedia.org/wiki/Chris_Olah","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["No completed formal degree found in public record; self-taught path through Google Brain research"],"canonical_papers":["Zoom In: An Introduction to Circuits (2020)","A Mathematical Framework for Transformer Circuits (2021)","Toy Models of Superposition (2022)","Training a Helpful and Harmless Assistant with RLHF (2022)"],"first_verifiable_year":2016,"notable_systems":["Anthropic's mechanistic interpretability / Circuits research program","early contributions to TensorFlow"],"citations":14840,"h_index":20,"patents":0,"dossier_notes":"Dossier's OpenAlex earliest_year (2016, TensorFlow paper) understates his actual start in ML research (DeepDream / feature visualization work at Google Brain from around 2015); this is a minor dating gap, not a homonym issue. Wikipedia/Wikidata/OpenAlex identity all consistent and correctly matched.","years_language_modeling":5,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2020-2026","evidence":"transformer-circuits interpretability + RLHF/HHH co-authorship at Anthropic"}],"frontier_lineage":["Co-author of Anthropic's RLHF Helpful & Harmless assistant training (2022), an alignment/training method Claude-class models descend from","Mechanistic-interpretability framework for transformers (downstream analysis, not a training block)"],"technical_founder_roles":["Anthropic — co-founder, interpretability research lead — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":14,"lm_domain_depth":11,"lm_domain_breadth":5,"scientific_founder":12},"confidence":0.72,"duration_ms":50696},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":11,"lm_domain_breadth":5,"scientific_founder":11},"confidence":0.76,"duration_ms":52661}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Chris_Olah","wikipedia_title":"Chris Olah","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"illia-polosukhin","name":"Illia Polosukhin","title":"Co-Founder, NEAR Protocol / CEO, NEAR AI","company":"NEAR Protocol / NEAR AI","sector":"crypto","rank":27,"rubric_version":4,"dimensions":{"foundations":12,"vector_embeddings":14,"transformers_lm":18,"frontier_founder":18,"lm_domain_depth":11,"lm_domain_breadth":10,"hands_on_engineering":14,"industry_impact":15,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":69,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":69,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.88,"rationale":"Polosukhin holds a Master's in Applied Mathematics and Computer Science from Kharkiv Polytechnic Institute, and worked as a software/AI engineer at Google (2014-2017), including on the Search and Google Brain/Research teams. He is a confirmed co-author (8th of 8) on 'Attention Is All You Need' (Vaswani et al., 2017), the canonical paper that introduced the Transformer architecture underlying nearly all modern language models — direct, verifiable authorship in the exact lineage this rubric targets. He also co-authored pre-transformer QA/representation work (WikiReading 2016, Coarse-to-Fine QA 2017, Natural Questions 2019). Since 2017-2018 his technical output shifted from core ML research to blockchain infrastructure (NEAR Protocol); recent 'NEAR AI' work (2024-2025) touches decentralized AI agent infrastructure but is not core transformer/embedding research, so industry impact and foundations score below Sutskever-tier despite the canonical co-authorship.\n\nPolosukhin is a confirmed co-author of 'Attention Is All You Need' (2017), the transformer architecture that GPT/Claude/Gemini/Llama-class models are directly built on — the strongest possible frontier-lineage position, placing him squarely in the top frontier_founder band regardless of his 8th-of-8 author slot or the OpenAlex record's mis-dated 2025 re-index. His hands-on language-modeling record runs ~2014-2019 at Google Search and Brain/Research (QA systems, WikiReading 2016, Coarse-to-Fine QA 2017, the transformer, Natural Questions 2019), roughly 5-6 verifiable years of deep NL work, with later NEAR AI (2024+) being decentralized-agent infrastructure rather than core LM research. He has a verifiable record in two language-modeling domains: natural-language text (QA/transformer) and source code (Neural Program Search 2018, near.ai's original program-synthesis focus). He has operated as a technical co-founder of one entity (near.ai → NEAR Protocol → NEAR AI) since 2017, ~9 years, but that company's core is a proof-of-stake blockchain rather than language modeling, which caps the scientific_founder score below the tier reserved for founders of LM-core companies.","evidence":[{"claim":"Confirmed co-author of 'Attention Is All You Need' (arXiv:1706.03762), the original Transformer paper","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Worked as AI/software engineer at Google 2014-2017 (Search team, then Google Brain/Research), contributing to TensorFlow and QA systems","source_url":"https://en.wikipedia.org/wiki/Illia_Polosukhin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded NEAR Protocol with Alexander Skidanov in 2017/2018, originally as an AI/program-synthesis research effort (near.ai) before pivoting to a layer-1 blockchain","source_url":"https://en.wikipedia.org/wiki/NEAR_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Education: Master's in Applied Mathematics and Computer Science, Kharkiv Polytechnic Institute","source_url":"https://fourweekmba.com/illia-polosukhin/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Attention Is All You Need' (arXiv 1706.03762, 2017), listed as the eighth of eight authors alongside Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez and Kaiser","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (3SyxFIAAAAAJ), affiliation NEAR, research interests Deep Learning / Machine Learning / Program Synthesis: ~307,379 citations, h-index 41, i10-index 45; publications include WikiReading (2016), Coarse-to-Fine Question Answering for Long Documents (2017), Natural Questions (201","source_url":"https://scholar.google.com/citations?user=3SyxFIAAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ukrainian-born computer scientist; engineering manager at Google Research where he co-authored the 2017 transformer paper; co-founded NEAR with Alexander Skidanov, originally launched as Near.ai focused on AI and program synthesis before pivoting to blockchain","source_url":"https://en.wikipedia.org/wiki/Illia_Polosukhin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored post-Google AI research including 'Neural Program Search: Solving Programming Tasks from Description and Examples' (arXiv 1802.04335, 2018)","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NEAR is a public proof-of-stake blockchain founded in 2018 by Illia Polosukhin and Alexander Skidanov, mainnet 2020","source_url":"https://en.wikipedia.org/wiki/NEAR_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Confirmed co-author of 'Attention Is All You Need' (arXiv:1706.03762), the transformer paper underlying modern frontier LLMs","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Post-Google code/program-synthesis LM work: 'Neural Program Search: Solving Programming Tasks from Description and Examples' (arXiv 1802.04335, 2018); NEAR launched as Near.ai focused on AI/program synthesis","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded NEAR with Alexander Skidanov (2017/2018), technical founder of a proof-of-stake blockchain, mainnet 2020","source_url":"https://en.wikipedia.org/wiki/NEAR_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Attention Is All You Need' (arXiv:1706.03762, 2017), the original Transformer paper underlying modern frontier LLMs","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Post-Google program-synthesis (code) work: 'Neural Program Search: Solving Programming Tasks from Description and Examples' (arXiv:1802.04335, 2018)","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded NEAR (launched as Near.ai, AI/program-synthesis focus, before pivoting to a layer-1 blockchain) in 2018 with Alexander Skidanov; mainnet 2020","source_url":"https://en.wikipedia.org/wiki/NEAR_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Master's, Applied Mathematics and Computer Science, Kharkiv Polytechnic Institute (Ukraine)"],"canonical_papers":["Attention Is All You Need (2017)","Natural Questions: A Benchmark for Question Answering Research (2019)","WikiReading: A Novel Large-scale Language Understanding Task over Wikipedia (2016)"],"first_verifiable_year":2014,"notable_systems":["Transformer architecture (co-author)","Natural Questions dataset","NEAR Protocol","NEAR AI"],"citations":10103,"h_index":9,"patents":0,"dossier_notes":"Dossier's Wikipedia entry is for 'NEAR (blockchain platform)' rather than a personal biography page; OpenAlex canonical_flags correctly identifies the transformer_2017 paper but mis-dates it as 2025 (likely a re-indexed/updated record) rather than the true 2017 publication date — verified directly via arXiv. No homonym issue on the transformer paper itself; PubMed match (Polosukhin VV/IV, 2014, allergy research) is a different person and irrelevant to this profile.","years_language_modeling":6,"years_as_technical_founder":9,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2016-2019","evidence":"WikiReading, Coarse-to-Fine QA, Attention Is All You Need, Natural Questions at Google Research"},{"domain":"code","years":"2017-2018","evidence":"Neural Program Search (arXiv 1802.04335); near.ai original program-synthesis research effort"}],"frontier_lineage":["Transformer architecture (co-author, Attention Is All You Need 2017) — the direct architectural foundation of all frontier LLMs"],"technical_founder_roles":["NEAR (near.ai → NEAR Protocol → NEAR AI) — co-founder / technical lead — 2017-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":10,"scientific_founder":11},"confidence":0.82,"duration_ms":60629},{"pass":"pass_2","dimensions":{"frontier_founder":19,"lm_domain_depth":10,"lm_domain_breadth":9,"scientific_founder":12},"confidence":0.83,"duration_ms":64212}],"source":"seeded","profile_url":null,"wikipedia_title":"NEAR (blockchain platform)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"tom-brown","name":"Tom B. Brown","title":"Co-founder; Chief Compute Officer / pretraining lead","company":"Anthropic","sector":"general","rank":28,"rubric_version":4,"dimensions":{"foundations":10,"vector_embeddings":9,"transformers_lm":20,"frontier_founder":19,"lm_domain_depth":12,"lm_domain_breadth":5,"hands_on_engineering":19,"industry_impact":18,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":68,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":68,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.83,"rationale":"Brown is the verified first author of 'Language Models are Few-Shot Learners' (GPT-3, arXiv:2005.14165, submitted 28 May 2020) — first position on a 31-author training-run paper that ends with Sutskever and Amodei reflects principal responsibility for the run itself, which is the canonical anchor on the transformer/LM dimension. He is also an author of 'Scaling Laws for Neural Language Models' (2020) and, earlier at Google Brain, the verified first author of 'Adversarial Patch' (arXiv:1712.09665, with Dandelion Mane, Aurko Roy, Martin Abadi and Justin Gilmer), an optimization-based construction of physically realizable universal adversarial examples — real personal research, not only infrastructure work. Since co-founding Anthropic he appears on the RLHF line ('Training a Helpful and Harmless Assistant with RLHF', arXiv:2204.05862). His mathematical foundations, however, are the weakest verifiable part of his record and the dispute between the passes turns on it: reporting corroborated by his own account describes a B-minus in undergraduate linear algebra followed by roughly six months of self-study from Axler's 'Linear Algebra Done Right', Coursera and Kaggle before joining OpenAI, with no graduate degree and no authored work in linear algebra, optimization or statistical learning theory. Under the rubric's anchors that is below 'strong graduate training', so foundations sits at the top of the 8-12 band rather than in the PhD-level band. No embedding, contrastive or retrieval work exists under his name, so vector_embeddings is scored on implicit representation learning only.\n\nBrown is the verified first author of 'Language Models are Few-Shot Learners' (GPT-3, arXiv:2005.14165) and a co-author of 'Scaling Laws for Neural Language Models' — the in-context few-shot-learning result and the scaling-law framework that every GPT/Claude/Gemini/Llama-class frontier model built after 2020 directly descends from, putting him at the top of the frontier_founder lineage (a frontier report need not name him). His hands-on language-modeling record runs from GPT-3 at OpenAI (2019-2021) through Anthropic pretraining and the Claude line (2021-present), roughly 7 years — his Google Brain years (2016-2019, e.g. 'Adversarial Patch') were vision/adversarial-ML, not LM — placing lm_domain_depth in the 3-8-year band despite exceptional depth. His verifiable modeling work is confined to a single language-modeling domain, natural-language text (GPT-3, Claude); no biological, financial, code, or other domain LM work exists under his name, so breadth is a single-domain score. He is a co-founder of Anthropic (Jan 2021-present, ~5 years) operating as Chief Compute Officer leading the Core Resources/compute-and-pretraining org — a genuine technical/scientific founder role at one company in this field, matching the 3-8-year single-company band (his 2011 startup Grouper was outside this field).","evidence":[{"claim":"First author of 'Language Models are Few-Shot Learners' (GPT-3); verified author order begins Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan and ends Ilya Sutskever, Dario Amodei; submitted 28 May 2020","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Adversarial Patch' (2017) with Dandelion Mane, Aurko Roy, Martin Abadi and Justin Gilmer, Google Brain; implementation released in TensorFlow CleverHans","source_url":"https://arxiv.org/abs/1712.09665","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on 'Scaling Laws for Neural Language Models' (2020)","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on Anthropic's 'Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback' (2022)","source_url":"https://arxiv.org/abs/2204.05862","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No graduate ML training: received a B-minus in his linear algebra course, then self-studied for six months using Axler's 'Linear Algebra Done Right', Coursera and Kaggle before joining OpenAI via a Y Combinator connection; previously worked at the startup Grouper; co-founded Anthropic in 2021 with D","source_url":"https://ca.news.yahoo.com/anthropic-cofounder-tom-brown-networked-160107558.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Tom B. Brown is listed as first author of 'Language Models are Few-Shot Learners' (GPT-3 paper), 31 authors total including Kaplan, Sutskever, D. Amodei.","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Brown was engineering lead for GPT-3 at OpenAI and co-founded Anthropic in 2021 as Chief Compute Officer, leading the Core Resources (compute/infrastructure) team.","source_url":"https://x.com/ycombinator/status/1957815586744070653","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Brown's career timeline: Google Brain (2016-2019, research scientist) -> OpenAI (2019-Jan 2021, researcher/engineering lead on GPT-3) -> Anthropic (Jan 2021-present, co-founder).","source_url":"https://www.longtermwiki.com/wiki/E1258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Brown is a self-taught engineer without formal graduate ML training; studied computer science and cognitive science at MIT and reportedly got a B-minus in linear algebra before self-studying AI full time; previously co-founded startup Grouper (2011, YC-backed).","source_url":"https://ca.news.yahoo.com/anthropic-cofounder-tom-brown-networked-160107558.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Language Models are Few-Shot Learners' (GPT-3), submitted 28 May 2020","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Tom B. Brown is first author of 'Language Models are Few-Shot Learners' (GPT-3), submitted 28 May 2020 — the few-shot in-context-learning result frontier LMs descend from","source_url":"https://arxiv.org/abs/2005.14165","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Scaling Laws for Neural Language Models' (2020), the scaling framework underpinning frontier-model training","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Brown was engineering lead for GPT-3 at OpenAI (2019-2021) and co-founded Anthropic in Jan 2021 as Chief Compute Officer, leading the Core Resources (compute/infrastructure) team","source_url":"https://x.com/ycombinator/status/1957815586744070653","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career timeline Google Brain (2016-2019) -> OpenAI (2019-Jan 2021, GPT-3 lead) -> Anthropic (Jan 2021-present, co-founder); LM-specific work begins ~2019","source_url":"https://www.longtermwiki.com/wiki/E1258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on 'Scaling Laws for Neural Language Models' (2020), a foundational scaling result the frontier stack builds on","source_url":"https://arxiv.org/abs/2001.08361","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career timeline: Google Brain (2016-2019) -> OpenAI (2019-Jan 2021, GPT-3 lead) -> Anthropic (Jan 2021-present, co-founder)","source_url":"https://www.longtermwiki.com/wiki/E1258","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Computer science / cognitive science coursework, MIT; no graduate degree found; publicly reported and uncontested as largely self-taught in advanced ML"],"canonical_papers":["Language Models are Few-Shot Learners (GPT-3), 2020 — first author","Scaling Laws for Neural Language Models, 2020 — co-author","Adversarial Patch, 2017 — first author","Training a Helpful and Harmless Assistant with RLHF, 2022 — co-author"],"first_verifiable_year":2016,"notable_systems":["GPT-3 training run (OpenAI)","Anthropic pretraining / Claude model line","RLHF helpful-and-harmless assistant training"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Every programmatic block in the dossier is a homonym and none was used. OpenAlex A5066170084 (173 works, h-index 38, affiliations Karlsruhe Institute of Technology / FZI / Goethe University Frankfurt, topics 'Integrated Energy Systems Optimization', top work on cost-optimised renewable European energy systems and PyPSA-Eur) is Tom Brown the energy-systems researcher. Semantic Scholar 'Tom H. Brown' (8 papers, 725 citations, name_exact false) is a third person; Wikidata Q114052496 ('data scientist') is unconfirmed; the PubMed block is flagged homonym_risk on 7,954 hits. The timeline's first_verifiable_year 1990 / 36 years active belongs to the energy researcher — his real first verifiable year in this lineage is ~2016-2017 (Google Brain). Citation and h-index are left at 0 because no correctly disambiguated bibliometric profile exists for him (he has no Google Scholar profile), not because his work is uncited; GPT-3 alone carries 82,085 citations on Dario Amodei's verified profile.","adjudication_notes":["foundations (pass_1 8, pass_2 15, final 10): pass_1 was right and pass_2 over-credited. Pass_2 inferred graduate-level mathematical foundations from the Adversarial Patch optimization result; the deciding verified fact is his own account — a B-minus in undergraduate linear algebra and six months of self-study from Axler before joining OpenAI, with no graduate degree and no authored foundations paper. That is below the rubric's 13-17 'PhD-level work' band. Raised from pass_1's 8 to 10 because Adversarial Patch is a genuine first-authored optimization result.","vector_embeddings (pass_1 10, pass_2 14, final 9): both passes over-credited; neither cited a single embedding, contrastive or retrieval paper by him, and none exists. Under the rubric an uncorroborated implicit contribution scores as near-absent, so this sits in the 'senior engineering adjacent to the core' band at best.","transformers_lm (pass_1 19, pass_2 20, final 20): pass_2 right. Verified arXiv author order confirms first authorship of GPT-3, which is principal-builder status on the system the field builds on.","hands_on_engineering (pass_1 19, pass_2 19, final 19): agreed, no dispute; he led the GPT-3 training infrastructure and now Anthropic's compute org."],"years_language_modeling":7,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2019-2026","evidence":"GPT-3 first author at OpenAI; Anthropic pretraining and Claude model line as Chief Compute Officer"}],"frontier_lineage":["GPT-3 (Language Models are Few-Shot Learners) — few-shot in-context learning, the direct ancestor of modern LLMs","Scaling Laws for Neural Language Models — the scaling framework frontier training follows","GPT-3 training-run infrastructure and Anthropic pretraining/compute stack behind Claude"],"technical_founder_roles":["Anthropic — co-founder & Chief Compute Officer (Core Resources / compute & pretraining) — 2021-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":20,"lm_domain_depth":12,"lm_domain_breadth":5,"scientific_founder":12},"confidence":0.82,"duration_ms":57108},{"pass":"pass_2","dimensions":{"frontier_founder":18,"lm_domain_depth":12,"lm_domain_breadth":5,"scientific_founder":12},"confidence":0.83,"duration_ms":49141}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"edo-liberty","name":"Edo Liberty","title":"Founder & Chief Scientist (former CEO)","company":"Pinecone","sector":"general","rank":29,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":20,"transformers_lm":8,"frontier_founder":14,"lm_domain_depth":13,"lm_domain_breadth":4,"hands_on_engineering":16,"industry_impact":16,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":67,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":67,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.83,"rationale":"Liberty holds a PhD in Computer Science from Yale (postdoc in Yale's Program in Applied Mathematics) and a BSc in Physics/CS from Tel Aviv University, giving deep foundations in linear algebra and matrix methods. He authored canonical streaming/matrix-sketching work — 'Randomized algorithms for the low-rank approximation of matrices' (PNAS 2007) and 'Frequent Directions: Simple and Deterministic Matrix Sketching' (2016), both squarely in the mathematical foundations of vector/embedding representation and widely cited (PNAS paper alone has 575+ citations). He led Amazon AI Labs and Amazon SageMaker research before founding Pinecone in 2019, the company that established and now leads the vector-database category underlying modern embedding-based retrieval/RAG systems — direct hands-on engineering and industry impact centered on vector embeddings. His record in the specific attention/transformer/LM lineage is thinner; his contribution is concentrated upstream in the vector-representation and retrieval-infrastructure side of the field rather than authoring transformer/LM research himself.\n\nLiberty's lineage into frontier systems is real but sits on the retrieval/representation side rather than the attention→transformer core: his randomized low-rank / matrix-sketching and fast dimension-reduction math (PNAS 2007, Frequent Directions 2016) is foundational to the vector-embedding and dimensionality-reduction machinery that dense retrieval depends on, and Pinecone (2019) established the vector-database category that today's RAG/embedding pipelines around GPT/Claude/Gemini-class models build on — a documented retrieval component the frontier stack draws from, not a named training block like word2vec or the transformer. His verifiable language-modeling record is in vector-space/embedding search systems rather than neural/statistical LMs proper: custom vector-search work at Yahoo (~2010-2017), Amazon AI Labs/SageMaker, and Pinecone (2019-present), roughly 14 hands-on years centered on embedding-based text retrieval — deep on the vector-space side, thin on n-gram/seq2seq/transformer authorship. That work is essentially a single domain (natural-language / general-purpose text embeddings); the dossier shows no distinct biological, financial, code or chemistry language-modeling record, so cross-domain breadth is low. As Founder & Chief Scientist of Pinecone since 2019 (~7 years, one company) he personally sets and authors the core technical direction, a genuine scientific-founder role of one company in the 3-8-year band.","evidence":[{"claim":"PhD Computer Science, Yale University; postdoctoral fellow, Yale Program in Applied Mathematics; BSc Physics & CS, Tel Aviv University","source_url":"https://www.frederick.ai/blog/edo-liberty-pinecone","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Director of Research / Head of Amazon AI Labs and Senior Manager of Research for Amazon SageMaker before founding Pinecone","source_url":"https://www.linkedin.com/in/edoliberty/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Pinecone in 2019, the company credited with establishing the vector-database category, now used by over a million engineers","source_url":"https://edoliberty.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5042783525: 77 works, 2,986 citations, h-index 26; affiliations Yale (2007-2010), Yahoo (2010-2017), Amazon (2016-2020), Tel Aviv University (2003); topics are sparse/compressive sensing, stochastic gradient optimization and algorithms","source_url":"https://api.openalex.org/authors/A5042783525","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Randomized algorithms for the low-rank approximation of matrices', PNAS 2007, Liberty, Woolfe, Martinsson, Rokhlin, Tygert","source_url":"https://pubmed.ncbi.nlm.nih.gov/18056803/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Pinecone company page: Edo Liberty is Founder & Chief Scientist, company founded 2019; previously research director at AWS and at Yahoo, where he worked on custom vector search systems","source_url":"https://www.pinecone.io/company/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Fast Dimension Reduction Using Rademacher Series on Dual BCH Codes', Discrete & Computational Geometry 2008","source_url":"https://doi.org/10.1007/s00454-008-9110-x","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder & Chief Scientist of Pinecone (founded 2019), the company credited with establishing the vector-database category underlying embedding-based retrieval","source_url":"https://www.pinecone.io/company/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Randomized algorithms for the low-rank approximation of matrices' (PNAS 2007) and 'Frequent Directions: Simple and Deterministic Matrix Sketching' (SIAM 2016) — foundational dimension-reduction/sketching math for vector representations","source_url":"https://pubmed.ncbi.nlm.nih.gov/18056803/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Head of Amazon AI Labs and research lead for Amazon SageMaker, and vector-search systems work at Yahoo, before founding Pinecone","source_url":"https://www.linkedin.com/in/edoliberty/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5042783525: topics are sparse/compressive sensing, stochastic gradient optimization and algorithms; affiliations Yahoo (2010-2017), Amazon (2016-2020)","source_url":"https://api.openalex.org/authors/A5042783525","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder & Chief Scientist of Pinecone (founded 2019); previously research director at AWS and at Yahoo, where he worked on custom vector search systems","source_url":"https://www.pinecone.io/company/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fast dimension-reduction (fast Johnson-Lindenstrauss) via Rademacher series on dual BCH codes — foundational for ANN/embedding retrieval","source_url":"https://doi.org/10.1007/s00454-008-9110-x","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Randomized algorithms for the low-rank approximation of matrices, PNAS 2007 — canonical sketching math underlying dimensionality reduction","source_url":"https://pubmed.ncbi.nlm.nih.gov/18056803/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5042783525: affiliations Yahoo (2010-2017) and Amazon (2016-2020); topics sparse/compressive sensing, stochastic gradient optimization, ML algorithms — vector-space/retrieval systems work","source_url":"https://api.openalex.org/authors/A5042783525","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BSc Physics & Computer Science, Tel Aviv University","PhD Computer Science, Yale University"],"canonical_papers":["Randomized algorithms for the low-rank approximation of matrices (PNAS, 2007)","A fast randomized algorithm for the approximation of matrices (2007)","Frequent Directions: Simple and Deterministic Matrix Sketching (2016)"],"first_verifiable_year":2003,"notable_systems":["Pinecone vector database","Amazon SageMaker research","Amazon AI Labs"],"citations":4090,"h_index":29,"patents":0,"dossier_notes":"PubMed sample includes clearly unrelated homonym hits (breast biopsy, jugular vein thrombosis, 1998 Hebrew-language case reports) alongside the genuine 2007 PNAS matrix-approximation paper — the scorer should only count the matrix-sketching paper as verified. OpenAlex/Semantic Scholar identity match (name_exact, ambiguous:false) appears correct and is corroborated by independent web sources.","years_language_modeling":14,"years_as_technical_founder":7,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2010-2026","evidence":"vector-space/embedding text search systems at Yahoo, Amazon SageMaker, and Pinecone vector database (general-purpose dense retrieval over text embeddings)"}],"frontier_lineage":["Randomized SVD / fast matrix sketching / dimension-reduction math underpinning embedding and dense-retrieval representations","Pinecone vector database — canonical retrieval infrastructure the RAG/embedding layer of frontier LLM systems builds on"],"technical_founder_roles":["Pinecone — Founder & Chief Scientist — 2019-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":4,"scientific_founder":12},"confidence":0.7,"duration_ms":62228},{"pass":"pass_2","dimensions":{"frontier_founder":14,"lm_domain_depth":13,"lm_domain_breadth":5,"scientific_founder":11},"confidence":0.7,"duration_ms":65553}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"kasian-franks","name":"Kasian Franks","title":"Founder & CEO","company":"Vector Space Biosciences / Vectorspace AI (also Cymetica/EventTrader)","sector":"general","rank":30,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":16,"transformers_lm":6,"frontier_founder":14,"lm_domain_depth":18,"lm_domain_breadth":16,"hands_on_engineering":13,"industry_impact":10,"scientific_founder":17},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":67,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":67,"tier":"deep_practitioner","tier_label":"Deep Practitioner","confidence":0.8,"rationale":"Both passes correctly treated the self-authored profile page as untrusted and scored only corroborated items; they diverged because pass 1 was blocked by Google Patents rate-limiting and could verify only one patent, while pass 2 reached the full patent record. I re-verified each disputed item directly. The patents pass 1 could not confirm are real: US20030204496A1 'Inter-term relevance analysis for large libraries' (inventors Sandip Ray, Raf Podowski, Kasian Franks; assignee X-MINE Inc.; filed 29 April 2002) claims second-order term co-occurrence to surface relationships not explicitly stated in a corpus; US7987191B2 and its continuation US9026543B2 'System and method for generating a relationship network' (Franks, Myers, Podowski; assigned to the University of California; priority 6 June 2005) build variable-length data vectors from framed inter-term distance scores with direct and indirect relationships and thematic context filtering — a hand-built distributional-semantics engine filed eight years before word2vec, which the rubric explicitly treats as foundational lineage rather than dated. Further patents assigned to Intertrust (US9177044B2 on relationships extracted from human-generated lists, US9600533B2 on media matching, priority 2006-2007) show this was shipped product engineering across several companies over fifteen years, not a single filing. His one substantive paper is corroborated by Crossref: he is second author, between D.M. Blei and M.I. Jordan, on 'Statistical modeling of biomedical corpora' (BMC Bioinformatics 2006), applying Latent Dirichlet Allocation to biomedical text — real co-authorship inside the statistical-learning tradition, though a single paper. The LBNL tech-transfer honour for SeeqPod is corroborated by Berkeley Lab's own history site rather than by his page. What is absent is any transformer-era record: no paper, patent or public model on attention, pretraining, scaling or alignment, and academic citation is modest (54 citations, h-index 1). Pass 2's scores are closer to the evidence, but its foundations of 13 over-reads a patent portfolio plus one co-authored paper as graduate-level mathematics training, and there is no verified degree beyond an undergraduate one.\n\nFranks holds verifiable pre-word2vec vector-space / distributional-semantics patents that sit squarely in the lineage today's embedding-based language models descend from: US20030204496A1 'Inter-term relevance analysis for large libraries' (X-MINE, filed 29 Apr 2002, second-order term co-occurrence to surface unstated relationships) and US7987191B2 / US9026543B2 'System and method for generating a relationship network' (Regents of the University of California, priority 6 Jun 2005), whose abstract literally claims generating 'variable length data vectors to represent the relationships between the terms' — hand-built distributional semantics roughly a decade before word2vec, plus a Blei/Jordan LDA co-authorship (BMC Bioinformatics 2006); this is documented lineage rather than a named frontier block, so it anchors in the 13-17 pre-word2vec-precursor band. His language-modeling record is continuous and long: first verifiable year 2002 (X-MINE) through the 2005 UC patents, SeeqPod (2005-2009) and Vectorspace AI / Vector Space Biosciences (2016-present, molecular-sequence and financial-correlation models), ~20 years spanning the pre-word2vec era into the transformer era, still active. Distinct hands-on language-modeling domains with a verifiable record are biological (X-MINE gene/drug/disease vectors, BMC 2006, LBNL genomics — deep, multi-year), media/music (SeeqPod search & recommendation corroborated by LBNL tech-transfer, plus Intertrust media-matching patent US9600533B2), and n","evidence":[{"claim":"Inventor on US20030204496A1 'Inter-term relevance analysis for large libraries' (inventors Sandip Ray, Raf Podowski, Kasian Franks; assignee X-MINE Inc.; filed/priority 29 April 2002) — second-order term-proximity correlation to detect previously unidentified relationships in large text libraries; v","source_url":"https://patents.google.com/patent/US20030204496A1/en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Continuation US9026543B2 'System and method for generating a relationship network' (inventors Kasian Franks, Cornelia A. Myers, Raf M. Podowski; assignee University of California San Diego; priority 6 June 2005, granted 5 May 2015) — abstract states the system 'generates variable length data vectors","source_url":"https://patents.google.com/patent/US9026543B2/en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First-named inventor on US7987191B2 'System and method for generating a relationship network' (priority June 2005, granted 26 July 2011), assignee The Regents of the University of California","source_url":"https://patentimages.storage.googleapis.com/pdfs/US7987191.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Inventor on US9600533B2 'Matching and recommending relevant videos and media to individual search engine results' (inventors Kasian Franks, Raf Podowski; assignee Intertrust Technologies Corp; priority 8 November 2006, granted 21 March 2017)","source_url":"https://patents.google.com/patent/US9600533B2/en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Crossref record confirms author order D.M. Blei, K. Franks, M.I. Jordan, I.S. Mian on 'Statistical modeling of biomedical corpora: mining the Caenorhabditis Genetic Center Bibliography for genes related to life span', BMC Bioinformatics 2006 — he is second author between the author of Latent Dirichl","source_url":"https://api.crossref.org/works/10.1186/1471-2105-7-250","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Berkeley Lab's own history site records that Kasian Franks and colleagues created SeeqPod, 'a search engine technology company whose roots were in a patent at the Lab' — independent corroboration of the LBNL tech-transfer claim","source_url":"https://history.lbl.gov/Publications/today/2007/Dec/13-Thu/tech-transfer-jump.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named first inventor on US Patent 7,987,191 B2 'System and Method for Generating a Relationship Network' (filed Nov 2007, continuation of PCT filed Jun 2006 / provisional Jun 2005, granted Jul 26 2011), assignee The Regents of the University of California, co-inventors Cornelia A. Myers and Raf M. P","source_url":"https://patentimages.storage.googleapis.com/pdfs/US7987191.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author (with David Blei and Michael I. Jordan) of 'Statistical modeling of biomedical corpora: mining the Caenorhabditis Genetic Center Bibliography for genes related to life span', BMC Bioinformatics, 2006, applying Latent Dirichlet Allocation to biomedical text","source_url":"https://api.semanticscholar.org/graph/v1/paper/DOI:10.1186/1471-2105-7-250","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Recognized as a Lawrence Berkeley National Laboratory tech-transfer success story for SeeqPod at the 2007 Excellence in Technology Transfer Awards ceremony","source_url":"https://history.lbl.gov/Publications/today/2007/Dec/13-Thu/tech-transfer-jump.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Second author, after D.M. Blei and before M.I. Jordan, on 'Statistical modeling of biomedical corpora: mining the Caenorhabditis Genetic Center Bibliography for genes related to life span', BMC Bioinformatics 2006 - verified author order from the Crossref record","source_url":"https://api.crossref.org/works/10.1186/1471-2105-7-250","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Berkeley Lab history site records Franks and colleagues created SeeqPod, 'a search engine technology company whose roots were in a patent at the Lab' — independent corroboration of him as a repeat technical founder building on his own patents (scientific_founder)","source_url":"https://history.lbl.gov/Publications/today/2007/Dec/13-Thu/tech-transfer-jump.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author with David Blei and Michael I. Jordan on 'Statistical modeling of biomedical corpora... genes related to life span', BMC Bioinformatics 2006 — LDA applied to biomedical text, biological language-modeling domain depth","source_url":"https://api.crossref.org/works/10.1186/1471-2105-7-250","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author with D.M. Blei and M.I. Jordan on 'Statistical modeling of biomedical corpora...' (BMC Bioinformatics 2006), applying LDA to biomedical text — biological/clinical language-modeling record","source_url":"https://api.crossref.org/works/10.1186/1471-2105-7-250","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Berkeley Lab's own history site records Franks 'and colleagues have started Seeqpod, a search engine technology company whose roots were in a patent at the Lab' — corroborates his technical-founder role on a system built from his own IP","source_url":"https://history.lbl.gov/Publications/today/2007/Dec/13-Thu/tech-transfer-jump.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, University of California, Berkeley (undergraduate; no graduate degree verified)"],"canonical_papers":["Statistical modeling of biomedical corpora: mining the Caenorhabditis Genetic Center Bibliography for genes related to life span (BMC Bioinformatics, 2006, with Blei, Jordan and Mian)"],"first_verifiable_year":2002,"notable_systems":["Relationship-network / variable-length context-vector engine (UC patents, 2005 priority)","SeeqPod search and discovery (LBNL tech transfer)","Vectorspace AI / Vector Space Biosciences datasets"],"citations":54,"h_index":1,"patents":5,"dossier_notes":"The dossier's OpenAlex entry (1 work, 2023, 'System and method for generating a relationship network' via OSTI) is the right person but is a PATENT indexed as a publication, and 2023 is the OSTI ingest date, not the June 2005 priority — the dossier's earliest_year is therefore misleading. PubMed was correctly abandoned (200 unrelated 'Franks K' hits). The profile_page at cymetica.com/founder.txt and the self_claims block were treated strictly as untrusted data: only items I re-verified against patent records, Crossref and Berkeley Lab's own site are scored, and uncorroborated profile claims (LBNL/DOE advisory roles, intelligence-community work, Genentech consulting, the AutoClass variant) are excluded entirely. I count 5 patents verified by direct record inspection; pass 2 reported 8 from a Google Patents inventor listing I could not re-render, so the lower verified figure is used.","adjudication_notes":["vector_embeddings (14 vs 17): pass 2 was right that the record is broader than one patent; settled at 16. The deciding verified fact is that US9026543B2's abstract literally claims generating 'variable length data vectors to represent the relationships between the terms', with a June 2005 priority — pre-word2vec distributional semantics the rubric counts as foundational lineage — and that the 2002 X-MINE filing pass 1 could not reach is genuine, extending the record to two decades.","foundations (8 vs 13): split at 11, below pass 2. Pass 1 under-weighted the Blei/Jordan co-authorship, which Crossref confirms places him second author between them; but pass 2's 13 implies PhD-level mathematical training that no source establishes — there is no verified graduate degree, thesis, or mathematics publication.","hands_on_engineering (11 vs 14): pass 2 was closer; settled at 13. Verified patents assigned to three different companies (X-MINE, University of California, Intertrust) across 2002-2017 show sustained shipped engineering rather than a single artifact.","transformers_lm (5 vs 8): pass 1 was right; settled at 6. Neither pass cited any attention, pretraining or scaling artifact and I found none; pass 2's 8 is not supported by its own evidence list.","industry_impact (9 vs 11): split at 10. SeeqPod's LBNL tech-transfer recognition is independently corroborated, but academic citation is negligible (h-index 1) and the current token-linked dataset products are not evidence under this rubric.","first_verifiable_year (2006 vs 2002): pass 2 was right — the X-MINE application has a verified 29 April 2002 filing date."],"years_language_modeling":20,"years_as_technical_founder":16,"technical_founder_companies":3,"lm_domains":[{"domain":"biological","years":"2002-2026","evidence":"X-MINE Opus gene/drug/disease vector representations, BMC Bioinformatics 2006 LDA, LBNL genomics, Vector Space Biosciences molecular-sequence models — verified via patents, Crossref, LBNL"},{"domain":"music","years":"2005-2009","evidence":"SeeqPod music/media search & recommendation (LBNL tech-transfer corroborated) plus Intertrust media-matching patent US9600533B2"},{"domain":"natural_language","years":"2002-2009","evidence":"X-MINE LSI/NLP term-relationship and inference/Q&A text mining; UC relationship-network patent operates over corpus terms"},{"domain":"financial","years":"2016-2026","evidence":"Vectorspace AI financial-market correlation datasets and Cymetica/EventTrader — lighter, less independently verified than the others"}],"frontier_lineage":["Pre-word2vec (2002/2005) term-relationship 'variable length data vector' patents in the distributional-semantics lineage that word2vec-era embeddings descend from","LSI / inter-term co-occurrence relevance analysis (X-MINE 2002)","LDA applied to text corpora (BMC 2006, with Blei/Jordan)"],"technical_founder_roles":["X-MINE Inc. — Director, Genomic Search / creator of Opus, first-named vector-space patent — 2001-2002","SeeqPod Inc. — Founder, CEO, CVO (core tech from his LBNL patents) — 2005-2009","Vector Space Biosciences / Vectorspace AI — CEO, Co-Founder — 2016-present","Cymetica (EventTrader) — Founder, CEO — present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":14,"lm_domain_depth":18,"lm_domain_breadth":16,"scientific_founder":18},"confidence":0.6,"duration_ms":143378},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":18,"lm_domain_breadth":16,"scientific_founder":16},"confidence":0.6,"duration_ms":90427}],"source":"community","profile_url":"https://cymetica.com/founder.txt","wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"david-luan","name":"David Luan","title":"Co-founder & CEO","company":"Adept AI","sector":"general","rank":31,"rubric_version":4,"dimensions":{"foundations":10,"vector_embeddings":9,"transformers_lm":16,"frontier_founder":15,"lm_domain_depth":14,"lm_domain_breadth":10,"hands_on_engineering":16,"industry_impact":15,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":64,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":64,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.78,"rationale":"Luan holds a BS in Applied Mathematics & Political Science from Yale (not a graduate research degree), so foundations is scored as strong undergraduate quantitative training rather than PhD-level theory. As VP of Engineering at OpenAI (2017-2020) he led the teams that shipped GPT-2, GPT-3, CLIP and DALL-E, and he is a listed co-author on 'PaLM: Scaling Language Modeling with Pathways' (2022, 2,136+ citations) and 'Generative Pretraining From Pixels' (iGPT, 2020) — both are substantive transformer/LM-lineage contributions with organizational leadership, not just authorship credit. He subsequently led Google Brain's large-model effort before co-founding Adept AI (with Ashish Vaswani and Niki Parmar, co-authors of the original Transformer paper) to build agentic AI systems, so his industry impact is a direct extension of hands-on LM engineering. His personal-paper record is thin (OpenAlex lists only 2 works under his name, reflecting an engineering-leadership rather than first-author-research career), which caps foundations and vector_embeddings.\n\nLuan's frontier-lineage position is real but as an engineering leader rather than a named-block author: he ran the OpenAI engineering org (VP Eng, 2017–2020) that shipped GPT-2 and GPT-3 — systems today's frontier models directly descend from — and is a listed co-author on PaLM (2022, ~2,136 citations), a canonical dense-transformer scaling study, and on Generative Pretraining from Pixels (iGPT, ICML 2020). His hands-on language-modeling record runs continuously from joining OpenAI in December 2017 through Google Brain's large-model effort and co-founding Adept, ~8–9 years, all in the transformer/LLM era. His distinct language-modeling domains are natural-language text (GPT-2/3, PaLM) and autoregressive image/pixel sequence modeling (iGPT) — two domains with a verifiable record, none biological or financial. He is a technical co-founder of exactly one company (Adept AI, ~2022–2024) but co-founded it alongside the actual Transformer scientists (Vaswani, Parmar) who supplied much of the core science, and his tenure was short, so scientific_founder sits at the low end of the one-company band.","evidence":[{"claim":"BS Applied Mathematics & Political Science, Yale University (2009-2013)","source_url":"https://news.ycombinator.com/item?id=37415797","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (listed as VP Engineering, OpenAI): ~39,503 citations, h-index 8; papers include GPT-2 'Language Models are Unsupervised Multitask Learners', PaLM, Generative Pretraining from Pixels, and Scratchpads","source_url":"https://scholar.google.com/citations?user=cItVg2MAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata identifies David Luan as an AI researcher and VP of engineering at OpenAI, linking Google Scholar id cItVg2MAAAAJ","source_url":"https://www.wikidata.org/wiki/Q115923706","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'PaLM: Scaling Language Modeling with Pathways' (2022), a 540B-parameter dense transformer scaling study","source_url":"https://arxiv.org/abs/2204.02311","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Generative Pretraining from Pixels' (Image GPT), ICML 2020, applying autoregressive transformer pretraining to images","source_url":"https://proceedings.mlr.press/v119/chen20s.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Generative Pretraining from Pixels' (Image GPT), ICML 2020 — autoregressive transformer pretraining on image pixel sequences (a distinct sequence-modeling domain)","source_url":"https://proceedings.mlr.press/v119/chen20s.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (VP Engineering, OpenAI): GPT-2 'Language Models are Unsupervised Multitask Learners', PaLM, iGPT, Scratchpads — continuous LM-era record","source_url":"https://scholar.google.com/citations?user=cItVg2MAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Adept AI (2022) as CEO with Transformer authors Ashish Vaswani and Niki Parmar to build agentic AI (ACT-1)","source_url":"https://news.ycombinator.com/item?id=37415797","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Yale University, BS Applied Mathematics & Political Science (2013)"],"canonical_papers":["PaLM: Scaling Language Modeling with Pathways (2022)","Generative Pretraining From Pixels (2020)"],"first_verifiable_year":2017,"notable_systems":["GPT-2/GPT-3 engineering org at OpenAI","Google Brain large-model effort","Adept AI (Adept / ACT-1 agentic systems)"],"citations":2704,"h_index":2,"patents":0,"dossier_notes":"Dossier's OpenAlex match flags ambiguous=true with 10 candidates; the works listed (PaLM, iGPT) are consistent with the correct David Luan (OpenAI/Adept), confirmed against independent sources. No Wikipedia page exists for him; Wikidata Q115923706 correctly identifies him as OpenAI VP of engineering. His first_verifiable_year in the dossier (2020, from OpenAlex) understates his actual start in the field — he joined OpenAI in December 2017, three years earlier than the dossier's earliest_year.","years_language_modeling":9,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2017-2026","evidence":"GPT-2/GPT-3 engineering org at OpenAI; PaLM co-author; Adept agentic LM"},{"domain":"other","years":"2020","evidence":"Generative Pretraining from Pixels (iGPT) — autoregressive image/pixel sequence modeling, ICML 2020"}],"frontier_lineage":["Led OpenAI engineering org that produced GPT-2 and GPT-3","Co-author of PaLM dense-transformer scaling study","Co-author of iGPT autoregressive pixel pretraining"],"technical_founder_roles":["Adept AI — co-founder & CEO — 2022-2024"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":14,"lm_domain_breadth":10,"scientific_founder":9},"confidence":0.6,"duration_ms":64448},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":9,"scientific_founder":11},"confidence":0.68,"duration_ms":55689}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"fei-fei-li","name":"Fei-Fei Li","title":"Co-founder & CEO, World Labs; Sequoia Professor of Computer Science, Stanford University","company":"World Labs","sector":"general","rank":32,"rubric_version":4,"dimensions":{"foundations":17,"vector_embeddings":15,"transformers_lm":9,"frontier_founder":14,"lm_domain_depth":8,"lm_domain_breadth":5,"hands_on_engineering":16,"industry_impact":19,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":61,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":61,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.9,"rationale":"Li holds a Caltech PhD (2005) under Pietro Perona and Christof Koch following a Princeton physics degree, and her early record is Bayesian statistical learning rather than applied tooling: 'A Bayesian Hierarchical Model for Learning Natural Scene Categories' (CVPR 2005, Fei-Fei Li and P. Perona, verified via Crossref, 5,247 citations) and 'Learning Generative Visual Models from Few Training Examples' (2004, 7,047 citations) are generative-model and one-shot-learning mathematics — that is the deciding evidence on the foundations dispute, and pass_1's 15 undervalued it. She created and led ImageNet (102,864 citations) and ILSVRC (56,952), supplying the data and benchmark the deep-learning era was built on, and Visual Genome grounds language to image regions. Her representation-learning work is genuine but visual: 'Deep Visual-Semantic Alignments for Generating Image Descriptions' (Karpathy and Fei-Fei, 2015, 8,318 citations) aligns CNN region features with bidirectional-RNN sentence encodings through a multimodal embedding. Her verified Google Scholar profile shows 372,989 citations, h-index 181 and i10-index 467 — among the largest verified records in the field. Transformers/LM is her weakest dimension by a wide margin: she is a co-author of the 2021 foundation-models position paper, but that is a multi-author survey, and she has authored no attention, pretraining, scaling or alignment result — her lineage is vision and spatial intelligence, and the rubric measures the language-modeling core specifically. She now leads World Labs building generative 3D world models (Marble), an organization whose technical core she leads rather than funds.\n\nLi's foundational position in the frontier stack is real but sits in the VISION branch: ImageNet (2009, ~63k–102k citations) and ILSVRC are the canonical large-scale dataset/benchmark that catalyzed the deep-learning era, and today's multimodal frontier models (vision towers, image-text grounding) descend from that lineage and from her vision-language grounding work (Visual Genome, Deep Visual-Semantic Alignments with Karpathy, 2014-2017) — datasets/benchmarks are named frontier blocks, so 14, but she authored no attention, tokenizer, pretraining, scaling or alignment result in the language line. Her language-modeling record is adjacent rather than core: image-captioning / vision-language sequence models (CNN + bidirectional RNN) from roughly 2014-2017 touch natural-language generation but are not neural/statistical text language modeling, and she has no pre-2013 vector-space/LSI text lineage — hence ~5 NL-adjacent years and depth of 7 (adjacent). The distinct language-modeling domains reduce to a single, thin one — natural-language text via visual grounding — with no biological, financial, code or other sequence-model record, so breadth is 5. As a scientific founder she is the genuine technical/scientific co-founder of ONE company, World Labs (founded Feb 2024, building generative 3D world models Marble/Atlas), where she sets and executes the technical direction — but that is only ~2.5 years and one company, below the 3-8-year single-company band, so 8; SAIL and Google Cloud","evidence":[{"claim":"Google Scholar rDfyQnIAAAAJ (Professor of Computer Science, Stanford): 372,989 citations, h-index 181, i10-index 467; top works ImageNet (102,864), ILSVRC (56,952), Perceptual Losses (15,448), foundation models report (12,379), Deep Visual-Semantic Alignments (8,318), Visual Genome (8,062), Learning","source_url":"https://scholar.google.com/citations?user=rDfyQnIAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author with Pietro Perona of 'A Bayesian Hierarchical Model for Learning Natural Scene Categories', CVPR 2005, pp. 524-531 — generative Bayesian statistical-learning work","source_url":"https://api.crossref.org/works/10.1109/cvpr.2005.16","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author with Andrej Karpathy of 'Deep Visual-Semantic Alignments for Generating Image Descriptions' (2014/2015), aligning CNN image-region features with bidirectional-RNN sentence representations via a multimodal embedding","source_url":"https://arxiv.org/abs/1412.2306","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Caltech 2005 (advisors Pietro Perona and Christof Koch), BA physics Princeton 1999; established ImageNet; SAIL director 2013-2018; Chief Scientist of AI/ML at Google Cloud; co-founded World Labs 2024","source_url":"https://en.wikipedia.org/wiki/Fei-Fei_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"World Labs co-founded by Fei-Fei Li with Justin Johnson, Ben Mildenhall and Christoph Lassner, building generative 3D world models (Marble)","source_url":"https://www.worldlabs.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ImageNet (2009) has ~102,864 citations per Google Scholar and is Li's most-cited canonical work founding large-scale visual recognition datasets/benchmarks","source_url":"https://scholar.google.com/citations?user=rDfyQnIAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li's Wikipedia entry confirms she established ImageNet, is a Stanford CS professor, and is known as a founder of modern computer vision benchmarking","source_url":"https://en.wikipedia.org/wiki/Fei-Fei_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from Caltech (2005) with doctoral advisors Pietro Perona and Christof Koch, BA physics from Princeton (1999)","source_url":"https://www.wikidata.org/wiki/Q18686107","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"World Labs, founded by Li in Feb 2024 with Justin Johnson, Ben Mildenhall and Christoph Lassner (ML/graphics/vision researchers), raised $1B and shipped the Marble and Atlas world-model products","source_url":"https://www.worldlabs.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ImageNet: A Large-Scale Hierarchical Image Database (CVPR 2009), the canonical large-scale visual dataset/benchmark that catalyzed deep learning and underlies frontier multimodal vision","source_url":"https://doi.org/10.1109/cvpr.2009.5206848","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Deep Visual-Semantic Alignments for Generating Image Descriptions (Karpathy & Fei-Fei, 2014/2015) — vision-language sequence modeling aligning CNN region features with a bidirectional-RNN sentence encoder, her closest natural-language-adjacent work","source_url":"https://arxiv.org/abs/1412.2306","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations (IJCV 2017), grounding language to image regions","source_url":"https://doi.org/10.1007/s11263-016-0981-7","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"World Labs co-founded Feb 2024 by Fei-Fei Li (with Justin Johnson, Ben Mildenhall, Christoph Lassner) building generative 3D world models (Marble); she leads the technical/scientific direction","source_url":"https://www.worldlabs.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ImageNet: A Large-Scale Hierarchical Image Database (CVPR 2009) — the canonical dataset/benchmark that catalyzed the deep-learning scaling era; ~63k-102k citations","source_url":"https://doi.org/10.1109/cvpr.2009.5206848","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"World Labs co-founded by Fei-Fei Li (Feb 2024) with Justin Johnson, Ben Mildenhall and Christoph Lassner, building generative 3D world models (Marble); Li leads the technical/scientific direction","source_url":"https://www.worldlabs.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BA physics, Princeton University, 1999","MS, Caltech, 2001","PhD, Caltech, 2005 (advisors Pietro Perona, Christof Koch)"],"canonical_papers":["ImageNet: A Large-Scale Hierarchical Image Database (CVPR 2009)","ImageNet Large Scale Visual Recognition Challenge (IJCV 2015)","Learning Generative Visual Models from Few Training Examples (2004)","A Bayesian Hierarchical Model for Learning Natural Scene Categories (CVPR 2005)","Deep Visual-Semantic Alignments for Generating Image Descriptions (2015)","Visual Genome (IJCV 2017)"],"first_verifiable_year":2003,"notable_systems":["ImageNet / ILSVRC","Visual Genome","Stanford Vision & Learning Lab research stack","World Labs Marble"],"citations":372989,"h_index":181,"patents":0,"dossier_notes":"The dossier's PubMed block is worthless: the query '(Li F[Author]) AND (Stanford OR Illinois ...)' matched 48,951 records and the samples are colorectal-cancer chemistry, dairy-cow nutrition, methane observation and neuroimmune pharmacology — all different people named Li F. OpenAlex A5100450462 is the right works list but carries plainly wrong affiliations (Harbin University of Science and Technology, Jiangxi University of Traditional Chinese Medicine, Beijing Union University) and undercounts her (224,089 citations / h-index 137 against the verified Scholar record of 372,989 / 181). The Semantic Scholar match (id 2146326571, 10 papers, 110 citations) is a different person. She remains a Stanford professor alongside World Labs. Patents were skipped, not searched.","adjudication_notes":["foundations (pass_1 15, pass_2 18, final 17): pass_2 was substantially right. Pass_1 treated her foundations as generic PhD training; the deciding verified fact is the Crossref record for 'A Bayesian Hierarchical Model for Learning Natural Scene Categories' (Fei-Fei Li and Perona, CVPR 2005) plus the 2004 one-shot generative-model work — authored Bayesian statistical learning, the mathematics the rubric names. Set at 17 rather than 18 because her mathematical contributions are applied generative-vision models rather than results in optimization or learning theory itself.","transformers_lm (pass_1 6, pass_2 12, final 9): pass_1 was closer to right. Pass_2's only cited support is the multi-author foundation-models position paper, which is a survey, not authored attention, pretraining, scaling or alignment work; the rubric measures the language-modeling core specifically and warns against crediting fame. Raised from 6 to 9 only for the verified vision-language sequence-modeling content of Deep Visual-Semantic Alignments (CNN + bidirectional RNN).","vector_embeddings (pass_1 15, pass_2 16, final 15): effectively agreed; credit rests on the verified multimodal-embedding alignment work and Visual Genome, which is real representation learning but visual rather than lexical.","industry_impact (pass_1 19, pass_2 20, final 19): pass_1 marginally right. Verified 372,989 citations and ImageNet's role are canonical, but the rubric's top anchor is reserved for systems whose core is the embedding/transformer/LM lineage; ImageNet catalysed deep learning through vision data and benchmarking."],"years_language_modeling":5,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2014-2017","evidence":"image-captioning / vision-language grounding (Deep Visual-Semantic Alignments, Visual Genome) — NL generation via visual grounding, not text LM"}],"frontier_lineage":["ImageNet dataset/benchmark — foundation of the vision branch of frontier multimodal models","vision-language grounding (Visual Genome, image captioning) — precursor to multimodal image-text alignment"],"technical_founder_roles":["World Labs — co-founder & scientific/technical lead — 2024-present (~2.5y)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":14,"lm_domain_depth":7,"lm_domain_breadth":5,"scientific_founder":8},"confidence":0.82,"duration_ms":72781},{"pass":"pass_2","dimensions":{"frontier_founder":15,"lm_domain_depth":8,"lm_domain_breadth":5,"scientific_founder":9},"confidence":0.8,"duration_ms":64428}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Fei-Fei_Li","wikipedia_title":"Fei-Fei Li","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"liang-wenfeng","name":"Liang Wenfeng","title":"Founder & CEO, DeepSeek; co-founder, High-Flyer","company":"DeepSeek","sector":"general","rank":33,"rubric_version":4,"dimensions":{"foundations":10,"vector_embeddings":7,"transformers_lm":13,"frontier_founder":13,"lm_domain_depth":9,"lm_domain_breadth":12,"hands_on_engineering":16,"industry_impact":18,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":61,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":61,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.83,"rationale":"Liang holds a bachelor's (2006/07) and master's (2010) in electronic information / information and communication engineering from Zhejiang University, with a master's thesis on object tracking from a low-cost PTZ camera — real signal-processing training, but not graduate work in learning theory, embeddings or language modeling, and no authored mathematics. The fact that decides the hands_on_engineering and transformers_lm dispute is one pass_1 missed: he is a named author (position 17 of 52) and the submitting contact of 'Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning' (arXiv:2408.14158), a genuine systems paper on the 10,000-PCIe-A100 cluster, HFReduce allreduce acceleration, HaiScale, 3FS and a congestion-free computation-storage integrated network achieving DGX-A100-class performance at roughly half the cost and 40% less energy. That is a documented engineering artifact he drove, not a leadership credit, and it is the kind of training-infrastructure work the rubric credits under hands_on_engineering. He is likewise a listed author and corresponding contact on the DeepSeek-V3 Technical Report (671B MoE with Multi-head Latent Attention and multi-token prediction) and DeepSeek-R1, which showed reasoning can be elicited by pure RL. Scores stay moderate on the research dimensions because these are very large corporate reports in which his individual contribution is not separable, he has no independent authored work in attention, embeddings or scaling, and no correctly-disambiguated citation record exists. Industry impact is high on the verifiable ground that DeepSeek-V2/V3/R1 are open models the field demonstrably builds on.\n\nLiang has no verifiable pre-2013 vector-space/distributional-semantics lineage work and no independently authored method, so he is not a foundational-block author; his frontier position rests on being the founder-CEO who set and drove the technical direction of DeepSeek, whose V3 architecture innovations (Multi-head Latent Attention, multi-token prediction, MoE efficiency) and the R1 pure-RL reasoning recipe are demonstrably built on by other frontier efforts today — a genuine but founder-level (not personal-authorship) lineage position, hence 14. His hands-on language-modeling record is recent and short: High-Flyer's deep-learning-for-trading and AI-HPC build from ~2019 and DeepSeek's LLM work from 2023, roughly 6-7 verifiable years, placing lm_domain_depth in the 3-8-year band. Across domains he has a strong multi-year financial-modeling record at High-Flyer (ML quant trading, ~2015-2023), natural-language modeling at DeepSeek (V2/V3/R1, 2023+), and code modeling (DeepSeek-Coder) — three domains, though only two with deep personal depth. He has operated as a technical/scientific founder-CEO of two companies (High-Flyer 2015/2016, DeepSeek 2023) over ~11 years, personally setting technical direction and listed as author/corresponding contact on the core reports, which places scientific_founder in the two-companies / 8-15-year band.","evidence":[{"claim":"Named author (17th of 52: 'Wenfeng Liang') and submitting contact of 'Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning' (submitted 26 August 2024) — Fire-Flyer 2 with 10,000 PCIe A100 GPUs, HFReduce, HaiScale, 3FS, HAI-Platform, DGX-A100-class performance at half the","source_url":"https://arxiv.org/abs/2408.14158","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author and corresponding contact on the DeepSeek-V3 Technical Report (671B-parameter MoE, 37B active, Multi-head Latent Attention, multi-token prediction, 14.8T tokens)","source_url":"https://arxiv.org/abs/2412.19437","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author on 'DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning'","source_url":"https://arxiv.org/abs/2501.12948","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BEng and MEng from Zhejiang University (master's thesis on object tracking with a low-cost PTZ camera); co-founded High-Flyer, a quantitative hedge fund applying machine learning to trading, and founded DeepSeek in 2023","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepSeek released the open-weight DeepSeek-V2, V3 and R1 models","source_url":"https://en.wikipedia.org/wiki/DeepSeek","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bachelor's degree (2006) and Master's degree (2010) in Information and Communication Engineering, Zhejiang University; master's thesis on target-tracking with PTZ cameras.","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded High-Flyer Capital Management in 2015, a quantitative hedge fund that applied machine learning to trading and built large-scale GPU compute clusters.","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder and CEO of DeepSeek (founded 2023), which released DeepSeek-V2, V3 and R1, open-weight models that had significant technical and market impact.","source_url":"https://en.wikipedia.org/wiki/DeepSeek","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as a contact author on the DeepSeek-V3 Technical Report (arXiv 2412.19437), a ~200-author paper credited to \"DeepSeek-AI\".","source_url":"https://arxiv.org/abs/2412.19437","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BEng Electronic Information Engineering (2007) and MEng Information & Communication Engineering (2010), Zhejiang University; master's thesis on object tracking with a low-cost PTZ camera; co-founded High-Flyer 2016, High-Flyer AI 2019, DeepSeek May 2023; began acquiring thousands of Nvidia GPUs in 2","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder and CEO of DeepSeek (2023); co-founded the quantitative hedge fund High-Flyer, which applied machine learning to trading and built large GPU compute clusters","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepSeek-V3 Technical Report — 671B MoE with Multi-head Latent Attention and multi-token prediction; Liang listed as author/corresponding contact — architecture the field now builds on","source_url":"https://arxiv.org/abs/2412.19437","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepSeek-R1: reasoning elicited via pure reinforcement learning — a recipe widely adopted across frontier reasoning models","source_url":"https://arxiv.org/abs/2501.12948","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DeepSeek-Coder — a distinct code language-modeling line shipped by DeepSeek","source_url":"https://arxiv.org/abs/2401.02954","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fire-Flyer AI-HPC — Liang named author (17th) and submitting contact; documents the 10,000-A100 cluster, HFReduce, 3FS training stack","source_url":"https://arxiv.org/abs/2408.14158","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named author (position 17 of 52, 'Wenfeng Liang') and submitting contact of 'Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning', documenting the 10,000-GPU Fire-Flyer 2 cluster, HFReduce and 3FS","source_url":"https://arxiv.org/abs/2408.14158","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed author / corresponding contact on the DeepSeek-V3 Technical Report (671B MoE, Multi-head Latent Attention, multi-token prediction) — a foundation the frontier ecosystem draws on","source_url":"https://arxiv.org/abs/2412.19437","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of quantitative hedge fund High-Flyer (ML-driven trading, large GPU clusters) and founder/CEO of DeepSeek (2023), the two companies where he sets technical direction","source_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BEng Electronic Information Engineering, Zhejiang University","MEng Information & Communication Engineering, Zhejiang University (2010)"],"canonical_papers":["Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning (arXiv:2408.14158) — named author and submitting contact","DeepSeek-V3 Technical Report (arXiv:2412.19437) — listed author / corresponding contact","DeepSeek-R1 (arXiv:2501.12948) — listed author"],"first_verifiable_year":2010,"notable_systems":["Fire-Flyer / Fire-Flyer 2 GPU cluster","HFReduce","HaiScale","3FS","High-Flyer quantitative trading stack","DeepSeek-V2/V3/R1"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"OpenAlex A5003469271 is the WRONG PERSON: its works are 'The Art of Linux Kernel Design' and its chapters (Auerbach Publications, 2014-2018), with topics Business Process Modeling, Embedded Systems and Legal and Regulatory Analysis; its 14 works / 1 citation / h-index 1 and the derived earliest_year 2011 must all be discarded. The Semantic Scholar match resolved to 'W. Liang' (name_exact false, 8 candidates), so its 13,772 citations / h-index 14 cannot be attributed to him. The PubMed block (honeybee symbiosis, fowl adenovirus, lung cancer) is unrelated homonyms. Wikidata's educated_at field is malformed, leaking raw qualifier codes. Citation and h-index are left at 0 because no correctly-disambiguated profile exists, not because the record is empty. No evidence of family or inherited wealth funding DeepSeek was found — High-Flyer was built from trading returns — so capital_without_competence is 0 rather than asserted.","adjudication_notes":["hands_on_engineering (pass_1 9, pass_2 18, final 16): pass_2 was right that pass_1 missed decisive evidence — the verified arXiv author list for Fire-Flyer AI-HPC contains 'Wenfeng Liang' at position 17 and names him the submitting contact, documenting the 10,000-GPU cluster, HFReduce and 3FS as an engineering artifact rather than a management claim. Set at 16 rather than 18 because he is one of 52 authors and his individual contribution within that stack is not separable.","transformers_lm (pass_1 10, pass_2 16, final 13): split. Authorship on the V3 and R1 reports is verified, but both are ~200-author corporate reports, Multi-head Latent Attention is publicly attributed to a junior researcher's initiative rather than to him, and he has no independent authored modeling work — that supports credit above pass_1's 10 but well below pass_2's 16.","foundations (pass_1 9, pass_2 13, final 10): pass_1 was right. A master's in information and communication engineering with a PTZ-camera tracking thesis is real signal-processing training but not the graduate mathematics band; no authored work in linear algebra, optimization or statistical learning exists.","vector_embeddings (pass_1 6, pass_2 9, final 7): pass_1 closer; no embedding, retrieval or representation-learning work of any kind is verifiable under his name.","industry_impact (pass_1 17, pass_2 18, final 18): effectively agreed — DeepSeek-V3/R1 are open models with demonstrable technical influence on the field, which is impact whose core is these systems."],"years_language_modeling":7,"years_as_technical_founder":11,"technical_founder_companies":2,"lm_domains":[{"domain":"financial","years":"2015-2023","evidence":"High-Flyer ML/quant trading models on market time-series, technically driven as co-founder"},{"domain":"natural_language","years":"2023-2026","evidence":"DeepSeek-V2/V3/R1 LLMs, listed author/corresponding contact on technical reports"},{"domain":"code","years":"2023-2026","evidence":"DeepSeek-Coder and code capabilities of V2/V3, shipped by DeepSeek under his direction"}],"frontier_lineage":["DeepSeek-V3 Multi-head Latent Attention + multi-token prediction (efficiency methods adopted downstream)","DeepSeek-R1 pure-RL reasoning recipe (widely built on / distilled)","Fire-Flyer AI-HPC training-stack co-design"],"technical_founder_roles":["High-Flyer — co-founder, technical/ML direction — 2015-2023","DeepSeek — founder & CEO setting technical direction, report author — 2023-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":14,"lm_domain_depth":10,"lm_domain_breadth":12,"scientific_founder":14},"confidence":0.6,"duration_ms":78320},{"pass":"pass_2","dimensions":{"frontier_founder":12,"lm_domain_depth":8,"lm_domain_breadth":12,"scientific_founder":13},"confidence":0.6,"duration_ms":96820}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Liang_Wenfeng","wikipedia_title":"Liang Wenfeng","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"greg-brockman","name":"Greg Brockman","title":"Co-founder & President","company":"OpenAI","sector":"general","rank":34,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":4,"transformers_lm":12,"frontier_founder":15,"lm_domain_depth":13,"lm_domain_breadth":12,"hands_on_engineering":17,"industry_impact":16,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":60,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":60,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.82,"rationale":"Brockman has no ML/CS graduate degree and no first-author research papers in the core lineage; he studied at Harvard and MIT (leaving MIT without completing a degree per Wikipedia) and built his early career as a software engineer/CTO at Stripe before co-founding OpenAI. He is listed as a co-author on several major OpenAI systems papers — Evaluating Large Language Models Trained on Code (Codex, 2021), Robust Speech Recognition via Large-Scale Weak Supervision (Whisper, 2022), Dota 2 with Large-Scale Deep RL (2019), and OpenAI Gym (2016) — reflecting large-team engineering-leadership co-authorship on canonical systems rather than personal authorship of the core mathematical/architectural ideas, so transformers_lm and foundations are scored as engineering-adjacent, not principal-investigator-level. His genuine strength is hands-on infrastructure and engineering: he was OpenAI's founding CTO, personally built early engineering culture/infra, and is widely credited as a strong low-level programmer (e.g., early payments infra at Stripe, OpenAI's compute/training infrastructure). Industry impact is high as a co-founder and president of OpenAI, one of the organizations that produced canonical transformer/RLHF-era systems, though that impact is organizational/leadership rather than personally authored research.\n\nBrockman is not the author of a named foundational block (transformer, attention, word2vec, RLHF or scaling laws), but as OpenAI's founding CTO and President he personally built and led the training/inference infrastructure and engineering that GPT-class frontier models actually run on, and is a co-author on Codex, Whisper and the GPT-4 technical report — a documented builder of the frontier training stack rather than the underlying math, placing him in the 13-17 'component the frontier labs build on' band. His hands-on language-modeling record spans OpenAI's GPT line from roughly 2018 through 2026 (~8 years), always as an engineering leader who codes rather than as a first-author researcher. Across that work he has a verifiable hands-on record in three language-modeling domains — natural-language text (GPT/GPT-4), source code (Codex, 2021), and speech/audio sequence modeling (Whisper, 2022) — two of them outside plain NL text, though each contribution is large-team co-authorship; a possible biological genome-LM (Evo 2, 2026) co-authorship exists in PubMed under 'Brockman G' but is homonym-flagged and unverified, so it is not counted. He operates as the technical/engineering co-founder of one company, OpenAI (2015-2026, ~11 years), personally setting execution and building core infrastructure; his earlier Stripe CTO tenure was a senior-hire role, not a founder role, so it is not counted here.","evidence":[{"claim":"Brockman began his career at Stripe in 2010 after leaving MIT, became CTO in 2013, left in 2015 to co-found OpenAI where he was CTO and later became President.","source_url":"https://en.wikipedia.org/wiki/Greg_Brockman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed co-author on 'Evaluating Large Language Models Trained on Code' (Codex), arXiv 2107.03374, 2021.","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar profile lists 40 papers and 55,611 citations under his name (co-authorship on large OpenAI team papers).","source_url":"https://www.semanticscholar.org/author/Greg-Brockman/2065151121","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Enrolled at Harvard 2008 and left after a year, briefly enrolled at MIT and dropped out in 2010 to join Stripe; no degree completed; Stripe's first CTO from 2013 to May 2015; co-founded OpenAI December 2015, led recruiting of the founding team, served as CTO and President; led OpenAI Gym and OpenAI","source_url":"https://en.wikipedia.org/wiki/Greg_Brockman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of Evaluating Large Language Models Trained on Code / Codex (2021), Robust Speech Recognition via Large-Scale Weak Supervision / Whisper (2022), Dota 2 with Large Scale Deep Reinforcement Learning (2019), OpenAI Gym (2016) and the GPT-4 Technical Report (2023)","source_url":"https://api.openalex.org/authors/A5040311065","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar records 55,611 citations and h-index 11 for the profile, with OpenAI Gym at 5,645 citations and Dota 2 with Large Scale Deep RL at 2,204","source_url":"https://api.semanticscholar.org/graph/v1/author/2065151121?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists occupations entrepreneur, investor, programmer, researcher, education at Harvard (2008-2009) and MIT (2009-2010) with no degree recorded, employers OpenAI (from 2015) and Stripe (2010-2015)","source_url":"https://www.wikidata.org/wiki/Q108398183","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder, CTO and President of OpenAI (from 2015), the org that built GPT/GPT-4; Wikidata records OpenAI employer from 2015 and occupations programmer/researcher.","source_url":"https://en.wikipedia.org/wiki/Greg_Brockman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Evaluating Large Language Models Trained on Code' (Codex, 2021) — source-code language modeling.","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Robust Speech Recognition via Large-Scale Weak Supervision' (Whisper, 2022) — speech/audio sequence modeling.","source_url":"https://arxiv.org/abs/2212.04356","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed on GPT-4 Technical Report (2023) and OpenAI Gym (2016) via OpenAlex author record, reflecting building the frontier training/benchmark stack.","source_url":"https://api.openalex.org/authors/A5040311065","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder, founding CTO and president of OpenAI (co-founded Dec 2015), which built the frontier GPT-class systems.","source_url":"https://en.wikipedia.org/wiki/Greg_Brockman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on Evaluating Large Language Models Trained on Code (Codex, 2021), Robust Speech Recognition via Large-Scale Weak Supervision (Whisper, 2022) and OpenAI Gym (2016) — spanning code, speech and RL system stacks.","source_url":"https://api.openalex.org/authors/A5040311065","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author on the Codex systems paper, arXiv 2107.03374 (2021).","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar attributes 40 papers / 55,611 citations to the profile, reflecting co-authorship on OpenAI's canonical LM system papers.","source_url":"https://www.semanticscholar.org/author/Greg-Brockman/2065151121","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended Harvard University (2008-2009)","Attended MIT (2009-2010, left without completing degree)"],"canonical_papers":[],"first_verifiable_year":2016,"notable_systems":["OpenAI Gym","OpenAI Codex","Whisper","OpenAI Five (Dota 2)","OpenAI engineering/training infrastructure"],"citations":5268,"h_index":11,"patents":0,"dossier_notes":"The dossier's OpenAlex block (17 works, earliest_year 1977, topics 'Graph Labeling'/'Advanced Combinatorial Mathematics') is contaminated by an unrelated graph-theory academic homonym predating Brockman's birth (1987) — disregarded. The dossier's PubMed block (earliest_year 1948, medical/dermatology papers under 'Brockman G') is entirely homonym noise from other people and unrelated to OpenAI's Greg Brockman; one 2026 Nature genomics paper ('Evo 2') in that list may be a genuine OpenAI-affiliate co-authorship but was not independently verified this pass. No canonical first-authored math/ML papers were found for Brockman; his record is engineering leadership and large-team co-authorship, consistent with the rubric's 'senior engineering adjacent to the core' anchor rather than principal-author status.","years_language_modeling":8,"years_as_technical_founder":11,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2018-2026","evidence":"GPT line / GPT-4 technical report co-author, OpenAI CTO/President"},{"domain":"code","years":"2021","evidence":"Codex — Evaluating Large Language Models Trained on Code, arXiv 2107.03374"},{"domain":"other","years":"2022","evidence":"Whisper speech/audio sequence model, arXiv 2212.04356"}],"frontier_lineage":["OpenAI GPT-4 training/inference infrastructure and engineering leadership","OpenAI Gym RL benchmark","Codex code-LM and Whisper speech-LM as OpenAI systems"],"technical_founder_roles":["OpenAI — co-founder, CTO then President — 2015-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":14,"lm_domain_depth":13,"lm_domain_breadth":12,"scientific_founder":14},"confidence":0.6,"duration_ms":80434},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":13,"lm_domain_breadth":13,"scientific_founder":14},"confidence":0.72,"duration_ms":60442}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Greg_Brockman","wikipedia_title":"Greg Brockman","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"aravind-srinivas","name":"Aravind Srinivas","title":"Co-founder & CEO","company":"Perplexity AI","sector":"general","rank":35,"rubric_version":4,"dimensions":{"foundations":14,"vector_embeddings":16,"transformers_lm":14,"frontier_founder":10,"lm_domain_depth":8,"lm_domain_breadth":5,"hands_on_engineering":15,"industry_impact":14,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":59,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":59,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.82,"rationale":"Aravind Srinivas holds a PhD in Computer Science from UC Berkeley (2021, advisor Pieter Abbeel) and authored well-cited contrastive-representation-learning and reinforcement-learning papers, including CURL (Contrastive Unsupervised Representations for Reinforcement Learning) and Data-Efficient Image Recognition with Contrastive Predictive Coding (CPCv2, with DeepMind during a 2019 internship) — genuine vector-embeddings/representation-learning work with real citation counts (935 and multiple hundreds per paper per the dossier's OpenAlex data, name-matched exactly). He also interned at OpenAI on policy-gradient RL and at Google Brain/DeepMind. He founded Perplexity AI in 2022, an LLM-powered search product built directly on retrieval + language-model integration (Sonar, built on Llama), which is hands-on engineering leadership of an AI-core product, not merely business leadership. His direct authored contributions to the transformer/language-modeling lineage itself (vs. RL/contrastive vision) are thinner, so transformers_lm is scored lower than vector_embeddings and hands_on_engineering.\n\nSrinivas authored transformer-lineage work the frontier stack draws on — Bottleneck Transformers (2021, with transformer co-author Ashish Vaswani), Decision Transformer sequence modeling (2021) and CPCv2/CURL contrastive representation learning — but none is a named foundational block (no attention/transformer/word2vec/optimizer/tokenizer/scaling/RLHF authorship) nor a pre-2013 vector-space precursor, so this is published lineage work rather than a component frontier models directly descend from (8-12 band). His verifiable language-modeling record is thin and recent: his PhD-era output is RL, contrastive vision and sequence-modeling-for-control, not text LMs; his hands-on LM work is the ~4 years (2022-2026) building Perplexity's retrieval-augmented search on fine-tuned Llama (Sonar), placing years_language_modeling in the low 3-8 band. The LM domain is essentially a single one — natural-language text search — so breadth is single-domain (3-7). As scientific founder he is the technical founder-CEO of ONE company (Perplexity, ~4 years, PhD-credentialed, sets technical direction), matching the 3-8-years / one-company band (8-12).","evidence":[{"claim":"Authored CURL (Contrastive Unsupervised Representations for Reinforcement Learning) and Reinforcement Learning with Augmented Data; interned at OpenAI (2018), DeepMind London (2019, CPCv2 contrastive self-supervised learning), Google Research (2020-21)","source_url":"https://en.wikipedia.org/wiki/Aravind_Srinivas","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex record (name-exact match) shows 32 works, 2910 citations, h-index 15, top paper 'Data-Efficient Image Recognition with Contrastive Predictive Coding' (935 citations)","source_url":"https://scholar.google.com/citations?user=GhrKC1gAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar record: 35 papers, 11876 citations, h-index 20","source_url":"https://www.semanticscholar.org/author/A.-Srinivas/41207614","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile lists him as Cofounder and CEO, Perplexity AI with ~15,408 citations, h-index 15; top works include Decision Transformer (3,417), Data-Efficient Image Recognition with CPC (1,916), Bottleneck Transformers (1,801) and CURL (1,698)","source_url":"https://scholar.google.com/citations?user=GhrKC1gAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CURL: Contrastive Unsupervised Representations for Reinforcement Learning — Aravind Srinivas first author with Michael Laskin and Pieter Abbeel (UC Berkeley)","source_url":"https://arxiv.org/abs/2004.04136","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bottleneck Transformers for Visual Recognition — Aravind Srinivas first author with Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel and Ashish Vaswani","source_url":"https://arxiv.org/abs/2101.11605","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Decision Transformer: Reinforcement Learning via Sequence Modeling — Srinivas co-author","source_url":"https://arxiv.org/abs/2106.01345","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Perplexity's Sonar search engine is based on Meta's Llama model","source_url":"https://en.wikipedia.org/wiki/Perplexity_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Perplexity's Sonar search engine is based on Meta's Llama model; company founded 2022 by Srinivas as co-founder & CEO","source_url":"https://en.wikipedia.org/wiki/Perplexity_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Computer Science, UC Berkeley (2021, advisor Pieter Abbeel); first verifiable publication year 2015","source_url":"https://scholar.google.com/citations?user=GhrKC1gAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bottleneck Transformers for Visual Recognition — Srinivas first author with Ashish Vaswani et al.","source_url":"https://arxiv.org/abs/2101.11605","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Perplexity AI is an LLM-powered search product founded 2022; its Sonar engine is based on Meta's Llama model (built on, not trained by, Perplexity)","source_url":"https://en.wikipedia.org/wiki/Perplexity_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Srinivas holds a PhD in Computer Science from UC Berkeley (2021, advisor Pieter Abbeel) and is co-founder and CEO of Perplexity AI","source_url":"https://en.wikipedia.org/wiki/Aravind_Srinivas","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BTech Electrical Engineering, IIT Madras","PhD Computer Science, UC Berkeley (2021, advisor Pieter Abbeel)"],"canonical_papers":["CURL: Contrastive Unsupervised Representations for Reinforcement Learning (2020)","Data-Efficient Image Recognition with Contrastive Predictive Coding (2019)","Decision Transformer: Reinforcement Learning via Sequence Modeling (2021)"],"first_verifiable_year":2015,"notable_systems":["Perplexity AI (Sonar search engine)"],"citations":2910,"h_index":15,"patents":0,"dossier_notes":"OpenAlex and Semantic Scholar matches both are name-exact and topically consistent (RL, contrastive learning, vision) so the dossier's identification is trustworthy for this person, unlike several others in this batch. Semantic Scholar citation count (11876) is notably higher than OpenAlex (2910) likely due to different coverage; both are plausibly this person given topic/affiliation match, but exact figure not independently reconciled.","years_language_modeling":4,"years_as_technical_founder":4,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2022-2026","evidence":"Perplexity AI retrieval-augmented LLM search product (Sonar, on Llama), co-founder & CEO"}],"frontier_lineage":["Bottleneck Transformers (transformer-lineage vision, with A. Vaswani)","Decision Transformer (sequence modeling)","CPCv2/CURL contrastive representation learning"],"technical_founder_roles":["Perplexity AI — co-founder & CEO / technical founder — 2022-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":10,"lm_domain_depth":8,"lm_domain_breadth":5,"scientific_founder":10},"confidence":0.7,"duration_ms":51919},{"pass":"pass_2","dimensions":{"frontier_founder":9,"lm_domain_depth":8,"lm_domain_breadth":5,"scientific_founder":11},"confidence":0.72,"duration_ms":52982}],"source":"seeded","profile_url":null,"wikipedia_title":"Perplexity AI","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"shane-legg","name":"Shane Legg","title":"Co-founder & Chief AGI Scientist","company":"Google DeepMind","sector":"general","rank":36,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":6,"transformers_lm":12,"frontier_founder":14,"lm_domain_depth":6,"lm_domain_breadth":3,"hands_on_engineering":14,"industry_impact":19,"scientific_founder":16},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":59,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":59,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.82,"rationale":"Legg holds a PhD from IDSIA under Marcus Hutter (thesis 'Machine Super Intelligence', 2008) on theoretical models of general intelligence (AIXI-adjacent universal-intelligence formalism), giving him genuine PhD-level foundations in statistical learning, information theory and optimization applied to agents. His personally authored canonical contribution is 'Universal Intelligence: A Definition of Machine Intelligence' (Legg & Hutter, 2007), a foundational but non-mainstream (RL/AGI-theory, not embeddings or transformer) piece of the field's math. He is a listed co-author on landmark deep-RL systems papers built at DeepMind (the 2015 Nature DQN paper 'Human-level control through deep reinforcement learning', and 'Deep reinforcement learning from human preferences', 2017, which seeded RLHF techniques now used to align transformer LMs), which supports hands_on_engineering and a link into the transformers_lm/alignment lineage, though his personal authorship role on these large-team papers is not lead-author/architect level and he has no personally authored vector-embeddings or transformer-architecture papers. industry_impact is very high: he co-founded DeepMind (2010, with Demis Hassabis and Mustafa Suleyman), which produced AlphaGo, AlphaFold, and (post 2023 merger) Gemini, and he has served as Chief AGI Scientist directing research strategy — this is leadership of a lab that produced canonical work, not merely branding.\n\nLegg's clearest line into today's frontier models is 'Deep reinforcement learning from human preferences' (Christiano, Leike, Brown, Martic, Legg, Amodei, 2017), a canonical RLHF paper whose preference-based reward-learning technique directly underlies the alignment/instruction-tuning stage of GPT/Claude/Gemini-class LLMs — a documented, named alignment block he co-authored (5th of 6), which places him in the 13-17 frontier_founder band; his co-authorship of the 2015 Nature DQN paper is deep-RL lineage rather than an LM block. His personal record is in AGI theory and reinforcement learning, NOT language modeling: he has no authored vector-space/LSI/n-gram/neural-LM/embedding/transformer papers, so his hands-on language-modeling years are essentially limited to the RL-from-preferences alignment work (adjacent, ~2 years), keeping lm_domain_depth in the 3-7 band, and he has no distinct hands-on language-modeling domains (breadth 2). As scientific founder he is a genuine co-founder of DeepMind (2010) and its Chief AGI Scientist, personally authoring core research (Universal Intelligence, DQN, RLHF) and setting research direction across ~16 years at one company — matching the 13-17 band (8-15+ years, one company).","evidence":[{"claim":"PhD at IDSIA (Dalle Molle Institute for Artificial Intelligence Research) under advisor Marcus Hutter, thesis 'Machine Super Intelligence' (2008)","source_url":"https://en.wikipedia.org/wiki/Shane_Legg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded DeepMind in 2010 with Demis Hassabis and Mustafa Suleyman; DeepMind acquired by Google 2014, merged with Google Brain in 2023 to form Google DeepMind; Legg serves as Chief AGI Scientist","source_url":"https://en.wikipedia.org/wiki/Shane_Legg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Universal Intelligence: A Definition of Machine Intelligence' (Legg & Hutter, Minds and Machines, 2007) — a formal mathematical definition of machine intelligence combining Kolmogorov complexity and reinforcement learning theory","source_url":"https://doi.org/10.1007/s11023-007-9079-x","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on 'Human-level control through deep reinforcement learning' (Nature, 2015), the DQN paper, and 'IMPALA: Scalable Distributed Deep-RL' (2018) and 'Massively Parallel Methods for Deep Reinforcement Learning' (2015)","source_url":"https://doi.org/10.1038/nature14236","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author (5th of 6) on 'Deep reinforcement learning from human preferences' (Christiano, Leike, Brown, Martic, Legg, Amodei, 2017), an early RLHF paper whose technique underlies later LM alignment work","source_url":"https://arxiv.org/abs/1706.03741","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MSc Auckland (thesis 'Solomonoff Induction'); PhD IDSIA 2008 'Machine Super Intelligence' under Marcus Hutter; co-founded DeepMind 2010 with Hassabis and Suleyman; Chief AGI Scientist at Google DeepMind","source_url":"https://en.wikipedia.org/wiki/Shane_Legg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Universal Intelligence: A Definition of Machine Intelligence — Shane Legg and Marcus Hutter, Minds and Machines, 2007","source_url":"https://arxiv.org/abs/0712.3329","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Deep reinforcement learning from human preferences — Christiano, Leike, Brown, Martic, Legg, Amodei (2017); Legg is fifth author","source_url":"https://arxiv.org/abs/1706.03741","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Human-level control through deep reinforcement learning, Nature 2015 (DQN)","source_url":"https://doi.org/10.1038/nature14236","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Legg co-authored 'Deep reinforcement learning from human preferences' (Christiano, Leike, Brown, Martic, Legg, Amodei, 2017), an early RLHF paper whose preference-reward technique underlies modern LLM alignment","source_url":"https://arxiv.org/abs/1706.03741","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Legg co-founded DeepMind (2010) with Demis Hassabis and Mustafa Suleyman and serves as Chief AGI Scientist at Google DeepMind, directing AGI research","source_url":"https://en.wikipedia.org/wiki/Shane_Legg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Legg is a co-author on 'Human-level control through deep reinforcement learning' (Nature, 2015), the DQN paper — deep-RL lineage work","source_url":"https://doi.org/10.1038/nature14236","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Legg is fifth author on 'Deep reinforcement learning from human preferences' (Christiano, Leike, Brown, Martic, Legg, Amodei, 2017), an early RLHF paper whose technique underlies modern LM alignment","source_url":"https://arxiv.org/abs/1706.03741","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded DeepMind in 2010 with Demis Hassabis and Mustafa Suleyman and serves as Chief AGI Scientist at Google DeepMind, directing AGI research strategy","source_url":"https://en.wikipedia.org/wiki/Shane_Legg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD at IDSIA under Marcus Hutter, thesis 'Machine Super Intelligence' (2008); author of 'Universal Intelligence: A Definition of Machine Intelligence' (Legg & Hutter, 2007) — AGI theory, not language modeling","source_url":"https://doi.org/10.1007/s11023-007-9079-x","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD, IDSIA (advisor: Marcus Hutter), 'Machine Super Intelligence', 2008","MSc thesis on Solomonoff Induction, 1996 (per Wikipedia)"],"canonical_papers":["Legg & Hutter, 'Universal Intelligence: A Definition of Machine Intelligence', Minds and Machines, 2007","Mnih et al. (incl. Legg), 'Human-level control through deep reinforcement learning', Nature, 2015","Christiano, Leike, Brown, Martic, Legg, Amodei, 'Deep reinforcement learning from human preferences', 2017"],"first_verifiable_year":2007,"notable_systems":["DeepMind (co-founder)","DQN","AlphaGo/AlphaFold/Gemini (as organizational leadership, not personal authorship)"],"citations":35890,"h_index":26,"patents":0,"dossier_notes":"The dossier's OpenAlex match (A5008987732, affiliations Google/Google DeepMind/University of Auckland/Dalle Molle Institute) is correctly identified — no homonym issue here, unlike richard-socher/thomas-wolf. However, the dossier's OpenAlex 'earliest_year: 1979' and timeline.first_verifiable_year: 1979 are almost certainly a database artifact (Legg was born ~1973/74, making 1979 implausible as his first publication year); I used 2007 (the Universal Intelligence paper, his first well-documented significant work) as the realistic first_verifiable_year instead. Citation/h-index figures (35,890 citations, h-index 26) are plausible and were retained from the dossier's OpenAlex block since that match is not disputed. profile_page and self_claims were both empty/unchecked in the dossier, so no self-claims needed discounting.","years_language_modeling":2,"years_as_technical_founder":16,"technical_founder_companies":1,"frontier_lineage":["RLHF / deep RL from human preferences (2017, co-author) — preference-based reward learning that underlies LLM alignment and instruction tuning","DQN / deep reinforcement learning (Nature 2015, co-author) — deep-RL lineage"],"technical_founder_roles":["DeepMind / Google DeepMind — co-founder & Chief AGI Scientist — 2010-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":15,"lm_domain_depth":5,"lm_domain_breadth":2,"scientific_founder":15},"confidence":0.82,"duration_ms":54236},{"pass":"pass_2","dimensions":{"frontier_founder":13,"lm_domain_depth":8,"lm_domain_breadth":4,"scientific_founder":16},"confidence":0.7,"duration_ms":67941}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Shane_Legg","wikipedia_title":"Shane Legg","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"alexander-long","name":"Alexander Long","title":"Founder (Pluralis Research)","company":"Pluralis Research","sector":"crypto","rank":37,"rubric_version":4,"dimensions":{"foundations":14,"vector_embeddings":13,"transformers_lm":16,"frontier_founder":8,"lm_domain_depth":9,"lm_domain_breadth":4,"hands_on_engineering":16,"industry_impact":11,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":56,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":56,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.78,"rationale":"Both passes correctly discarded the dossier, every block of which is a different person, and both verified his Pluralis authorship; they diverged because pass_2 found a pre-Pluralis research record that pass_1 could not. That record checks out. The verified arXiv author list for 'Retrieval Augmented Classification for Long-Tail Visual Recognition' (arXiv:2202.11233, 2022) has Alexander Long as FIRST author, with Ajanthan, Garg, Blair, Shen and van den Hengel; its method is a retrieval branch querying a non-parametric external memory of pre-encoded images and text snippets — authored dense-retrieval/embedding work, which is precisely what pass_1 recorded as absent and scored a 6 for. His first verifiable ML paper is 'Multi-hop Reading Comprehension via Deep Reinforcement Learning based Document Traversal' (23 May 2019, Alex Long, Joel Mason, Alan Blair, Wei Wang), giving about seven years active, not the dossier's 31. At Pluralis he is the senior/last author of a coherent programme on training transformers over low-bandwidth networks: 'Subspace Networks' (arXiv:2506.01260, verified last author) confines activations and gradients to a predefined low-dimensional subspace, achieving up to 99% compression of model-parallel communication and training billion-parameter models over 80Mbps links — a linear-algebra result applied to transformer structure — plus asynchronous pipeline-parallel optimization and the Agora permissionless 8.6B-parameter pretraining run. Industry impact stays low: the programme is young, citations are minimal, and its influence is still prospective. No Google Scholar profile or verified citation count could be retrieved.\n\nLong's verifiable contribution to the frontier lineage is training-stack research on decentralized/communication-efficient transformer pretraining — 'Subspace Networks' (up to 99% compression of model-parallel communication), asynchronous pipeline-parallel optimization (Nesterov, ICML 2025), and the Agora / Pluralis-8B permissionless 8.6B-parameter run — genuine pretraining-infrastructure work in the lineage, but its adoption by GPT/Claude/Gemini/Llama-class stacks is prospective, not demonstrated, so it sits below the 'named block the frontier builds on' band. His language-modeling record is thin and recent: NLP reading-comprehension in 2019, then LLM-pretraining infrastructure 2024-2026, for roughly 3-4 hands-on LM years (his 2022 work is vision retrieval and RL-from-pixels, different modalities, not language modeling). He has a verifiable record in only one language-modeling domain — natural-language text — so cross-domain breadth is low. He is the clear scientific/technical founder of Pluralis Research: sole author of the foundational 'Protocol Learning' paper (email alexander@pluralis.ai) and senior/last author on the company's core technical programme, but the company is ~2 years old (founded c. 2024), so this is one company across a short tenure.","evidence":[{"claim":"First author of 'Retrieval Augmented Classification for Long-Tail Visual Recognition' (2022); verified author order Alexander Long, Wei Yin, Thalaiyasingam Ajanthan, Vu Nguyen, Pulak Purkait, Ravi Garg, Alan Blair, Chunhua Shen, Anton van den Hengel; method fuses a base image encoder with a retrieva","source_url":"https://arxiv.org/abs/2202.11233","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Earliest verifiable ML paper: 'Multi-hop Reading Comprehension via Deep Reinforcement Learning based Document Traversal', submitted 23 May 2019, authors Alex Long, Joel Mason, Alan Blair, Wei Wang","source_url":"http://export.arxiv.org/api/query?search_query=all:%22Multi-hop%20Reading%20Comprehension%20via%20Deep%20Reinforcement%20Learning%20based%20Document%20Traversal%22&start=0&max_results=5","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Last author of 'Subspace Networks: Scaling Decentralized Training with Communication-Efficient Model Parallelism' (submitted 2 June 2025); verified author order Sameera Ramasinghe, Thalaiyasingam Ajanthan, Gil Avraham, Yan Zuo, Alexander Long; up to 99% compression with no convergence degradation, b","source_url":"https://arxiv.org/abs/2506.01260","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sole author of 'Protocol Learning, Decentralized Frontier Risk and the No-Off Problem' (10 December 2024), affiliation Pluralis Research, email alexander@pluralis.ai — the identity-confirming source","source_url":"https://arxiv.org/abs/2412.07890","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Fast and Data Efficient Reinforcement Learning from Pixels via Non-Parametric Value Approximation' (7 March 2022) with Alan Blair and Herke van Hoof — non-parametric/nearest-neighbour value estimation","source_url":"http://export.arxiv.org/api/query?search_query=au:%22Alexander_Long%22+AND+cat:cs.LG&start=0&max_results=40&sortBy=submittedDate&sortOrder=ascending","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Last author of 'Agora: Collective and Permissionless Internet-Scale Pretraining of Large Language Models', reporting Pluralis-8B, an 8.6B-parameter permissionless pretraining run","source_url":"https://arxiv.org/abs/2607.13332","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alexander Long is the sole author of 'Protocol Learning, Decentralized Frontier Risk and the No-Off Problem' (Dec 2024), with affiliation 'Pluralis Research' and email 'alexander@pluralis.ai' listed in the paper header — this is the identity-confirming source tying this Alexander Long to Pluralis Re","source_url":"https://arxiv.org/abs/2412.07890","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alexander Long is a co-author (with the same Pluralis research team: Avraham, Shevchenko, Dolatabadi, Pajak, Snewin, Xi, O'Donnell, Ajanthan, Ramasinghe, Koneputugodage, Siriwardhana) of 'Agora: Collective and Permissionless Internet-Scale Pretraining of Large Language Models', which reports trainin","source_url":"https://arxiv.org/abs/2607.13332","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prior to Pluralis, Alexander Long was affiliated with 'Amazon, Australia' as of the ICML 2024 paper 'A Sampling Theory Perspective on Activations for Implicit Neural Representations' (with co-authors Hemanth Saratchandran, Sameera Ramasinghe, Violetta Shevchenko, Simon Lucey of Univ. of Adelaide) —","source_url":"https://arxiv.org/abs/2402.05427","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author profile (ID 2283309100, matched by co-author overlap and subject matter, distinct from 7 other same-name candidates checked and ruled out) lists 17 papers, h-index 3, citation count 39, spanning 2024-2026, all on distributed/pipeline-parallel/asynchronous LLM training, decent","source_url":"https://api.semanticscholar.org/graph/v1/author/2283309100","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Earliest verifiable NLP/LM paper 'Multi-hop Reading Comprehension via Deep Reinforcement Learning based Document Traversal' (23 May 2019) — first LM-adjacent work, anchoring the language-modeling year count","source_url":"https://arxiv.org/abs/2202.11233","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior/last author of 'Subspace Networks: Scaling Decentralized Training with Communication-Efficient Model Parallelism' (2025) — a low-dimensional-subspace method for compressing model-parallel transformer communication; a training-stack contribution in the frontier lineage but not a named frontier","source_url":"https://arxiv.org/abs/2506.01260","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Earliest verifiable language-modeling paper: 'Multi-hop Reading Comprehension via Deep Reinforcement Learning based Document Traversal' (23 May 2019) — natural-language reading comprehension, first year of the LM record","source_url":"https://arxiv.org/abs/2412.07890","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD-track ML research at UNSW Sydney (2019-2022 publications with Alan Blair and Wei Wang as supervisor-pattern co-authors); degree award year not verified from a primary source"],"canonical_papers":["Retrieval Augmented Classification for Long-Tail Visual Recognition (2022) — first author","Protocol Learning, Decentralized Frontier Risk and the No-Off Problem (2024) — sole author","Subspace Networks (NeurIPS 2025) — last author","Nesterov Method for Asynchronous Pipeline Parallel Optimization (ICML 2025)","Agora: Collective and Permissionless Internet-Scale Pretraining of LLMs (2026) — last author"],"first_verifiable_year":2019,"notable_systems":["Pluralis Protocol Models","Subspace Networks compression for model-parallel training","Pluralis-8B permissionless 8.6B-parameter pretraining run"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Every programmatic block is the wrong person. Wikipedia/Wikidata Q2642579 is Alexander Long the American politician (1816-1886). OpenAlex A5078125817 (147 works, h=33, affiliations Notre Dame / Biogen / Amgen / Los Alamos, topics 'Nuclear Physics and Applications', top work a 1998 Nature paper on bacteriophage T7 DNA replication) conflates at least two more unrelated people, and is wrongly flagged ambiguous=false. Semantic Scholar 'A. Long' (76 papers) is name_exact=false. PubMed ran an unusable query ('Long A' AND 'Research[Affiliation]') with 2,050 hits. timeline.first_verifiable_year 1995 / 31 years active is spurious; the verified figure is 2019 / ~7 years. No Google Scholar profile for him could be retrieved, so citations and h-index are left at 0 rather than guessed.","adjudication_notes":["vector_embeddings (pass_1 6, pass_2 14, final 13): pass_2 was right — pass_1 missed a paper that exists. Verified arXiv record shows he is FIRST author of 'Retrieval Augmented Classification for Long-Tail Visual Recognition' (2022), whose core is a retrieval module over a non-parametric memory of pre-encoded images and text. That is authored retrieval/embedding work, not adjacency. Set at 13 not 14: it is one paper, in vision rather than text retrieval, with minimal citation weight.","foundations (pass_1 11, pass_2 15, final 14): pass_2 closer. Pass_1 could not verify graduate training; the verified 2019-2022 UNSW first-author series with Alan Blair and Wei Wang establishes a doctoral-track research record, and Subspace Networks is a genuine low-dimensional-subspace linear-algebra result on transformer activations. Held at 14 because no thesis or degree record could be confirmed from a primary source.","transformers_lm (pass_1 15, pass_2 16, final 16): near-agreement; the verified Subspace Networks and Agora author positions show he leads an actual pretraining programme rather than writing about one.","industry_impact (pass_1 10, pass_2 12, final 11): split. Pluralis is young, no verified citation record exists, and the rubric bars fundraising and branding from counting — impact remains prospective."],"years_language_modeling":4,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2019, 2024-2026","evidence":"Multi-hop reading comprehension (2019, UNSW); LLM pretraining infrastructure at Pluralis — Subspace Networks, Nesterov pipeline-parallel optimization, Agora/Pluralis-8B (2024-2026)"}],"frontier_lineage":["Communication-efficient model-parallel training (Subspace Networks) and asynchronous pipeline-parallel optimization for decentralized transformer pretraining — training-stack lineage, adoption by frontier labs prospective, not demonstrated"],"technical_founder_roles":["Pluralis Research — founder / lead researcher (scientific founder, sole author of foundational paper, senior author on core programme) — c. 2024-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":8,"lm_domain_depth":8,"lm_domain_breadth":4,"scientific_founder":9},"confidence":0.6,"duration_ms":79943},{"pass":"pass_2","dimensions":{"frontier_founder":8,"lm_domain_depth":10,"lm_domain_breadth":5,"scientific_founder":10},"confidence":0.6,"duration_ms":66447}],"source":"seeded","profile_url":null,"wikipedia_title":"Alexander Long","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"charles-xie","name":"Charles Xie","title":"Founder & CEO","company":"Zilliz (creator of Milvus)","sector":"general","rank":38,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":18,"transformers_lm":6,"frontier_founder":7,"lm_domain_depth":10,"lm_domain_breadth":4,"hands_on_engineering":16,"industry_impact":15,"scientific_founder":13},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":55,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":55,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.8,"rationale":"The two passes diverged almost entirely because pass_1 exhausted its search budget and could find no authored record, scoring him as a founder of a real product with no personal research, while pass_2 located peer-reviewed authorship. That authorship is verified here independently through Crossref: he is the last author of 'Milvus: A Purpose-Built Vector Data Management System' (SIGMOD 2021, DOI 10.1145/3448016.3457550), listed 22nd of 22 authors with the affiliation 'Zilliz, Shanghai, China', and last author of 'Manu: A Cloud Native Vector Database Management System' (PVLDB 2022, DOI 10.14778/3554821.3554843), 15th of 15 with the same affiliation. Last-author position on both system papers, with the company affiliation printed, is the senior/responsible role and is exactly the 'authored, built or shipped' standard the vector_embeddings dimension names — approximate-nearest-neighbour indexing, quantization and distributed similarity search over high-dimensional embeddings is the vector-space retrieval layer of the rubric's own lineage, and he built and shipped it rather than consuming it. Milvus is a graduated LF AI & Data Foundation project under the Linux Foundation with core contributors from multiple hardware and platform vendors. Foundations is scored mid-band and deliberately conservatively: the work demands applied linear algebra and ANN index mathematics, but no thesis, degree or first-principles mathematical publication under his name could be verified — his Wikidata ORCID record is an empty stub with zero works, educations or employments. transformers_lm is low: his systems serve retrieval-augmented and embedding workloads for language models, but he has authored no modeling, pretraining or scaling work.\n\nXie founded Zilliz in 2017 and invented/open-sourced the Milvus vector database (released Oct 19, 2019), and is the senior/last author of the peer-reviewed Milvus (SIGMOD 2021) and Manu (PVLDB 2022) system papers — so he is a verifiable technical/scientific founder of ONE company across ~9 years (2017-2026), previously a founding engineer on the Oracle 12c cloud database. His frontier-lineage position is real but downstream: Milvus is the vector-search/ANN retrieval layer that RAG pipelines use alongside GPT/Claude/Gemini-class models, not a block those models are built or trained on, and the core ANN methods it implements (HNSW, IVF, PQ) were invented by others — so it sits below the named-block bar. His language-modeling record is ~7 years and confined to the retrieval/vector-database layer over general embeddings, not modeling, pretraining, or embedding training. Milvus is a domain-general vector engine rather than hands-on modeling across distinct domains, so cross-domain breadth is effectively a single general-purpose natural-language/multimodal retrieval domain.","evidence":[{"claim":"Last author (22nd of 22) of 'Milvus: A Purpose-Built Vector Data Management System', Proceedings of the 2021 ACM SIGMOD International Conference on Management of Data, with the affiliation 'Zilliz, Shanghai, China'; co-authors include Jianguo Wang (Zilliz & Purdue), Xiaomeng Yi, Rentong Guo and Hai","source_url":"https://api.crossref.org/works/10.1145/3448016.3457550","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Last author (15th of 15) of 'Manu: A Cloud Native Vector Database Management System', Proceedings of the VLDB Endowment, 2022, affiliation Zilliz","source_url":"https://api.crossref.org/works/10.14778/3554821.3554843","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus is an open-source high-performance vector database developed by Zilliz and donated to the LF AI & Data Foundation under the Linux Foundation (Apache 2.0), with core contributors from Zilliz, ARM, NVIDIA, AMD, Intel, Meta, IBM, Salesforce, Alibaba and Microsoft","source_url":"https://milvus.io/docs/overview.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him only as 'researcher' with ORCID 0009-0000-1713-8696; the ORCID public record contains no works, education or employment entries, so no degree or prior role is verifiable","source_url":"https://pub.orcid.org/v3.0/0009-0000-1713-8696/record","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus is an actively maintained open-source vector database for AI applications, per its GitHub organization","source_url":"https://github.com/milvus-io","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus is an actively maintained, widely used open-source vector database whose GitHub organization describes it as 'the open source vector database designed for AI applications.'","source_url":"https://github.com/milvus-io","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Zilliz's live engineering blog shows substantive, ongoing technical work on the Milvus vector database (e.g. 'Announcing Milvus 3.0: Lake-Native Vector Search and a More Powerful Retrieval Engine', 'How Force Merge Compaction Nearly Doubled Milvus Search QPS'), evidencing a real, technically deep ve","source_url":"https://milvus.io/blog","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's Semantic Scholar match for 'Charles Xie' (author ID 30749452, 44 papers, h-index 16, citation_count 1017) consists entirely of engineering-design-education, CAD, physics-classroom, and STEM-pedagogy papers (e.g. in The Physics Teacher, Computers & Education, Journal of Mechanical Desig","source_url":"https://api.semanticscholar.org/graph/v1/author/30749452?fields=name,affiliations,papers.title,papers.year,papers.venue","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus was released October 19, 2019 and has been developed by Zilliz since 2017","source_url":"https://en.wikipedia.org/wiki/Milvus_(vector_database)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Last author of 'Milvus: A Purpose-Built Vector Data Management System' (SIGMOD 2021) and 'Manu: A Cloud Native Vector Database Management System' (PVLDB 2022), affiliation Zilliz","source_url":"https://api.crossref.org/works/10.1145/3448016.3457550","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus is a general vector search engine (ANN indexing/similarity search) used for AI/RAG retrieval workloads, a graduated LF AI & Data project","source_url":"https://milvus.io/docs/overview.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Milvus is a graduated LF AI & Data / Linux Foundation open-source vector database for AI applications, used as the retrieval layer for embedding/RAG workloads","source_url":"https://milvus.io/docs/overview.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["Milvus: A Purpose-Built Vector Data Management System (SIGMOD 2021) — last author","Manu: A Cloud Native Vector Database Management System (PVLDB 2022) — last author"],"first_verifiable_year":2021,"notable_systems":["Milvus open-source vector database (LF AI & Data project)","Zilliz Cloud managed vector search","Manu cloud-native vector DBMS"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier is nearly empty and its one research signal is a homonym: Semantic Scholar id 30749452 ('Charles Xie', 44 papers, 1,017 citations, h=16) is the educational-technology physicist known for Molecular Workbench and Energy3D, whose papers appear in The Physics Teacher, Computers & Education and the Journal of Mechanical Design — not the Zilliz founder; its citation and h-index figures are not imported, which is why those fields read 0 despite his real work being well cited. profile_id is null, Wikipedia and OpenAlex found nothing, patents were skipped, PubMed returned 10,312 hits flagged homonym_risk, and Wikidata Q130826501 is a bare stub whose ORCID is an empty claimed profile. The timeline is entirely null; the verified first year in the published record is 2021 (Zilliz was founded earlier and Milvus released in 2019, but neither date could be confirmed from a primary source). Education and prior roles (the Oracle/IBM claim that circulates in secondary sources) remain unverified and are not asserted.","adjudication_notes":["vector_embeddings (pass_1 12, pass_2 19, final 18): pass_2 was right; pass_1 scored on absence of evidence, not evidence of absence. Verified independently via Crossref: he is last author with the Zilliz affiliation on BOTH the SIGMOD 2021 Milvus paper and the PVLDB 2022 Manu paper — peer-reviewed senior authorship of the vector-search system itself, which pass_1 never found. Set at 18 rather than 19 because his contribution is the database/retrieval layer over embeddings rather than the embedding models themselves.","foundations (pass_1 4, pass_2 12, final 11): pass_2 closer. Pass_1's 4 ('uses the tools, no personal record') is refuted by peer-reviewed senior authorship of two ANN/vector-DBMS systems papers at top database venues, which require applied linear algebra and index mathematics. Held at 11, not higher, because the verified ORCID record is empty and no degree, thesis or mathematical publication exists.","hands_on_engineering (pass_1 9, pass_2 17, final 16): pass_2 right. He founded the company that built Milvus and is senior author on its two system papers; Milvus is a graduated Linux Foundation project with multi-vendor contributors — built and shipped infrastructure, not managed procurement.","transformers_lm (pass_1 2, pass_2 9, final 6): split toward pass_1. Neither pass cited any modeling, pretraining or scaling work by him, and none exists; the credit is only that his systems serve RAG and embedding workloads for LMs.","industry_impact (pass_1 11, pass_2 16, final 15): pass_2 closer — Milvus is widely deployed open-source infrastructure whose core IS vector search, which the rubric credits directly; discounted from 16 for the absence of any verifiable personal citation or patent record."],"years_language_modeling":7,"years_as_technical_founder":9,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2019-2026","evidence":"Milvus general-purpose vector search / ANN retrieval over embeddings, serving RAG and embedding workloads; SIGMOD 2021 + PVLDB 2022 system papers"}],"frontier_lineage":["Milvus open-source vector database — the ANN retrieval/vector-search serving layer used in RAG pipelines alongside frontier LLMs (downstream infrastructure, not a block the models are trained/built on)"],"technical_founder_roles":["Zilliz — Founder & CEO / inventor of Milvus, senior author on its system papers — 2017-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":7,"lm_domain_depth":9,"lm_domain_breadth":4,"scientific_founder":13},"confidence":0.6,"duration_ms":114638},{"pass":"pass_2","dimensions":{"frontier_founder":7,"lm_domain_depth":10,"lm_domain_breadth":4,"scientific_founder":13},"confidence":0.62,"duration_ms":103892}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ben-goertzel","name":"Ben Goertzel","title":"CEO","company":"SingularityNET / ASI Alliance","sector":"crypto","rank":39,"rubric_version":4,"dimensions":{"foundations":15,"vector_embeddings":7,"transformers_lm":5,"frontier_founder":5,"lm_domain_depth":12,"lm_domain_breadth":15,"hands_on_engineering":12,"industry_impact":10,"scientific_founder":18},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":54,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":54,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.83,"rationale":"Goertzel holds a genuine PhD in mathematics from Temple University (1989, advisor Avi Lin), confirmed via Mathematics Genealogy Project and his own institutional record, and has a 35+ year publication record (417 works, h-index 32-44 depending on source) with real academic faculty positions (University of Nevada Las Vegas, City University of New York, Xiamen University). However his core research lineage is symbolic AI, probabilistic logic networks, and cognitive architectures (the OpenCog / OpenCog Hyperon project) — a lineage distinct from, and largely predating or running parallel to, the vector-embedding and transformer/LM lineage this rubric targets; his post-2017 transformer-adjacent work (e.g. 'Guiding symbolic natural language grammar induction via transformer-based sequence probabilities', AGI 2020) integrates transformers into symbolic systems rather than contributing to core transformer/LM research. He founded SingularityNET via a 2017 public ICO ($36M in one minute), a legitimate public token sale with no evidence of family/inherited funding. Strong math foundations and long hands-on research career, but limited direct contribution to the specific embeddings/transformer lineage.\n\nGoertzel's core lineage is symbolic AI, probabilistic logic networks and cognitive architectures (Webmind, Novamente, OpenCog, OpenCog Hyperon) — a branch that runs PARALLEL to, not into, the vector-space/word2vec/transformer foundation frontier models (GPT/Claude/Gemini/Llama) descend from; his NLP work integrates transformer sequence probabilities into symbolic grammar induction (AGI 2020) rather than authoring any transformer, embedding, optimizer, tokenizer or scaling building block, so frontier_founder is low. He has a genuine but non-central language-modeling record of roughly a decade — OpenCog natural-language understanding (link-grammar/RelEx), transformer-guided symbolic grammar induction, plus biomedical-literature and gene-expression text/data mining and clinical-notes text mining — placing lm_domain_depth in the mid range. Distinct hands-on language-modeling domains with verifiable records: natural-language text (OpenCog/grammar induction), biological (first-authored gene-expression and genomics-interrelationship mining, PubMed 2006–2014), and clinical (suicide-risk prediction from clinical notes, PLoS One 2014) — three domains, two outside natural-language text. As a scientific/technical founder he is exceptionally deep: he has served as chief scientist/CTO/founder personally authoring the core research and code across Webmind (~1998), Biomind, Novamente, Aidyia and SingularityNET over ~25+ years — well past the 15-year, 3-company threshold — earning a top scien","evidence":[{"claim":"PhD in Mathematics, Temple University, 1989, doctoral advisor Avi Lin","source_url":"https://goertzel.org/bio.htm","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"417 works, 4,672+ citations, h-index 32 per OpenAlex; chief architect of the OpenCog symbolic/cognitive-architecture project","source_url":"https://en.wikipedia.org/wiki/Ben_Goertzel","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'Artificial General Intelligence: Concept, State of the Art, and Future Prospects' (2014) and editor of the 'Artificial General Intelligence' book series","source_url":"https://doi.org/10.2478/jagi-2014-0001","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded SingularityNET in 2017 (with Simone Giacomelli and David Hanson); the AGIX token ICO raised $36M in a public sale, not family/private funding","source_url":"https://www.nextbigfuture.com/2017/12/ai-researcher-ben-goertzel-launches-singularitynet-marketplace-and-agi-coin-cryptocurrency.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in mathematics, Temple University, 1990, under Avi Lin, dissertation 'A Multilevel Approach to Global Optimization'","source_url":"https://www.genealogy.math.ndsu.nodak.edu/id.php?id=40053","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Computer scientist and AI researcher who helped popularize the term artificial general intelligence; leading developer of the OpenCog framework; founder and CEO of SingularityNET; former Chief Scientist at Hanson Robotics, whose Sophia claims were criticised by researchers including Yann LeCun; left","source_url":"https://en.wikipedia.org/wiki/Ben_Goertzel","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First-authored machine-learning bioinformatics research: 'Identifying the genes and genetic interrelationships underlying the impact of calorie restriction on maximum lifespan: an artificial intelligence-based approach' (Rejuvenation Res, 2008) and related chronic fatigue syndrome data-mining papers","source_url":"https://pubmed.ncbi.nlm.nih.gov/18729806/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Lead author of 'OpenCog Hyperon: A Framework for AGI at the Human Level and Beyond' (arXiv 2310.18318, 2023), describing the AtomSpace/MeTTa neural-symbolic architecture","source_url":"https://arxiv.org/abs/2310.18318","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First-authored biological language/data-mining research: gene-interrelationships underlying calorie restriction (Rejuvenation Res 2008), MUTIC gene-expression exploration in prostate cancer (2007), and SNP combinations predicting chronic fatigue syndrome (Pharmacogenomics 2006)","source_url":"https://pubmed.ncbi.nlm.nih.gov/18729806/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Clinical text-mining: 'Predicting the risk of suicide by analyzing the text of clinical notes' (PLoS One 2014), co-authored with Goertzel","source_url":"https://pubmed.ncbi.nlm.nih.gov/24489669/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder and CEO of SingularityNET and chief architect of OpenCog; former Chief Scientist at Hanson Robotics — a multi-decade record of setting and executing the technical direction of companies he founded","source_url":"https://en.wikipedia.org/wiki/Ben_Goertzel","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder & CEO and chief architect of SingularityNET; leading developer of OpenCog; former Chief Scientist at Hanson Robotics; prior AI ventures — the same person personally authoring the core research the companies run on","source_url":"https://en.wikipedia.org/wiki/Ben_Goertzel","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First-authored biological language/data-mining: gene and genetic-interrelationship discovery for calorie-restriction lifespan effects (Rejuvenation Res, 2008), plus prostate-cancer gene-expression exploration (2007)","source_url":"https://pubmed.ncbi.nlm.nih.gov/18729806/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Clinical text/language modeling: co-author of 'Predicting the risk of suicide by analyzing the text of clinical notes' (PLoS One, 2014)","source_url":"https://pubmed.ncbi.nlm.nih.gov/24489669/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Mathematics, Temple University (1989)","Undergraduate/early graduate work, Bard College at Simon's Rock and NYU Courant Institute (no degree completed at NYU)"],"canonical_papers":["Artificial General Intelligence: Concept, State of the Art, and Future Prospects (2014)","Artificial General Intelligence (2007 book)","Mapping the Landscape of Human-Level Artificial General Intelligence (2012)"],"first_verifiable_year":1991,"notable_systems":["OpenCog / OpenCog Hyperon cognitive architecture","SingularityNET decentralized AI marketplace"],"citations":4672,"h_index":32,"patents":1,"dossier_notes":"The dossier's patent list is contaminated by homonyms/unrelated ventures: several 2007 patents ('Note Overlay System', 'Merchant Tool for Embedding Advertisement Hyperlinks', 'Self Teaching Thesaurus', various 'Intellectual property' filing-tool patents) could not be independently verified as Ben Goertzel's and appear to belong to an unrelated patent-filing business; only 'Deep Learning Neuromorphic Engineering' (US20180005109A1, priority 2016) is plausibly his given subject-matter fit, though it also could not be independently confirmed via USPTO/Google Patents search in this pass — reported patent count of 1 reflects this uncertainty rather than the dossier's raw count of 14. His research lineage (symbolic AI/AGI/cognitive architecture) is real and substantial but is a different branch of AI than the vector-embedding/transformer lineage the rubric weights most heavily, which caps vector_embeddings and transformers_lm scores despite strong foundations and industry_impact.","years_language_modeling":10,"years_as_technical_founder":25,"technical_founder_companies":5,"lm_domains":[{"domain":"natural_language","years":"2001-2023","evidence":"OpenCog NLP (link grammar/RelEx), transformer-guided symbolic grammar induction (AGI 2020), OpenCog Hyperon MeTTa"},{"domain":"biological","years":"2006-2014","evidence":"first-authored gene-expression (MUTIC), calorie-restriction gene-interrelationship, and CFS SNP-prediction mining (PubMed; Biomind)"},{"domain":"clinical","years":"2014","evidence":"suicide-risk prediction from the text of clinical notes, PLoS One 2014"}],"frontier_lineage":["No foundational transformer/embedding/word2vec-lineage contribution; symbolic/cognitive-architecture branch parallel to the frontier LM stack; only integrates transformer sequence probabilities into symbolic NLP"],"technical_founder_roles":["Webmind Inc. — CTO/Chief Scientist — ~1998-2001","Biomind LLC — founder/Chief Scientist (bioinformatics) — ~2000s","Novamente LLC — founder/Chief Scientist — ~2000s","Aidyia Ltd — co-founder/Chief Scientist (AI hedge fund) — ~2010s","SingularityNET — founder & CEO / chief architect (OpenCog Hyperon) — 2017-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":5,"lm_domain_depth":10,"lm_domain_breadth":14,"scientific_founder":18},"confidence":0.72,"duration_ms":79488},{"pass":"pass_2","dimensions":{"frontier_founder":5,"lm_domain_depth":13,"lm_domain_breadth":16,"scientific_founder":18},"confidence":0.6,"duration_ms":117501}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Ben_Goertzel","wikipedia_title":"Ben Goertzel","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"kai-fu-lee","name":"Kai-Fu Lee","title":"Chairman & CEO","company":"01.AI","sector":"general","rank":40,"rubric_version":4,"dimensions":{"foundations":15,"vector_embeddings":8,"transformers_lm":10,"frontier_founder":9,"lm_domain_depth":13,"lm_domain_breadth":5,"hands_on_engineering":14,"industry_impact":16,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":54,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":54,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.8,"rationale":"Lee earned a PhD in Computer Science from Carnegie Mellon (1988) under Raj Reddy, and his doctoral dissertation created SPHINX, the first large-vocabulary speaker-independent continuous speech-recognition system using statistical/neural methods — genuinely canonical, foundational statistical-ML work that predates but feeds into the modern deep-learning lineage (h-index 20-35 across OpenAlex/Semantic Scholar, confirmed). He built and led engineering/research organizations at Apple, SGI, Microsoft (founding director of Microsoft Research China/Asia), and Google (president of Google China) — senior technical leadership of real research labs, not just business roles. However his direct hands-on research output ends in the early 1990s; from the mid-1990s onward his record is executive/investor (Sinovation Ventures). In 2023 he founded 01.AI, which built and shipped the open-weight Yi series of LLMs — but Lee's personal authorship role on the Yi models themselves is not established (he leads the company; the model-building credit sits with 01.AI's research team). No specific vector-embeddings or transformer-authorship record was found; his direct contribution to the modern transformer/LM lineage is as an org-builder/funder rather than an author.\n\nLee's own hands-on lineage work is statistical speech recognition — his CMU dissertation SPHINX (1988) and SPHINX-II (1993) were pioneering HMM-based large-vocabulary continuous recognition systems that used statistical n-gram language models, genuine foundation-era statistical sequence modeling but not the vector-space/distributional-semantics precursor line that word2vec descends from, nor a named block (attention, transformers, embeddings, optimizers) inside today's frontier LM stack; his 01.AI shipped the open-weight Yi LLM series, but the modeling credit sits with the company's research team, not with him personally, so his frontier position is org-builder rather than author. His verifiable hands-on language-modeling record runs roughly 1988–1996 (CMU dissertation, SPHINX-II, Apple speech products PlainTalk/Casper) — about 8 years — after which his record is executive and investor; that places lm_domain_depth in the low 8–15y band. It is a single language-modeling domain, natural-language speech and text; no biological, financial, code, or chemistry language-modeling record exists, so breadth is a one-domain score. As a technical founder he has ~3 years at one company (01.AI, 2023–present) whose core LM science is done by others, plus lab-founding leadership (Microsoft Research China/Asia) that is not a company founded on his own authored code — a founder-CEO of an AI company whose modeling is executed by his team.","evidence":[{"claim":"PhD Computer Science, Carnegie Mellon University (1988), advisor Raj Reddy; doctoral dissertation built SPHINX, a pioneering large-vocabulary speaker-independent continuous speech-recognition system","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Microsoft Research China (later Microsoft Research Asia), 1998-2000; president of Google China, 2005-2009","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded 01.AI in March 2023; the company released the open-weight Yi-34B model in November 2023","source_url":"https://www.turingpost.com/p/01ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar lm2nBYkAAAAJ: 13,143 citations, h-index 41, i10 65; top works are 'Readings in speech recognition' (1990), 'Speaker-independent phone recognition using hidden Markov models' (1989), 'Automatic speech recognition: the development of the SPHINX system' (1988), 'Speaker adaptation throug","source_url":"https://scholar.google.com/citations?user=lm2nBYkAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS Computer Science Columbia 1983 summa cum laude; PhD Computer Science CMU 1988 under Raj Reddy, dissertation on large-vocabulary speaker-independent continuous speech recognition (SPHINX); Apple 1990-1996 (PlainTalk, Casper, GalaTea); SGI 1996-1998; founding director of Microsoft Research China/As","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5064910846: 55 works, earliest 1988, topics are speech recognition and synthesis, speech and audio processing, NLP; affiliations include Microsoft, Google, Microsoft Research Asia, Apple","source_url":"https://api.openalex.org/authors/A5064910846","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'The SPHINX-II speech recognition system: an overview', Computer Speech & Language 1993","source_url":"https://doi.org/10.1006/csla.1993.1007","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'The development of a world class Othello program', Artificial Intelligence 1990 (Lee & Mahajan) — early search/evaluation-function AI work","source_url":"https://doi.org/10.1016/0004-3702(90)90068-b","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CMU PhD 1988 under Raj Reddy; SPHINX was the first large-vocabulary speaker-independent continuous speech recognition system, using HMM statistical acoustic + n-gram language modeling","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded 01.AI in 2023; the company released the open-weight Yi-34B LLM in November 2023 (Lee as founder/CEO, model authorship credited to the 01.AI research team)","source_url":"https://www.turingpost.com/p/01ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founding director of Microsoft Research China/Asia (1998–2000); Apple speech products 1990–1996 (PlainTalk); OpenAlex topics are speech recognition/synthesis and NLP only","source_url":"https://api.openalex.org/authors/A5064910846","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CMU PhD 1988 under Raj Reddy; dissertation built SPHINX, a large-vocabulary speaker-independent continuous speech-recognition system combining HMM acoustic modeling with statistical language models","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"SPHINX-II speech recognition system overview, Computer Speech & Language 1993 — statistical ASR/LM research","source_url":"https://doi.org/10.1006/csla.1993.1007","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Apple 1990-1996 building speech systems (PlainTalk, Casper), continuing hands-on speech/language work","source_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded 01.AI in 2023; company released the open-weight Yi-34B LLM in November 2023, built by the 01.AI research team","source_url":"https://www.turingpost.com/p/01ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, Columbia University","PhD Computer Science, Carnegie Mellon University (1988)"],"canonical_papers":["Automatic Speech Recognition: The Development of the SPHINX System (1988)","The SPHINX-II speech recognition system: an overview (1993)"],"first_verifiable_year":1988,"notable_systems":["SPHINX speech recognition system","01.AI / Yi model series (as company founder/leader, not sole author)"],"citations":6487,"h_index":35,"patents":0,"dossier_notes":"OpenAlex identity flagged ambiguous (10 candidates) but the top works (SPHINX papers, Othello program) plausibly do belong to this Kai-Fu Lee and are corroborated independently (Wikipedia, CMU dissertation record). PubMed sample is essentially noise (63,593 raw hits, no usable homonym-filtered records) and was not relied on.","years_language_modeling":8,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"1988-1996; 2023-2026","evidence":"SPHINX/SPHINX-II speech recognition with n-gram LMs (CMU, Apple) hands-on; Yi open-weight LLMs shipped by 01.AI as founder (not personal authorship)"}],"frontier_lineage":["Statistical HMM speech recognition + n-gram language modeling (SPHINX, 1988-1993) — foundation-era statistical sequence modeling, not a named block in the transformer/embedding stack","01.AI Yi open-weight LLM series — founder/leader, not personal author"],"technical_founder_roles":["01.AI — founder & CEO — 2023-2026","Microsoft Research China/Asia — founding director — 1998-2000"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":9,"lm_domain_depth":12,"lm_domain_breadth":5,"scientific_founder":7},"confidence":0.72,"duration_ms":67851},{"pass":"pass_2","dimensions":{"frontier_founder":9,"lm_domain_depth":14,"lm_domain_breadth":5,"scientific_founder":7},"confidence":0.72,"duration_ms":62447}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Kai-Fu_Lee","wikipedia_title":"Kai-Fu Lee","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"pramod-viswanath","name":"Pramod Viswanath","title":"Co-founder (Sentient); Forrest G. Hamrick Professor in Engineering, Princeton University","company":"Sentient","sector":"crypto","rank":41,"rubric_version":4,"dimensions":{"foundations":18,"vector_embeddings":13,"transformers_lm":12,"frontier_founder":8,"lm_domain_depth":9,"lm_domain_breadth":4,"hands_on_engineering":10,"industry_impact":13,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":53,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":53,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.85,"rationale":"Viswanath is a career information theorist with the deepest mathematical record in this batch: a UC Berkeley PhD under Venkat Anantharam and David Tse, IEEE Fellow 2013, co-author with Tse of 'Fundamentals of Wireless Communication' (13,408 citations in OpenAlex alone), and author of foundational multiuser information-theory results — opportunistic beamforming, vector Gaussian broadcast channel capacity and uplink-downlink duality, the diversity-multiplexing tradeoff. The dispute that matters is vector_embeddings, and pass_2 found evidence pass_1 missed: 'All-but-the-Top: Simple and Effective Postprocessing for Word Representations' (arXiv:1702.01417, Jiaqi Mu, Suma Bhat, Pramod Viswanath, ICLR 2018) is verified authored word-embedding geometry — removing the common mean vector and top dominating directions improves word2vec and GloVe on similarity, analogy and semantic textual similarity. That is squarely inside the rubric's vector-space lineage and cannot be scored at pass_1's 6. His LM work is real but recent and collaborative: he is the sixth of seven authors on 'Scalable Fingerprinting of Large Language Models' (arXiv:2502.07760, with Nasery, Hayase, Brooks, Sheng, Tyagi and Oh) — LLM security rather than modeling, pretraining or scaling. Hands-on engineering is his weakest dimension and pass_2 over-scored it at 13: his contribution is algorithm design and analysis carried out with students and co-authors, and no personally built or shipped training system, model or infrastructure could be verified. Sentient is early and, per the rubric, its fundraising and branding count for nothing.\n\nViswanath's verifiable lineage into today's frontier models is his authored word-embedding geometry paper 'All-but-the-Top' (Mu, Bhat, Viswanath, ICLR 2018), which improves word2vec and GloVe representations — published lineage work the embedding stack draws on, but a postprocessing result rather than a named foundational block, and his pre-2013 record is wireless-communications information theory (opportunistic beamforming, vector Gaussian broadcast channel, the Tse–Viswanath textbook), not vector-space or distributional-semantics precursor work, so this sits in the mid band. His hands-on language-modeling record runs roughly 2017–present (~8 years, intermittent): the 2018 word-embedding paper and, at Sentient, the 2025 'Scalable Fingerprinting of Large Language Models' (sixth of seven authors); this is real but thin and collaborative, and the PubMed hits are a homonym (Pavithra Viswanath, UCSF), so there is no biological-LM record. The distinct language-modeling domains reduce to a single one — natural-language text (word embeddings, LLM fingerprinting); no financial, biological, code or chemistry LM record is verifiable. As scientific founder he is a genuine technical co-founder of one company, Sentient (co-founded ~2024 with researcher Himanshu Tyagi and Sandeep Nailwal), authoring core OML/fingerprinting research, but for only about two years, placing him at the lower end of the technical-founder anchor.","evidence":[{"claim":"Co-author of 'All-but-the-Top: Simple and Effective Postprocessing for Word Representations' (2017/ICLR 2018); verified author list Jiaqi Mu, Suma Bhat, Pramod Viswanath; eliminates the common mean vector and top dominating directions from word vectors, improving word2vec and GloVe on similarity, ca","source_url":"https://arxiv.org/abs/1702.01417","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sixth of seven authors on 'Scalable Fingerprinting of Large Language Models' (2025); verified author list Anshul Nasery, Jonathan Hayase, Creston Brooks, Peiyao Sheng, Himanshu Tyagi, Pramod Viswanath, Sewoong Oh","source_url":"https://arxiv.org/abs/2502.07760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD UC Berkeley EECS (advisors Venkat Anantharam and David Tse); Professor of ECE at Princeton; IEEE Fellow 2013 for contributions to the theory and practice of wireless communications","source_url":"https://en.wikipedia.org/wiki/Pramod_Viswanath","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile lPycXNcAAAAJ is the identifier recorded for him in Wikidata","source_url":"https://www.wikidata.org/wiki/Q29387745","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD UC Berkeley EECS, advisors Venkat Anantharam and David Tse; IEEE Fellow (2013) for wireless communications theory","source_url":"https://en.wikipedia.org/wiki/Pramod_Viswanath","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author, 'Scalable Fingerprinting of Large Language Models' (arXiv 2502.07760, 2025) with Peiyao Sheng, Himanshu Tyagi and others","source_url":"https://arxiv.org/html/2505.16723","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Sentient (with Himanshu Tyagi and Sandeep Nailwal), which raised $85M seed led by Founders Fund/Pantera/Framework to build the OML cryptographic protocol for decentralized AI","source_url":"https://www.dailyprincetonian.com/article/2025/05/princeton-features-profiles-sentient-ai-loyal-alignment","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (Forrest G. Hamrick Professor in Engineering, Princeton): ~40,148 citations, h-index 64, i10-index 171; research areas blockchains and wireless communication","source_url":"https://scholar.google.com/citations?user=lPycXNcAAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'All-but-the-Top: Simple and Effective Postprocessing for Word Representations' (Mu, Bhat, Viswanath, ICLR 2018) is authored word-embedding-geometry work improving word2vec and GloVe — his verifiable link into the embeddings lineage frontier models build on","source_url":"https://arxiv.org/abs/1702.01417","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Scalable Fingerprinting of Large Language Models' (2025), Viswanath sixth of seven authors — LLM security work, his only recent LLM-core paper, part of a ~2017-2026 natural-language LM record","source_url":"https://arxiv.org/abs/2502.07760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Sentient (with Himanshu Tyagi and Sandeep Nailwal), raised $85M seed to build the OML cryptographic protocol for decentralized AI; he authors the core research (LLM fingerprinting) — genuine technical-founder role, ~2 years","source_url":"https://www.dailyprincetonian.com/article/2025/05/princeton-features-profiles-sentient-ai-loyal-alignment","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored word-embedding geometry paper 'All-but-the-Top: Simple and Effective Postprocessing for Word Representations' improving word2vec and GloVe (Mu, Bhat, Viswanath, ICLR 2018) — verified lineage into embedding methods frontier models descend from","source_url":"https://arxiv.org/abs/1702.01417","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Sentient with Himanshu Tyagi and Sandeep Nailwal, building the OML cryptographic protocol for decentralized AI (founded ~2024)","source_url":"https://www.dailyprincetonian.com/article/2025/05/princeton-features-profiles-sentient-ai-loyal-alignment","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career information theorist; PhD UC Berkeley under Anantharam and Tse, IEEE Fellow 2013 for wireless communications — the pre-2013 record is information theory, not distributional-semantics precursor work","source_url":"https://en.wikipedia.org/wiki/Pramod_Viswanath","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Electrical Engineering and Computer Sciences, UC Berkeley (advisors Venkat Anantharam, David Tse)"],"canonical_papers":["Fundamentals of Wireless Communication (Tse & Viswanath, Cambridge University Press, 2005)","Opportunistic Beamforming Using Dumb Antennas (IEEE Trans. Inf. Theory, 2002)","Sum Capacity of the Vector Gaussian Broadcast Channel and Uplink-Downlink Duality (2003)","All-but-the-Top: Simple and Effective Postprocessing for Word Representations (ICLR 2018)","Scalable Fingerprinting of Large Language Models (2025)"],"first_verifiable_year":1996,"notable_systems":["Prism blockchain protocol","Sentient OML model-fingerprinting stack (co-founder)"],"citations":28981,"h_index":50,"patents":0,"dossier_notes":"Wikidata records date_of_birth as '+2000-00-00', which is an error — that is his PhD year. The PubMed block is a complete homonym: all 34 refined records belong to Pavithra Viswanath, a UCSF cancer/hyperpolarized-imaging researcher (co-authors Batsios, Taglang, Gillespie), so earliest_pubmed_year 2015 is meaningless. The OpenAlex affiliation list is noisy (Clarion University, Capital University, University of New Mexico, Urbana University are spurious) though the works are correctly his. Both OpenAlex and the dossier's top-works list miss his AI-lineage output entirely — no embeddings or LLM papers appear — which understates him; his ICLR 2018 word-embedding paper and 2025 LLM-fingerprinting work were verified directly on arXiv. Citations/h-index are reported from OpenAlex because pass_2's higher Google Scholar figures (40,148 / h-index 64) could not be re-verified in this session. The dossier does not mention Sentient at all.","adjudication_notes":["vector_embeddings (pass_1 6, pass_2 14, final 13): pass_2 was right and pass_1 missed a paper that exists. Verified on arXiv: 'All-but-the-Top' (Mu, Bhat, Viswanath) is authored word-embedding geometry improving word2vec and GloVe — the rubric's named lineage. Set at 13 rather than 14 as it is a postprocessing result rather than an embedding method or system he built.","foundations (pass_1 16, pass_2 18, final 18): pass_2 right. A Berkeley information-theory PhD under Anantharam and Tse, IEEE Fellowship, the Tse-Viswanath textbook and the vector Gaussian broadcast-channel capacity results are canonical mathematics of exactly the kind the dimension names.","transformers_lm (pass_1 10, pass_2 14, final 12): split. The verified LLM-fingerprinting paper is real but he is the sixth of seven authors and the subject is model security, not modeling, pretraining, scaling or alignment; neither pass could cite authored work in the modeling core itself.","hands_on_engineering (pass_1 8, pass_2 13, final 10): pass_1 was closer. Pass_2 itself conceded his contribution is 'algorithm design and analysis carried out with students rather than personally shipping large training systems'; no personally built or shipped AI system, model or infrastructure is verifiable, which is the 3-7/8-12 boundary, not the PhD-production band.","industry_impact (pass_1 10, pass_2 16, final 13): split toward pass_1. Pass_2's 16 leans on Sentient, which is early-stage, and the rubric bars fundraising and branding; the defensible basis is his verified citation record (28,981 / h-index 50) and the influence of the Tse-Viswanath textbook, not a shipped product whose core is these systems."],"years_language_modeling":8,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2017-2026","evidence":"'All-but-the-Top' word-embedding postprocessing (ICLR 2018) and 'Scalable Fingerprinting of Large Language Models' (2025)"}],"frontier_lineage":["Word-embedding geometry postprocessing improving word2vec/GloVe representations (All-but-the-Top, ICLR 2018)"],"technical_founder_roles":["Sentient — co-founder / technical (Chief Scientist-type) — 2024-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":8,"lm_domain_depth":9,"lm_domain_breadth":5,"scientific_founder":8},"confidence":0.62,"duration_ms":66660},{"pass":"pass_2","dimensions":{"frontier_founder":9,"lm_domain_depth":9,"lm_domain_breadth":4,"scientific_founder":7},"confidence":0.72,"duration_ms":58056}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Pramod_Viswanath","wikipedia_title":"Pramod Viswanath","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jonathan-ross","name":"Jonathan Ross","title":"Founder & CEO","company":"Groq","sector":"general","rank":42,"rubric_version":4,"dimensions":{"foundations":12,"vector_embeddings":4,"transformers_lm":8,"frontier_founder":16,"lm_domain_depth":6,"lm_domain_breadth":2,"hands_on_engineering":18,"industry_impact":16,"scientific_founder":14},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":52,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":52,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.76,"rationale":"Ross studied mathematics and computer science at NYU's Courant Institute. At Google he started the Tensor Processing Unit (TPU) as a 20%-time project in 2013, personally designing and implementing core elements of the first-generation chip, and is a listed co-author on the canonical 'In-Datacenter Performance Analysis of a Tensor Processing Unit' (ISCA 2017, arXiv:1704.04760) among ~75 co-authors; the TPU became the hardware backbone for a majority of Google's ML compute and powered AlphaGo. He founded Groq in 2016 to build inference-optimized LPU chips, co-authoring hardware architecture papers ('Think Fast: A TSP for Accelerating Deep Learning Workloads', ISCA 2020). This is strong, verifiable hands-on chip-engineering depth and real industry impact on the infrastructure AI models run on — but it is systems/hardware engineering, not authorship of the math, embeddings, or transformer/LM research lineage itself, so those dimensions score low.\n\nRoss's contribution to the frontier stack is real but sits at the silicon layer: he initiated Google's TPU (2013) and is a named co-author of the canonical 'In-Datacenter Performance Analysis of a Tensor Processing Unit' (ISCA 2017), and TPUs are now the primary training/inference substrate for Google's frontier models (Gemini), while his Groq LPU is a shipped LLM-inference accelerator — foundational infrastructure the frontier lineage runs on, though not the math/embedding/transformer methods themselves. His language-modeling record is adjacent, not modeling: he designs the accelerators that run language models rather than authoring LMs, with roughly 6 hands-on years since Groq's chip became explicitly LLM-inference-focused (the 'Language Processing Unit'), so lm_domain_depth sits in the low-adjacent band. Breadth is a single domain (natural-language inference hardware, no biological/financial/code modeling record), scoring at the single-domain floor. As scientific/technical founder he founded and led Groq from 2016 through 2025 (~9 years, one company), personally setting the chip architecture and co-authoring 'Think Fast: A TSP for Accelerating Deep Learning Workloads' (ISCA 2020) under Groq affiliation, which places scientific_founder in the 8–15-year / one-company band.","evidence":[{"claim":"Co-author on 'In-Datacenter Performance Analysis of a Tensor Processing Unit' (ISCA 2017), the canonical TPU paper","source_url":"https://arxiv.org/abs/1704.04760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Started the Google TPU as a 20% project in 2013, personally designed/implemented core elements of the first-gen chip; team took it to production in 15 months","source_url":"https://en.wikipedia.org/wiki/Tensor_Processing_Unit","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Groq in 2016 to build the LPU (Language Processing Unit), an inference-optimized AI chip; NVIDIA received a perpetual license to Groq's patent portfolio in a Dec 2025 deal","source_url":"https://en.wikipedia.org/wiki/Groq","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Studied mathematics and computer science at NYU's Courant Institute","source_url":"https://www.linkedin.com/in/ross-jonathan/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named author on 'In-Datacenter Performance Analysis of a Tensor Processing Unit', ISCA 2017, with Jouppi, Young, Patil, Patterson, Dean et al.","source_url":"https://arxiv.org/abs/1704.04760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Second author on 'Think Fast: A Tensor Streaming Processor (TSP) for Accelerating Deep Learning Workloads', ISCA 2020, pp. 145-158, all authors affiliated Groq Inc.","source_url":"https://dblp.org/rec/conf/isca/AbtsRSWBHBTKKHL20.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross was one of the designers of the TPU at Google; co-founded Groq in 2016; the chip was renamed from Tensor Streaming Processor to Language Processing Unit; joined Nvidia in December 2025 as part of a licensing deal while Groq continues operating","source_url":"https://en.wikipedia.org/wiki/Groq","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross started the Google TPU and is a named co-author of 'In-Datacenter Performance Analysis of a Tensor Processing Unit' (ISCA 2017), the canonical accelerator underlying Google's ML/frontier-model compute","source_url":"https://arxiv.org/abs/1704.04760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross founded Groq in 2016 and is the technical founder behind the LPU inference chip; second author on 'Think Fast: A TSP for Accelerating Deep Learning Workloads' (ISCA 2020), all authors affiliated Groq Inc.","source_url":"https://dblp.org/rec/conf/isca/AbtsRSWBHBTKKHL20.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross was one of the designers of the TPU at Google and co-founded Groq in 2016; the chip was renamed Tensor Streaming Processor to Language Processing Unit (LPU), an inference stack used for LLMs","source_url":"https://en.wikipedia.org/wiki/Groq","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross initiated the Google TPU as a 20% project (2013) and co-authored the canonical TPU paper; TPUs are the backbone of Google's ML compute including frontier model training","source_url":"https://arxiv.org/abs/1704.04760","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross founded Groq in 2016 to build the LPU (Language Processing Unit), an inference-optimized accelerator for language models; joined Nvidia Dec 2025 in a licensing deal while Groq continues","source_url":"https://en.wikipedia.org/wiki/Groq","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross is second author on 'Think Fast: A Tensor Streaming Processor for Accelerating Deep Learning Workloads' (ISCA 2020), all authors affiliated Groq Inc.","source_url":"https://dblp.org/rec/conf/isca/AbtsRSWBHBTKKHL20.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ross was one of the designers of the TPU at Google, the hardware that powered AlphaGo and Google's ML production compute","source_url":"https://en.wikipedia.org/wiki/Tensor_Processing_Unit","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Mathematics and Computer Science coursework, NYU Courant Institute (no confirmed degree completion found)"],"canonical_papers":["In-Datacenter Performance Analysis of a Tensor Processing Unit (ISCA 2017)","Think Fast: A Tensor Streaming Processor (TSP) for Accelerating Deep Learning Workloads (ISCA 2020)"],"first_verifiable_year":2013,"notable_systems":["Google TPU v1","Groq LPU / Tensor Streaming Processor"],"citations":4515,"h_index":0,"patents":46,"dossier_notes":"CRITICAL: the dossier's Wikipedia/Wikidata match (Q40054, 'English broadcaster and comedian born 1960') and virtually the entire OpenAlex record (349 works, medical/STD topics, h-index 44) are for a DIFFERENT person — the British TV presenter Jonathan Ross, not the Groq/TPU founder. The dossier's own PubMed and Semantic Scholar sections flag high homonym risk and low name-match confidence, correctly. Only the OpenAlex 'tpu_2017' canonical_flag (the TPU paper) is verifiably the correct Jonathan Ross; everything else in the dossier's programmatic sections should be disregarded for this profile. Patent count (46) is a self-reported LinkedIn figure, not independently verified against USPTO/Google Patents in this pass — treated as unverified and excluded from citations/h_index metadata, reported separately.","years_language_modeling":6,"years_as_technical_founder":9,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2016-2026","evidence":"Groq LPU inference accelerator built and shipped for large-language-model inference; hardware substrate, not modeling"}],"frontier_lineage":["Google TPU — training/inference accelerator that frontier models (incl. Gemini) run on","Groq LPU / Tensor Streaming Processor — shipped low-latency LLM-inference stack"],"technical_founder_roles":["Groq — founder & CEO / chip architect — 2016-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":16,"lm_domain_depth":5,"lm_domain_breadth":2,"scientific_founder":15},"confidence":0.7,"duration_ms":71362},{"pass":"pass_2","dimensions":{"frontier_founder":16,"lm_domain_depth":8,"lm_domain_breadth":3,"scientific_founder":14},"confidence":0.72,"duration_ms":63231}],"source":"seeded","profile_url":null,"wikipedia_title":"Jonathan Ross","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"johannes-hagemann","name":"Johannes Hagemann","title":"Co-founder & Head of Research","company":"Prime Intellect","sector":"general","rank":43,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":3,"transformers_lm":15,"frontier_founder":11,"lm_domain_depth":10,"lm_domain_breadth":10,"hands_on_engineering":16,"industry_impact":10,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":50,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":50,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.85,"rationale":"Both passes identified him correctly and both discarded the dossier, which is matched throughout to a DESY X-ray nanoscience physicist of the same name (ORCID 0000-0003-2768-9496, born 1988, 261 works back to 1985) with no connection to Prime Intellect. The substantive disagreement is about how far back his record runs and how much it earns. Pass 1 found the earlier work and pass 2 missed it: I verified that he is first author of 'Efficient Parallelization Layouts for Large-Scale Distributed Model Training' (arXiv:2311.05610, November 2023), with Aleph Alpha co-founder Samuel Weinbach and Gerard de Melo among the co-authors, an ablation study of training configurations reporting 70.5% Model FLOPs utilisation on a Llama 13B — first-author work on LLM pretraining efficiency, done at a foundation-model company before Prime Intellect existed. Pass 2's first_verifiable_year of 2024 is therefore wrong and its foundations score partly rests on that error. On the Prime Intellect papers pass 2 is the more accurate reader: he is the final author of the INTELLECT-1 technical report (verified author order ending '...Max Ryabinin, Johannes Hagemann'), the first 10B-parameter model trained collaboratively across 14 nodes on 3 continents with 30 compute providers, and of INTELLECT-2, the first globally distributed RL run of a 32B reasoning model, plus senior author on OpenDiLoCo. Final-author position across the series is the senior-technical-lead slot, and the substance — ElasticDeviceMesh fault tolerance, a hybrid DiLoCo-FSDP2 implementation cutting communication bandwidth 400x, MFU figures, PRIME-RL and SHARDCAST — is exactly what the hands-on engineering anchor rewards, so pass 2's 17 is closer than pass 1's 15. What neither pass could establish is any degree, thesis or mathematics publication, and he has no record at all in embeddings or retrieval. His track record is also genuinely short (2023 onward), which the rubric says should lower rather than raise the depth-sensitive dimensions, so industry_impact stays mid-band despite the visibility of the INTELLECT series.\n\nHagemann's contribution to the frontier lineage is training-infrastructure rather than architecture: first-author LLM-parallelization-efficiency work at Aleph Alpha (arXiv:2311.05610, 70.5% MFU on a Llama-13B) and senior authorship of OpenDiLoCo / DiLoCo-style low-communication distributed training (SHARDCAST, ElasticDeviceMesh, hybrid DiLoCo-FSDP2 with ~400x bandwidth reduction) — published lineage work the training stack draws on, but not a named block (attention, tokenizer, optimizer, scaling law, RLHF) that GPT/Claude/Gemini/Llama directly descend from, and nothing pre-2013, so this sits mid-band, not at the top. His verifiable language-modeling record runs from joining Aleph Alpha in early 2022 through the INTELLECT-1/INTELLECT-2 runs of 2024-2025, ~4 continuous hands-on years — a real but short record placing lm_domain_depth in the 3-8-year band. He has a hands-on record in two distinct LM domains: natural-language text (parallelization paper, INTELLECT-1/2) and biological language modeling (METAGENE-1, a metagenomic foundation model), which is a two-domain, 8-12 breadth. As a technical founder he co-founded ONE company, Prime Intellect (with Vincent Weisser, end of 2023), operating as its scientific/technical lead — final-author slot across the INTELLECT series — for ~2.7 years, just under the 3-year threshold, so scientific_founder lands at the very bottom of the one-company band.","evidence":[{"claim":"First author of 'Efficient Parallelization Layouts for Large-Scale Distributed Model Training' (arXiv:2311.05610, Nov 2023) — verified author order Johannes Hagemann, Samuel Weinbach, Konstantin Dobler, Maximilian Schall, Gerard de Melo; reports 70.5% Model FLOPs utilization training a Llama 13B","source_url":"https://arxiv.org/abs/2311.05610","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final (senior) author of the 'INTELLECT-1 Technical Report' (arXiv:2412.01152, Dec 2024) — verified author order ends 'Max Ryabinin, Johannes Hagemann'; first 10B-parameter LM trained collaboratively across 14 nodes on 3 continents with 30 compute providers, with ElasticDeviceMesh and a hybrid DiLoC","source_url":"https://arxiv.org/abs/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of 'OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training' (arXiv:2407.07852, July 2024), trained across continents at 90-95% compute utilisation","source_url":"https://arxiv.org/abs/2407.07852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final author of 'INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning' (arXiv:2505.07291, May 2025), contributing PRIME-RL, TOPLOC rollout verification and SHARDCAST weight distribution","source_url":"https://arxiv.org/abs/2505.07291","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His Google Scholar profile (qlCqOBEAAAAJ) shows ~203 citations and h-index 8, with top works Intellect-1, OpenDiLoCo, Synthetic-1, Metagene-1, Intellect-2 and the parallelization-layouts paper — consistent with the arXiv record and distinct from the DESY physicist's profile","source_url":"https://scholar.google.com/citations?user=qlCqOBEAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hagemann joined Aleph Alpha (German foundation-model company) in early 2022 as an AI Research Engineer focused on large-scale parallelization and distributed-systems engineering for LLM training.","source_url":"https://nextomoro.com/johannes-hagemann/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hagemann is first author of 'Efficient Parallelization Layouts for Large-Scale Distributed Model Training' (arXiv:2311.05610, Nov 2023), with co-authors including Aleph Alpha co-founder Samuel Weinbach; selected for oral presentation at WANT@NeurIPS 2023, later published at COLM 2024.","source_url":"https://arxiv.org/abs/2311.05610","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hagemann co-founded Prime Intellect with Vincent Weisser at the end of 2023, building infrastructure for globally distributed LLM training and inference.","source_url":"https://hagemann.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hagemann is a listed author on Prime Intellect's OpenDiLoCo (arXiv:2407.07852), INTELLECT-1 technical report (arXiv:2412.01152, a 1-trillion-token LLM trained across 14 nodes / 30 compute providers on 3 continents), and INTELLECT-2 (arXiv:2505.07291, decentralized RL training of a 32B reasoning mode","source_url":"https://arxiv.org/pdf/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hagemann's real Google Scholar profile (qlCqOBEAAAAJ) shows 203 total citations, h-index 8, with top works being Intellect-1, OpenDiLoco, Synthetic-1, Metagene-1, Intellect-2, and the parallelization-layouts paper — consistent with the arXiv record.","source_url":"https://scholar.google.com/citations?user=qlCqOBEAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of OpenDiLoCo (arXiv:2407.07852, 2024), an open-source globally-distributed low-communication training framework in the DiLoCo lineage","source_url":"https://arxiv.org/abs/2407.07852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Final (senior-technical-lead) author of the INTELLECT-1 Technical Report (arXiv:2412.01152, Dec 2024) and INTELLECT-2 (arXiv:2505.07291, May 2025), contributing ElasticDeviceMesh, hybrid DiLoCo-FSDP2, PRIME-RL and SHARDCAST — end-to-end frontier-scale LLM training and RL post-training","source_url":"https://arxiv.org/abs/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author on METAGENE-1, a metagenomic (biological-sequence) foundation model from Prime Intellect — a second, non-natural-language LM domain listed among his top Google Scholar works","source_url":"https://scholar.google.com/citations?user=qlCqOBEAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Prime Intellect with Vincent Weisser at the end of 2023, building infrastructure for globally distributed LLM training and inference; operates as its technical/research lead","source_url":"https://hagemann.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Senior author of OpenDiLoCo (arXiv:2407.07852, 2024) and final author of the INTELLECT-1 (arXiv:2412.01152, 2024) and INTELLECT-2 (arXiv:2505.07291, 2025) technical reports, contributing ElasticDeviceMesh, hybrid DiLoCo-FSDP2, PRIME-RL and SHARDCAST — open decentralized-training components","source_url":"https://arxiv.org/abs/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Joined Aleph Alpha in early 2022 as an AI Research Engineer for large-scale parallelization / distributed-systems LLM training — first verifiable language-modeling year","source_url":"https://nextomoro.com/johannes-hagemann/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Prime Intellect with Vincent Weisser at the end of 2023, setting and executing the distributed-training technical direction; his real Google Scholar profile (qlCqOBEAAAAJ) lists METAGENE-1, a metagenomic (biological-sequence) foundation model, among top works — a second language-modeling","source_url":"https://scholar.google.com/citations?user=qlCqOBEAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["Efficient Parallelization Layouts for Large-Scale Distributed Model Training (arXiv:2311.05610, 2023; COLM 2024)","OpenDiLoCo (arXiv:2407.07852, 2024)","INTELLECT-1 Technical Report (arXiv:2412.01152, 2024)","INTELLECT-2 (arXiv:2505.07291, 2025)"],"first_verifiable_year":2022,"notable_systems":["INTELLECT-1 / INTELLECT-2 decentralized training runs","OpenDiLoCo","PRIME-RL","Aleph Alpha training stack"],"citations":203,"h_index":8,"patents":0,"dossier_notes":"The dossier's Wikidata, OpenAlex and PubMed blocks are all a different Johannes Hagemann — a DESY/Göttingen X-ray nanoscience physicist (ORCID 0000-0003-2768-9496, Scholar KU5yapUAAAAJ, X-ray holography and phase retrieval, works from 1985-1992 among them) plus unrelated medical papers. None of it is the Prime Intellect co-founder and all of it was excluded. The correct profile is Google Scholar qlCqOBEAAAAJ.","adjudication_notes":["first_verifiable_year (2022 vs 2024) and foundations (6 vs 9): pass 1 was right about the timeline. Verified fact: arXiv:2311.05610 is dated November 2023 with Hagemann as FIRST author alongside Aleph Alpha co-founder Samuel Weinbach, so pass 2's claim that his verifiable record 'begins in 2024' is wrong and its foundations score rests partly on that error. Settled at 8 — the parallelization paper is a real first-author systems-efficiency study, but no degree, thesis or mathematics publication is verifiable for him.","hands_on_engineering (15 vs 17): pass 2 was right; settled at 16. Verified final-author position on both INTELLECT-1 and INTELLECT-2 is the senior-technical-lead slot, and the contributions cited (ElasticDeviceMesh, hybrid DiLoCo-FSDP2 with 400x bandwidth reduction, SHARDCAST) are training-stack engineering of exactly the kind the anchor names.","transformers_lm (13 vs 15): pass 2 was closer; settled at 15. He has first-author pretraining-efficiency work and senior authorship on training two frontier-scale models end to end, which is training and post-training rather than authoring architecture or scaling-law results.","industry_impact (9 vs 12): split at 10. The open decentralized-training stack is genuinely built on by others, but the track record is about three years old and the company has produced no canonical result; fundraising and token are not evidence.","vector_embeddings (3 vs 5): pass 1 was right; settled at 3. Neither pass cited any embedding or retrieval artifact and I found none."],"years_language_modeling":4,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2022-2026","evidence":"arXiv:2311.05610 parallelization efficiency; INTELLECT-1/INTELLECT-2 pretraining and RL"},{"domain":"biological","years":"2025","evidence":"METAGENE-1 metagenomic foundation model, Prime Intellect"}],"frontier_lineage":["OpenDiLoCo / DiLoCo-style low-communication distributed training","large-scale LLM parallelization-layout efficiency (MFU tuning)","decentralized training stack: ElasticDeviceMesh, SHARDCAST, PRIME-RL"],"technical_founder_roles":["Prime Intellect — co-founder & technical/research lead — 2023-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":10,"lm_domain_depth":11,"lm_domain_breadth":10,"scientific_founder":9},"confidence":0.72,"duration_ms":61357},{"pass":"pass_2","dimensions":{"frontier_founder":12,"lm_domain_depth":9,"lm_domain_breadth":9,"scientific_founder":8},"confidence":0.7,"duration_ms":71181}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"bob-van-luijt","name":"Bob van Luijt","title":"Co-founder & CEO","company":"Weaviate","sector":"general","rank":44,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":14,"transformers_lm":6,"frontier_founder":6,"lm_domain_depth":12,"lm_domain_breadth":4,"hands_on_engineering":14,"industry_impact":12,"scientific_founder":13},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":46,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":46,"tier":"technically_fluent","tier_label":"Technically Fluent","confidence":0.75,"rationale":"Van Luijt has no formal computer science or mathematics degree — his education is in music (ArtEZ, Berklee College of Music) plus a Harvard Business School executive program, and he is a self-taught programmer who started a software company at 15, so foundations scores low on credentialed math/stats grounds. He is, however, the verified original architect and CEO of Weaviate, an open-source vector database he started in March 2016 — this is real, personally-built, shipped infrastructure squarely in the 'vector databases & search' portion of the vector_embeddings dimension, predating the post-2022 vector-DB boom by several years, which supports meaningful vector_embeddings and hands_on_engineering credit despite the lack of formal training. Weaviate integrates embedding models and GraphQL-based semantic search rather than Van Luijt personally authoring embedding/transformer research papers, so transformers_lm is scored low-moderate for applied systems integration rather than research authorship. His Semantic Scholar record (3 papers, 38 citations) suggests some light technical writing but not a research career. Industry impact is real (Weaviate is a widely-used production vector database, $67M+ raised) but modest relative to labs that produced canonical LM research.\n\nVan Luijt's foundational contribution to today's frontier stack is indirect: Weaviate is a widely-used open-source vector database (HNSW ANN, hybrid vector+BM25, pluggable OpenAI/Cohere/HuggingFace vectorizers, RAG/reranking) that sits in the retrieval/inference layer many deployments wrap around frontier models — but he authored none of the load-bearing methods (word2vec, attention, transformers, the HNSW algorithm itself, tokenizers, scaling or alignment), and there is no pre-2013 vector-space/distributional-semantics patent or shipped system in his record, so he is a consumer-and-integrator of the lineage, not part of its foundation. His language-modeling depth rests on ~10 years (2016→present) building and shipping a production vector-space text-retrieval system he originally architected himself before hiring a CTO — genuine hands-on semantic-search/vector-space work, though systems integration rather than model authorship. The breadth is essentially one domain — natural-language text semantic search; Weaviate is a horizontal tool users apply to code/other domains, but that is downstream application, not his own modeling in those domains. As scientific/technical founder he counts one company in this field (SeMI Technologies, renamed Weaviate, founded 2016) where he personally wrote the initial engine and set technical direction, ~10 years, sharing later engineering with co-founder/CTO Etienne Dilocker.","evidence":[{"claim":"Started the open-source vector search engine Weaviate in March 2016, predating the post-ChatGPT vector-database wave","source_url":"https://en.wikipedia.org/wiki/Bob_van_Luijt","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar profile: 3 papers, 38 citations, h-index 1","source_url":"https://www.semanticscholar.org/author/2030042874","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Weaviate is a Go vector database using HNSW approximate nearest-neighbour search, hybrid semantic + BM25 keyword search, vector compression/quantization, pluggable vectorizers (OpenAI, Cohere, HuggingFace), RAG and reranking; ~16.8k GitHub stars","source_url":"https://github.com/weaviate/weaviate","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'StructuredRAG: JSON Response Formatting with Large Language Models' (2024), benchmarking Gemini 1.5 Pro and Llama 3 8B-instruct on structured output following across 24 experiments","source_url":"https://arxiv.org/abs/2408.11061","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record lists only three works: IRPAPERS (2026, visual document benchmark for scientific retrieval and QA), StructuredRAG (2024), and a 2020 Journal of Creating Value interview — 3 papers, 38 citations, h-index 1","source_url":"https://api.semanticscholar.org/graph/v1/author/2030042874/papers?fields=title,year,venue,externalIds,authors&limit=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Studied jazz and electronic composition at ArtEZ and Berklee College of Music (no technical degree); launched Weaviate as an open-source project in March 2016 and founded SeMI Technologies, later renamed Weaviate; authored an IEEE Software article (2020) 'Bringing Semantic Knowledge Graph Technology","source_url":"https://en.wikipedia.org/wiki/Bob_van_Luijt","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bob van Luijt is co-founder and CEO of Weaviate, an open-source vector database; started the project in March 2016 and founded SeMI Technologies, later renamed Weaviate","source_url":"https://en.wikipedia.org/wiki/Bob_van_Luijt","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record lists three works (IRPAPERS 2026, StructuredRAG 2024, a 2020 interview) — 3 papers, 38 citations, h-index 1; no OpenAlex or patent record","source_url":"https://www.semanticscholar.org/author/2030042874","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Weaviate, an open-source vector database; original architect who started the open-source project in March 2016","source_url":"https://en.wikipedia.org/wiki/Bob_van_Luijt","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Weaviate is a Go vector database using HNSW ANN search, hybrid semantic + BM25 search, quantization, pluggable vectorizers (OpenAI/Cohere/HuggingFace), RAG and reranking — retrieval infrastructure over embeddings, not a model frontier LMs are trained on","source_url":"https://github.com/weaviate/weaviate","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar record is 3 papers / 38 citations / h-index 1 (StructuredRAG 2024, IRPAPERS, a 2020 interview) — applied RAG/benchmark writing, no distributional-semantics or transformer research authorship, and no OpenAlex/PubMed/patent record","source_url":"https://www.semanticscholar.org/author/2030042874","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["ArtEZ University of the Arts (music)","Berklee College of Music","Harvard Business School, Program of Management Excellence"],"canonical_papers":[],"first_verifiable_year":2016,"notable_systems":["Weaviate open-source vector database"],"citations":38,"h_index":1,"patents":0,"dossier_notes":"Dossier has no OpenAlex match (found:false) and no PubMed hits, consistent with him having no peer-reviewed academic research record; Wikipedia/Wikidata identity is unambiguous and correctly matched (Q25346162, notable_work Weaviate). His only real credential-adjacent claim is founding/building Weaviate itself starting 2016, which is genuine hands-on engineering directly in the vector_embeddings space even without formal academic training — scored on that basis rather than on business success or funding raised, per the rubric's instruction not to reward fundraising.","years_language_modeling":10,"years_as_technical_founder":10,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2016-2026","evidence":"Weaviate open-source vector database for semantic text search / RAG, originally architected and built by Van Luijt"}],"frontier_lineage":["Weaviate vector database — retrieval/RAG inference-layer infrastructure the ecosystem uses around frontier models, but built on others' methods (HNSW, external embedding models); not a foundational LM building block"],"technical_founder_roles":["SeMI Technologies / Weaviate — founder, original architect, later CEO — 2016-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":6,"lm_domain_depth":12,"lm_domain_breadth":5,"scientific_founder":13},"confidence":0.68,"duration_ms":73211},{"pass":"pass_2","dimensions":{"frontier_founder":5,"lm_domain_depth":13,"lm_domain_breadth":4,"scientific_founder":13},"confidence":0.76,"duration_ms":71945}],"source":"seeded","profile_url":null,"wikipedia_title":"Bob van Luijt","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jacob-steeves","name":"Jacob Steeves","title":"Co-founder","company":"Bittensor / Opentensor Foundation","sector":"crypto","rank":45,"rubric_version":4,"dimensions":{"foundations":7,"vector_embeddings":8,"transformers_lm":12,"frontier_founder":6,"lm_domain_depth":10,"lm_domain_breadth":4,"hands_on_engineering":14,"industry_impact":9,"scientific_founder":11},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":44,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":44,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.82,"rationale":"This is the batch's clearest case of one pass simply missing the record. Pass 1 scored him from a dossier that found a single 2025 OpenAlex work and concluded his first verifiable year was 2025 with essentially no publication history; pass 2 pulled the Semantic Scholar author record and found six more papers, all of which I verified. He is first author of 'BitTensor: An Intermodel Intelligence Measure' (2020) and a named co-author of 'BitTensor: A Peer-to-Peer Intelligence Market' (Rao, Steeves, Shaabana, Attevelt, McAteer, arXiv:2003.03917, March 2020), which prices model contributions by the information-theoretic value one neural network adds to another's representations — though the arXiv version was withdrawn by the authors as incomplete and obsolete, which caps the credit it earns. His strongest credential is co-authorship of 'BTLM-3B-8K: 7B Parameter Performance in a 3B Parameter Model' (arXiv:2309.11568, 2023) with the Cerebras team, where he appears as 'Jacob Robert Steeves': that is direct participation in pretraining a real 3B-parameter open-weights language model with 8K context, not commissioning one, and it is what moves transformers_lm decisively out of pass 1's band of 4. His 2025 paper 'Incentivizing Permissionless Distributed Learning of LLMs' introduces Gauntlet, filtering peers by loss improvement attributable to each peer's pseudo-gradient, and reports training a 1.2B-parameter model that way — genuine distributed-training engineering. Against this: no verifiable degree, no thesis, no foundational mathematics record, a small citation footprint (43 citations, h-index 3), and work that is predominantly mechanism design over machine learning rather than contributions to attention, embeddings or scaling themselves. Pass 2's scores are directionally right but uniformly a notch generous for a record this thin in citations and absent any credentialed mathematics.\n\nSteeves is a genuine hands-on participant in the language-modeling lineage but not a foundational building block of it: his verifiable record starts in March 2020 with the two BitTensor papers, runs through named co-authorship of the BTLM-3B-8K 3B-parameter open-weights LM with the Cerebras team (2023), and continues into permissionless distributed LLM pretraining (Gauntlet, 2025) — roughly six years, all in natural-language modeling and the incentive/mechanism-design layer around it, no pre-2013 vector-space precursor and no named block (attention, tokenizer, optimizer, scaling law) that frontier GPT/Claude/Gemini/Llama systems descend from, so frontier_founder sits low in the 3-7 band. His language-modeling depth is a real ~6-year record with a small citation footprint (43 citations, h-index 3), placing lm_domain_depth in the 3-8-year (8-12) band. His personal hands-on modeling is confined to a single domain, natural-language text (BTLM-3B-8K and distributed LLM training); Bittensor's subnets span many domains but that is the network's community output, not his own modeling work, so lm_domain_breadth is a single-domain 3-7. As co-founder/CTO of the Opentensor Foundation he personally authored the core BitTensor whitepapers and mechanism, operating as the technical founder of one company for about six years, which lands scientific_founder in the 3-8-year single-company (8-12) band.","evidence":[{"claim":"Semantic Scholar author 1557385586 (name_exact, 1 candidate) lists 7 papers, 43 citations, h-index 3: 'Incentivizing Permissionless Distributed Learning of LLMs' (2025, ICDAI), 'Poster: Solving the Free-rider Problem in Bittensor' (CCS 2024), 'BTLM-3B-8K' (2023), 'BitTensor: An Intermodel Intelligen","source_url":"https://api.semanticscholar.org/graph/v1/author/1557385586/papers?fields=title,year,venue,externalIds,authors,citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'BTLM-3B-8K: 7B Parameter Performance in a 3B Parameter Model' (arXiv:2309.11568, 2023) — author list verified as including 'Jacob Robert Steeves' with the Cerebras team (Dey, Soboleva, Al-Khateeb, Vassilieva, Hestness); the paper introduces the Bittensor Language Model, a 3B-parameter","source_url":"https://arxiv.org/abs/2309.11568","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'BitTensor: A Peer-to-Peer Intelligence Market' (arXiv:2003.03917, submitted 9 March 2020, authors Yuma Rao, Jacob Steeves, Ala Shaabana, Daniel Attevelt, Matthew McAteer), proposing peers ranking one another through trained neural networks; withdrawn by the authors as incomplete and ob","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Incentivizing Permissionless Distributed Learning of LLMs' (arXiv:2505.21684, 2025), introducing Gauntlet for permissionless distributed pretraining, reporting a 1.2B-parameter model trained this way","source_url":"https://arxiv.org/abs/2505.21684","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar clean match: 7 papers, 43 citations, h-index 3, name_exact match with 1 candidate (low homonym risk)","source_url":"https://www.semanticscholar.org/author/Jacob-Steeves/1557385586","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bittensor's 2021 founding whitepaper (peer-ranking/incentive mechanism for distributed ML) was published under the pseudonym 'Yuma Rao,' a collective pseudonym publicly reported to include the founding team","source_url":"https://bittensor.com/whitepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PubMed search for Steeves J with Bittensor/Opentensor affiliation returned 227 results flagged as high homonym risk — not usable as evidence for this person","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Steeves+J","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'BitTensor: A Peer-to-Peer Intelligence Market' (Rao, Steeves, Shaabana, Attevelt, McAteer, arXiv:2003.03917, March 2020), proposing peers pricing each other's model contributions; subsequently withdrawn by the authors as incomplete","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 1557385586 (Jacob Steeves, name_exact, 1 candidate): 7 papers, 43 citations, h-index 3, including 'BitTensor: An Intermodel Intelligence Measure' (2020), 'Poster: Solving the Free-rider Problem in Bittensor' (CCS 2024), 'Stake-Based Consensus for Utility Scoring' and 'Incenti","source_url":"https://api.semanticscholar.org/graph/v1/author/1557385586/papers?fields=title,year,venue,externalIds,authors,citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author (as 'Jacob Robert Steeves') of BTLM-3B-8K, a real 3B-parameter open-weights pretrained language model built with the Cerebras team — direct participation in LM pretraining, the frontier lineage","source_url":"https://arxiv.org/abs/2309.11568","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Incentivizing Permissionless Distributed Learning of LLMs' (2025), reporting a 1.2B-parameter model trained via the Gauntlet mechanism — distributed LLM training work, natural-language domain","source_url":"https://arxiv.org/abs/2505.21684","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 1557385586 (name_exact, 1 candidate): 7 papers, 43 citations, h-index 3 — small footprint, all in the LM/ML-incentive area, natural-language text only","source_url":"https://www.semanticscholar.org/author/Jacob-Steeves/1557385586","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bittensor's founding whitepaper (peer-ranking/incentive mechanism for distributed ML) under the collective pseudonym 'Yuma Rao'; Steeves is a co-founder/technical founder of the Opentensor Foundation running the network","source_url":"https://bittensor.com/whitepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["BitTensor: An Intermodel Intelligence Measure (2020)","BitTensor: A Peer-to-Peer Intelligence Market (arXiv:2003.03917, 2020, withdrawn by authors)","BTLM-3B-8K: 7B Parameter Performance in a 3B Parameter Model (arXiv:2309.11568, 2023)","Incentivizing Permissionless Distributed Learning of LLMs (arXiv:2505.21684, ICDAI 2025)"],"first_verifiable_year":2020,"notable_systems":["Bittensor / TAO network and subnet incentive mechanism","BTLM-3B-8K language model","Gauntlet permissionless distributed training"],"citations":43,"h_index":3,"patents":0,"dossier_notes":"The dossier materially undercounts him: its OpenAlex match A5121026310 found only the 2025 paper and set first_verifiable_year to 2025 with 1 year active; the correct floor is 2020 via the two BitTensor papers, about six years active. The OpenAlex 'Open Society Foundations' affiliation is spurious. The Semantic Scholar record (7 papers) was available but not reflected in the timeline. PubMed's 227 hits are correctly flagged homonym risk.","adjudication_notes":["All five dimensions (pass 1 uniformly 7-9 points lower): pass 2 was right on the facts and pass 1 was scoring a record it could not see. The deciding verified fact is BTLM-3B-8K (arXiv:2309.11568), where 'Jacob Robert Steeves' is a named co-author of a real 3B-parameter pretrained language model built with the Cerebras team — pass 1 explicitly recorded first_verifiable_year 2025 and 'no personal academic publication record before this year', which the March 2020 arXiv submission and the 2023 Cerebras paper both disprove.","transformers_lm (4 vs 13): settled at 12. Participation in pretraining a real open-weights LM is squarely in the lineage, but he is one of thirteen authors and contributed no architecture or scaling result.","foundations (3 vs 10): settled at 7, below pass 2. No degree, thesis or mathematics publication was found by either pass or by me; pass 2's 10 credits the information-theoretic framing of the BitTensor papers, one of which its own evidence notes was withdrawn as incomplete. That does not reach the 'strong graduate training' anchor.","vector_embeddings (3 vs 11): settled at 8. The intermodel-intelligence measure does operate on representations, which pass 1 missed, but he has shipped no embedding, retrieval or vector-search system, so pass 2's 11 overshoots.","industry_impact (5 vs 12): settled at 9. Bittensor is a real network with real technical output, but it has produced no work the mainstream field builds on, and token market cap is explicitly not evidence."],"years_language_modeling":6,"years_as_technical_founder":6,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2020-2026","evidence":"BTLM-3B-8K 3B open-weights LM (arXiv:2309.11568, 2023) and distributed LLM pretraining / Gauntlet (arXiv:2505.21684, 2025)"}],"frontier_lineage":["Participated in pretraining the BTLM-3B-8K open-weights language model (downstream LM, not a component frontier labs build on)","Distributed/permissionless LLM training mechanism (Gauntlet) — niche, not in the frontier stack"],"technical_founder_roles":["Bittensor / Opentensor Foundation — co-founder & CTO, authored the core BitTensor whitepapers and mechanism — ~2020-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":6,"lm_domain_depth":10,"lm_domain_breadth":4,"scientific_founder":11},"confidence":0.62,"duration_ms":51228},{"pass":"pass_2","dimensions":{"frontier_founder":5,"lm_domain_depth":10,"lm_domain_breadth":5,"scientific_founder":11},"confidence":0.6,"duration_ms":64741}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jeff-huber","name":"Jeff Huber","title":"Co-founder & CEO","company":"Chroma","sector":"general","rank":46,"rubric_version":4,"dimensions":{"foundations":5,"vector_embeddings":15,"transformers_lm":6,"frontier_founder":5,"lm_domain_depth":8,"lm_domain_breadth":4,"hands_on_engineering":14,"industry_impact":12,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":43,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":43,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.8,"rationale":"Both passes correctly rejected the dossier, which is worthless for this person: its only identity anchor, Wikidata Q115655617, is the CEO of Home Instead Inc., a home-care company, and its Semantic Scholar match (24 papers, 4 candidates, no affiliation) cannot be tied to him — there is also a separate well-known Jeff Huber who was a Google SVP and Grail co-founder, so none of those papers are counted. Identification is via GitHub and Chroma's own materials. The dispute is about how hands-on he is, and pass 2 supplied the measurement pass 1 lacked: GitHub's commit search API attributes 434 commits in chroma-core/chroma to his account jeffchuber, which I re-verified directly. That is the difference between managing builders and being one, and it settles hands_on_engineering and vector_embeddings in pass 2's favour — Chroma is an AI-native open-source embeddings database, ~29k stars, whose whole purpose is storing and querying embeddings for retrieval, and he is a substantial contributor to it, not merely its spokesperson. Chroma also publishes retrieval research (Embedding Adapters, Evaluating Chunking Strategies, Generative Benchmarking, Context Rot) that sits in the dense-retrieval and long-context line, but the reports do not list him as an author, so that counts as leading a group producing relevant work rather than personal authorship. Where pass 2 goes too far is foundations and transformers_lm: it scored both at 8 while its own rationale concedes no peer-reviewed publication, no verifiable degree and no personal contribution to attention, pretraining or scaling. An 8 sits in the 'strong graduate training' band, which nothing in the record supports; the rubric's instruction is to score lower when unsure. Prior work at Standard Cyborg (3D scanning and computer vision) is real applied engineering but outside the lineage.\n\nHuber co-founded Chroma (2022) and is a hands-on builder of it (434 verified commits to chroma-core/chroma), an open-source embeddings/vector database in the dense-retrieval line — but a vector DB is infrastructure used ALONGSIDE frontier models (RAG), not a method, architecture, dataset or named block that GPT/Claude/Gemini-class models descend from, and there is no pre-2013 vector-space/distributional-semantics patent or shipped system in his record, so his frontier-lineage position is that of an applied retrieval-infra builder, not a foundation author. His verifiable language-modeling record is ~4 years, all vector-space/embeddings text retrieval via Chroma (2022→2026); Chroma publishes retrieval/long-context research (Embedding Adapters, Context Rot) but he is not listed as an author, so that is org leadership, not personal LM authorship. His hands-on domain is a single language-modeling domain — general natural-language text embeddings — while his prior company Standard Cyborg (YC W15, 3D scanning/computer vision) is outside the language-modeling domains, giving one distinct LM domain. As a technical founder he is genuine and personally writes the core code: ~10 total years across two companies he founded in a technical/CEO role (Standard Cyborg 2015 CV, Chroma 2022 embeddings), but only Chroma (~4 yrs) has an in-field core, placing scientific_founder in the strong end of the one-in-field band rather than the multi-company/15-year top band.","evidence":[{"claim":"GitHub's commit search API attributes 434 commits in chroma-core/chroma to author jeffchuber — verified total_count directly","source_url":"https://api.github.com/search/commits?q=author:jeffchuber+repo:chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub user jeffchuber is Jeff Huber of San Francisco, associated with the StandardCyborg organization, with chroma-core/chroma ('Search infrastructure for AI', ~29k stars) pinned to his profile","source_url":"https://github.com/jeffchuber","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chroma is an AI-native open-source embeddings database founded by Jeff and Anton to store and query embeddings with filtering for embedding-based document retrieval","source_url":"https://www.trychroma.com/blog/seed","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"chroma-core/chroma repository created 2022-10-05, ~29k stars, primary language Rust","source_url":"https://api.github.com/repos/chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chroma publishes technical retrieval research including Embedding Adapters (2024), Evaluating Chunking Strategies for Retrieval (2024), Generative Benchmarking (2025) and Context Rot (2025); Huber is not listed as an author on these reports","source_url":"https://www.trychroma.com/research","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's Wikidata match Q115655617 is 'President and Chief Executive Officer - Home Instead Inc.', a home-care company unrelated to Chroma — a clear homonym","source_url":"https://www.wikidata.org/wiki/Q115655617","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chroma is described as 'Search infrastructure for AI', an open-source embedding database; Jeff Huber's GitHub profile (jeffchuber) has it pinned with ~29k stars, and lists 34 public repositories and active contribution badges (Pull Shark, Pair Extraordinaire)","source_url":"https://github.com/jeffchuber","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jeff Huber's GitHub profile lists prior affiliation with the organization StandardCyborg, a 3D-scanning/computer-vision company","source_url":"https://github.com/jeffchuber","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dossier's Wikidata match for 'Jeff Huber' (Q115655617) is labeled 'President and Chief Executive Officer - Home Instead Inc.', a senior home-care company unrelated to Chroma or AI — a clear homonym, not this person","source_url":"https://www.wikidata.org/wiki/Q115655617","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dossier's Semantic Scholar match for 'Jeff Huber' shows 24 papers / 138 citations / h-index 2 with 4 candidates and no affiliation data, an unresolved identity match not corroborated as this Jeff Huber by any independent source found","source_url":"https://www.semanticscholar.org/author/40441754","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jeff Huber co-founded Chroma, an open-source embeddings/vector database ('Search infrastructure for AI', ~29k stars), and has 434 verified commits to chroma-core/chroma","source_url":"https://api.github.com/search/commits?q=author:jeffchuber+repo:chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chroma is an AI-native open-source embeddings database (dense-retrieval infra), created 2022-10-05 — post-word2vec application/infra, not a foundational block frontier models descend from","source_url":"https://api.github.com/repos/chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Huber co-founded Standard Cyborg (Y Combinator W2015), a 3D-scanning/computer-vision company — a technical founder role outside the language-modeling domains","source_url":"https://www.ycombinator.com/companies/standard-cyborg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chroma's published retrieval/long-context research (Embedding Adapters, Evaluating Chunking Strategies, Generative Benchmarking, Context Rot) does not list Huber as an author","source_url":"https://www.trychroma.com/research","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub commit search attributes 434 commits in chroma-core/chroma to author jeffchuber (Jeff Huber), the person's own hands-on record on the embeddings database","source_url":"https://api.github.com/search/commits?q=author:jeffchuber+repo:chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"chroma-core/chroma created 2022-10-05, ~29k stars — bounding Huber's verifiable years in the embeddings/retrieval field at ~4","source_url":"https://api.github.com/repos/chroma-core/chroma","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jeff Huber's GitHub profile lists prior affiliation with StandardCyborg, a 3D-scanning/computer-vision company — a technical-founder role outside language modeling","source_url":"https://github.com/jeffchuber","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2022,"notable_systems":["Chroma open-source embeddings database","Standard Cyborg (3D scanning / computer vision)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier is unusable for this person. Wikidata Q115655617 is the wrong person (Home Instead Inc. CEO). The Semantic Scholar match (id 40441754, 24 papers, h-index 2, 4 candidates, no affiliation) cannot be tied to him and is presumed homonym contamination, further confused by the separate Jeff Huber who was a Google SVP and Grail co-founder. PubMed returned 2,213 hits flagged homonym_risk. OpenAlex and Wikipedia found nothing and the timeline is entirely null. No education could be confirmed from any primary source by either pass or by me, so the field is left empty rather than guessed.","adjudication_notes":["hands_on_engineering (13 vs 14) and vector_embeddings (13 vs 15): pass 2 was right and settled at its numbers. The deciding verified fact is GitHub's commit API returning total_count 434 for author jeffchuber in chroma-core/chroma — pass 1 inferred hands-on involvement from a pinned repository and badges, while pass 2 measured it. Chroma's core product is an embeddings database, placing that contribution directly in the dimension.","foundations (4 vs 8): pass 1 was right; settled at 5. No degree, thesis or mathematics publication is verifiable from any source, and pass 2's own rationale concedes this while still scoring 8, which is inside the 'strong graduate training' band the evidence does not reach.","transformers_lm (5 vs 8): pass 1 was right; settled at 6. Chroma's published research is retrieval and long-context evaluation and does not list him as an author; he has no personal work on attention, pretraining or scaling. A small credit above pass 1 reflects leading an org whose Context Rot report addresses LLM long-context behaviour directly.","industry_impact (12 vs 12): agreed, held at 12."],"years_language_modeling":4,"years_as_technical_founder":10,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2022-2026","evidence":"Chroma open-source embeddings/vector database for text retrieval; 434 verified commits by jeffchuber"}],"frontier_lineage":["Chroma vector/embeddings database — dense-retrieval (RAG) infrastructure used alongside frontier LLMs; an application/infra layer, not a foundational method or block those models descend from"],"technical_founder_roles":["Standard Cyborg — co-founder/CEO (computer vision, 3D scanning), 2015-~2021 — outside the LM field","Chroma — co-founder/CEO and hands-on builder (embeddings database), 2022-2026 — in-field"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":5,"lm_domain_depth":8,"lm_domain_breadth":4,"scientific_founder":11},"confidence":0.55,"duration_ms":132904},{"pass":"pass_2","dimensions":{"frontier_founder":5,"lm_domain_depth":8,"lm_domain_breadth":4,"scientific_founder":10},"confidence":0.6,"duration_ms":57404}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"clement-delangue","name":"Clément Delangue","title":"Co-founder & CEO","company":"Hugging Face","sector":"general","rank":47,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":7,"transformers_lm":9,"frontier_founder":10,"lm_domain_depth":10,"lm_domain_breadth":4,"hands_on_engineering":10,"industry_impact":16,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":41,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":41,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.85,"rationale":"The passes agree on the shape of the record and differ by small margins on how much credit a non-first-author library-paper byline and a personal model-training habit earn. I verified the byline: the ACL Anthology record for 'Transformers: State-of-the-Art Natural Language Processing' (EMNLP 2020 demo, ~8.3k citations) lists him fifth of 22 authors, after Wolf, Debut, Sanh and Chaumond. That is genuine inclusion on a paper documenting a library the entire field uses, but the technical authorship sits with the engineering team, and his OpenAlex h-index of 5 is almost entirely these two library artifacts, so it cannot be read as a personal research record. He has no doctorate, no thesis and no paper in linear algebra, optimization or statistical learning; his education is a Master in Management from ESCP plus a non-degree Stanford extension programming course. Pass 2 adds one fact pass 1 lacked and I confirmed it: his personal Hugging Face account 'clem' publishes 13 models and 24 datasets, including SmolLM2 SFT and Qwen2.5-1.5B fine-tunes — modest but real hands-on model-training practice rather than pure delegation, which is why hands_on_engineering lands nearer pass 2's number than pass 1's. Pass 2's vector_embeddings of 9 is the one score I reduce: neither pass cited any embedding artifact authored by him, and the Hub hosting embedding models is organisational, not personal. His overwhelming strength is industry impact, and it is legitimate under this rubric for a reason distinct from his fame: the Transformers library, the Hub and Datasets are the distribution layer through which essentially every transformer and embedding model in the field now ships, and that core is exactly the systems the rubric measures. None of his valuation, fundraising or media presence is counted.\n\nDelangue is a listed co-author (5th of 22, after Wolf, Debut, Sanh and Chaumond) on the Transformers library paper and a co-author on the Datasets library paper — genuine lineage credit because those libraries are the dominant training/inference and data stack through which essentially every frontier and open transformer model now ships, but his personal contribution is founder/product leadership rather than the architecture, optimizer, tokenizer or objective, so this is shared organizational lineage work, not a named block he authored (frontier_founder in the 8-12 band). His verifiable personal hands-on language-modeling record runs from the 2019 Transformers/TransferTransfo work through recent personal SmolLM2 and Qwen2.5-1.5B fine-tunes on his 'clem' account — roughly seven years, all in natural-language text, with no biological, financial, code or chemistry modeling of his own, so depth sits at 3-8 years (11) and breadth is a single domain (5). He co-founded Hugging Face (2016) and is CEO, but the science and engineering are led by CSO Thomas Wolf and CTO Julien Chaumond — the classic 'founder/CEO of an AI company whose science was done by others' case — so he does not earn the scientific/technical-founder dimension beyond the 3-7 floor, with zero companies founded in a technical-founder role. The PubMed hits (2009/2017 emergency medicine) are a homonym clinician and are excluded from every count.","evidence":[{"claim":"ACL Anthology record for 'Transformers: State-of-the-Art Natural Language Processing' (EMNLP 2020 demo) lists Clement Delangue fifth among 22 authors, after Wolf, Debut, Sanh and Chaumond","source_url":"https://aclanthology.org/2020.emnlp-demos.6/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed among the 22 authors of 'HuggingFace's Transformers: State-of-the-art Natural Language Processing' (arXiv:1910.03771, 2019)","source_url":"https://arxiv.org/abs/1910.03771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His personal Hugging Face account 'clem' publishes 13 models and 24 datasets, including smollm2-135m-sft-tiny, cifar10-vit-poc and macron-style-qwen2.5-1.5B — verified as Clément Delangue's account via linked handles clementdelangue and clmnt","source_url":"https://huggingface.co/clem","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5009717413 (Hugging Face affiliation): 10 works, 13,262 citations, h-index 5, dominated by the Transformers and Datasets library papers","source_url":"https://api.openalex.org/authors/A5009717413","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hugging Face was founded in 2016 by Delangue (CEO), Julien Chaumond (CTO) and Thomas Wolf (CSO), starting as a chatbot before pivoting to an ML platform after open-sourcing the model","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on 'Datasets: A Community Library for Natural Language Processing' (EMNLP 2021 demo)","source_url":"https://aclanthology.org/2021.emnlp-demo.21/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Master in Management, ESCP Business School (2008-2012); non-degree Stanford intro-CS extension course (2011-2012)","source_url":"https://www.clay.com/dossier/hugging-face-ceo","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Hugging Face, became CEO in July 2016; company builds the Transformers open-source library","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed co-author on 'Transformers: State-of-the-Art Natural Language Processing' (EMNLP 2020), 8318 citations per OpenAlex","source_url":"https://doi.org/10.18653/v1/2020.emnlp-demos.6","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on 'HuggingFace's Transformers: State-of-the-art Natural Language Processing' arXiv 1910.03771 (2019), 3149 citations","source_url":"https://doi.org/10.48550/arxiv.1910.03771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ACL Anthology lists Clément Delangue 5th of 22 authors on 'Transformers: State-of-the-Art Natural Language Processing' (EMNLP 2020 demo), the paper documenting the Transformers library used across the field","source_url":"https://aclanthology.org/2020.emnlp-demos.6/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author on 'TransferTransfo: A Transfer Learning Approach for Neural Network Based Conversational Agents' (arXiv:1901.08149, 2019) — his earliest verifiable personal language-modeling work","source_url":"https://arxiv.org/abs/1901.08149","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Personal Hugging Face account 'clem' publishes fine-tuned natural-language models (SmolLM2 SFT, Qwen2.5-1.5B) — modest personal LM practice, all in natural-language text","source_url":"https://huggingface.co/clem","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hugging Face was co-founded in 2016 by Delangue (CEO), Julien Chaumond (CTO) and Thomas Wolf (CSO) — the science/engineering is led by the CTO/CSO, not the CEO","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ACL Anthology lists Delangue 5th of 22 authors on the Transformers library paper (EMNLP 2020 demo, ~8.3k citations); technical authorship is the engineering team (Wolf, Debut, Sanh, Chaumond)","source_url":"https://aclanthology.org/2020.emnlp-demos.6/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hugging Face founded 2016 by Delangue (CEO), Julien Chaumond (CTO) and Thomas Wolf (CSO); the Transformers open-source library is the org's core NLP product, placing science with Wolf/Chaumond","source_url":"https://en.wikipedia.org/wiki/Hugging_Face","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His personal 'clem' account publishes 13 models and 24 datasets including SmolLM2 and Qwen2.5-1.5B fine-tunes — modest hands-on natural-language modeling practice","source_url":"https://huggingface.co/clem","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5009717413: 10 works, h-index 5, dominated by the Transformers and Datasets library papers; earliest LM-lineage publication 2018-2019","source_url":"https://api.openalex.org/authors/A5009717413","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Master in Management, ESCP Business School","Non-degree introductory programming course, Stanford (extension)"],"canonical_papers":["Transformers: State-of-the-Art Natural Language Processing (EMNLP 2020 demo) — 5th of 22 authors","Datasets: A Community Library for Natural Language Processing (EMNLP 2021 demo)"],"first_verifiable_year":2019,"notable_systems":["Hugging Face Transformers library","Hugging Face Hub","Datasets library"],"citations":13262,"h_index":5,"patents":0,"dossier_notes":"The OpenAlex block A5009717413 is the correct person and correctly affiliated, though the phantom affiliations 'Bio Signal Group' and 'FACE Foundation' are string-matching noise on the name 'Face'. The Wikipedia and Wikidata blocks are for the COMPANY Hugging Face (Q108943604), not the person, so the dossier carries no biographical evidence about Delangue himself. The PubMed block is entirely a homonym — a French emergency-medicine clinician (morphine side effects, digoxin poisoning) — which also corrupts the timeline.","adjudication_notes":["hands_on_engineering (8 vs 11): pass 2 was right; settled at 10. The deciding verified fact is his personal Hugging Face account 'clem', which publishes 13 models and 24 datasets including SmolLM2 and Qwen2.5-1.5B fine-tunes — real if modest personal model-training practice that pass 1 did not find.","transformers_lm (8 vs 10): split at 9. Verified: the ACL Anthology record places him 5th of 22 on the Transformers demo paper — genuine inclusion, but the technical authorship is Wolf, Debut, Sanh and the engineering team, so neither pass's extreme is right.","vector_embeddings (6 vs 9): pass 1 was right; settled at 7. Neither pass cited an embedding or retrieval artifact authored by him; hosting embedding models on the Hub is organisational credit already captured in industry_impact.","foundations (3 vs 5): split at 4. Verified education is a management master's plus a non-degree programming course, with no mathematics or statistical-learning record of any kind.","industry_impact (15 vs 16): settled at 16. The Transformers library and Hub are the distribution layer for the field's models, which is core to what this rubric measures — credited for the infrastructure, not for valuation or media presence."],"years_language_modeling":7,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2019-2026","evidence":"Transformers & Datasets library co-authorship, TransferTransfo conversational-agent paper, personal SmolLM2/Qwen2.5 fine-tunes on 'clem' account"}],"frontier_lineage":["Co-author of the Transformers library — the dominant training/inference stack through which frontier and open transformer models ship","Co-author of the Datasets library used for LM pretraining and evaluation data"],"technical_founder_roles":["Hugging Face — co-founder & CEO (business/product founder; CSO Thomas Wolf and CTO Julien Chaumond lead the science) — 2016-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":10,"lm_domain_depth":11,"lm_domain_breadth":5,"scientific_founder":5},"confidence":0.72,"duration_ms":86007},{"pass":"pass_2","dimensions":{"frontier_founder":9,"lm_domain_depth":10,"lm_domain_breadth":4,"scientific_founder":4},"confidence":0.8,"duration_ms":65630}],"source":"seeded","profile_url":null,"wikipedia_title":"Hugging Face","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"robin-li","name":"Robin Li (Li Yanhong)","title":"Co-founder, Chairman & CEO","company":"Baidu","sector":"general","rank":48,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":8,"transformers_lm":3,"frontier_founder":11,"lm_domain_depth":9,"lm_domain_breadth":4,"hands_on_engineering":9,"industry_impact":12,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":41,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":41,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.82,"rationale":"Both passes discarded the dossier's research blocks as homonyms — correctly: its OpenAlex match is a University of Pennsylvania bioengineering researcher working on polyacrylamide gels and traction force microscopy, and the PubMed block draws on 44,906 'Li R' hits. Both also agree on the one genuinely personal technical artifact, and it is a real one: while at IDD Information Services in 1996 he invented the RankDex site-scoring algorithm and is the named inventor on US Patent 5,920,859, 'Hypertext Document Retrieval System and Method' (filed 5 February 1997, issued 6 July 1999), which ranks documents by hyperlink and anchor-text analysis and which Google's PageRank patent cites. He also published 'Toward a Qualitative Search Engine' in IEEE Internet Computing in 1998 and personally wrote search software at Dow Jones and worked as a staff engineer at Infoseek. The passes differ on how much this earns. Pass 2's framing is the more accurate one — link-analysis ranking is an eigenvector problem on the web graph and genuine first-principles information-retrieval work, which lifts foundations and hands-on engineering above the 'manages builders' band — but pass 1 is right that it is a single dated artifact: term- and graph-based retrieval, not learned vector representations, and his last hands-on technical contribution of record is roughly a quarter-century old, with everything since being executive. On transformers, pass 1 is right and pass 2 too generous: Baidu's ERNIE family is published by Baidu research staff (Sun, Wang, Li, Feng et al.), Deep Speech and PaddlePaddle likewise, and Baidu's own history shows dedicated technical leadership hired for AI research. Leading the company that funded those labs is industry impact, not personal authorship in the lineage, and fame and the scale of Baidu are explicitly not evidence.\n\nLi's one verifiable technical artifact in the lineage is the RankDex link-analysis ranking method (1996) and US Patent 5,920,859 'Hypertext Document Retrieval System and Method' (filed 1997, issued 1999), a pre-2013 shipped information-retrieval system that Google's PageRank patent references — genuine term/document-vector and web-graph retrieval work, but link-analysis ranking rather than the distributional-semantics/embedding lineage word2vec descends from, so it sits at the top of published-lineage / bottom of the precursor band rather than the named-block tier. His hands-on language-modeling record is a single domain — natural-language web search/IR (Dow Jones search software, Infoseek staff engineer, RankDex, early Baidu search) — spanning roughly 1996-2000 before he became a full-time executive; there is no verifiable biological, financial, code or other language-modeling domain in his personal record. As founder-CEO of Baidu (2000) he brought his own RankDex patent as the founding search core and set early technical direction for a few years, which earns credit as the technical founder of one company; but ERNIE, Deep Speech and PaddlePaddle are authored by Baidu research staff (Sun, Wang, Li, Feng et al.) with dedicated leadership hires such as Andrew Ng, not by Li personally. His last personal hands-on technical contribution of record is roughly a quarter-century old, which caps the depth and founder scores.","evidence":[{"claim":"Created the RankDex algorithm in 1996 at IDD Information Services; named inventor on US patent 5,920,859 'Hypertext Document Retrieval System and Method', filed 5 February 1997, issued 6 July 1999; Google's PageRank patent references it; earlier wrote search software at Dow Jones for the Wall Street","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"RankDex was the first search engine to use hyperlinks to measure site quality, predating PageRank by about two years, and the technology was carried into Baidu","source_url":"https://en.wikipedia.org/wiki/RankDex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Baidu's ERNIE pretrained language model family is published by Baidu research staff (Sun, Wang, Li, Feng et al.), not authored by Robin Li","source_url":"https://arxiv.org/abs/1904.09223","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Baidu appointed Andrew Ng as Chief Scientist in 2014 to lead AI research, indicating AI research leadership vested in dedicated technical hires rather than Li personally","source_url":"https://en.wikipedia.org/wiki/Baidu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Robin Li earned a master's degree in computer science from the University at Buffalo (SUNY), leaving the PhD program in 1994; prior degree (Bachelor of Management, Information Management) from Peking University.","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"While working at IDD Information Services (a Dow Jones & Company division, 1994-1997), Li created the RankDex site-scoring algorithm in 1996, which used hyperlink anchor-text analysis to rank search results; he later worked at Infoseek (1997-1999) as a staff engineer on image search.","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li published 'Toward a Qualitative Search Engine' in IEEE Internet Computing (July/August 1998), describing the RankDex approach; this predates and is referenced in relation to Larry Page's PageRank patent (filed 1998).","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Baidu appointed Andrew Ng as Chief Scientist in May 2014 to lead AI research, indicating Baidu's deep-learning/AI research leadership has been vested in dedicated technical hires rather than Robin Li personally; no source attributes personal authorship of ERNIE/Wenxin Yiyan LLM research to Li.","source_url":"https://en.wikipedia.org/wiki/Baidu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li created RankDex in 1996 and is the named inventor on US Patent 5,920,859 'Hypertext Document Retrieval System and Method' (filed 1997, issued 1999), a hyperlink/anchor-text ranking method referenced by Google's PageRank patent; the technology was carried into Baidu","source_url":"https://en.wikipedia.org/wiki/RankDex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li wrote search software at IDD/Dow Jones (1994-1997) and worked as a staff engineer at Infoseek (1997-1999), and published 'Toward a Qualitative Search Engine' in IEEE Internet Computing (1998)","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li co-founded Baidu in 2000 and serves as its CEO; Baidu's later ERNIE language-model family is authored by Baidu research staff, and Baidu appointed Andrew Ng as Chief Scientist in 2014 to lead AI research, indicating research leadership vested in dedicated hires rather than Li personally","source_url":"https://en.wikipedia.org/wiki/Baidu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"RankDex site-scoring algorithm (1996) and US Patent 5,920,859 'Hypertext Document Retrieval System and Method' (filed 1997, issued 1999), a pre-2013 shipped link-analysis retrieval system referenced by Google's PageRank patent; carried into Baidu","source_url":"https://en.wikipedia.org/wiki/RankDex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Li wrote search software at Dow Jones/IDD and was a staff engineer at Infoseek before co-founding Baidu in 2000, his personal hands-on IR/search work spanning ~1996-2000","source_url":"https://en.wikipedia.org/wiki/Robin_Li","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Baidu's ERNIE pretrained language-model family is authored by Baidu research staff (Sun, Wang, Li, Feng et al.), not by Robin Li","source_url":"https://arxiv.org/abs/1904.09223","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Baidu vested AI/deep-learning research leadership in dedicated hires such as Andrew Ng (Chief Scientist, 2014), not in Li personally","source_url":"https://en.wikipedia.org/wiki/Baidu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["MS Computer Science, University at Buffalo (SUNY); left PhD program 1994","BS Information Management, Peking University"],"canonical_papers":["Toward a Qualitative Search Engine (IEEE Internet Computing, 1998)"],"first_verifiable_year":1996,"notable_systems":["RankDex link-analysis ranking algorithm","Infoseek / Go.com search engineering","Baidu search"],"citations":0,"h_index":0,"patents":1,"dossier_notes":"Both research blocks are homonyms and were discarded. OpenAlex A5084605211 (2 works, 88 citations, University of Pennsylvania, topics 'Cellular Mechanics and Interactions', top paper on polyacrylamide gels for traction force microscopy, earliest year 2020) is a bioengineering researcher. The Semantic Scholar match (4 papers, 144 citations, 8 ambiguous candidates) could not be corroborated by co-author, venue or affiliation and is very likely the same wrong person. The PubMed block draws 44,906 raw 'Li R' hits. The dossier's timeline (first verifiable year 2020, 6 years active) is the bioengineer's; Li's real first verifiable technical year is 1996.","adjudication_notes":["transformers_lm (1 vs 5): pass 1 was right; settled at 3. The deciding verified fact is that the ERNIE papers are authored by Baidu research staff (Sun, Wang, Li, Feng et al.) with no authorship by Robin Li, and Baidu vested AI research leadership in dedicated hires such as Andrew Ng. Funding and directing a lab is industry impact, not personal work in the lineage.","foundations (6 vs 9): pass 2 was closer; settled at 8. The RankDex patent US 5,920,859 is a personally-invented link-analysis ranking method — an eigenvector/matrix problem on the web graph, and genuine first-principles retrieval work — but it is one artifact from 1996-97 by someone who left his PhD program, not a body of mathematical training.","hands_on_engineering (7 vs 11): split at 9. He verifiably wrote search software at Dow Jones and engineered at Infoseek, which pass 1 under-credits; but the record of personal building ends about 1999 and the rubric rewards depth of continuous hands-on work, which here stopped a quarter-century ago.","vector_embeddings (8 vs 8): agreed and held. The patent's dot-product ranking of query vectors against link vectors is pre-2013 vector-space lineage, which the rubric counts as foundational rather than dated.","industry_impact (12 vs 13): settled at 12. Baidu's search core and the labs that produced Deep Speech, PaddlePaddle and ERNIE are real impact, but the company's scale and his fame are not evidence under this rubric."],"years_language_modeling":4,"years_as_technical_founder":4,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"1996-2000","evidence":"RankDex link-analysis retrieval (US 5,920,859), Dow Jones/Infoseek search engineering, early Baidu web search"}],"frontier_lineage":["RankDex hyperlink/anchor-text ranking — a pre-2013 web-scale retrieval precursor (referenced by Google's PageRank patent); retrieval is a component frontier stacks build on, but link-analysis rather than the distributional-semantics embedding lineage"],"technical_founder_roles":["Baidu — founder-CEO, brought own RankDex patent as founding search core — 2000-~2004 hands-on technical direction, executive thereafter"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":10,"lm_domain_depth":9,"lm_domain_breadth":4,"scientific_founder":10},"confidence":0.72,"duration_ms":69934},{"pass":"pass_2","dimensions":{"frontier_founder":12,"lm_domain_depth":9,"lm_domain_breadth":4,"scientific_founder":10},"confidence":0.7,"duration_ms":70825}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Robin_Li","wikipedia_title":"Robin Li","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"himanshu-tyagi","name":"Himanshu Tyagi","title":"Co-founder","company":"Sentient","sector":"crypto","rank":49,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":8,"transformers_lm":8,"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"hands_on_engineering":10,"industry_impact":10,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":40,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":40,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.74,"rationale":"The real Himanshu Tyagi (co-founder of Sentient) is a Professor of Electrical Communication Engineering at the Indian Institute of Science (IISc) Bangalore, whose research home field is information theory, applied to cryptography, statistics, privacy, and federated learning — including authored work such as 'Wyner-Ziv Gradient Compression for Federated Learning' and distributed mean-estimation/communication-efficient federated-learning papers, and a Cambridge University Press book on information-theoretic cryptography. This gives strong mathematical/statistical-learning foundations (information theory is core to statistical learning) and a genuine, if narrow, machine-learning-adjacent research record (federated learning, distributed estimation) rather than direct transformer/LM or embeddings authorship. He co-founded Sentient (a decentralized/open-source AGI protocol, 'The Grid') while remaining an active IISc professor — real technical leadership of an AI infrastructure venture, though Sentient's core LLM/agent technology is a team effort, not solely his personal engineering output. Scored moderately: strong math foundations, real but adjacent (not core-lineage) ML research, and company leadership without a personally-authored canonical transformer/embeddings paper.\n\nTyagi's verifiable research home is information theory and its applications to distributed/federated machine learning (Wyner-Ziv gradient compression, communication-efficient distributed mean estimation, inference under information constraints) and information-theoretic cryptography — genuine, deep math, but NOT the attention→transformer→LM lineage, with no named block (embeddings, optimizer, tokenizer, attention, scaling/alignment) and no pre-2013 vector-space/distributional-semantics precursor work that frontier LLMs descend from, hence a low frontier_founder. His language-modeling record is thin and recent: it runs through decentralized-AI/model-ownership work (SAKSHI 2023, OML 2024) and Sentient (~2024), roughly 2 years, and is more model-ownership/infrastructure than hands-on language modeling — placing lm_domain_depth in the under-3-years/adjacent band and lm_domain_breadth at a single weak natural-language domain. He is, however, a real SCIENTIFIC/technical founder: a sitting IISc professor who personally co-authors the core research (OML, SAKSHI) his ventures run on, and a co-founder of TWO deep-tech companies — Witness Chain and Sentient — across roughly 3 years, which lifts scientific_founder into the lower-middle band while discounting for short tenure and the fact that neither company's core is language modeling per se.","evidence":[{"claim":"Professor of Electrical Communication Engineering, Indian Institute of Science (IISc), Bengaluru; research focus is information theory applied to cryptography, statistics, privacy, federated learning, and networks","source_url":"https://ece.iisc.ac.in/~htyagi/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored federated-learning research including communication-efficient distributed mean estimation and Wyner-Ziv gradient compression for federated learning","source_url":"https://par.nsf.gov/servlets/purl/10415421","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of a Cambridge University Press book on Information Theoretic Cryptography; his research group has won best-paper awards at IEEE ISIT (International Symposium on Information Theory) twice","source_url":"https://scholar.google.com/citations?user=OWMi2AQAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"IISc faculty page: Associate Professor, Department of Electrical Communication Engineering; PhD Electrical and Computer Engineering, University of Maryland 2013, advisor Prakash Narayan; dual degree IIT Delhi 2007; research in information theory, cryptography, privacy, federated learning, distribute","source_url":"https://ece.iisc.ac.in/~htyagi/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 143725642: 102 papers, 1,809 citations, h-index 26; works include Inference Under Information Constraints I/II/III, Estimating Renyi Entropy of Discrete Distributions (2014), Test without Trust (2018), RATQ universal fixed-length quantizer (2019), Secret Key Agreement (2014),","source_url":"https://api.semanticscholar.org/graph/v1/author/143725642","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution' — authors include Sandeep Nailwal, Sewoong Oh, Himanshu Tyagi and Pramod Viswanath; proposes AI-native model fingerprinting with crypto-economic enforcement","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'SAKSHI: Decentralized AI Platforms' (2023) — authors include Sreeram Kannan, Himanshu Tyagi and Pramod Viswanath; affiliations Princeton, UIUC, Tsinghua, HKUST, Witness Chain, EigenLayer","source_url":"https://arxiv.org/abs/2307.16562","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sentient describes itself as an open-source AI reasoning lab building OML (Open, Monetizable, Loyal AI), Arena and EvoSkill, and emphasises peer-reviewed research","source_url":"https://sentient.xyz/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sentient's own account: Himanshu Tyagi is a core contributor/co-founder of Sentient, Professor in ECE at IISc working on information theory and its applications to ML, cryptography and statistics, and prior co-founder of Witness Chain","source_url":"https://x.com/sentient_agi/status/1834648836528099552","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution — co-authored by Himanshu Tyagi with Sandeep Nailwal, Sewoong Oh, Pramod Viswanath (the core research Sentient is built on; model-ownership, not the transformer/embedding lineage)","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"SAKSHI: Decentralized AI Platforms (2023) — co-authored by Himanshu Tyagi; decentralized AI serving infrastructure, not language-model architecture","source_url":"https://arxiv.org/abs/2307.16562","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"IISc ECE faculty homepage: research in information theory, cryptography, privacy, federated learning and distributed estimation — not embeddings/transformers/LM pretraining","source_url":"https://ece.iisc.ac.in/~htyagi/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Professor of Electrical Communication Engineering, IISc Bengaluru; research focus is information theory applied to cryptography, statistics, privacy, federated learning and networks — not transformers/embeddings/LM","source_url":"https://ece.iisc.ac.in/~htyagi/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sentient is an open-source AI reasoning lab building OML, Arena and EvoSkill; Tyagi is a co-founder providing the academic/scientific direction alongside Pramod Viswanath","source_url":"https://sentient.xyz/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD, Arizona State University (per dossier Wikidata, unverified independently)","BTech, Indian Institute of Technology Delhi (per dossier Wikidata, unverified independently)"],"canonical_papers":["Wyner-Ziv Gradient Compression for Federated Learning","Communication-efficient distributed mean estimation with side information (federated learning)"],"first_verifiable_year":2008,"notable_systems":["Sentient / The Grid (decentralized AI agent network)"],"citations":null,"h_index":null,"patents":0,"dossier_notes":"CRITICAL: the dossier's OpenAlex match (id A5056763953, h-index 29, works on nanofluids/solar-thermal collectors/diesel-fuel ignition at IIT Ropar) is the WRONG homonym — confirmed via web search that the real Sentient co-founder Himanshu Tyagi is the IISc Bangalore ECE professor working on information theory/cryptography/federated learning, a completely different research area from the dossier's nanofluids match. The dossier's PubMed sample (tuberculosis prevalence, IoT trust management, JWST ice observations) is also unrelated noise/homonyms. Scorer relied on IISc's own faculty page and independent web search instead; could not independently verify the dossier's Wikidata education claims (Arizona State PhD, IIT Delhi BTech) against IISc's own bio, so citations/h_index left null pending a directly-fetched Google Scholar page rather than assumed from the wrong OpenAlex match.","years_language_modeling":2,"years_as_technical_founder":3,"technical_founder_companies":2,"lm_domains":[{"domain":"natural_language","years":"2023-2025","evidence":"Decentralized-AI model serving/ownership work (SAKSHI 2023, OML 2024) and Sentient open-model protocol — building around LLMs, weak/adjacent hands-on modeling record"}],"frontier_lineage":["No verifiable foundational block; federated/distributed-learning methods (gradient compression, distributed mean estimation) are ML-adjacent but not part of the transformer/embedding stack frontier LLMs descend from"],"technical_founder_roles":["Witness Chain — co-founder (decentralized infrastructure/DePIN) — ~2022/23-present","Sentient — co-founder / core research author (OML) — ~2024-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":12},"confidence":0.7,"duration_ms":92914},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":8},"confidence":0.7,"duration_ms":66714}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sreeram-kannan","name":"Sreeram Kannan","title":"Founder","company":"EigenLayer (Eigen Labs)","sector":"crypto","rank":50,"rubric_version":4,"dimensions":{"foundations":16,"vector_embeddings":9,"transformers_lm":8,"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":4,"hands_on_engineering":12,"industry_impact":12,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":40,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":40,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.81,"rationale":"Kannan holds a PhD in Information Theory and Wireless Networks from University of Illinois Urbana-Champaign (2008-2012, plus an MS in Mathematics), and was (is, as affiliate) an Associate Professor of Electrical & Computer Engineering at the University of Washington — solid graduate-level mathematical/statistical foundations. His authored/co-authored papers include genuine ML research: 'ClusterGAN: Latent Space Clustering in Generative Adversarial Networks' (AAAI 2019, 533 citations) which is directly relevant to vector-embeddings/representation-learning, and 'Improving Federated Learning Personalization via Model-Agnostic Meta-Learning' (2019, 900 citations) and 'Communication Algorithms via Deep Learning' (2018) — real, well-cited, hands-on ML research, not just blockchain theory. His verified Google Scholar h-index is 37 with 7,027 citations, confirming a substantive, continuous research record since ~2008. He founded EigenLayer and is credited with originating the 'restaking' mechanism — strong hands-on systems/protocol engineering, though this is a distributed-systems/cryptoeconomic contribution rather than core transformer/LM research. Overall: real academic ML depth plus applied protocol engineering, but not canonical transformer/LM authorship.\n\nKannan's ML record — ClusterGAN latent-space clustering (AAAI 2019), federated-learning meta-learning (2019), 'Communication Algorithms via Deep Learning' (ICLR 2018), and single-cell causal gene-network inference (Scribe, Cell Systems 2020) — is genuine representation-learning and information-theory work, but none of it is in the attention→transformer→LM lineage, and there is no verifiable pre-2013 vector-space / distributional-semantics / LSI / term-relationship work, so his position in the frontier-LM foundation is minimal (frontier_founder ~3). He has essentially zero verifiable years in language modeling proper (no n-gram/neural-LM/seq2seq/transformer/embedding-for-text record), placing lm_domain_depth in the under-3-year band. Across language-modeling domains he has at most one adjacent, borderline record — genomics relationship-network inference (biological), which is network causal inference rather than a DNA/RNA/protein sequence LM — with no natural-language, financial, code, or other LM domain, so lm_domain_breadth sits at the low single-domain/adjacent band. He is, however, a real scientific/technical founder: he personally originated the restaking mechanism and authored the EigenLayer whitepaper for Eigen Labs (founded ~2021, ~5 years), but that company's core is cryptoeconomic restaking / distributed systems, NOT language modeling, so under the anchors this is 'a technical founder outside this field' (3-7), scored 7 for the strength and hands-on nature of that fo","evidence":[{"claim":"Google Scholar profile shows h-index 37 and 7,027 total citations","source_url":"https://scholar.google.com/citations?user=RrYw5jkAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'ClusterGAN: Latent Space Clustering in Generative Adversarial Networks' (AAAI 2019) and 'Improving Federated Learning Personalization via Model Agnostic Meta-Learning' (2019)","source_url":"https://scholar.google.com/citations?user=RrYw5jkAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 2404352759: 105 papers, 4,903 citations, h-index 30; dominant areas blockchain/consensus, machine learning and bioinformatics/information theory","source_url":"https://api.semanticscholar.org/graph/v1/author/2404352759","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Deconstructing the Blockchain to Approach Physical Limits' (Prism) — authors Vivek Bagaria, Sreeram Kannan, David Tse, Giulia Fanti, Pramod Viswanath","source_url":"https://arxiv.org/abs/1810.08092","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'ClusterGAN: Latent Space Clustering in Generative Adversarial Networks' — authors Sudipto Mukherjee, Himanshu Asnani, Eugene Lin, Sreeram Kannan (AAAI 2019)","source_url":"https://arxiv.org/abs/1809.03627","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'SAKSHI: Decentralized AI Platforms' (2023) lists Sreeram Kannan among the authors, with EigenLayer among the affiliations","source_url":"https://arxiv.org/abs/2307.16562","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ClusterGAN published in AAAI Proceedings vol 33 (2019)","source_url":"https://doi.org/10.1609/aaai.v33i01.33014610","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5100722219: 135 works, 3,106 citations, h-index 28, affiliations include University of Washington and University of Illinois Urbana-Champaign; topics blockchain, distributed systems, cryptography, network coding","source_url":"https://api.openalex.org/authors/A5100722219","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ClusterGAN: Latent Space Clustering in GANs (AAAI 2019) — representation learning, not the transformer/LM lineage","source_url":"https://arxiv.org/abs/1809.03627","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Scribe: inferring causal gene regulatory networks from single-cell expression dynamics (Cell Systems 2020) — genomics network inference, adjacent to biological modeling","source_url":"https://doi.org/10.1016/j.cels.2020.02.003","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record: 105 papers dominated by blockchain/consensus, ML and bioinformatics/information theory — no language-modeling corpus","source_url":"https://api.semanticscholar.org/graph/v1/author/2404352759","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sreeram Kannan founded EigenLayer/Eigen Labs and originated the restaking mechanism, personally authoring the protocol design","source_url":"https://arxiv.org/abs/2307.16562","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"ClusterGAN (latent-space clustering in GANs, AAAI 2019) and single-cell gene-regulatory-network inference (Scribe, Cell Systems 2020) are his ML/bioinformatics record — representation learning and genomics, not language modeling","source_url":"https://doi.org/10.1016/j.cels.2020.02.003","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author profile: dominant areas are blockchain/consensus, information theory and bioinformatics — no transformer/LM authorship","source_url":"https://api.semanticscholar.org/graph/v1/author/2404352759","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BE Electronics & Communication Engineering, College of Engineering Guindy","ME Telecommunications, Indian Institute of Science (IISc)","PhD Information Theory/Wireless Networks, University of Illinois Urbana-Champaign; MS Mathematics, UIUC"],"canonical_papers":["ClusterGAN: Latent Space Clustering in Generative Adversarial Networks (AAAI 2019)","Improving Federated Learning Personalization via Model Agnostic Meta-Learning (2019)","Communication Algorithms via Deep Learning (ICLR 2018)"],"first_verifiable_year":2008,"notable_systems":["EigenLayer restaking protocol","Prism blockchain scaling protocol"],"citations":7027,"h_index":37,"patents":0,"dossier_notes":"Dossier's OpenAlex identity was flagged ambiguous (6 candidates) but independent Google Scholar verification (scholar.google.com/citations?user=RrYw5jkAAAAJ) confirms the ClusterGAN/MAML/Prism papers all belong to this Sreeram Kannan, with a real h-index of 37 — notably higher than the dossier's OpenAlex figure of 28, suggesting OpenAlex undercounts his work. This is a correctly-resolved identity, not a homonym issue.","years_language_modeling":0,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"biological","years":"2020-2020","evidence":"Scribe single-cell causal gene regulatory network inference (Cell Systems 2020) — network inference, adjacent to biological sequence modeling, not a sequence LM"}],"technical_founder_roles":["EigenLayer / Eigen Labs — founder, originator of restaking mechanism and whitepaper author (cryptoeconomics/distributed systems, not LM) — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":3,"scientific_founder":7},"confidence":0.72,"duration_ms":51402},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":4,"scientific_founder":6},"confidence":0.72,"duration_ms":57055}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ben-fielding","name":"Ben Fielding","title":"Co-founder & CEO","company":"Gensyn","sector":"crypto","rank":51,"rubric_version":4,"dimensions":{"foundations":12,"vector_embeddings":4,"transformers_lm":10,"frontier_founder":6,"lm_domain_depth":4,"lm_domain_breadth":3,"hands_on_engineering":13,"industry_impact":10,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":39,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":39,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.82,"rationale":"The two passes split because each held half of the record. Pass 1 credited a doctoral-level evolutionary neural-architecture-search corpus but missed his 2025 language-model papers; pass 2 found the LM papers but refused to attribute the academic corpus to him, having failed to confirm identity. Both halves are his. The Google Scholar profile carrying the Northumbria evolutionary-NAS publications is titled 'Co-Founder of Gensyn' and is verified on a gensyn.ai email address, which resolves the identity question pass 2 could not: the PSO/GA neural-architecture-search and image-classification work with Li Zhang and Kamlesh Mistry (IEEE Transactions on Cybernetics 2017, IEEE Access, 2016-2020, 790 citations, h-index 7) is the same person. That corpus is genuine graduate-level optimization and statistical-learning work applied to deep networks, which supports foundations well above the practitioner band, though it is evolutionary search over CNN architectures for computer vision, not representation learning. Separately and verifiably he is a named co-author of 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (arXiv:2509.08721, 2025), introducing SAPO for decentralized RL post-training of language models, and of 'Verde: Verification via Refereed Delegation for Machine Learning Programs' (arXiv:2502.19405, 2025) on verifying delegated LLM inference and training — real, current authorship inside the post-training half of the lineage, which is why pass 1's transformers_lm of 4 is too low and pass 2's 12 slightly too generous for co-authored team papers with no architecture or scaling contribution. He has no work at all in vector embeddings, retrieval or vector search, so that dimension stays near the floor. Gensyn's core is verifiable distributed deep-learning compute, a genuine ML-systems company rather than an AI label, but it has produced no canonical result, so industry impact is mid-band.\n\nFielding's authored contribution to the frontier lineage is recent and indirect: he is a named co-author of SAPO (arXiv:2509.08721, 2025), a decentralized swarm RL post-training method for language models, and Verde (arXiv:2502.19405, 2025), a verification protocol for delegated ML/LLM training — genuine post-training/verification-infrastructure work that the frontier stack could draw on, but neither a named foundational block (attention, transformers, embeddings, optimizers, RLHF) nor a pre-2013 vector-space precursor, so it sits in the 'published lineage work' band rather than the foundation frontier models directly descend from. His verifiable language-modeling record is narrow and short: the 2016-2020 corpus is evolutionary neural-architecture search for computer vision (image classification, facial-emotion recognition), which is not language modeling, and the only LM-specific work is 2024-2025 through Gensyn, giving roughly 1-2 hands-on years in a single domain — natural-language text only, with no biological, financial, code, chemistry or other LM domain on record. As scientific/technical founder he is stronger: he co-founded Gensyn (decentralized deep-learning compute) around 2020-2021 and personally authors the research the protocol runs on (SAPO, Verde, the RL Swarm system), so ~5 years as the technical founder of ONE company, which places scientific_founder in the 3-8-year single-company band.","evidence":[{"claim":"Google Scholar profile 'Ben Fielding', listed as Co-Founder of Gensyn with a verified email at gensyn.ai, carries the evolutionary/PSO neural-architecture-search publications (h-index 7, 790 citations; top works 'A micro-GA embedded PSO feature selection approach to intelligent facial emotion recogn","source_url":"https://scholar.google.com/citations?user=B9lV7zUAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Crossref record for 'A Micro-GA Embedded PSO Feature Selection Approach to Intelligent Facial Emotion Recognition', IEEE Transactions on Cybernetics 2017, authors Kamlesh Mistry, Li Zhang, Siew Chin Neoh, Chee Peng Lim, Ben Fielding","source_url":"https://api.crossref.org/works/10.1109/tcyb.2016.2549639","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named co-author of 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (arXiv:2509.08721, 10 Sep 2025), introducing SAPO, a decentralized swarm-sampling policy-optimization method for RL post-training of language models; author list verified as including Ben Fieldin","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Verde: Verification via Refereed Delegation for Machine Learning Programs' (arXiv:2502.19405, 2025), on verifying LLM inference, fine-tuning and training delegated to untrusted compute; author list verified as Arun, St. Arnaud, Titov, Wilcox, Kolobaric, Brinkmann, Ersoy, Fielding, Bonn","source_url":"https://arxiv.org/abs/2502.19405","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn's litepaper specifies a protocol for verifying distributed deep-learning compute via probabilistic proof-of-learning and graph-based pinpoint protocols","source_url":"https://docs.gensyn.ai/litepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fielding holds a PhD in Computer Science from Northumbria University, completed Dec 2019, on evolutionary optimization of deep neural architectures","source_url":"https://iq.wiki/wiki/ben-fielding","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (verified gensyn.ai email) shows h-index 7, ~790 total citations, publications on evolutionary/PSO-based neural architecture search and image classification (2016-2020), co-authored with Li Zhang (Royal Holloway) and Kamlesh Mistry (Northumbria)","source_url":"https://scholar.google.com/citations?user=B9lV7zUAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Top-cited paper: 'A Micro-GA Embedded PSO Feature Selection Approach to Intelligent Facial Emotion Recognition', IEEE Transactions on Cybernetics, 2016","source_url":"https://doi.org/10.1109/tcyb.2016.2549639","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (arXiv:2509.08721, 10 Sep 2025), introducing SAPO for decentralized RL post-training of language models.","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn litepaper specifies a protocol for verifiable distributed deep-learning compute (proof-of-learning, graph-based pinpoint), the ML-training-infrastructure company Fielding co-founded and technically leads","source_url":"https://docs.gensyn.ai/litepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fielding holds a PhD in Computer Science from Northumbria University (evolutionary optimization of deep neural architectures, computer vision — not language modeling), and is listed as Co-Founder of Gensyn","source_url":"https://iq.wiki/wiki/ben-fielding","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile 'Ben Fielding' verified on a gensyn.ai email and titled Co-Founder of Gensyn, carrying the 2016-2020 evolutionary-NAS computer-vision publications","source_url":"https://scholar.google.com/citations?user=B9lV7zUAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn litepaper specifies the distributed-ML-compute verification protocol Fielding co-founded and builds on","source_url":"https://docs.gensyn.ai/litepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, Northumbria University (evolutionary optimization of deep neural architectures / neural architecture search)"],"canonical_papers":["A Micro-GA Embedded PSO Feature Selection Approach to Intelligent Facial Emotion Recognition (IEEE Trans. Cybernetics, 2017)","Evolving ensemble models for image segmentation using enhanced particle swarm optimization (2019)","Verde: Verification via Refereed Delegation for Machine Learning Programs (arXiv:2502.19405, 2025)","Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing (arXiv:2509.08721, 2025)"],"first_verifiable_year":2016,"notable_systems":["Gensyn protocol (verifiable distributed ML compute)","RL Swarm / SAPO"],"citations":790,"h_index":7,"patents":0,"dossier_notes":"The dossier's Wikipedia/Wikidata match (Q86968598) is an Australian worship songwriter — the wrong person entirely; discarded. The PubMed block (240 hits) is homonym noise. The OpenAlex/Semantic Scholar academic match, flagged 'ambiguous' with 3 candidates, IS correct: the gensyn.ai-verified Google Scholar profile carries exactly those publications. The dossier misses both 2025 Gensyn papers because neither indexed source captured the author lists.","adjudication_notes":["foundations (11 vs 11, agreed but for opposite reasons): pass 1's reasoning is upheld. The Northumbria evolutionary-NAS corpus IS his — the Google Scholar profile carrying it is verified on a gensyn.ai address and titled 'Co-Founder of Gensyn'. Pass 2 declined to score it on unconfirmed identity; that confirmation now exists, so the score rests on the doctoral corpus and rises slightly to 12.","transformers_lm (4 vs 12): pass 2 was right that authored LM work exists and pass 1 simply missed it. Verified: he is a named co-author of arXiv:2509.08721 (SAPO, RL post-training of language models) and arXiv:2502.19405 (Verde, verifying LLM training). Settled at 10 rather than 12 because both are multi-author team papers contributing verification and decentralization machinery, not attention, architecture or scaling results.","vector_embeddings (3 vs 6): pass 1 was right. Neither pass cited a single embedding, retrieval or vector-search artifact, and I found none; pass 2's 6 is unsupported by its own evidence list. Scored 4.","industry_impact (10 vs 9): effectively agreed; Gensyn is a genuine ML-training-infrastructure company with no canonical output yet. Held at 10.","first_verifiable_year (2016 vs 2025): pass 1 was right — the 2016 IEEE publications are his, per the verified-email Scholar profile."],"years_language_modeling":2,"years_as_technical_founder":5,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2024-2025","evidence":"SAPO decentralized RL post-training of LMs (arXiv:2509.08721) and Verde verification of LLM training (arXiv:2502.19405); Gensyn RL Swarm"}],"frontier_lineage":["Decentralized RL post-training method (SAPO) for language models — post-training/decentralization machinery, not a foundational architecture, optimizer or alignment block","Verde verification protocol for delegated ML/LLM training — training-infrastructure lineage, not cited as a named frontier block"],"technical_founder_roles":["Gensyn — Co-founder (technical, authors core research: SAPO, Verde, RL Swarm) — ~2020-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":7,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":10},"confidence":0.62,"duration_ms":61032},{"pass":"pass_2","dimensions":{"frontier_founder":6,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":11},"confidence":0.6,"duration_ms":53568}],"source":"seeded","profile_url":null,"wikipedia_title":"Ben Fielding","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ali-ghodsi","name":"Ali Ghodsi","title":"Co-founder & CEO","company":"Databricks","sector":"general","rank":52,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":4,"transformers_lm":5,"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":1,"hands_on_engineering":14,"industry_impact":14,"scientific_founder":12},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":36,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":36,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.85,"rationale":"The passes diverged because pass 1 scored from the dossier's OpenAlex record and pass 2 caught that the record is a conflation. I verified this and pass 2 is right: OpenAlex A5040035859 is 'Ali Reza Ghodsi', last known institution University of Waterloo, whose largest topic is 'Face and Expression Recognition' with 58 works. That is a different person — Ali Ghodsi, Professor of Statistics and Actuarial Science at the University of Waterloo, who teaches statistical learning and deep learning and works on dimensionality reduction. The dossier's h-index 59, 17,626 citations and earliest year 1984 are therefore contaminated and cannot be used, and the confusion is compounded by the Waterloo professor's field being closer to this rubric's subject matter than the Databricks CEO's actually is. The correct record is the Google Scholar profile verified on a cs.berkeley.edu address: ~39,955 citations, h-index 51, i10-index 84, top works OpenFlow (13,898), Apache Spark (4,373), Mesos (2,777), Spark SQL (2,172) and Dominant Resource Fairness (1,927). That is a genuinely distinguished record in distributed systems, scheduling, networking and data management, built on a KTH PhD (2006) on distributed hash tables under Seif Haridi and years at the Berkeley AMPLab. It sits one layer beneath AI rather than inside it. He has no authored work on attention, transformers, pretraining, scaling laws or alignment, and none on embeddings, dense retrieval or vector search; Databricks' LLM artifacts (Dolly, MosaicML/DBRX) are outputs of the Mosaic team, and the DBRX announcement does not name him as a technical contributor. Pass 2's vector_embeddings of 8 and transformers_lm of 7 credit him for organisational proximity to work he did not author, which the rubric forbids; pass 1's 3/3 are nearer the mark, adjusted slightly for the MLflow and Dolly co-authorships that do appear on his verified profile.\n\nGhodsi's verifiable authored corpus is distributed systems and data infrastructure — Apache Mesos (NSDI 2011), Spark SQL (SIGMOD 2015), Apache Spark (CACM 2016) and Delta Lake (VLDB 2020) — not the attention/transformer/embedding lineage; his work sits one layer beneath AI as data-processing and cluster-scheduling infrastructure that ML pipelines can run on, which is adjacent to, not part of, the named frontier building blocks (attention, transformers, tokenizers, pretraining objectives, scaling/alignment), and there is no pre-2013 vector-space/distributional-semantics/LSI patent or shipped system in his record, so frontier_founder is low. He has zero verifiable hands-on years authoring language-modeling work: Databricks' LLM artifacts (Dolly, and DBRX built by the acquired Mosaic team) are not authored by him, so years_language_modeling is 0 and lm_domain_depth/breadth score at the floor. He is, however, a genuine scientific/technical founder — he personally co-authored the core Spark SQL and Delta Lake papers and code that Databricks commercializes and has led it as co-founder since 2013 (CEO since 2016), ~13 years — but of one company whose core is data/analytics infrastructure rather than the embedding/transformer systems this rubric measures, placing scientific_founder mid-band. Distinct language-modeling domains with a hands-on modeling record: none.","evidence":[{"claim":"Google Scholar profile verified on a cs.berkeley.edu email, affiliation UC Berkeley and Databricks: ~39,955 citations, h-index 51, i10-index 84; top works OpenFlow (2008, 13,898 citations), Apache Spark (2016, 4,373), Mesos (2011, 2,777), Spark SQL (2015, 2,172), Dominant Resource Fairness (2011, 1,","source_url":"https://scholar.google.com/citations?user=YsXNU78AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5040035859 is 'Ali Reza Ghodsi' with last known institution University of Waterloo (active 2002-2025) and top topic 'Face and Expression Recognition' (58 works) — a conflated record, not cleanly the Databricks CEO","source_url":"https://api.openalex.org/authors/A5040035859","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A different Ali Ghodsi is Professor in the Department of Statistics and Actuarial Science at the University of Waterloo, teaching Statistical Learning and Deep Learning — confirming two distinct people share the name","source_url":"https://uwaterloo.ca/statistics-and-actuarial-science/profile/aghodsib","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from KTH Royal Institute of Technology (2006), dissertation 'Distributed k-ary System: Algorithms for Distributed Hash Tables', advised by Seif Haridi; co-founded Databricks in 2013, CEO from 2016","source_url":"https://en.wikipedia.org/wiki/Ali_Ghodsi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DBRX was built by the Mosaic team that previously built MPT; Ghodsi is not named as an author or technical contributor in the announcement","source_url":"https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD from KTH Royal Institute of Technology, thesis 'Distributed k-ary System: Algorithms for Distributed Hash Tables' (2006), advisor Seif Haridi","source_url":"https://en.wikipedia.org/wiki/Ali_Ghodsi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Databricks, founded 2013 to commercialize Apache Spark","source_url":"https://www.databricks.com/dataaisummit/speaker/ali-ghodsi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Adjunct professor at UC Berkeley, worked with Scott Shenker and Ion Stoica at AMPLab","source_url":"https://kitrum.com/blog/the-inspiring-story-ali-ghodsi-ceo-of-databricks/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (UC Berkeley and Databricks): ~39,955 citations, h-index 51, i10-index 84; top works are OpenFlow, Apache Spark, Mesos, Spark SQL, Dominant Resource Fairness, Delta Lake","source_url":"https://scholar.google.com/citations?user=YsXNU78AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Databricks in 2013 to commercialize Apache Spark; CEO since 2016; personally co-authored Spark SQL and Delta Lake, the platform's core","source_url":"https://en.wikipedia.org/wiki/Ali_Ghodsi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DBRX was built by the Mosaic team (previously MPT); Ghodsi is not named as an author or technical contributor in the announcement","source_url":"https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Verified Google Scholar top works are OpenFlow, Apache Spark, Mesos, Spark SQL, Dominant Resource Fairness, Delta Lake — distributed systems / data infrastructure, no attention/transformer/embedding work","source_url":"https://scholar.google.com/citations?user=YsXNU78AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Databricks (2013), CEO since 2016, adjunct professor at UC Berkeley; KTH PhD 2006 on distributed hash tables under Seif Haridi — a distributed-systems/data-management record, not a language-modeling one","source_url":"https://en.wikipedia.org/wiki/Ali_Ghodsi","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"DBRX was built by the Mosaic team; Ghodsi is not named as an author or technical contributor, so Databricks' LLM work is not his authored frontier-lineage contribution","source_url":"https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"cs.berkeley.edu-verified Google Scholar profile: top works are OpenFlow, Apache Spark, Mesos, Spark SQL, Dominant Resource Fairness, Delta Lake — no attention/transformer/embedding/pretraining authorship","source_url":"https://scholar.google.com/citations?user=YsXNU78AAAAJ","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, KTH Royal Institute of Technology, 2006 (advisor Seif Haridi)"],"canonical_papers":["Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center (NSDI 2011)","Dominant Resource Fairness: Fair Allocation of Multiple Resource Types (NSDI 2011)","Spark SQL: Relational Data Processing in Spark (SIGMOD 2015)","Apache Spark: A Unified Engine for Big Data Processing (CACM 2016)","Delta Lake (VLDB 2020)"],"first_verifiable_year":2006,"notable_systems":["Apache Mesos","Apache Spark / Spark SQL","Delta Lake","Databricks Lakehouse platform"],"citations":39955,"h_index":51,"patents":0,"dossier_notes":"The dossier's OpenAlex record A5040035859 is a CONFLATION of at least two people: the Databricks CEO and Ali Ghodsi the statistics professor at the University of Waterloo (dimensionality reduction, statistical learning), which I confirmed against the Waterloo faculty page. Its h-index 59, 17,626 citations, 'Face and Expression Recognition' topic and earliest_year of 1984 (impossible for a person born in 1978) all derive from that conflation, so the dossier's timeline of first_verifiable_year 1984 / 42 years active is wrong; 2006 is the correct floor. The PubMed block is unrelated homonym noise. Metrics here are taken from the cs.berkeley.edu-verified Google Scholar profile instead.","adjudication_notes":["The dossier's OpenAlex conflation is the root of the split: pass 1 scored partly from a record that belongs substantially to a different Ali Ghodsi, whose field (dimensionality reduction, statistical learning) is closer to this rubric than the real subject's. Verified: OpenAlex A5040035859 is 'Ali Reza Ghodsi' at the University of Waterloo, top topic 'Face and Expression Recognition', and the Waterloo statistics faculty page confirms a distinct person of that name. Pass 2 caught this; pass 1 did not.","vector_embeddings (3 vs 8) and transformers_lm (3 vs 7): pass 1 was right; settled at 4 and 5. Neither pass cited a single authored artifact by him in either area, and pass 2's own evidence establishes the opposite — that DBRX names the Mosaic team and not him. Organisational proximity to models his company acquired is not authorship.","foundations (9 vs 12): split at 11. The KTH PhD and the distributed-systems corpus are real first-principles work (resource allocation, consensus, distributed data structures), but they are not the linear-algebra/optimization/statistical-learning mathematics the dimension names, and the inflated h-index that supported pass 2's 12 is partly another person's.","hands_on_engineering (12 vs 15): pass 2 was closer; settled at 14. He personally co-authored and built Mesos, Spark SQL and Delta Lake — verified on the cs.berkeley.edu Scholar profile — which is genuine infrastructure that large-scale ML training runs on.","industry_impact (13 vs 15): settled at 14. Databricks is major infrastructure for ML workloads, but its core is data and analytics platforms rather than the embedding/transformer systems this rubric measures, and fundraising and valuation are not evidence."],"years_language_modeling":0,"years_as_technical_founder":13,"technical_founder_companies":1,"frontier_lineage":["Apache Spark / Databricks as general large-scale data-processing infrastructure that ML pipelines can run on — adjacent infra, not a named LM building block; no pre-2013 vector-space/distributional-semantics precursor"],"technical_founder_roles":["Databricks — co-founder & CEO, co-author of core Spark SQL / Delta Lake — 2013-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":5,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":12},"confidence":0.8,"duration_ms":69830},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":12},"confidence":0.8,"duration_ms":84347}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Ali_Ghodsi","wikipedia_title":"Ali Ghodsi","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jed-mccaleb","name":"Jed McCaleb","title":"Founder and chairman, Vast; co-founder and CTO, Stellar","company":"Vast","sector":"crypto","rank":53,"rubric_version":4,"dimensions":{"foundations":5,"vector_embeddings":4,"transformers_lm":10,"frontier_founder":6,"lm_domain_depth":5,"lm_domain_breadth":4,"hands_on_engineering":13,"industry_impact":8,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":33,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":33,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.88,"rationale":"McCaleb is an unusual case: a self-taught programmer with no completed degree who nonetheless has a genuine and recent co-authorship record in language-model research that the dossier does not surface at all. Querying the arXiv API directly for his name returns four papers, all of which I confirmed carry him as a listed author: 'Thinking agents for zero-shot generalization to qualitatively novel tasks' (2503.19815, March 2025, with Miconi, McKee and Zheng), 'Goal-Directed Search Outperforms Goal-Agnostic Memory Compression in Long-Context Memory Tasks' (2511.21726, November 2025), 'End-to-End Test-Time Training for Long Context' (2512.23675, December 2025, with Dalal, Leskovec, Koyejo, Hashimoto, Guestrin, Choi and Yu Sun) and 'Learning to Discover at Test Time' (2601.16175, January 2026, TTT-Discover). These sit squarely in the modern long-context, memory and test-time-adaptation lineage — the first formulates long-context language modelling as continual learning over a sliding-window-attention Transformer — and four papers across a year is a sustained line rather than a one-off byline, though he appears mid-list among senior academics rather than leading. His engineering record is deep and hands-on across decades: he personally wrote eDonkey2000 and the Overnet peer-to-peer network, built Mt. Gox, wrote the original Ripple consensus implementation, and co-authored the Stellar Consensus Protocol and 'Fast and secure global payments with Stellar' (SOSP 2019). He has no verifiable training in linear algebra, optimization or statistical learning and no embedding or pretraining work of his own, so foundations and vector_embeddings stay low. His earlier AI involvement was funding — donations to MIRI and OpenAI — which is not evidence under this rubric; only the authored 2025-26 work counts, and its recency is discounted under the depth-of-experience rule.\n\nMcCaleb's language-modeling record is entirely 2025-26: mid-list co-authorship on ~3-4 arXiv papers in the long-context / memory / test-time-training lineage (2503.19815, 2511.21726, 2512.23675, 2601.16175) with Stanford/NVIDIA faculty. That is published lineage work the frontier stack draws on but neither a named foundational block (transformers, word2vec, RLHF are not his) nor a verifiable pre-2013 vector-space/distributional-semantics precursor — he has zero such patents or shipped LM systems, so frontier_founder sits at 6, not the 13-17 precursor band. His verifiable years in language modeling are ~1 (all 2025-26), placing lm_domain_depth in the under-3-year band; his P2P/crypto/payments/aerospace decades are not LM work. The LM papers cover a single domain (general natural-language long-context modeling), so lm_domain_breadth is a single-domain 4. He is a genuine, prolific hands-on technical founder — he personally wrote eDonkey2000/Overnet, Mt. Gox, the original Ripple implementation, co-authored the Stellar Consensus Protocol, and founded Vast — but none of these companies' cores are language models, so scientific_founder scores in the 'technical founder outside this field' band (5), with zero in-field LM companies founded.","evidence":[{"claim":"arXiv API query for 'Jed McCaleb' returns four papers with him as a listed author: 'Thinking agents for zero-shot generalization to qualitatively novel tasks' (2503.19815, Mar 2025), 'Goal-Directed Search Outperforms Goal-Agnostic Memory Compression in Long-Context Memory Tasks' (2511.21726, Nov 202","source_url":"http://export.arxiv.org/api/query?search_query=all:%22Jed%20McCaleb%22&max_results=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'End-to-End Test-Time Training for Long Context' (arXiv:2512.23675) lists Jed McCaleb as twelfth of fourteen authors alongside Karan Dalal, Jure Leskovec, Sanmi Koyejo, Tatsunori Hashimoto, Carlos Guestrin, Yejin Choi and Yu Sun; it formulates long-context language modelling as continual learning ov","source_url":"https://arxiv.org/abs/2512.23675","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Learning to Discover at Test Time' (arXiv:2601.16175, TTT-Discover) lists Jed McCaleb as fifth of eleven authors with Mert Yuksekgonul, Jan Kautz, James Zou, Carlos Guestrin and Yu Sun; it applies reinforcement learning at test time so an LLM continues training on test-problem-specific experience","source_url":"https://arxiv.org/abs/2601.16175","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 2077048287 (Jed McCaleb) holds six papers: four machine-learning papers (2025-2026) plus 'Fast and secure global payments with Stellar' (SOSP 2019) and 'The Stellar Consensus Protocol' (2018)","source_url":"https://api.semanticscholar.org/graph/v1/author/2077048287/papers?fields=title,year,venue,authors,externalIds&limit=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"McCaleb left UC Berkeley without completing a degree; created eDonkey2000 and the Overnet peer-to-peer networks, founded Mt. Gox in 2010, was founder and CTO of Ripple until 2013, co-founder and CTO of Stellar, and founder of the aerospace company Vast","source_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Left UC Berkeley to work as a programmer in New York; no completed degree found","source_url":"https://www.bitnovo.com/blog/en/who-is-jed-mccaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Created eDonkey2000 and Overnet peer-to-peer file-sharing networks; founded Mt. Gox bitcoin exchange in 2010","source_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded and served as CTO of Ripple until 2013; co-founder and CTO of Stellar; founder/ex-CEO/chairman of aerospace startup Vast","source_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"End-to-End Test-Time Training for Long Context (arXiv 2512.23675) lists Jed McCaleb among the authors alongside Jure Leskovec, Sanmi Koyejo, Tatsunori Hashimoto, Carlos Guestrin, Yejin Choi and Yu Sun","source_url":"https://arxiv.org/abs/2512.23675","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Learning to Discover at Test Time (arXiv 2601.16175, TTT-Discover) lists Jed McCaleb as fifth author with Mert Yuksekgonul, Jan Kautz, James Zou and Yu Sun","source_url":"https://arxiv.org/abs/2601.16175","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'End-to-End Test-Time Training for Long Context' formulates long-context language modeling as continual learning; McCaleb is 12th of 14 authors — his only natural-language LM domain and its full extent (2025)","source_url":"https://arxiv.org/abs/2512.23675","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"McCaleb founded/was CTO of Ripple (until 2013), co-founded and was CTO of Stellar, created eDonkey2000/Overnet, founded Mt. Gox and founder/chairman of aerospace startup Vast — technical founder across multiple companies, none with an AI/LM core","source_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"McCaleb appears as a mid-list co-author on 'End-to-End Test-Time Training for Long Context' alongside Karan Dalal, Jure Leskovec, Tatsunori Hashimoto, Carlos Guestrin, Yejin Choi and Yu Sun — long-context LM as continual learning over a sliding-window Transformer","source_url":"https://arxiv.org/abs/2512.23675","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"McCaleb is fifth of eleven authors on 'Learning to Discover at Test Time' (TTT-Discover), RL at test time so an LLM continues training on test-problem-specific experience","source_url":"https://arxiv.org/abs/2601.16175","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"McCaleb founded/was CTO of Ripple until 2013, co-founded and was CTO of Stellar, created eDonkey2000/Overnet and Mt. Gox, and founded aerospace startup Vast — technical founder across decades, none of them language-model companies","source_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 2077048287 (Jed McCaleb): 6 papers, 225 citations, h-index 4 — four 2025-26 ML papers plus Stellar consensus/payments work, no embedding or pretraining authorship","source_url":"https://api.semanticscholar.org/graph/v1/author/2077048287/papers?fields=title,year,venue,authors,externalIds&limit=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended UC Berkeley; left without completing a degree"],"canonical_papers":["End-to-End Test-Time Training for Long Context (arXiv:2512.23675, 2025) — co-author","Learning to Discover at Test Time (arXiv:2601.16175, 2026) — co-author","Goal-Directed Search Outperforms Goal-Agnostic Memory Compression in Long-Context Memory Tasks (arXiv:2511.21726, 2025) — co-author","Thinking agents for zero-shot generalization to qualitatively novel tasks (arXiv:2503.19815, 2025) — co-author","Fast and secure global payments with Stellar (SOSP 2019)","The Stellar Consensus Protocol (2018)"],"first_verifiable_year":2018,"notable_systems":["eDonkey2000 / Overnet P2P network","Mt. Gox","Ripple protocol (original implementation)","Stellar / Stellar Consensus Protocol","Vast (aerospace)"],"citations":225,"h_index":4,"patents":0,"dossier_notes":"The PubMed block is entirely HOMONYMS — J.A. McCaleb (tai chi and knee osteoarthritis), J.L. McCaleb (adenosine receptors and atherosclerosis in ApoE-deficient mice at Vanderbilt) and a North Carolina primary-care physician — none of whom is the programmer. This corrupts the timeline: first_verifiable_year 1989 and years_active 37 derive from a 1989 rural-health-centres article by a different person; his first indexed publication is 2018. OpenAlex returned nothing for him, which is a collection gap rather than a correct absence: the four 2025-26 arXiv papers are real and verifiable through the arXiv API and Semantic Scholar. The dossier should be amended to carry these four papers, which are the single most decision-relevant facts about him for this index.","adjudication_notes":["transformers_lm — PASS_2 RIGHT AND STILL INCOMPLETE; final 10 vs pass_1's 0 and pass_2's 11. The decisive verified fact: a direct arXiv API query returns FOUR papers authored by Jed McCaleb in the long-context/memory/test-time-training lineage (2503.19815, 2511.21726, 2512.23675, 2601.16175); I confirmed the author lists on the two largest individually. Pass_1 scored 0 having consulted only OpenAlex (which indexes none of them) and Wikipedia, and explicitly flagged the Semantic Scholar record as possibly a different person — it is not; it is him. Pass_2 found three of the four. Four papers in twelve months is a sustained line, but he is mid-list among senior academics with no evidence of leading the work, and all of it is 2025-26, which the depth-of-experience rule discounts — so 10 rather than pass_2's 11.","vector_embeddings — SPLIT, final 4 (pass_1 0, pass_2 6). The verified long-context-memory paper (2511.21726) concerns memory compression versus goal-directed search over long contexts, which touches representation storage and retrieval, so pass_1's 0 is too absolute; but no authored work on vector-space models, embeddings, contrastive or dense retrieval, or vector search exists, so pass_2's 6 is not supported either.","foundations — SPLIT, final 5 (pass_1 1, pass_2 6). He has no completed degree and no verifiable training in linear algebra, optimization or statistical learning, so pass_2's 6 slightly overstates; but co-authoring the Stellar Consensus Protocol and now appearing on ML papers with Stanford faculty demonstrates more mathematical capability than pass_1's 1, which is the 'nothing verifiable' anchor and is contradicted by the verified record.","hands_on_engineering — PASS_2 RIGHT, final 13 (pass_1 8, pass_2 13). He personally wrote eDonkey2000 and Overnet, built Mt. Gox, wrote the original Ripple consensus implementation and co-authored the Stellar Consensus Protocol — decades of systems he wrote himself, which is squarely the 13-17 'production systems built and led personally' band. Pass_1's 8 understates a record both passes describe identically.","industry_impact — SPLIT, final 8 (pass_1 2, pass_2 9). Pass_1's 2 is the 'nothing verifiable' anchor and is wrong on its own evidence: 225 citations, h-index 4, and organizations he built that the payments world runs on. But those orgs' core is consensus and payments, not language models, and the AI papers are recent and not led by him, so 8 rather than 9.","penalties — both passes agree at 0/0 and both are right: Vast is an aerospace company, his wealth is self-made from businesses he personally built, and no citable evidence of paid coverage or family funding of an AI company exists."],"years_language_modeling":1,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2025-2026","evidence":"mid-list co-author on long-context / memory / test-time-training arXiv papers (2503.19815, 2511.21726, 2512.23675, 2601.16175)"}],"frontier_lineage":["Recent mid-list co-authorship in the long-context / test-time-training research line frontier labs draw on; no named foundational block and no pre-2013 vector-space precursor of his own"],"technical_founder_roles":["Ripple — founder/CTO — until 2013 (crypto payments, not LM)","Stellar — co-founder/CTO — 2014+ (consensus/payments, not LM)","Mt. Gox / eDonkey2000 / Overnet — creator (P2P/exchange, not LM)","Vast — founder/ex-CEO/chairman — 2021+ (aerospace, not LM)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":6,"lm_domain_depth":5,"lm_domain_breadth":4,"scientific_founder":7},"confidence":0.72,"duration_ms":71066},{"pass":"pass_2","dimensions":{"frontier_founder":6,"lm_domain_depth":5,"lm_domain_breadth":4,"scientific_founder":5},"confidence":0.7,"duration_ms":61672}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Jed_McCaleb","wikipedia_title":"Jed McCaleb","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"mira-murati","name":"Mira Murati","title":"Co-founder & CEO","company":"Thinking Machines Lab","sector":"general","rank":54,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":3,"transformers_lm":6,"frontier_founder":6,"lm_domain_depth":7,"lm_domain_breadth":7,"hands_on_engineering":8,"industry_impact":12,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":32,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":32,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.83,"rationale":"Murati holds a B.Eng in mechanical engineering (Dartmouth, 2012) and a B.A. from Colby College — no graduate degree, thesis, or coursework record in linear algebra/optimization/statistical learning, and no personal authored papers in vector embeddings or the attention/transformer lineage. Her one identified individual publication is a non-technical essay, 'Language & Coding Creativity' (Daedalus, 2022), about creativity rather than ML research. Her OpenAlex record (3 works, h-index 3) consists entirely of large collaborative OpenAI system-card papers (Codex eval 2021, GPT-4o System Card, o1 System Card) where she appears as one of dozens-to-hundreds of co-authors in a corporate-report capacity typical of a CTO signing off on org output, not as a research contributor with a distinguishable technical contribution. Her career trajectory — Tesla product manager, Leap Motion product/engineering lead, OpenAI VP then CTO — is consistently product/engineering leadership rather than personal hands-on model-building; Wikipedia explicitly frames her OpenAI role as overseeing 'research, product and safety teams' rather than authoring research. She does get meaningful hands_on_engineering/industry_impact credit for having led (not built) the engineering org that shipped ChatGPT, DALL-E, Codex, and Sora, and for now running Thinking Machines Lab, but per the rubric's explicit instruction, fame/leadership/company-building is not itself research depth, and no verifiable personal foundational-math, embeddings, or transformer research record exists.\n\nMurati's own work is not a foundational building block that today's frontier models descend from: she authored no attention/transformer/embedding/optimizer/tokenizer/scaling/alignment method, and her only lineage-adjacent authorship is as one of ~58 co-authors on the 2021 Codex evaluation paper in a corporate-sign-off (CTO) capacity, so frontier_founder sits low. Her language-modeling record is ~6 years of product/engineering LEADERSHIP over LM systems at OpenAI (VP then CTO, 2018–2024, ChatGPT/GPT-4/DALL-E) plus ~1.5 years running Thinking Machines Lab (2025–), not personal hands-on modeling — the PubMed 2011 signal is a confirmed radiology homonym, so first verifiable LM year is 2018. Distinct LM domains with a verifiable record are two — natural-language text (org leadership of GPT/ChatGPT) and code (co-author on the Codex paper) — both via leadership/co-authorship rather than solo modeling. As a founder she is co-founder & CEO of one AI company, but its technical publications (Defeating Nondeterminism, Modular Manifolds, LoRA Without Regret, On-Policy Distillation) are authored by Horace He, Jeremy Bernstein, John Schulman and Kevin Lu — the science is done by technical co-founders, so she is a business/product founder-CEO, not the scientific/technical founder, earning zero technical-founder years.","evidence":[{"claim":"B.Eng mechanical engineering, Dartmouth Thayer School, 2012; B.A. Colby College, 2011 — no graduate ML/math degree","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career path was product/engineering management: PM at Tesla (2013-16), product+eng lead at Leap Motion (2016-18), VP then CTO at OpenAI (2018-2024) overseeing teams rather than authoring research","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Only identified personal-authorship publication is a non-technical essay on creativity, not ML research","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BA Colby College (2011), BEng mechanical engineering Dartmouth Thayer School (2012); Tesla Model X product manager 2013-2016; Leap Motion product/engineering 2016-2018; OpenAI 2018-2024 rising to CTO, led work on ChatGPT, DALL-E, Codex and Sora; co-founder and CEO of Thinking Machines Lab from Febru","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of 58 authors on 'Evaluating Large Language Models Trained on Code' (Codex, arXiv:2107.03374, July 2021)","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thinking Machines Lab technical publications (Defeating Nondeterminism in LLM Inference, Modular Manifolds, LoRA Without Regret, On-Policy Distillation) are authored by Horace He, Jeremy Bernstein, John Schulman and Kevin Lu; Murati is not an author on any of them","source_url":"https://thinkingmachines.ai/blog/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records her education at Colby, Dartmouth/Thayer and Pearson College UWC and her notable work as ChatGPT, DALL-E, GPT-4 and Thinking Machines Lab","source_url":"https://www.wikidata.org/wiki/Q116706551","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Thinking Machines Lab (2025); CTO of OpenAI 2022–2024, overseeing research/product/safety teams rather than authoring research","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of ~58 authors on 'Evaluating Large Language Models Trained on Code' (Codex), arXiv:2107.03374, July 2021 — a corporate-capacity co-authorship, code-LM lineage","source_url":"https://arxiv.org/abs/2107.03374","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thinking Machines Lab technical blog posts are authored by Horace He, Jeremy Bernstein, John Schulman and Kevin Lu; Murati is not an author on any of them, indicating she is not the company's scientific founder","source_url":"https://thinkingmachines.ai/blog/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records her notable work as ChatGPT, DALL-E and GPT-4 (org leadership) and education as mechanical engineering (Dartmouth/Thayer) — no graduate ML/research record","source_url":"https://www.wikidata.org/wiki/Q116706551","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAI VP then CTO 2018–2024 overseeing research, product and safety teams that shipped ChatGPT, GPT-4, DALL-E and Codex — leadership of language-modeling work rather than personal authorship","source_url":"https://en.wikipedia.org/wiki/Mira_Murati","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Thinking Machines Lab (founded Feb 2025); the lab's technical papers (Defeating Nondeterminism in LLM Inference, Modular Manifolds, LoRA Without Regret, On-Policy Distillation) are authored by Horace He, Jeremy Bernstein, John Schulman and Kevin Lu — Murati is not an author on","source_url":"https://thinkingmachines.ai/blog/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["B.A., Colby College (2011)","B.Eng, Mechanical Engineering, Dartmouth College / Thayer School of Engineering (2012)"],"canonical_papers":[],"first_verifiable_year":2011,"notable_systems":["ChatGPT (org leadership)","DALL-E (org leadership)","Codex (org leadership)","Sora (org leadership)","Thinking Machines Lab (co-founder/CEO)"],"citations":1668,"h_index":3,"patents":0,"dossier_notes":"PubMed block (23 results, earliest 2011) is a clear homonym — those are pediatric radiology/oncology case reports by unrelated clinicians named Murati M/Murati MA, not this Mira Murati; the dossier's 'first_verifiable_year: 2011' and 'years_active: 15' derived from that PubMed match are WRONG and should not be trusted. Semantic Scholar's 26 papers / 39,639 citations / h-index 6 is also almost certainly a merged/ambiguous profile (far exceeds the clean OpenAlex match of 3 works/1,668 citations) and should be discounted in favor of the verified OpenAlex set. No profile_page/self_claims were available to filter.","years_language_modeling":6,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2018-2024","evidence":"Led OpenAI product/engineering shipping ChatGPT, GPT-4, DALL-E as VP then CTO — leadership, not personal modeling"},{"domain":"code","years":"2021","evidence":"Co-author (1 of ~58) on the Codex evaluation paper, arXiv:2107.03374 — corporate-capacity co-authorship"}],"frontier_lineage":["No named foundational block authored; only corporate-capacity co-authorship on the 2021 Codex evaluation paper, part of the code-LLM lineage"],"technical_founder_roles":["Thinking Machines Lab — co-founder & CEO (business/product founder; technical science authored by co-founders) — 2025-"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":5,"lm_domain_depth":7,"lm_domain_breadth":8,"scientific_founder":5},"confidence":0.72,"duration_ms":68701},{"pass":"pass_2","dimensions":{"frontier_founder":6,"lm_domain_depth":7,"lm_domain_breadth":6,"scientific_founder":5},"confidence":0.83,"duration_ms":69988}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Mira_Murati","wikipedia_title":"Mira Murati","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"evan-cheng","name":"Evan Cheng","title":"Co-founder & CEO","company":"Mysten Labs (Sui)","sector":"crypto","rank":55,"rubric_version":4,"dimensions":{"foundations":7,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":13,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":14,"industry_impact":10,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":31,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":31,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.68,"rationale":"Cheng is a genuine, decorated hands-on systems engineer: he shared the 2012 ACM Software System Award with Chris Lattner and Vikram Adve for 'designing and implementing LLVM,' the compiler infrastructure that underlies most modern ML training/inference backends (XLA, MLIR, Triton, PyTorch's compiler stack). He was part of Meta's Diem blockchain project before co-founding Mysten Labs (2021), which built the Sui Layer-1 blockchain (per Wikipedia's list of Mysten Labs' original authors). This is compiler/systems engineering depth, not AI-specific research: no verifiable personal record in vector embeddings, statistical learning theory, or transformer/LM authorship was found, and the dossier's OpenAlex match (neuroscience/genetics papers, single-cell transcriptomics) is very likely a different Evan Cheng and was disregarded as an unconfirmed homonym. His hands_on_engineering score reflects real, award-recognized infrastructure engineering that AI systems depend on, while the AI-specific research dimensions remain low absent direct evidence.\n\nCheng's verifiable foundational contribution is compiler infrastructure — he is a co-designer of LLVM (2012 ACM Software System Award) and, per his own bio, worked on Clang, Swift, MLIR and CIRCT; MLIR/LLVM are the canonical IR/codegen layers underneath today's ML training and inference compiler stacks (XLA, Triton, PyTorch/Glow, TensorFlow), so his work sits in the frontier training/inference-stack lineage — general-purpose, not language-modeling-specific, placing him at the bottom of the 13-17 band. He has NO verifiable language-modeling record (no vector-space/embedding/transformer/LM papers, patents or shipped LM systems), so both lm_domain_depth (0 LM years) and lm_domain_breadth (0 LM domains) score near the floor. As technical founder he is co-founder & CEO of Mysten Labs (founded Sept 2021, ~5 years), a real engineer setting technical direction, but Mysten's core is blockchain (Move/cryptography, done by co-founders Blackshear, Danezis, Chalkias), i.e. a technical founder OUTSIDE the language-modeling field — one company, ~5 years — landing in the 3-7 scientific_founder band. He led Novi Research and programming-languages/runtimes engineering at Meta, but no verifiable LM-specific research emerged from that.","evidence":[{"claim":"Co-recipient (with Chris Lattner and Vikram Adve) of the ACM Software System Award 2012 for designing and implementing LLVM","source_url":"https://en.wikipedia.org/wiki/LLVM","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"One of Mysten Labs' five original founders, formerly part of Meta's Diem project team, which left Meta to found Mysten Labs in September 2021","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evan Cheng shared the 2012 ACM Software System Award with Vikram Adve and Chris Lattner for designing and implementing LLVM","source_url":"https://en.wikipedia.org/wiki/LLVM","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"2012 ACM Software System Award list: LLVM - Vikram S. Adve, Evan Cheng, Chris Lattner","source_url":"https://en.wikipedia.org/wiki/ACM_Software_System_Award","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Mysten Labs founded September 2021 by five former Meta/Diem engineers including Evan Cheng, Sam Blackshear (creator of the Move language), Adeniyi Abiodun, George Danezis and Kostas Chalkias, to build the Sui Layer 1","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evan Cheng is NOT an author of 'Glow: Graph Lowering Compiler Techniques for Neural Networks' (arXiv 1805.00907), Facebook's neural-network compiler","source_url":"https://arxiv.org/abs/1805.00907","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evan Cheng shared the 2012 ACM Software System Award for designing and implementing LLVM, the compiler infrastructure underlying modern ML compiler/inference stacks","source_url":"https://en.wikipedia.org/wiki/ACM_Software_System_Award","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cheng's bio lists work on LLVM, Clang, Swift, MLIR and CIRCT; MLIR is the multi-dialect IR built for ML frameworks and accelerators that lowers to LLVM IR","source_url":"https://mlir.llvm.org/users/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evan Cheng is Co-Founder & CEO of Mysten Labs (founded September 2021), having led Novi Research and programming-languages/runtimes engineering at Meta; Sam Blackshear is CTO","source_url":"https://www.mystenlabs.com/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Mysten Labs / Sui was founded in 2021 by former Meta Diem engineers; its core is a Layer-1 blockchain and the Move language, not language modeling","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Evan Cheng shared the 2012 ACM Software System Award with Vikram Adve and Chris Lattner for designing and implementing LLVM, the compiler infrastructure that underlies modern ML compilation stacks (XLA, MLIR, Triton, torch.compile)","source_url":"https://en.wikipedia.org/wiki/ACM_Software_System_Award","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cheng is a co-founder of Mysten Labs (September 2021), which built the Sui Layer-1 blockchain — a blockchain company, not a language-modeling company","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2012,"notable_systems":["LLVM compiler infrastructure","Diem (Meta)","Sui blockchain / Mysten Labs"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's OpenAlex/Semantic Scholar matches (single-cell transcriptomics, Alzheimer's, yeast meiosis papers at NYU/Stony Brook) are a different Evan Cheng — ambiguous match (9 candidates), topics unrelated to compilers/blockchain/AI infrastructure — and were disregarded as an unverified homonym. Wikipedia/Wikidata blocks are correctly populated but only cover the Sui platform, not a dedicated biography page for Cheng himself; his LLVM award and Mysten Labs co-founding were confirmed via LLVM's own Wikipedia page and the Sui blockchain page. Could not independently verify commonly-cited claims about his prior roles leading ML/AI infrastructure at Apple or Facebook (PyTorch-adjacent); these were excluded from scoring for lack of a citable source.","years_language_modeling":0,"years_as_technical_founder":5,"technical_founder_companies":1,"frontier_lineage":["LLVM compiler infrastructure — codegen/backend layer beneath ML training & inference compiler stacks","MLIR/CIRCT compiler work feeding modern ML compilation pipelines (XLA, Triton, PyTorch/Glow)"],"technical_founder_roles":["Mysten Labs — Co-founder & CEO — 2021-2026 (blockchain, outside language-modeling field)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":13,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.8,"duration_ms":100620},{"pass":"pass_2","dimensions":{"frontier_founder":13,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":6},"confidence":0.6,"duration_ms":52199}],"source":"seeded","profile_url":null,"wikipedia_title":"Sui (blockchain platform)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jensen-huang","name":"Jensen Huang","title":"Founder, President & CEO","company":"NVIDIA","sector":"general","rank":56,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":10,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":10,"industry_impact":15,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":31,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":31,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.89,"rationale":"Huang holds a BS in Electrical Engineering (Oregon State, 1984) and an MS in Electrical Engineering (Stanford, 1992), and worked as a hands-on chip designer at AMD and LSI Logic (1984-1993) before founding NVIDIA in 1993. No personal peer-reviewed papers, patents, or authored research in AI/ML math, embeddings, or transformers were found; the dossier's own OpenAlex match (a single 'Foreword' to a book chapter) is not genuine research output. NVIDIA's GPUs and CUDA platform are foundational AI infrastructure, but CUDA and NVIDIA's ML-relevant architectures were built by teams of NVIDIA engineers over decades, not personally designed by Huang, whose role since 1993 has been founder/CEO. This is a case of a company built years before the deep-learning/transformer era later becoming critical infrastructure for it — strong organizational industry impact, but no personal research or engineering record in the core dimensions this rubric measures.\n\nHuang's lineage position is real but organizational, not personal: he founded and has run NVIDIA since 1993, whose GPUs and the CUDA stack are the canonical training/inference substrate every frontier GPT/Claude/Gemini/Llama-class model trains and serves on — yet CUDA, the tensor-core architectures and the deep-learning libraries were authored by NVIDIA engineering and research teams, not by Huang, who holds no papers, patents or code in the language-modeling lineage. He has zero verifiable hands-on years in language modeling specifically (statistical/neural LMs, vector-space text, seq2seq, transformers, pretraining/alignment) — his BSEE/MSEE and 1984–1993 chip-design work at AMD/LSI predate and are orthogonal to that field — so lm_domain_depth and lm_domain_breadth show no distinct LM domain with a personal record. As a founder he was the business/CEO principal among technical co-founders (Curtis Priem and Chris Malachowsky architected the early graphics silicon), which the rubric places in the 'founder whose science was done by others' band rather than founder-Chief-Scientist authorship. One company, ~33 years, but not as the personal author of its core research or code.","evidence":[{"claim":"BS Electrical Engineering, Oregon State University (1984)","source_url":"https://engineering.oregonstate.edu/alumni-partners/oregon-stater-awards/searchable-awards-database/jen-hsun-huang-engineering-hall","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MS Electrical Engineering, Stanford University (1992)","source_url":"https://engineering.stanford.edu/about/history/heroes/2018-heroes/jensen-huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chip design engineer at AMD then LSI Logic (1984-1993) before co-founding NVIDIA in 1993","source_url":"https://nvidianews.nvidia.com/bios/jensen-huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No IEEE/ACM peer-reviewed publications found under his name; OpenAlex record shows only a single non-research 'Foreword' (2022)","source_url":"https://corporate-awards.ieee.org/recipient/jensen-huang/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BSEE Oregon State University 1984 with highest honours; MSEE Stanford University 1992; designed AMD microprocessors; joined LSI Logic as a technical officer where he worked with Malachowsky and Priem on graphics accelerators; co-founded Nvidia 1993 and has been president and CEO since day one","source_url":"https://en.wikipedia.org/wiki/Jensen_Huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Official Nvidia biography describes him as founder, president and CEO since 1993, holding BSEE (Oregon State) and MSEE (Stanford), with no research publications or patents listed","source_url":"https://nvidianews.nvidia.com/bios/jensen-huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder, president and CEO of NVIDIA since 1993; the company's GPUs and CUDA platform are the dominant compute stack for training and serving frontier AI models","source_url":"https://en.wikipedia.org/wiki/Jensen_Huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NVIDIA was co-founded by Jensen Huang with Curtis Priem and Chris Malachowsky, the engineers who architected the early graphics chips; Huang served as CEO","source_url":"https://en.wikipedia.org/wiki/Nvidia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder, president and CEO of NVIDIA since 1993; NVIDIA GPUs/CUDA are the dominant AI training and inference platform","source_url":"https://en.wikipedia.org/wiki/Jensen_Huang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NVIDIA was co-founded in 1993 by Jensen Huang with Chris Malachowsky and Curtis Priem, who led the core graphics/silicon engineering","source_url":"https://en.wikipedia.org/wiki/Nvidia","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prior assessment: no personal peer-reviewed AI/ML publications, embeddings, or transformer authorship; OpenAlex shows only a non-research 'Foreword' (2022)","source_url":"https://corporate-awards.ieee.org/recipient/jensen-huang/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Electrical Engineering, Oregon State University (1984)","MS Electrical Engineering, Stanford University (1992)"],"canonical_papers":[],"first_verifiable_year":1984,"notable_systems":["NVIDIA GPU architectures (organizational leadership)","CUDA platform (organizational leadership)"],"citations":0,"h_index":0,"patents":1,"dossier_notes":"Dossier's OpenAlex match (works_count=1, a book 'Foreword') is not a genuine research record and should not be used to infer any research depth. PubMed sample is entirely homonym noise (unrelated Huang J. authors in botany, materials science, etc.) with high homonym_risk correctly flagged; none of it is this Jensen Huang. No credible personal patent record for GPU/AI hardware was found.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":1,"frontier_lineage":["NVIDIA GPUs + CUDA + tensor-core/cuDNN training and inference stack that frontier models train and serve on — organizational, via the company he founded and leads, not personal authorship"],"technical_founder_roles":["NVIDIA — founder, president & CEO (business/CEO principal among technical co-founders who did the chip science) — 1993–present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":10,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":6},"confidence":0.83,"duration_ms":65293},{"pass":"pass_2","dimensions":{"frontier_founder":11,"lm_domain_depth":3,"lm_domain_breadth":2,"scientific_founder":8},"confidence":0.82,"duration_ms":72873}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Jensen_Huang","wikipedia_title":"Jensen Huang","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"shaw-walters","name":"Shaw Walters","title":"Founder & lead developer","company":"Eliza Labs / elizaOS (formerly ai16z)","sector":"crypto","rank":57,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":7,"transformers_lm":6,"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":4,"hands_on_engineering":13,"industry_impact":8,"scientific_founder":10},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":31,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":31,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.8,"rationale":"Both passes correctly threw out the dossier, whose OpenAlex match is 'Warren Shaw. Walters', author of a 1973 Naval Postgraduate School thesis on Marine Corps officer assignment, producing a nonsensical timeline of 53 years active for a person whose record starts this decade. The dispute is that pass 2 found a publication pass 1 declared did not exist, and it is real: I verified that he is the first author of 'Eliza: A Web3 friendly AI Agent Operating System' (arXiv:2501.06781, January 2025), ahead of thirteen co-authors. That is a genuine first-author paper, but its content is decisive for how much it earns — it describes a TypeScript agent runtime that lets language-model agents read and write blockchain data and call smart contracts. It is a systems and framework paper, not a contribution to the mathematics, to representation learning, or to the attention/transformer lineage, so it lifts hands-on engineering rather than the research dimensions. The elizaOS repository is a substantial shipped artifact (~19.3k stars, MIT-licensed) whose core includes memory and knowledge management, local embedding generation via a local-inference plugin with runtime-managed GGUF models, and retrieval-augmented generation, which is personally shipped embedding and retrieval machinery and justifies pass 2's direction on vector_embeddings over pass 1's 4 — though wiring existing embedding models into an agent memory store is integration, not authorship of retrieval methods, so 9 overshoots. He has no degree, no thesis, no mathematics or ML publication, and he trains no models: he orchestrates other people's. Extensive media coverage concerns the ai16z DAO and token and is explicitly not evidence here.\n\nWalters sits downstream of the frontier stack, not in its foundation: elizaOS is a TypeScript runtime that wires existing LLMs (via multiple external model providers) into blockchain reads/writes and agent memory — he applies and orchestrates frontier models rather than contributing any architecture, optimizer, embedding method, pretraining objective, dataset or scaling result they descend from, and there is no pre-2013 vector-space or distributional-semantics record, placing him in the 'applies/fine-tunes frontier models' band. His verifiable hands-on record in anything language-modeling adjacent begins ~2021-2022 (Webaverse AI characters, Magick) and runs through elizaOS's shipped embedding/RAG/memory machinery and the Jan-2025 Eliza paper — roughly three years of applied, integration-level work, not research or model training, so lm_domain_depth is thin. That work is confined to a single domain, general natural-language agents (Web3 agents are LLMs applied to crypto, which the rubric excludes as domain language modeling; the pinned 'autocoder' code tool is at most a weak second surface), so breadth is a single-domain score. Where he earns real credit is as scientific/technical founder: he is the verified creator and lead developer of the ~19.3k-star elizaOS framework and first author of arXiv:2501.06781, personally authoring the core code his company (Eliza Labs, formerly ai16z) runs on — roughly four years as a hands-on technical founder of one such company, landing in th","evidence":[{"claim":"First author of 'Eliza: A Web3 friendly AI Agent Operating System' (arXiv:2501.06781, January 2025) — verified author order begins Shaw Walters, Sam Gao, Shakker Nerd, ahead of 13 co-authors; describes an open-source TypeScript agentic framework integrating LLM agents with blockchain reads, writes a","source_url":"https://arxiv.org/abs/2501.06781","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"elizaOS/eliza is an MIT-licensed TypeScript agentic operating system (~19.3k stars, 5.7k forks) providing memory and knowledge workflows, local embeddings via a local-inference plugin with runtime-managed GGUF models, RAG, and multiple model providers","source_url":"https://raw.githubusercontent.com/elizaOS/eliza/main/README.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub profile 'lalalune' (Shaw, San Francisco), member of the Eliza Labs and elizaOS organizations, creator of elizaOS, with pinned AI projects Magick, CharacterStudio and autocoder","source_url":"https://github.com/lalalune","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder of Eliza Labs and lead developer of the ElizaOS open-source AI agent framework, formerly known as ai16z","source_url":"https://iq.wiki/wiki/shaw-walters","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Path into AI ran through blockchain/game development (NFT projects, Webaverse metaverse AI characters) where he learned Solidity before building agent infrastructure","source_url":"https://www.blockchaingamer.biz/features/interviews/41648/shaw-walters-elizaos-ai-agents-blockchain-gaming/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Early GPT-3 tester before ChatGPT's release, cited as basis for his interest in AI agents","source_url":"https://joetechnologist.com/elizaos-the-age-of-ai-agents-a-conversation-with-shaw-walters/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Eliza: A Web3 friendly AI Agent Operating System' (2025), an open-source agentic framework integrating LLM agents with blockchain operations as a TypeScript program","source_url":"https://arxiv.org/abs/2501.06781","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"elizaOS/eliza is an MIT-licensed TypeScript agentic operating system with ~19.3k stars providing memory/knowledge workflows, local embeddings via a local-inference plugin with runtime-managed GGUF models, RAG, and multiple model providers","source_url":"https://raw.githubusercontent.com/elizaOS/eliza/main/README.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"elizaOS/eliza is an MIT-licensed TypeScript agentic OS (~19.3k stars) providing memory/knowledge workflows, local embeddings via a local-inference plugin with runtime-managed GGUF models, RAG, and multiple external model providers — integration of existing embedding/LLM models, a single natural-lang","source_url":"https://raw.githubusercontent.com/elizaOS/eliza/main/README.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub 'lalalune' (Shaw, San Francisco): creator of elizaOS, member of Eliza Labs / elizaOS orgs, with prior AI projects Magick, CharacterStudio and autocoder — the hands-on technical-founder record","source_url":"https://github.com/lalalune","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder of Eliza Labs and lead developer of the elizaOS open-source AI agent framework, formerly known as ai16z","source_url":"https://iq.wiki/wiki/shaw-walters","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"elizaOS/eliza is an MIT-licensed TypeScript agentic OS providing memory/knowledge workflows, local embeddings via a local-inference plugin with runtime-managed GGUF models, RAG and multiple model providers — integration of existing models, not authorship of LM methods","source_url":"https://raw.githubusercontent.com/elizaOS/eliza/main/README.md","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder of Eliza Labs and lead developer of the elizaOS open-source agent framework (formerly ai16z) — verifiable technical founder who personally wrote the core code","source_url":"https://iq.wiki/wiki/shaw-walters","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub 'lalalune' (Shaw), creator of elizaOS, member of Eliza Labs/elizaOS orgs, with pinned AI projects Magick, CharacterStudio and autocoder — hands-on building record traceable to ~2021-2022","source_url":"https://github.com/lalalune","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["Eliza: A Web3 friendly AI Agent Operating System (arXiv:2501.06781, 2025) — first author"],"first_verifiable_year":2021,"notable_systems":["elizaOS / Eliza agent framework","Magick AI builder toolkit","CharacterStudio","Webaverse metaverse AI characters"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's academic blocks are all the wrong person. OpenAlex A5078073134 ('Warren Shaw. Walters', 1973 Naval Postgraduate School thesis) produced a false timeline of first_verifiable_year 1973 and 53 years active. The Semantic Scholar match ('S. Walters', 4 papers) and the PubMed records (immunology, medulloblastoma, islet transplant) are unrelated homonyms. The dossier entirely misses his one real publication, the Eliza arXiv paper on which he is first author, and misses his GitHub engineering record. His earliest verifiable AI-agent work traces to roughly 2021-2022 via Webaverse and Magick, predating the 2025 paper.","adjudication_notes":["vector_embeddings (4 vs 9): pass 2 was right that shipped embedding machinery exists; settled at 7. Verified: the elizaOS README documents local embedding generation via a local-inference plugin with runtime-managed GGUF models plus RAG and memory management, which pass 1 missed — but this is integrating existing embedding models into an agent memory store, not authoring retrieval or representation methods, so 9 overshoots.","hands_on_engineering (11 vs 14): pass 2 was right; settled at 13. He is verified first author of arXiv:2501.06781 and the creator of a ~19.3k-star shipped agent runtime, a substantial personally-built system.","transformers_lm (5 vs 6): settled at 6. The Eliza paper is an agent-runtime systems paper; he trains no models and has no work on attention, pretraining or scaling. Pass 1's assertion that no publication exists at all is corrected, but the paper's content does not move this dimension much.","foundations (3 vs 4): pass 1 was right; settled at 3. No degree, thesis, or mathematics/ML publication is verifiable from any source; he is self-taught and the Eliza paper contains no foundational contribution.","industry_impact (8 vs 9): settled at 8. elizaOS is genuinely widely adopted, but its core is agent orchestration over other people's models, and the extensive coverage of the ai16z DAO and token is explicitly not evidence."],"years_language_modeling":3,"years_as_technical_founder":4,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2022-2025","evidence":"elizaOS agent framework: memory/RAG/embedding integration over external LLMs; general-purpose NL agents"}],"technical_founder_roles":["Eliza Labs (ai16z / elizaOS) — founder & lead developer, author of core framework code and first author of arXiv:2501.06781 — ~2022-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":5,"lm_domain_breadth":4,"scientific_founder":10},"confidence":0.7,"duration_ms":61427},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":9},"confidence":0.72,"duration_ms":58685}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"alexandr-wang","name":"Alexandr Wang","title":"Chief AI Officer, Meta / Co-Founder (former CEO), Scale AI","company":"Meta Superintelligence Labs (formerly Scale AI)","sector":"general","rank":58,"rubric_version":4,"dimensions":{"foundations":5,"vector_embeddings":3,"transformers_lm":6,"frontier_founder":7,"lm_domain_depth":6,"lm_domain_breadth":4,"hands_on_engineering":6,"industry_impact":11,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":30,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":30,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.81,"rationale":"Wang has no completed degree in a technical field and no personal record of authored research in linear algebra/optimization/statistical learning, embeddings, or the transformer/LM lineage: he attended MIT intending to study machine learning but dropped out after one year (2016) to found Scale AI, and prior to that worked as a software engineer at Quora and Addepar, plus strong pre-college competitive math/programming results (Math Olympiad Program, US Physics Team, USACO) which show aptitude but are not verifiable research. Scale AI itself is a data-labeling/annotation and model-evaluation infrastructure business — Wang built and led the company (CEO for ~9 years) but the searched record shows him as a business/organizational leader rather than a hands-on model builder; OpenAlex/Semantic Scholar list him as a co-author on a handful of recent (2024-2026) multi-author policy/benchmark papers (WMDP unlearning benchmark, a Nature academic-benchmark paper, and a 'Superintelligence Strategy' position paper), which are large-consortium outputs consistent with an executive-sponsor/co-author role rather than a first-author technical contribution, and none rise to canonical original transformer/embedding research. His industry_impact score reflects that he built and scaled a company whose core product (human-in-the-loop data labeling and RLHF/eval infrastructure for foundation-model training) genuinely sits adjacent to LM training pipelines, and he now leads Meta's Superintelligence Labs, but per the rubric this is leadership/infrastructure-provisioning impact, not personal foundational research.\n\nWang authored no method, architecture, optimizer, tokenizer or pretraining objective that frontier models descend from; his lineage claim rests on Scale AI having supplied human-labeled RLHF and instruction-tuning data plus LLM evaluation benchmarks (WMDP, Humanity's Last Exam) that frontier labs consumed — an organizational data-pipeline contribution to the training stack, not a personal foundational block, and nothing pre-2013 exists in the record. His verifiable language-modeling record is roughly 2018-2026 building LLM data-labeling and evaluation infrastructure at Scale AI (~5 years), which is adjacent to language modeling rather than model-building, and it is confined to a single domain — general natural-language text — so breadth is thin. He co-founded and ran Scale AI (~2016-2025, one company) as CEO, but the searched record shows him as a business/organizational leader whose core science and engineering were done by others (dropped out of MIT after one year, no first-author technical papers, no patents), which is the 3-7 'founder/CEO of an AI company, science by others' band. No pre-2013 vector-space or distributional-semantics work, no multi-domain modeling, and no personal authorship of frontier building blocks are verifiable.","evidence":[{"claim":"Wang briefly attended MIT intending to study machine learning, dropped out after his freshman year (summer 2016) to found Scale AI with Lucy Guo — no degree completed","source_url":"https://en.wikipedia.org/wiki/Alexandr_Wang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Before MIT, Wang worked as a software engineer at Quora and at Addepar; his pre-college record includes Math Olympiad Program (2013), US Physics Team (2014), USACO finalist (2012, 2013)","source_url":"https://www.entrepreneur.com/business-news/who-is-alexandr-wang-the-founder-of-scale-ai-joining-meta/493281","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Scale AI's business is AI data labeling / annotation and model evaluation services (Remotasks, Outlier), not model research; Wang co-founded it in 2016 and served as CEO until stepping down in 2025 to become Meta's Chief AI Officer","source_url":"https://en.wikipedia.org/wiki/Scale_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The WMDP Benchmark paper (arXiv 2403.03218) is a large multi-institution consortium paper led by researchers at the Center for AI Safety (Nathaniel Li, Dan Hendrycks et al.), consistent with Wang appearing as an organizational/sponsoring co-author rather than lead technical author","source_url":"https://arxiv.org/abs/2403.03218","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Briefly attended MIT and dropped out to co-found Scale AI in 2016 with Lucy Guo; prior roles as software engineer at Addepar, programmer at Quora, intern at Hudson River Trading; Scale AI provides data labeling and LLM evaluation services; Chief AI Officer of Meta and head of Superintelligence Labs","source_url":"https://en.wikipedia.org/wiki/Alexandr_Wang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed among 56 authors of 'The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning' (2024) — a hazardous-knowledge evaluation benchmark and unlearning method.","source_url":"https://arxiv.org/abs/2403.03218","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed among 1,155+ authors of 'Humanity's Last Exam' (2025), a 2,500-question LLM evaluation benchmark.","source_url":"https://arxiv.org/abs/2501.14249","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Scale AI's core business is AI data labeling/annotation and LLM evaluation (Remotasks, Outlier) supplying RLHF/eval data used in foundation-model training pipelines; Wang co-founded it in 2016 and was CEO until 2025","source_url":"https://en.wikipedia.org/wiki/Scale_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wang appears among 56 authors of the WMDP unlearning benchmark (2024) and 1,155+ authors of 'Humanity's Last Exam' (2025) — multi-institution LLM evaluation benchmarks, consistent with an organizational/sponsor co-author role rather than lead technical authorship","source_url":"https://arxiv.org/abs/2403.03218","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wang attended MIT for ~1 year and dropped out in 2016 to found Scale AI; no completed technical degree and no first-author transformer/embedding/LM research in the record","source_url":"https://en.wikipedia.org/wiki/Alexandr_Wang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Scale AI provides data labeling and LLM evaluation/RLHF-data services used to train and evaluate foundation models — infrastructure adjacent to frontier training stacks, founded and led by Wang (CEO 2016-2025)","source_url":"https://en.wikipedia.org/wiki/Scale_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wang's earliest verifiable LM-lineage authorship is 2024+ multi-institution eval/policy work (WMDP unlearning benchmark, arXiv 2403.03218), as one of dozens of co-authors rather than a first-author technical contributor","source_url":"https://arxiv.org/abs/2403.03218","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wang co-founded Scale AI in 2016 after dropping out of MIT after one year; he is a founder-CEO/business leader, with no completed technical degree and no personal authored transformer/embedding research","source_url":"https://en.wikipedia.org/wiki/Alexandr_Wang","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended Massachusetts Institute of Technology, ~1 year, no degree completed (dropped out 2016)"],"canonical_papers":[],"first_verifiable_year":2024,"notable_systems":["Scale AI data-labeling/annotation platform","Scale AI RLHF/model-evaluation infrastructure","Meta Superintelligence Labs (leadership role)"],"citations":813,"h_index":4,"patents":0,"dossier_notes":"The dossier's OpenAlex match is explicitly flagged ambiguous (4 candidates) and its listed topics (materials science, network security, topic modeling) do not cohere with a single-person research identity, indicating residual mixing with other 'Alexandr/Alexander Wang' authors — treated the 3-4 identifiable AI-safety/policy papers as the plausible genuine set (consistent with his post-2024 public AI-policy role) but did not credit the unrelated topics. The dossier's PubMed block is entirely unrelated homonym contamination (veterinary, dairy science, radiotherapy, neutrino physics papers by other 'Wang A' authors) and was disregarded. Found no credible, specifically-citable source documenting either paid/purchased media coverage or family/inherited-wealth funding of Scale AI's founding (reporting indicates Y Combinator and standard VC funding rounds, not family money); per the brief, absent a citable source both penalties are scored 0, but the underlying concern the brief anticipates — a founder with no personal language-modeling research record building an 'AI company' — is real and reflected in the low core-research dimension scores rather than via an uncited penalty.","years_language_modeling":5,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2018-2026","evidence":"Scale AI human-labeled RLHF/instruction data and LLM evaluation benchmarks (WMDP, Humanity's Last Exam) for natural-language foundation models"}],"frontier_lineage":["Scale AI supplied human-labeled RLHF/instruction-tuning data and LLM evaluation benchmarks consumed by frontier labs — org-level data-pipeline contribution, not a personally authored method/architecture"],"technical_founder_roles":["Scale AI — co-founder & CEO (business/organizational, not scientific) — 2016-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":7,"lm_domain_depth":6,"lm_domain_breadth":3,"scientific_founder":6},"confidence":0.8,"duration_ms":63667},{"pass":"pass_2","dimensions":{"frontier_founder":7,"lm_domain_depth":6,"lm_domain_breadth":4,"scientific_founder":6},"confidence":0.72,"duration_ms":77709}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Alexandr_Wang","wikipedia_title":"Alexandr Wang","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"balaji-srinivasan","name":"Balaji Srinivasan","title":"Angel Investor / Author (former CTO, Coinbase)","company":"Network School","sector":"crypto","rank":59,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":4,"transformers_lm":3,"frontier_founder":2,"lm_domain_depth":7,"lm_domain_breadth":5,"hands_on_engineering":8,"industry_impact":8,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":30,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":30,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.75,"rationale":"Balaji Srinivasan holds a BS/MS/PhD in Electrical Engineering plus an MS in Chemical Engineering from Stanford, and taught Stanford courses in statistics, bioinformatics, and a popular 'Startup Engineering' course, giving him genuine graduate-level quantitative/statistical-learning training (foundations) — but this is bioinformatics/genomics-era statistical learning (pre-2013), not vector embeddings or the transformer lineage specifically. He co-founded the genomics company Counsyl (bioinformatics/statistics engineering) and later served as CTO of Coinbase and General Partner at a16z — these are technical-leadership and investing roles, not personal contributions to language modeling, embeddings, or transformers. There is no evidence of authored work in vector-space models, word embeddings, attention, or transformer architectures; his public output since ~2013 is largely commentary, books ('The Network State'), and media/podcast presence, which per rubric should not be rewarded. This is a case of strong general quantitative/statistics graduate training but a thin-to-absent record in the specific AI-lineage dimensions, correctly kept low on vector_embeddings/transformers_lm despite decent foundations.\n\nNo verifiable work of Srinivasan's sits in the lineage today's frontier language models descend from: his primary-research record (Graemlin multiple-network alignment, Genome Research 2006; automatic parameter learning for network alignment, J Comput Biol 2009; pharmacogenomics/carrier-screening, Hum Mutat 2009 and Reprod Biomed Online 2010) is biological interaction-network alignment and genomics statistics, with no pre-2013 distributional-semantics/LSI/vector-space/term-relationship text patents or systems, so frontier_founder stays at floor. His only credible language-modeling-lineage record is the one biological domain the rubric explicitly admits — genomics relationship networks — roughly 2006–2010 (~4 hands-on years); this is network alignment rather than sequence/text modeling, so lm_domain_depth sits at the boundary of the adjacent and real-record bands and lm_domain_breadth stays in the single-domain band (no natural-language, financial, code or other authored LM record; Coinbase CTO, a16z and Network School are leadership, investing and commentary). As a technical founder he genuinely authored the core science of Counsyl (first author on its carrier-screening method, ~2007–2013) and co-founded 21 Inc/Earn.com and Teleport — three technical-founder roles across ~8 years — but all in genomics and crypto, not these AI systems, capping scientific_founder in the 'technical founder outside this field' band. Coinbase CTO (2018–2019) was a hired role, not a founding one.","evidence":[{"claim":"BS/MS/PhD Electrical Engineering, MS Chemical Engineering, Stanford University; taught Stanford courses in statistics and bioinformatics","source_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'The Network State: How to Start a New Country'; founder of Network School retreat in Malaysia","source_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PubMed refined match under Stanford/Coinbase/a16z affiliation filter shows bioinformatics/pharmacogenomics papers (e.g., 'A universal carrier test for the long tail of Mendelian disease', 2010) consistent with Counsyl-era genomics work, not AI/ML","source_url":"https://pubmed.ncbi.nlm.nih.gov/20729146/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS, MS and PhD in electrical engineering plus MS in chemical engineering from Stanford; taught statistics and bioinformatics at Stanford; co-founded Counsyl (acquired by Myriad Genetics for $375M), 21 Inc/Earn.com and Teleport; Coinbase CTO 2018-2019; founded Network School September 2024","source_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 39363815 (Balaji S. Srinivasan): 21 papers, 1,881 citations, h-index 14, all in computational biology / network alignment / pharmacogenomics","source_url":"https://api.semanticscholar.org/graph/v1/author/39363815?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Graemlin: general and robust alignment of multiple large interaction networks (Genome Research 2006) — Srinivasan BS co-author with Batzoglou","source_url":"https://pubmed.ncbi.nlm.nih.gov/16899655/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Automatic parameter learning for multiple local network alignment (J Comput Biol 2009) — Srinivasan BS with Flannick, Novak, Do and Batzoglou","source_url":"https://pubmed.ncbi.nlm.nih.gov/19645599/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q87684934 lists doctoral advisors Brad Osgood and Serafim Batzoglou and occupations angel investor, entrepreneur, university teacher","source_url":"https://www.wikidata.org/wiki/Q87684934","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A universal carrier test for the long tail of Mendelian disease (2010) — Srinivasan BS first author, the Counsyl screening method","source_url":"https://pubmed.ncbi.nlm.nih.gov/20729146/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 39363815 (Balaji S. Srinivasan): 21 papers in computational biology/network alignment/pharmacogenomics — none in embeddings, attention or transformers","source_url":"https://api.semanticscholar.org/graph/v1/author/39363815?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded genomics company Counsyl and later 21 Inc/Earn.com and Teleport; CTO (hired) of Coinbase 2018-2019; founder of Network School","source_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Srinivasan BS is first author on 'A universal carrier test for the long tail of Mendelian disease' (2010), the Counsyl genetic-screening method — genomics, not language modeling","source_url":"https://pubmed.ncbi.nlm.nih.gov/20729146/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Graemlin: general and robust alignment of multiple large interaction networks' (Genome Research 2006) with Batzoglou — biological network alignment, not text/vector-space modeling","source_url":"https://pubmed.ncbi.nlm.nih.gov/16899655/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of genomics company Counsyl; CTO of Coinbase; GP at a16z; founder of Network School — no role authored embeddings/transformer/LM work","source_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 39363815: 21 papers in computational biology/network alignment/pharmacogenomics — the credible non-homonym subset, none in language modeling","source_url":"https://api.semanticscholar.org/graph/v1/author/39363815","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS/MS/PhD Electrical Engineering, Stanford University","MS Chemical Engineering, Stanford University"],"canonical_papers":["Automatic parameter learning for multiple local network alignment (2009)","A universal carrier test for the long tail of Mendelian disease (2010)"],"first_verifiable_year":2004,"notable_systems":["Counsyl genetic screening platform"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's broad OpenAlex match (103 works, ambiguous, 10 candidates, topics like ocular surface/biosensors/glaucoma) is largely homonym pollution unrelated to this Balaji Srinivasan (bioengineering/genomics PhD, not ophthalmology); the refined PubMed sample (Counsyl/Stanford/Batzoglou-affiliated genomics papers) is the credible subset and was used for scoring instead. No evidence found of any authored work touching vector embeddings, attention, or transformers specifically; his AI commentary is prolific but per rubric counts as media, not research.","years_language_modeling":4,"years_as_technical_founder":8,"technical_founder_companies":3,"lm_domains":[{"domain":"biological","years":"2006-2010","evidence":"Multiple biological interaction-network alignment papers (Graemlin, Genome Research 2006; J Comput Biol 2009) and genomics/pharmacogenomics screening (Hum Mutat 2009; Reprod Biomed Online 2010) — genomics relationship-network work with advisor S. Batzoglou; network alignment rather than sequence LM"}],"technical_founder_roles":["Counsyl — co-founder, authored core carrier-screening method — ~2007-2013","21 Inc/Earn.com — co-founder — ~2013-2015","Teleport — co-founder — ~2013"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.82,"duration_ms":61890},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":8,"lm_domain_breadth":5,"scientific_founder":6},"confidence":0.68,"duration_ms":77986}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Balaji_Srinivasan","wikipedia_title":"Balaji Srinivasan","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"alex-skidanov","name":"Alex Skidanov","title":"Co-founder, NEAR Protocol (formerly NEAR.ai)","company":"NEAR Protocol","sector":"crypto","rank":60,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":2,"transformers_lm":6,"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":4,"hands_on_engineering":13,"industry_impact":7,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":29,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":29,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.85,"rationale":"Skidanov has one genuine entry in the lineage, which the dossier and one assessor missed entirely. I verified directly on arXiv that he is the second author, with Illia Polosukhin, of 'Neural Program Search: Solving Programming Tasks from Description and Examples' (arXiv:1802.04335, 12 February 2018), which trains a Seq2Tree neural model to guide search over a domain-specific language and reports outperforming a sequence-to-sequence-with-attention baseline. That is hands-on work training attention-era neural models, not merely using them, and it was the technical substance of NEAR.ai, the program-synthesis startup he and Polosukhin founded in 2017 before pivoting to the NEAR sharded blockchain in late 2018. It is, however, a single workshop-level paper from a roughly one-year AI phase, with no pretraining, scaling or alignment work following it, so it supports the lower-middle band and not the PhD-level anchor. His deeper and better-documented strength is systems engineering: Director of Engineering at MemSQL/SingleStore, co-author of 'A column store engine for real-time streaming analytics' (ICDE 2016), and principal designer of NEAR's Nightshade sharding — demanding distributed-systems work that is not AI infrastructure. I found no publication, patent or shipped system by him on vector embeddings, dense retrieval or vector search, and no verifiable formal training in optimization or matrix methods. His indexed citations total in the tens, so there is no citation-based industry impact; the transformer pedigree often associated with NEAR belongs to co-founder Polosukhin, an 'Attention Is All You Need' co-author, and must not be transferred to Skidanov.\n\nSkidanov's only language-modeling lineage entry is arXiv:1802.04335 (Neural Program Search, Feb 2018), a Seq2Tree program-synthesis model benchmarked against a seq2seq-with-attention baseline — genuine attention-era neural work, but a single workshop-level paper on code generation that is neither a named frontier building block (attention/transformers/word2vec/scaling/RLHF, all of which belong to co-founder Polosukhin, not him) nor a pre-2013 vector-space precursor, so today's GPT/Claude-class models do not descend from it. His verifiable hands-on language-modeling record is the ~2017-2018 NEAR.ai program-synthesis phase (~2 years) before the pivot to a blockchain, giving him roughly two years and one distinct domain — source code / program synthesis (natural-language descriptions map to programs, but the modeling target is code). He co-founded NEAR (2017) and personally sets/executes technical direction (Nightshade sharding, prior MemSQL Director of Engineering, ICDE 2016 column store), so he is a real technical founder, but the company's core is a sharded blockchain and distributed systems, not language modeling — its AI-core period was ~one year. No pretraining, scaling, alignment, embedding or dense-retrieval work exists under his name, and the PubMed 'Skidanov A' orthopaedics papers are a confirmed homonym.","evidence":[{"claim":"Co-author with Illia Polosukhin of 'Neural Program Search: Solving Programming Tasks from Description and Examples', arXiv:1802.04335, submitted 12 February 2018; combines deep learning and program synthesis via a Seq2Tree model and outperforms a sequence-to-sequence-with-attention baseline","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'A column store engine for real-time streaming analytics', ICDE 2016 — his only OpenAlex-indexed work, 13 citations","source_url":"https://doi.org/10.1109/icde.2016.7498332","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NEAR Protocol was founded in early 2017 by Illia Polosukhin and Alexander Skidanov, initially as Near.ai, an AI/program-synthesis startup, before pivoting to a sharded blockchain; Skidanov was previously Director of Engineering at MemSQL and a software engineer at Microsoft","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record 3419692: 2 papers, 28 citations, h-index 2","source_url":"https://www.semanticscholar.org/author/Alex-Skidanov/3419692","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Director of Engineering at MemSQL (now SingleStore), software engineer at Microsoft prior to NEAR","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded NEAR Protocol in early 2017 with Illia Polosukhin","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'A column store engine for real-time streaming analytics' (ICDE 2016), a database-systems paper unrelated to AI/ML","source_url":"https://doi.org/10.1109/icde.2016.7498332","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NEAR Protocol founded early 2017 by Illia Polosukhin and Alexander Skidanov; initially established as Near.ai, an AI startup focused on program synthesis, pivoting to sharded blockchain from late 2018; Skidanov was previously Director of Engineering at MemSQL and a software engineer at Microsoft","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record: 2 papers, 28 citations, h-index 2","source_url":"https://www.semanticscholar.org/author/Alex-Skidanov/3419692","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alexander Skidanov is co-author (one of only two authors, with Illia Polosukhin) of 'Neural Program Search: Solving Programming Tasks from Description and Examples', arXiv:1802.04335, 12 Feb 2018 — a Seq2Tree neural model over a DSL benchmarked against a seq2seq-with-attention baseline","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NEAR was founded in early 2017 by Illia Polosukhin and Alexander Skidanov, initially as Near.ai focused on AI program synthesis, pivoting to a sharded blockchain from late 2018; Skidanov was previously Director of Engineering at MemSQL and a software engineer at Microsoft","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Skidanov co-authored 'A column store engine for real-time streaming analytics' (ICDE 2016) — his systems/database record, unrelated to language modeling","source_url":"https://doi.org/10.1109/icde.2016.7498332","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author (with Illia Polosukhin, only two authors) of 'Neural Program Search: Solving Programming Tasks from Description and Examples', arXiv:1802.04335, 12 Feb 2018 — a Seq2Tree model over a DSL benchmarked against a seq2seq-with-attention baseline; his sole language-modeling publication, an appli","source_url":"https://arxiv.org/abs/1802.04335","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"NEAR was co-founded in 2017 by Illia Polosukhin and Alexander Skidanov, initially as Near.ai (an AI/program-synthesis startup) before pivoting to a sharded blockchain from late 2018; Skidanov is the technical co-founder (previously Director of Engineering at MemSQL, engineer at Microsoft) — one comp","source_url":"https://en.wikipedia.org/wiki/NEAR_Protocol","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":["Neural Program Search: Solving Programming Tasks from Description and Examples (arXiv:1802.04335, 2018)","A column store engine for real-time streaming analytics (ICDE 2016)"],"first_verifiable_year":2016,"notable_systems":["MemSQL/SingleStore column store engine","NEAR.ai program-synthesis system","NEAR Protocol Nightshade sharding"],"citations":41,"h_index":2,"patents":0,"dossier_notes":"Two material dossier defects. (1) The PubMed block is a complete HOMONYM: 'Skidanov A' on the rat paraspinal-muscle and lumbar-spine degeneration papers (co-authors Radchenko, Ashukina, Maltseva) is a Ukrainian orthopaedic researcher, unrelated to the NEAR co-founder; the dossier's earliest_pubmed_year of 2022 belongs to that person and must be discarded. (2) OpenAlex indexes only the ICDE 2016 column-store paper and MISSES arXiv:1802.04335, his single most relevant publication, which I verified directly — the dossier therefore materially understates his AI record. No Wikipedia or Wikidata entry exists for him individually; identity was confirmed via the NEAR Protocol article and the verified arXiv co-authorship with Polosukhin. Education could not be verified from any primary source, so foundations is scored conservatively. Patents were skipped by the collector, not verified as zero.","adjudication_notes":["transformers_lm — PASS_2 RIGHT on existence, final 6 between pass_1's 0 and pass_2's 9. The decisive verified fact: arXiv:1802.04335 (12 Feb 2018) lists Illia Polosukhin and Alexander Skidanov as its only two authors, trains a Seq2Tree model and benchmarks against a seq2seq-with-attention baseline. Pass_1 scored 0 because it relied on OpenAlex, which does not index this paper; that is a miss, not a judgement. Pass_2's 9 over-credits: one workshop-level paper from a ~12-month AI phase, with no pretraining, scaling or alignment work, cannot reach the band that requires strong graduate training or senior engineering in the core. 6 places him at the top of 'uses the tools' / bottom of adjacency, which the single verified paper supports.","foundations — pass_1 closer, final 6 (pass_1 4, pass_2 8). No degree, thesis or coursework in linear algebra, optimization or statistical learning is verifiable from any primary source; pass_2 explicitly credits foundations 'on the strength of the database/distributed-systems work rather than on any formal statistical-learning training', which is the wrong dimension for that evidence. 6 gives modest credit for the demonstrable technical rigor of the ICDE paper and the Seq2Tree work without asserting training he has no record of.","vector_embeddings — pass_1 right, final 2 vs pass_2's 4. Both passes searched and neither found any embedding, dense-retrieval or vector-search publication or system; pass_2's own rationale concedes 'I found no publications, patents or shipped systems by him on vector embeddings, dense retrieval or vector search', which does not support 4.","hands_on_engineering — PASS_2 RIGHT, final 13 (pass_1 11, pass_2 15). He personally built production systems across three domains — the MemSQL column-store engine, the NEAR.ai Seq2Tree program-synthesis system he co-authored, and NEAR's Nightshade sharding — which is the 13-17 'production systems built and led personally' band. 13 rather than 15 because only one of the three is an AI system.","industry_impact — near-agreement, final 7 (pass_1 6, pass_2 9). Total indexed citations are in the tens, so there is no citation-based impact; the org he co-founded is a blockchain, not an AI system. Critically, the transformer pedigree associated with NEAR belongs to Polosukhin, not Skidanov, and cannot be transferred."],"years_language_modeling":2,"years_as_technical_founder":2,"technical_founder_companies":1,"lm_domains":[{"domain":"code","years":"2017-2018","evidence":"Neural Program Search (arXiv:1802.04335) Seq2Tree program synthesis at NEAR.ai"}],"frontier_lineage":["No named frontier building block; a single 2018 Seq2Tree program-synthesis paper in the attention era that frontier LMs do not directly descend from"],"technical_founder_roles":["NEAR.ai / NEAR Protocol — co-founder & technical lead (AI phase ~2017-2018, then blockchain/Nightshade sharding 2018-present)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":4,"scientific_founder":6},"confidence":0.7,"duration_ms":70628},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":8},"confidence":0.72,"duration_ms":83037}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ala-shaabana","name":"Ala Shaabana","title":"Co-founder, Opentensor Foundation (Bittensor)","company":"Bittensor (Opentensor Foundation)","sector":"crypto","rank":61,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":3,"transformers_lm":5,"frontier_founder":4,"lm_domain_depth":5,"lm_domain_breadth":2,"hands_on_engineering":9,"industry_impact":7,"scientific_founder":9},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":28,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":28,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.78,"rationale":"Shaabana has a real but narrow academic record that sits outside the lineage this index measures. I confirmed directly against OpenAlex that author A5058692441 is the correct person and not a homonym — the affiliations are University of Windsor (2013-2014), McMaster University (2014-2019) and University of Waterloo (2023), matching his known career — and that the 16 indexed works are on wearable electromyography, sensor networks, textiles, thermoregulation and green/optical networking. None of that is mathematics of learning, embeddings or language modelling; the profile's 'artificial intelligence' concept tag at 0.70 is a topic-classifier artifact, not authored AI work. His single document in the lineage is co-authorship of 'BitTensor: A Peer-to-Peer Intelligence Market' (Rao, Steeves, Shaabana, Attevelt, McAteer, arXiv:2003.03917), a mechanism-design proposal for a market in which peers rank one another by training neural networks — and I verified on arXiv that v3 (10 November 2021) is a WITHDRAWAL whose authors' note states the paper 'is incomplete', that one author has been removed, and that it is 'now obsolete from both a content and an author perspective'. A self-disowned whitepaper is weak evidence and cannot support scores in the PhD-level or canonical band. He is a genuine hands-on co-founder and builder of the Bittensor protocol, and the network does host language-model training subnets, which is real if indirect industry impact. His graduate-level CS research training is verified by the publication record itself, so foundations sits at the base of the strong-graduate-training band; the embedding and transformer dimensions stay low because no authored or shipped work in vector representations, attention or pretraining is verifiable under his name.\n\nNothing of Shaabana's own authorship sits in the lineage today's frontier models (GPT/Claude/Gemini/Llama) descend from: his sole lineage document, the BitTensor whitepaper (arXiv:2003.03917), is a mechanism-design proposal for a peer-ranking market that its own authors WITHDREW on 10 Nov 2021 as 'incomplete' and 'obsolete from both a content and an author perspective' — not an architecture, objective, optimizer, dataset or training/inference stack that frontier labs build on. He has no verifiable personal language-modeling record — no embeddings, seq2seq, attention, pretraining or alignment work under his name; the 16-work OpenAlex corpus is wearable EMG, sensor networks, textiles, thermoregulation and optical/green networking, so LM years count as adjacency (building the Bittensor network that HOSTS LM subnets, ~2020-2026) rather than his own modeling, and there are no distinct LM domains with a hands-on modeling record. He is, however, a genuine hands-on technical co-founder of one company (Opentensor Foundation / Bittensor, ~2020-present, ≈6 years), which anchors scientific_founder in the single-company band — held down because the incentive-protocol and subnet internals are attributed in primary sources to Jacob Steeves and Yuma Rao, not to him.","evidence":[{"claim":"OpenAlex author A5058692441 'Ala Shaabana': 16 works, 32 citations, h-index 3; affiliations University of Waterloo (2023), University of Windsor (2013, 2014), McMaster University (2014, 2015, 2017, 2019); primary topics are building-energy optimization, textile materials and thermoregulation, not ma","source_url":"https://api.openalex.org/authors/A5058692441","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'BitTensor: A Peer-to-Peer Intelligence Market' with Yuma Rao, Jacob Steeves, Daniel Attevelt and Matthew McAteer (arXiv:2003.03917, v1 March 2020)","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"arXiv v3 (10 November 2021) is a withdrawal; the authors' note states 'This paper is incomplete... one of the authors (daniel attevelt) has been removed from the work and so this paper is now obsolete from both a content and an author perspective'","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Portable Electromyography: A Case Study on Ballistic Finger Movement Recognition', IEEE Sensors Journal 2019 — representative of his indexed corpus (wearable sensing, not AI core)","source_url":"https://doi.org/10.1109/jsen.2019.2908312","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The active Bittensor/subtensor codebase is maintained under the opentensor GitHub organization he co-founded","source_url":"https://github.com/opentensor/bittensor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ala Shaabana is a listed co-author of 'BitTensor: A Peer-to-Peer Intelligence Market' alongside Yuma Rao, Jacob Steeves, Daniel Attevelt, and Matthew McAteer.","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The BitTensor arXiv paper (v3, Nov 2021) was withdrawn by Ala Shaabana, with an authors' note describing it as incomplete and obsolete from both a content and an author perspective.","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex profile for 'Ala Shaabana' (id A5058692441) lists 16 works, h-index 3, cited_by_count 32, affiliations at University of Waterloo/Windsor/McMaster, topics in wearable sensing, optical networks, textiles and thermoregulation -- none in vector embeddings or transformer/LM research -- with the","source_url":"https://doi.org/10.1109/jsen.2019.2908312","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar independently lists an 'A. Shaabana' author profile (id 1788032) with 20 papers, 61 citations, h-index 4, consistent in scale with the OpenAlex profile but not independently confirmed as the same Bittensor co-founder.","source_url":"https://www.semanticscholar.org/author/1788032","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The GitHub account 'unconst' (bio: opentensor/bittensor.com), which authors the active Bittensor/subtensor codebase, is associated with Jacob Steeves rather than Ala Shaabana, and no README or repository documentation names Shaabana's specific technical contributions to the current protocol.","source_url":"https://github.com/opentensor/bittensor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Shaabana's only lineage document, 'BitTensor: A Peer-to-Peer Intelligence Market' (arXiv:2003.03917), was withdrawn by its authors (v3, 10 Nov 2021) as incomplete and 'now obsolete from both a content and an author perspective' — no method or architecture frontier models build on","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5058692441 'Ala Shaabana': 16 works in wearable EMG, sensor networks, textiles, thermoregulation and optical networking — no embeddings, transformer or language-modeling authorship in any domain","source_url":"https://api.openalex.org/authors/A5058692441","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The active Bittensor/subtensor protocol he co-founded is maintained under the opentensor GitHub organization, establishing a hands-on technical co-founder role at one company","source_url":"https://github.com/opentensor/bittensor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BitTensor whitepaper (arXiv:2003.03917) v3 withdrawn 10 Nov 2021 by the authors as incomplete and obsolete — the only lineage item bearing his name, and self-disowned","source_url":"https://arxiv.org/abs/2003.03917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Active Bittensor/subtensor codebase maintained under the opentensor GitHub org he co-founded; network hosts language-model training subnets","source_url":"https://github.com/opentensor/bittensor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5058692441: 16 works on wearable EMG, sensor networks, textiles and optical/green networking — no embeddings, transformer or LM research","source_url":"https://api.openalex.org/authors/A5058692441","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Graduate-level computer science research at University of Windsor and McMaster University (2013-2019); degree titles and award years not confirmed from any primary source"],"canonical_papers":[],"first_verifiable_year":2013,"notable_systems":["Bittensor protocol / Opentensor Foundation","bittensor subnet software"],"citations":32,"h_index":3,"patents":0,"dossier_notes":"The OpenAlex match is verified correct and NOT a homonym: the Windsor/McMaster/Waterloo affiliation trail and the sensor-networks corpus are internally consistent and the BitTensor whitepaper appears in the same record. The dossier does not surface the single most load-bearing fact — that the BitTensor whitepaper has been WITHDRAWN by its own authors as incomplete and obsolete — and this should be added to the dossier. No Wikipedia or Wikidata entry exists, so his degree titles and award years could not be confirmed and are recorded as unverified rather than asserted. Semantic Scholar lists him as 'A. Shaabana' (name_exact false, 20 papers, h-index 4), a slightly wider but consistent match. Patents were skipped by the collector, not verified as zero. WebSearch budget was exhausted this session, so an independent McMaster thesis-repository confirmation could not be attempted; the degree therefore remains unverified in both directions.","adjudication_notes":["foundations — pass_1 closer, final 8 (pass_1 6, pass_2 10). I verified the OpenAlex record is the right person, so pass_1's homonym caution was over-applied and 6 is too low; but pass_2's 10 rests on 'verified graduate-level CS research training' whose subject matter is sensor networks and optical networking, with no thesis, degree or coursework in linear algebra, optimization or statistical learning confirmable from any primary source. 8 credits real graduate research training without crediting core-AI mathematics he has no record in.","vector_embeddings — pass_1 right, final 3 vs pass_2's 6. Pass_2 cited no embedding, representation-learning or retrieval work whatsoever; its own rationale concedes 'no authored or shipped work in vector representations... is verifiable', which contradicts a score of 6. The verified 16-work corpus contains none.","transformers_lm — pass_1 substantially right, final 5 vs pass_2's 8. The decisive verified fact is that the sole item in the lineage, arXiv:2003.03917, was withdrawn by the authors themselves on 10 Nov 2021 as 'incomplete' and 'obsolete from both a content and an author perspective'. Both passes found the withdrawal, but only pass_1 let it govern the score. A self-disowned mechanism-design whitepaper about a market for neural-network ranking cannot support the 8-12 band, which requires strong training or senior engineering in attention/pretraining.","hands_on_engineering — SPLIT, final 9 (pass_1 6, pass_2 11). Pass_2 is right that he is a genuine hands-on co-founder and builder of a live, non-trivial decentralized-ML network, which exceeds pass_1's 6; but no primary source attributes the incentive-protocol or subnet internals to him personally rather than to Jacob Steeves/Yuma Rao, so the top of the 8-12 band is not established.","industry_impact — near-agreement, final 7 (pass_1 6, pass_2 8). Bittensor is a real network hosting language-model subnets, which is genuine indirect impact; it is not an org whose canonical work the field builds on, and his citation record (32 citations, h-index 3) supplies no independent impact."],"years_language_modeling":3,"years_as_technical_founder":6,"technical_founder_companies":1,"lm_domains":[{"domain":"other","years":"2020-2026","evidence":"co-built the Bittensor decentralized-ML network which hosts language-model training subnets; adjacency, not personal LM modeling in a specific domain"}],"frontier_lineage":["No verifiable foundational contribution; sole lineage paper (arXiv:2003.03917) self-withdrawn as obsolete"],"technical_founder_roles":["Opentensor Foundation / Bittensor — co-founder / hands-on builder — ~2020-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":2,"scientific_founder":9},"confidence":0.7,"duration_ms":72840},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":6,"lm_domain_breadth":3,"scientific_founder":9},"confidence":0.68,"duration_ms":64778}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"lisa-su","name":"Lisa Su","title":"Chair & CEO","company":"AMD","sector":"general","rank":62,"rubric_version":4,"dimensions":{"foundations":10,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":5,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":14,"industry_impact":14,"scientific_founder":2},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":28,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":28,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.81,"rationale":"Su holds a PhD in electrical engineering from MIT (1994, advisors Dimitri Antoniadis and James Chung), with a dissertation on extreme-submicrometer silicon-on-insulator MOSFETs — genuine, verifiable PhD-level engineering/physics training, though in semiconductor device physics rather than the mathematics/statistical-learning core this rubric measures. She has a real, if modest, publication record (OpenAlex: 15 works, 236 citations, h-index 7) spanning multi-chip packaging and compute-efficiency papers, and she personally led engineering work at IBM (copper interconnects), Freescale, and now AMD. There is no verifiable record of her personally authoring or leading vector-embedding, attention, or language-model research — her impact on AI is through building the accelerator hardware (GPUs, Instinct MI-series) that AI training runs on, which is legitimate hands-on-engineering/industry-impact credit for 'the hardware under core systems' per the rubric, but does not itself constitute core-AI research depth. Scored as strong PhD-level hardware engineer and industry leader of AI-infrastructure hardware, not as an AI researcher.\n\nSu's verifiable record is in semiconductor device physics and compute hardware — MIT PhD on silicon-on-insulator MOSFETs, copper-interconnect work at IBM, and leadership of AMD's EPYC/Instinct accelerator lines — none of which is authored work in the language-modeling lineage (distributional semantics, embeddings, attention, transformers, pretraining, tokenizers, optimizers, alignment). Her only connection to today's frontier models is that AMD's Instinct MI-series GPUs are part of the training/inference hardware some models run on, a real but institutional and indirect contribution she leads as CEO rather than a method/architecture/objective she personally authored, so frontier_founder sits just above 'applies frontier models.' She has zero verifiable years of hands-on language-modeling research or systems work and no distinct language-modeling domains (biological, financial, natural-language text, code, etc.), so lm_domain_depth and lm_domain_breadth are at floor. She is the CEO of AMD, not a founder (AMD was founded in 1969) and has not operated as the scientific/technical founder of any company — years_as_technical_founder and technical_founder_companies are both zero — so scientific_founder is at floor.","evidence":[{"claim":"PhD electrical engineering, MIT (1994), dissertation on extreme-submicrometer silicon-on-insulator MOSFETs, advisors Dimitri Antoniadis and James Chung","source_url":"https://en.wikipedia.org/wiki/Lisa_Su","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Led IBM's Emerging Products group and helped drive the industry shift from aluminum to copper interconnects during 13 years at IBM","source_url":"https://www.clay.com/dossier/amd-ceo","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"President and CEO of AMD since 2014","source_url":"https://en.wikipedia.org/wiki/Lisa_Su","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MIT BS, MS (1991) and PhD (1994) in Electrical Engineering; doctoral work on extreme-submicrometer silicon-on-insulator MOSFETs under Dimitri Antoniadis and James Chung; copper interconnect work at IBM launched 1998; Cell processor contribution; over forty technical articles as of 2016","source_url":"https://en.wikipedia.org/wiki/Lisa_Su","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records MIT PhD (1994), doctoral advisors Dimitri A. Antoniadis and James E. Chung, employment at Texas Instruments (1994), IBM Research (1995), Freescale (2007), AMD (2012), CEO from 2014","source_url":"https://www.wikidata.org/wiki/Q18207172","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named author on 'Multi-chip technologies to unleash computing performance gains over the next decade' (IEDM 2017)","source_url":"https://doi.org/10.1109/iedm.2017.8268306","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named author on '1.1 Innovation For the Next Decade of Compute Efficiency' (ISSCC 2023)","source_url":"https://doi.org/10.1109/isscc42615.2023.10067810","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Su is president and CEO of AMD since 2014; the company was founded in 1969, so she is not a founder","source_url":"https://en.wikipedia.org/wiki/Lisa_Su","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Su is a named author on hardware/packaging papers 'Multi-chip technologies to unleash computing performance gains over the next decade' (IEDM 2017) — chip/compute-stack work, not language-modeling research","source_url":"https://doi.org/10.1109/iedm.2017.8268306","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"AMD Instinct GPU accelerators (led by Su as CEO) are used for AI training/inference, an infrastructure contribution to the frontier stack rather than an authored algorithmic block","source_url":"https://doi.org/10.1109/isscc42615.2023.10067810","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Su is president and CEO of AMD since 2014, which builds the EPYC/Instinct GPU accelerator lines used for AI training and inference","source_url":"https://en.wikipedia.org/wiki/Lisa_Su","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records Su as an employee/CEO of AMD (from 2012/2014), not a founder; prior employers Texas Instruments, IBM Research, Freescale","source_url":"https://www.wikidata.org/wiki/Q18207172","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["MIT, BS electrical engineering (1990)","MIT, MS electrical engineering (1991)","MIT, PhD electrical engineering (1994)"],"canonical_papers":["Multi-chip technologies to unleash computing performance gains over the next decade (2017)","1.1 Innovation For the Next Decade of Compute Efficiency (2023)"],"first_verifiable_year":1991,"notable_systems":["AMD EPYC / Instinct GPU accelerator lines used for AI training and inference"],"citations":236,"h_index":7,"patents":0,"dossier_notes":"Dossier's PubMed sample (magnetic particle imaging, nanozymes, cardiac studies, etc. under 'Su L') is entirely homonym noise given how common the surname/initial combination is; the dossier itself flags year_trusted:false and homonym_risk:true correctly. OpenAlex match is ambiguous:true (10 candidates) but the works shown (semiconductor/compute papers, AMD affiliation) are consistent with the real Lisa Su. No evidence found of personal AI/ML research (transformers, embeddings, statistical learning) beyond hardware-architecture papers; her AI relevance is entirely as the CEO who builds accelerator hardware, which the rubric explicitly separates from core-AI research depth.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"frontier_lineage":["AMD Instinct MI-series GPU accelerators used in some AI training/inference stacks — institutional, led as CEO, not personally authored"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":2},"confidence":0.86,"duration_ms":62473},{"pass":"pass_2","dimensions":{"frontier_founder":6,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":2},"confidence":0.85,"duration_ms":47985}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Lisa_Su","wikipedia_title":"Lisa Su","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"mustafa-suleyman","name":"Mustafa Suleyman","title":"CEO, Microsoft AI","company":"Microsoft AI","sector":"general","rank":63,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":4,"transformers_lm":6,"frontier_founder":5,"lm_domain_depth":8,"lm_domain_breadth":4,"hands_on_engineering":5,"industry_impact":12,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":28,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":28,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.82,"rationale":"Suleyman has no completed degree -- he enrolled in philosophy and theology at Oxford and dropped out at 19 -- and no PhD, publications record, or personal engineering track record in the mathematics, vector-embeddings, or transformer research core. At DeepMind (co-founded 2010) his own role was Head of Applied AI / Chief Product Officer: integrating DeepMind's technology into Google products, launching DeepMind Health, and founding DeepMind Ethics & Society, i.e. applied/business/policy leadership, not research. He appears as the 6th of 7 authors on 'Teaching Machines to Read and Comprehend' (Hermann et al. 2015), a genuinely important reading-comprehension/attention paper, but the ordering and his documented applied/ops role at the time (not a research scientist) indicate a leadership/co-founder authorship credit rather than personal technical authorship of the method. He co-founded Inflection AI (2022, chatbot 'Pi') and now runs Microsoft AI as an executive. Industry impact is scored moderately for building/leading consequential AI organizations (DeepMind co-founder, Inflection AI co-founder, Microsoft AI CEO), but the rubric explicitly excludes fame/business success from the research dimensions, and no verifiable personal research or engineering record was found to support higher scores there.\n\nSuleyman's own position in the frontier-model lineage is thin: he holds a leadership-credit co-authorship (6th of 7) on the attention-based reading-comprehension paper 'Teaching Machines to Read and Comprehend' (NIPS 2015), genuine NLP-lineage work, but he authored none of the load-bearing blocks (attention, transformers, word2vec, scaling, alignment) and has no pre-2013 vector-space/distributional-semantics patents or systems — the foundational science at DeepMind and Inflection was done by his technical co-founders (Hassabis/Legg/Simonyan). His language-modeling involvement runs from that 2015 paper through co-founding Inflection AI (Pi, 2022) and leading Microsoft AI/Copilot (2024+), roughly 8 verifiable years, but at applied/product/executive altitude rather than hands-on modeling. He has a real record in only one language-modeling domain — natural-language text — since his celebrated healthcare work (breast-cancer screening, retinal disease, Kinetics) is medical imaging/vision and general ML, not language modeling. He is a repeat co-founder of consequential AI companies, but as a product/applied/ethics leader whose science and engineering were executed by others, which is the 'AI-company founder, science done by others' band for scientific_founder rather than a technical/scientific founder role.","evidence":[{"claim":"Dropped out of Oxford at 19, no completed degree","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"At DeepMind (co-founded 2010) served as Head of Applied AI / Chief Product Officer, launched DeepMind Health and DeepMind Ethics & Society -- applied/business/ethics roles, not research scientist","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as 6th of 7 authors on 'Teaching Machines to Read and Comprehend' (Hermann, Kocisky, Grefenstette, Espeholt, Kay, Suleyman, Blunsom, 2015)","source_url":"https://arxiv.org/abs/1506.03340","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Inflection AI (2022) and is now CEO of Microsoft AI (since March 2024)","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Enrolled in philosophy and theology at Oxford, dropped out at 19; co-founded DeepMind 2010 as chief product officer, later head of applied AI; launched DeepMind Health 2016 and DeepMind Ethics & Society; placed on administrative leave in 2019 over bullying allegations, then moved to a Google policy","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (hkDGEVQAAAAJ): ~29,404 citations, h-index 26, i10 35, with no first-author papers; top entries are Kinetics dataset, Teaching Machines to Read and Comprehend, breast-cancer screening, clinical-impact commentary, retinal disease","source_url":"https://scholar.google.com/citations?user=hkDGEVQAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sixth of seven authors on 'Teaching Machines to Read and Comprehend' (Hermann, Kocisky, Grefenstette, Espeholt, Kay, Suleyman, Blunsom), NIPS 2015","source_url":"https://arxiv.org/abs/1506.03340","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex profile A5001712647: 51 works, h-index 20, dominant topics are healthcare AI, digital innovation and interdisciplinary technology-and-society rather than machine-learning methods","source_url":"https://api.openalex.org/authors/A5001712647","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded DeepMind (2010) as chief product officer / head of applied AI, and co-founded Inflection AI (2022, Pi chatbot); now CEO of Microsoft AI — leadership roles, with the core science led by technical co-founders","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5001712647 (51 works, 18,560 citations, h-index 20): dominant topics are healthcare AI, digital innovation and technology-and-society — medical imaging/clinical ML rather than language-modeling methods","source_url":"https://api.openalex.org/authors/A5001712647","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed 6th of 7 authors on 'Teaching Machines to Read and Comprehend' (Hermann, Kocisky, Grefenstette, Espeholt, Kay, Suleyman, Blunsom, 2015) — a co-founder/applied credit, not method authorship","source_url":"https://arxiv.org/abs/1506.03340","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded DeepMind (2010) as chief product officer / head of applied AI, and Inflection AI (2022, Pi chatbot); now CEO of Microsoft AI — applied/product/business roles, technical science done by co-founders","source_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex profile A5001712647: 51 works dominated by healthcare-AI/imaging and technology-and-society topics, not LM methods, reflecting organizational co-authorship rather than a personal LM research record","source_url":"https://api.openalex.org/authors/A5001712647","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["University of Oxford (philosophy and theology, did not complete degree)"],"canonical_papers":["Teaching Machines to Read and Comprehend (Hermann et al., 2015) -- co-author, not lead"],"first_verifiable_year":2010,"notable_systems":["DeepMind (co-founder, applied/product side)","Inflection AI / Pi chatbot (co-founder)","Microsoft AI (CEO)"],"citations":18560,"h_index":20,"patents":0,"dossier_notes":"OpenAlex record (51 works, 18560 citations, h-index 20, not ambiguous) is real but reflects group/organizational co-authorship as a DeepMind co-founder and applied-AI executive (e.g. large collaborative Nature/Nature Medicine healthcare-AI papers, the Kinetics dataset paper, Teaching Machines to Read and Comprehend) rather than a personal hands-on research record. Author order and his documented non-research applied/product/ethics role at DeepMind at the time support scoring these as leadership-adjacent rather than personally-authored core technical contributions.","years_language_modeling":8,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2015-2026","evidence":"co-author on 'Teaching Machines to Read and Comprehend' (2015); co-founder of Inflection AI / Pi (2022); CEO of Microsoft AI / Copilot (2024+) — leadership-level, not hands-on modeling"}],"frontier_lineage":["Leadership-credit co-authorship on an attention-based reading-comprehension paper (2015) in the NLP/attention lineage; co-founded DeepMind and Inflection AI (orgs producing lineage work) but authored none of the foundational blocks himself"],"technical_founder_roles":["DeepMind — co-founder, head of applied AI / chief product officer (not scientific founder), 2010-2019","Inflection AI — co-founder & CEO, product/business (science led by Karén Simonyan), 2022-2024","Microsoft AI — CEO, 2024-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":6,"lm_domain_depth":10,"lm_domain_breadth":5,"scientific_founder":4},"confidence":0.82,"duration_ms":92145},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":7,"lm_domain_breadth":4,"scientific_founder":5},"confidence":0.82,"duration_ms":71707}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Mustafa_Suleyman","wikipedia_title":"Mustafa Suleyman","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ron-bodkin","name":"Ron Bodkin","title":"Founder & CEO","company":"Theoriq (ChainML)","sector":"crypto","rank":64,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":6,"transformers_lm":4,"frontier_founder":2,"lm_domain_depth":4,"lm_domain_breadth":2,"hands_on_engineering":10,"industry_impact":9,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":28,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":28,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.76,"rationale":"Bodkin holds CS degrees from McGill and MIT and has a 15+ year applied-ML engineering career: Technical Director of Applied AI at Google Cloud CTO office, engineering lead at the Vector Institute, and founder of Think Big Analytics (data science consultancy acquired by Teradata). He is a co-author (7th of 8 authors, engineering-lead position) on a 2022 SIGIR paper on VAE-based recommender systems mitigating filter bubbles, which is genuine applied representation-learning/ML work, though his position in the author list suggests an engineering-support rather than principal-researcher role. His earlier technical record (2003 OOPSLA paper on AspectJ/middleware) is software-engineering, not AI. No evidence of authored foundational math/optimization theory, embeddings research, or transformer/LM work — his AI record is applied engineering leadership, not core research. Theoriq/ChainML is an AI-agent infrastructure company built on top of existing LLMs, not a model-building lab.\n\nNothing in Bodkin's verifiable record places his own work in the frontier-model lineage: no pre-2013 vector-space / distributional-semantics / LSI patents or shipped systems, no authored attention, transformer, embedding, optimizer, tokenizer, or pretraining work — his 2003 output is AspectJ/middleware software engineering and his 2022 SIGIR paper (7th of 8 authors) is a VAE-based recommender for filter-bubble diversification, applied representation learning rather than language modeling. He has essentially no hands-on language-modeling record (no n-gram / neural LM / seq2seq / transformer authorship or shipped LM systems); his AI career is applied-ML and data-engineering leadership (Think Big Analytics, Teradata Kylo data lake, Google Cloud CTO applied-AI, Vector Institute engineering), which is adjacent general ML with no LM domain depth, so years_language_modeling is effectively 0 and no distinct LM domain is verifiable. He does have a genuine technical-founder history — founder-CEO of Think Big Analytics (~2010–2014, acquired by Teradata) and founder-CEO of ChainML/Theoriq (2022–present) — roughly 8 years across two companies, but neither company's core is these systems (Think Big was a data-science consultancy; Theoriq is an AI-agent / DeFi-curation layer built on existing LLMs), which places scientific_founder as a technical founder outside the language-modeling field. Theoriq applying autonomous agents to DeFi is use of vendor models, not language modeling in a domain.","evidence":[{"claim":"CS degrees from McGill and MIT; 15+ years in AI/big data before founding ChainML/Theoriq in 2022","source_url":"https://podcasts.apple.com/us/podcast/ron-bodkin-chainml-founder-and-ceo-and-ex-google/id1476885647?i=1000622925860","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Technical director on Google's applied AI team in the Cloud CTO office; previously founded and was CEO of Think Big Analytics (enterprise big data, data science, data engineering), which was acquired by Teradata where he led the Kylo open-source data lake framework and helped establish Teradata's AI","source_url":"https://www.oreilly.com/people/ron-bodkin/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5022442588, affiliation Vector Institute (Canada), 4 works, 60 citations, h-index 3; research topics are software system performance and reliability, software engineering methodologies, service-oriented architecture, with one recommender-systems entry","source_url":"https://api.openalex.org/authors/A5022442588","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar record (8 papers, 86 citations, h-index 4): 'Mitigating the Filter Bubble While Maintaining Relevance: Targeted Diversification with VAE-based Recommender Systems' (2022, with Gao, Shen, Mai, Bouadjenek, Waller, Anderson, Sanner), 'Using AspectJ for component integration in middlewa","source_url":"https://api.semanticscholar.org/graph/v1/author/2261058?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount,papers.authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Theoriq describes itself as a DeFi strategy curator for tokenized assets using autonomous AI to monitor rates and liquidity, surface signal and validate executions, operating human-in-the-loop where 'Curators decide. AI streamlines'","source_url":"https://www.theoriq.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder/CEO of Think Big Analytics (enterprise big data/data engineering), acquired by Teradata where he led the Kylo open-source data-lake framework; later technical director of applied AI at Google Cloud CTO office; founded ChainML/Theoriq in 2022","source_url":"https://www.oreilly.com/people/ron-bodkin/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Theoriq is a DeFi strategy-curation layer using autonomous AI agents on top of existing LLMs ('Curators decide. AI streamlines') — not a model-building or language-modeling lab","source_url":"https://www.theoriq.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5022442588: 4 works, 60 citations, h-index 3; topics are software system performance/reliability, software engineering methodologies, service-oriented architecture and recommender systems — no LM / embeddings / transformer research","source_url":"https://api.openalex.org/authors/A5022442588","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bodkin founded and was CEO of Think Big Analytics (data science / data engineering consultancy, acquired by Teradata where he led the Kylo open-source data lake), a technical-founder role outside language modeling","source_url":"https://www.oreilly.com/people/ron-bodkin/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Theoriq/ChainML is an autonomous-AI DeFi strategy-curation layer ('Curators decide. AI streamlines'), agent infrastructure built on existing LLMs rather than a model-building lab","source_url":"https://www.theoriq.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["B.S./CS McGill University","S.M. Computer Science, MIT"],"canonical_papers":["Mitigating the Filter Bubble While Maintaining Relevance (SIGIR 2022)"],"first_verifiable_year":2003,"notable_systems":["Think Big Analytics","Theoriq (ChainML)"],"citations":86,"h_index":4,"patents":0,"dossier_notes":"The dossier's pubmed block (27 results, earliest 1946, medical/emergency-medicine authors like 'Rotoli J, Bodkin R') is a clear homonym — this Ron Bodkin has no medical publication record; discard entirely. The OpenAlex block (Vector Institute affiliation, AspectJ + SIGIR 2022 filter-bubble paper) does match the correct person and was corroborated via Semantic Scholar and direct DOI lookup.","years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":2,"technical_founder_roles":["Think Big Analytics — founder & CEO — ~2010-2014 (acquired by Teradata)","ChainML / Theoriq — founder & CEO — 2022-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":8},"confidence":0.72,"duration_ms":66398},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":3,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.7,"duration_ms":66934}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"amjad-masad","name":"Amjad Masad","title":"Founder & CEO","company":"Replit","sector":"general","rank":65,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":3,"transformers_lm":4,"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":4,"hands_on_engineering":11,"industry_impact":8,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":27,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":27,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.86,"rationale":"Masad is a genuine career software engineer with no verifiable personal record in the mathematics, embeddings or transformer/LM lineage. He holds a computer science degree from Princess Sumaya University for Technology, was a founding engineer at Codecademy (2011-2013), then ran the JavaScript infrastructure team at Facebook (2013-2016), before founding Replit in 2016 — real, substantial systems and developer-tooling engineering. Neither OpenAlex nor Semantic Scholar returns any author record for him, which is a correct absence rather than a collection failure, and the PubMed hits are Saudi and Palestinian clinician homonyms with no connection to him. The strongest AI item on his record is replit-code-v1-3b, a 2.7B-parameter causal code language model trained on 525B tokens across 256 A100-40GB GPUs; I retrieved the model card directly and it attributes the work to 'Replit, Inc.' as an organization with no individual named, so this counts as leading an organization that trained a model rather than personally authoring it. Replit's current product is an LLM-driven coding agent, which is genuine industry impact whose core is these systems, though the models underneath are largely third-party. He is best read as a strong systems and infrastructure engineer who manages builders of LM products: the research dimensions belong in the 3-7 'uses the tools, manages builders' band, and his engineering in the 8-12 band.\n\nMasad has no verifiable position in the frontier-model lineage: no pre-2013 vector-space/distributional-semantics work, no attention/transformer/embedding/optimizer/tokenizer/scaling/alignment authorship, and the one concrete LM artifact on his record — replit-code-v1-3b, a 2.7B code model — is a downstream consumer of the frontier stack (Flash Attention, ALiBi, LionW) attributed to 'Replit, Inc.,' not something frontier models descend from, so he sits in the 'applies/fine-tunes' band. His personal language-modeling record spans only the Replit code-LM era (~2022 onward, roughly 3 years) and is organizational rather than hands-on authored, which keeps lm_domain_depth in the under-3-years/thin-record band; the single verifiable LM domain is source code, so breadth is a one-domain score. As scientific/technical founder he is stronger: he founded Replit in 2016 (~9 years) and personally built its browser IDE and developer-tooling core, a genuine one-company technical-founder record — but the AI/LM science underneath the current agent product was executed by his team and largely on third-party models, so this lands in the mid 8-12 band rather than higher. Net: one company, ~9 years as a technical founder, one LM domain (code), ~3 years of organizational LM work, and no foundational lineage contribution.","evidence":[{"claim":"Computer science degree from Princess Sumaya University for Technology; founding engineer at Codecademy (2011-2013); software engineer overseeing the JavaScript infrastructure team at Facebook (2013-2016); founded Replit in 2016","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The replit-code-v1-3b model card attributes the model to 'Replit, Inc.' as an organization with no individual developer named; 2.7B parameters, 525B training tokens over Stack Dedup v1.2, trained on 256 x A100-40GB GPUs using Flash Attention, ALiBi positional embeddings and the LionW optimizer","source_url":"https://huggingface.co/replit/replit-code-v1-3b","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him as CEO of Replit, educated at Princess Sumaya University for Technology, with employers Replit, Codecademy and Meta Platforms — no research affiliation or academic post","source_url":"https://www.wikidata.org/wiki/Q113856785","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Amjad Masad holds a computer science degree from Princess Sumaya University for Technology in Jordan.","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Masad worked as a founding engineer at Codecademy (Nov 2011-Oct 2013), then as a software engineer at Facebook (Oct 2013-Apr 2016) where he led the JavaScript infrastructure team.","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Masad founded Replit, an online integrated development environment, in 2016 with his wife Haya Odeh and brother Faris Masad.","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Computer science degree, Princess Sumaya University for Technology; founding engineer at Codecademy 2011-2013; Facebook JavaScript infrastructure 2013-2016; founded Replit 2016","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"replit-code-v1-3b: 2.7B-parameter causal LM for code, 525B tokens, Stack Dedup v1.2, trained on 256 x A100-40GB; developer listed as Replit, Inc. with no individual author credited","source_url":"https://huggingface.co/replit/replit-code-v1-3b","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata: CEO of Replit; educated at Princess Sumaya University for Technology; employers Replit, Codecademy, Meta Platforms","source_url":"https://www.wikidata.org/wiki/Q113856785","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"replit-code-v1-3b is a 2.7B causal code LM using Flash Attention, ALiBi positional embeddings and the LionW optimizer, with the developer listed as 'Replit, Inc.' and no individual author — a downstream application of the frontier stack, not a precursor to it","source_url":"https://huggingface.co/replit/replit-code-v1-3b","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Masad founded Replit in 2016 and personally led its JavaScript-infrastructure and browser-IDE engineering; his only language-modeling domain of record is code","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him solely as CEO of Replit with employers Replit, Codecademy and Meta and no research affiliation, consistent with founder-CEO/technical-founder of a single company rather than an authored research lineage","source_url":"https://www.wikidata.org/wiki/Q113856785","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"replit-code-v1-3b is a 2.7B-parameter causal code language model trained on 525B tokens over 256 A100-40GB GPUs using Flash Attention, ALiBi positional embeddings and the LionW optimizer, with developer listed as 'Replit, Inc.' and no individual author credited","source_url":"https://huggingface.co/replit/replit-code-v1-3b","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Masad founded Replit in 2016 after being a founding engineer at Codecademy (2011-2013) and running Facebook's JavaScript infrastructure team (2013-2016); no research affiliation or academic post","source_url":"https://en.wikipedia.org/wiki/Amjad_Masad","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BSc Computer Science, Princess Sumaya University for Technology (Jordan)"],"canonical_papers":[],"first_verifiable_year":2011,"notable_systems":["Replit browser IDE","Replit Agent","replit-code-v1-3b (released by Replit, Inc.)","Codecademy learning runtime","Facebook JavaScript infrastructure"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"No research identity exists to confuse: OpenAlex and Semantic Scholar both return not-found, which is correct rather than a collection failure. The PubMed block is entirely homonyms — 'Al Masad AG' and 'Masad A' are Saudi and Palestinian clinical researchers publishing gastroenterology, oncology and toxicology case reports. That contamination propagates into timeline.earliest_pubmed_year 1992 and hence first_verifiable_year 1992 and years_active 34, which are wrong; his real first verifiable professional year is 2011 (Codecademy), about 15 years. Wikipedia calls him a 'billionaire entrepreneur', which is explicitly not evidence under this rubric and was not used. Patents were skipped by the collector, not verified as zero.","adjudication_notes":["vector_embeddings — pass_1 closer, final 3 (pass_1 1, pass_2 7). Pass_2's 7 is the largest unsupported score in this pair: its own rationale cites no embedding, retrieval or vector-search work and justifies the number on the ground that 'he ships and configures these systems directly rather than merely speaking about them', which is the definition of the 3-7 'uses the tools' band, not its top. The verified model card names ALiBi positional embeddings as a training detail of a model authored by 'Replit, Inc.', not by him. 3 credits genuine hands-on exposure without crediting authorship.","transformers_lm — SPLIT, final 4 (pass_1 2, pass_2 7). The deciding verified fact is the replit-code-v1-3b model card, which lists the developer as 'Replit, Inc.' and names no individual: Replit demonstrably trained a real 2.7B code LM on 256 A100s, which is more than pass_1's 2 allows, but no primary source attributes the architecture, training or evaluation to Masad personally, so pass_2's 7 credits organizational output as personal work.","foundations — near-agreement, final 4 (pass_1 3, pass_2 4). An undergraduate CS degree with no thesis, graduate training or publication in linear algebra, optimization or statistical learning; both passes land in the same band and pass_2's 4 is marginally better calibrated.","hands_on_engineering — PASS_2 closer, final 11 (pass_1 9, pass_2 12). Running Facebook's JavaScript infrastructure team, being a founding engineer at Codecademy and personally building Replit's browser IDE is real production engineering across fifteen years; it is general systems and developer-tooling work rather than AI infrastructure, which keeps it inside the 8-12 band.","industry_impact — SPLIT, final 8 (pass_1 6, pass_2 9). Replit's current core product is an LLM-driven coding agent used at scale and the company trained and released its own code LM, which exceeds pass_1's 6; but the underlying models are largely third-party, there is no citation, patent or canonical-paper record, and the company's valuation and reach are explicitly out of scope.","penalties — both passes agree at 0/0 and both are right: Replit was venture-funded on the strength of a shipped product and he has a verifiable fifteen-year engineering career, so capital_without_competence does not apply and no citable evidence of paid placement exists."],"years_language_modeling":3,"years_as_technical_founder":9,"technical_founder_companies":1,"lm_domains":[{"domain":"code","years":"2022-2025","evidence":"replit-code-v1-3b (2.7B code LM) and Replit Agent, released by Replit, Inc. under his leadership"}],"frontier_lineage":["None verifiable — consumes the frontier stack (Flash Attention/ALiBi/LionW) in a downstream code LM; no foundational block, precursor system, or pre-2013 vector-space work"],"technical_founder_roles":["Replit — founder & CEO / technical founder (browser IDE, JS infra) — 2016-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":5,"lm_domain_breadth":5,"scientific_founder":9},"confidence":0.8,"duration_ms":61007},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":4,"scientific_founder":8},"confidence":0.82,"duration_ms":70734}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Amjad_Masad","wikipedia_title":"Amjad Masad","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"andre-zayarni","name":"André Zayarni","title":"Co-founder & CEO","company":"Qdrant","sector":"general","rank":66,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":7,"transformers_lm":3,"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"hands_on_engineering":8,"industry_impact":10,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":26,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":26,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.78,"rationale":"The merge flagged an invariant violation on pass 2's vector_embeddings of 13 — no publications, patents or verified work, yet a score above 7 — and that flag is upheld. The dimension may exceed 7 only on cited verified engineering work, so I checked GitHub's commit API directly rather than inferring from profile badges as pass 2 did. Across the entire qdrant organisation his account azayarni has 46 commits, and in the core qdrant/qdrant engine repository just 6: 'Update logo.svg', a roadmap badge adjustment, two cloud-link edits, an ETA update and README spellcheck fixes. Those are documentation and branding changes, not vector-index, quantization or retrieval code. The rest of his organisation commits are merge commits on the landing page and docs repositories. His GitHub account has 7 public repositories and 80 followers. Pass 2's inference that he is 'a working engineer with a real GitHub presence' rather than a commercial founder does not survive inspection of what the commits actually contain, and its own evidence concedes that the engine-internals authorship at Qdrant belongs to co-founder and CTO Andrey Vasnetsov. Pass 1 reached the same conclusion about the division of technical labour and scored accordingly. There is no OpenAlex, Semantic Scholar, Wikipedia or Wikidata record for him, no paper on approximate nearest-neighbour search, quantization, contrastive learning or dense retrieval, and no patent. What he verifiably did is co-found and lead, as CEO, a company whose entire product is an open-source vector similarity search engine written in Rust with ~34.5k GitHub stars — genuinely core infrastructure for embedding-based retrieval, which supports real industry impact and a vector_embeddings score at the top of the permitted band, but as an org-builder rather than an author of the systems.\n\nQdrant is downstream retrieval infrastructure that serves embedding-based RAG for LLM applications — it is not part of the foundation today's frontier models are built on (no attention/transformer/optimizer/tokenizer/pretraining/scaling contribution, and no pre-2013 vector-space or distributional-semantics patents or systems), and Zayarni personally authored none of the engine anyway (the prior assessment's GitHub commit evidence shows only logo/badge/link/README docs commits, with engine authorship belonging to CTO Andrey Vasnetsov). His only verifiable position in the language-modeling lineage is as co-founder and CEO of a vector-search company since 2021 (~5 years), and that is org leadership adjacent to language modeling, not a hands-on record of building or training statistical/neural language, seq2seq, transformer or embedding models — before Qdrant his career (VZnet, Bigpoint, MoBerries) was general software engineering with no LM record. Qdrant is a single, domain-agnostic natural-language vector-infrastructure product, not distinct hands-on modeling across biological, financial, code or other domains. He is the CEO/business co-founder of exactly one AI company whose science and engineering were done by his technical co-founder, so he does not qualify as its scientific/technical founder (0 years, 0 companies in that role).","evidence":[{"claim":"GitHub's commit search attributes only 6 commits in qdrant/qdrant to author azayarni, and their messages are 'Update logo.svg', 'Roadmap Badge Title', 'adjusted the cloud link', 'Added cloud form link', 'Updated ETA' and 'README spellcheck fixes' — documentation and branding, not engine code","source_url":"https://api.github.com/search/commits?q=author:azayarni+repo:qdrant/qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Across the whole qdrant organisation his account has 46 commits, predominantly merge commits in qdrant/landing_page and qdrant/docs plus initial commits in .github and qdrant-dotnet","source_url":"https://api.github.com/search/commits?q=author:azayarni+org:qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub user azayarni is Andre Zayarni, company @qdrant, bio 'Co-founder at Qdrant', Berlin, with 7 public repositories and 80 followers","source_url":"https://api.github.com/users/azayarni","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Qdrant was founded by André Zayarni and Andrey Vasnetsov in 2021; Zayarni is CEO and co-founder; Qdrant is an open-source vector similarity search engine built in Rust","source_url":"https://qdrant.tech/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The qdrant/qdrant repository is a vector similarity search engine and vector database written in Rust with ~34.5k stars","source_url":"https://github.com/qdrant/qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"M.Sc. in Medieninformatik (Media Informatics), Karlsruhe University of Applied Sciences, 2007; career as Senior Software Developer/Engineer and technical product roles (VZnet Netzwerke, Bigpoint, Spreadsave, Choisr, MoBerries) before co-founding Qdrant","source_url":"https://theorg.com/org/qdrant/org-chart/andre-zayarni","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Qdrant in Berlin in 2021 as Co-founder and CEO; Qdrant is an open-source vector similarity search engine/database","source_url":"https://qdrant.tech/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CTO Andrey Vasnetsov (ML engineer, prior roles at Tinkoff Bank and MoBerries) proposed the pivot to neural search and built Qdrant's initial engine in Rust — the core vector-search technology's primary technical architect","source_url":"https://theorg.com/org/qdrant/org-chart/andrey-vasnetsov","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Qdrant was founded by André Zayarni and Andrey Vasnetsov, who collaborated in 2021 on a project leveraging vector similarity search to build a matching engine for unstructured data; Zayarni is CEO and co-founder; Qdrant is a vector search engine built in Rust, offered open-source and as a managed cl","source_url":"https://qdrant.tech/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub profile azayarni identifies André Zayarni as co-founder at Qdrant, Berlin, with work primarily in Rust and TypeScript and Pull Shark and Starstruck achievements","source_url":"https://github.com/azayarni","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Qdrant is an open-source vector similarity search engine/database in Rust, co-founded 2021 by André Zayarni (CEO) and Andrey Vasnetsov; Vasnetsov proposed the neural-search pivot and built the initial engine — the primary technical architect","source_url":"https://qdrant.tech/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub commit API shows only 6 commits by azayarni to core qdrant/qdrant, all documentation/branding (logo, badge title, cloud links, ETA, README spellcheck) — no vector-index or retrieval code","source_url":"https://api.github.com/search/commits?q=author:azayarni+repo:qdrant/qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No Wikipedia, Wikidata, OpenAlex, Semantic Scholar or patent record exists for Zayarni; first verifiable year is 2021 (Qdrant founding)","source_url":"https://github.com/azayarni","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Qdrant was founded in 2021 by André Zayarni and Andrey Vasnetsov; Zayarni is CEO/co-founder and Vasnetsov is CTO who built the Rust vector-search engine — the science/engineering belongs to the CTO, not Zayarni","source_url":"https://qdrant.tech/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub commit search attributes only 6 commits in the core qdrant/qdrant engine to azayarni (logo, badge title, two link edits, ETA, README spellcheck) — no vector-index, quantization or retrieval code, and no papers or patents under his name","source_url":"https://api.github.com/search/commits?q=author:azayarni+repo:qdrant/qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"qdrant/qdrant is a vector similarity search engine / vector database — retrieval infrastructure serving embedding-based RAG, downstream of frontier models rather than part of their training/inference foundation","source_url":"https://github.com/qdrant/qdrant","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Zayarni's pre-Qdrant career (VZnet Netzwerke, Bigpoint, Spreadsave, Choisr, MoBerries) was general senior software engineering and product roles, with no language-modeling research or systems record","source_url":"https://theorg.com/org/qdrant/org-chart/andre-zayarni","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2021,"notable_systems":["Qdrant vector search engine (co-founder and CEO, not engine author)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier is empty on every source: no Wikipedia, Wikidata, OpenAlex, Semantic Scholar or profile page, patents skipped, and a PubMed query returning 2,729 homonym hits that was correctly discarded. Everything here comes from sources I verified directly (Qdrant's about page, the qdrant/qdrant repository, his GitHub user record and GitHub's commit search API). I could not confirm his education from any primary source; pass 1 cited an M.Sc. in Medieninformatik from Karlsruhe University of Applied Sciences via a third-party org-chart aggregator, which is not a primary source, so no degree is recorded. first_verifiable_year is set to 2021, the founding of Qdrant, the earliest date I could verify.","adjudication_notes":["vector_embeddings (8 vs 13) — this is the flagged invariant: pass 1 was right and pass 2's 13 is rejected. The invariant permits a score above 7 only on cited verified engineering work (authored code, papers, patents). Verified fact: GitHub's commit API shows his 6 commits to the core qdrant/qdrant engine are a logo update, a badge title, two link edits, an ETA change and README spellcheck fixes — no vector-index or retrieval code. No paper or patent exists under his name. Scored 7, the ceiling the invariant allows, on the strength of having co-founded and led the vector-search product itself.","hands_on_engineering (8 vs 11): pass 1 was right; settled at 8. Pass 2 inferred hands-on engineering from a Pull Shark badge and Rust/TypeScript language stats; the underlying commits are documentation and merge commits, and pass 2's own evidence concedes engine authorship belongs to CTO Andrey Vasnetsov.","foundations (4 vs 6): pass 1 was right; settled at 4. No degree is confirmable from a primary source and there is no mathematics or ML publication of any kind.","transformers_lm (2 vs 5): split at 3. Qdrant serves LLM retrieval but trains no models and he has authored nothing in the lineage.","industry_impact (9 vs 11): settled at 10. Qdrant is genuinely core embedding-retrieval infrastructure with wide adoption, and leading the org that ships it is creditable even though he did not author the engine."],"years_language_modeling":5,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2021-2026","evidence":"CEO/co-founder of Qdrant, a general-purpose vector-search engine for embedding retrieval; org leadership, not hands-on modeling"}],"frontier_lineage":["None — Qdrant is downstream vector-retrieval infrastructure; no authored foundational block and no pre-2013 vector-space/distributional-semantics precursor work"],"technical_founder_roles":["Qdrant — CEO & business co-founder (technical/scientific direction held by CTO Andrey Vasnetsov) — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":5},"confidence":0.76,"duration_ms":60088},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":5,"lm_domain_breadth":3,"scientific_founder":4},"confidence":0.82,"duration_ms":76558}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"nat-friedman","name":"Nat Friedman","title":"Head of Product, Meta Superintelligence Labs; former CEO of GitHub","company":"Meta (Meta Superintelligence Labs)","sector":"general","rank":67,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":2,"transformers_lm":4,"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":11,"industry_impact":11,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":26,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":26,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.87,"rationale":"Friedman is a career software engineer and executive with a substantial building record and no personal research record in the AI lineage. He earned a BS from MIT in 1999 in computer science and mathematics — real quantitative training, but at undergraduate level and with no thesis, graduate coursework or publication in learning theory or optimization. I retrieved his Semantic Scholar author record directly and it confirms the absence: the only genuine entries are GNOME-era software writing ('The Bonobo component and document model', 1999; 'Bringing Usability to Open Source', 2006), while 'Carmina Gallo: Intertextual Metapoetics in Virgil's Eclogues' (2013) belongs to a different Nat Friedman. There is no paper, preprint or patent by him anywhere in embeddings, attention, pretraining or scaling. His engineering history is genuine and hands-on — co-founding Ximian with Miguel de Icaza and working on GNOME infrastructure, then co-founding Xamarin around Mono, acquired by Microsoft in 2016 — but that is developer tooling and systems software, not machine learning. As CEO of GitHub from 2018 to 2021 he shipped GitHub Copilot, the first mass-deployed LLM coding product, alongside Codespaces; that is the strongest entry on his record and the basis of the industry-impact score, but every available source documents it as product leadership, with no personal contribution to Codex's training, architecture or evaluation verifiable. His subsequent AI involvement — AI Grant and NFDG investing with Daniel Gross, advising Midjourney, nat.dev, and instigating and funding the Vesuvius Challenge, where his listed role is 'Instigator, Director & Founding Sponsor' while the ink-detection and virtual-unwrapping ML is done by competing teams — is capital allocation and convening, which the rubric excludes from research credit.\n\nFriedman authored no method, architecture, dataset, optimizer, tokenizer, embedding or training/inference component that today's frontier models descend from, and has no verifiable pre-2013 vector-space or distributional-semantics work — his frontier contact is deploying OpenAI's Codex as GitHub Copilot while CEO (2020-2021) and now heading product at Meta Superintelligence Labs (2025), which is product leadership, not lineage authorship, so frontier_founder sits at the '3-7 applies/fine-tunes frontier models' band. His hands-on language-modeling record is near-zero: no LM papers, patents or systems he personally modeled; Copilot is a vendor model applied to code, which the rubric expressly excludes from domain-modeling credit, so years_language_modeling is effectively 0 and lm_domain_depth stays in the under-3-years/adjacent band. The only LM-adjacent domain touch is code (applying Codex), an 'application of someone else's model with no modeling work' — a single applied domain, placing lm_domain_breadth in the 3-7 floor. He is a genuine, hands-on technical co-founder of Ximian (1999-2003) and Xamarin (2011-2016), but both are developer-tooling/systems-software companies whose core is not language modeling or these AI systems, so scientific_founder scores in the 3-7 'technical founder outside this field' band with zero in-field technical-founder companies.","evidence":[{"claim":"BS from MIT (1999) in computer science and mathematics; co-founded Ximian with Miguel de Icaza 1999-2003; CTO of Open Source at Novell 2003-2010; co-founded and led Xamarin 2011-2016 (acquired by Microsoft); CEO of GitHub 2018-2021, during which Copilot and Codespaces shipped; head of product at Met","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 41234023 contains only four items: 'The Bonobo component and document model' (1999), 'Eof: bringing usability to open source' (2006), 'Bringing Usability to Open Source' (2006) and 'Carmina Gallo: Intertextual Metapoetics in Virgil's Eclogues' (2013) — GNOME-era software writ","source_url":"https://api.semanticscholar.org/graph/v1/author/41234023/papers?fields=title,year,venue,authors&limit=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as 'Instigator, Director & Founding Sponsor' of the Vesuvius Challenge; the machine-learning work (ink detection, virtual unwrapping, segmentation) is performed by the competing technical teams","source_url":"https://scrollprize.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education at MIT, occupations programmer/engineer/computer scientist, employers GitHub and Xamarin, and GitHub username 'nat' — no academic degree beyond the MIT BS and no research affiliation","source_url":"https://www.wikidata.org/wiki/Q92955","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS Computer Science and Mathematics, MIT, 1999; co-founded Ximian in 1999, met Miguel de Icaza via LinuxNet IRC network","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CTO of Open Source at Novell 2003-2010; co-founded and was CEO of Xamarin 2011-2016 (acquired by Microsoft in 2016)","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CEO of GitHub 2018-2021, during which GitHub Copilot, Codespaces, and the native mobile app shipped; currently head of product at Meta Superintelligence Labs (2025) and advisor to Midjourney, board member at Arc Institute","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar 'Nat Friedman' match shows only 5 papers / 1 citation / h-index 1 with 3 candidates (unresolved homonym risk); PubMed sample for 'Friedman N' returns entirely unrelated authors (cannabis-use research, pediatric genetics, infection control) confirming those are different people, not","source_url":"https://www.semanticscholar.org/author/41234023","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as 'Instigator, Director & Founding Sponsor' of the Vesuvius Challenge with a $2,250,000 donation; the ML/CV work (segmentation, virtual unwrapping, ink detection via iterative pseudo-labeling) is performed by the technical research teams, with Friedman's contributions directorial and financi","source_url":"https://scrollprize.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 41234023 attributes only GNOME-era software papers to him ('The Bonobo component and document model', 1999; 'Bringing Usability to Open Source', 2006), with 5 papers and 1 total citation","source_url":"https://api.semanticscholar.org/graph/v1/author/41234023/papers?fields=title,year,venue,authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Shipped GitHub Copilot (built on OpenAI Codex) as CEO of GitHub 2018-2021 and is now head of product at Meta Superintelligence Labs (2025) — product leadership over deployment of others' models, not authorship of frontier methods.","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Complete Semantic Scholar author list (id 41234023) contains only GNOME-era software writing and a homonym classics paper — zero papers in embeddings, attention, pretraining, scaling or language modeling.","source_url":"https://api.semanticscholar.org/graph/v1/author/41234023/papers?fields=title,year,venue,authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Ximian (Mono/GNOME) 1999-2003 and Xamarin 2011-2016 (acquired by Microsoft) as a hands-on technical founder of developer-tooling / systems-software companies, not AI/language-modeling companies.","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Complete Semantic Scholar author record (id 41234023) contains only GNOME-era software items ('The Bonobo component and document model' 1999; 'Bringing Usability to Open Source' 2006) — no embeddings, attention, transformer, pretraining or scaling work; OpenAlex finds no research corpus.","source_url":"https://api.semanticscholar.org/graph/v1/author/41234023/papers?fields=title,year,venue,authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Ximian (1999) with Miguel de Icaza and Xamarin (2011, CEO to 2016 Microsoft acquisition) — developer tooling / systems software, not language modeling; CEO of GitHub 2018-2021 when Copilot shipped; head of product at Meta Superintelligence Labs.","source_url":"https://en.wikipedia.org/wiki/Nat_Friedman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records employers GitHub and Xamarin and occupations programmer/engineer/computer scientist, with no research affiliation or AI-lineage authorship.","source_url":"https://www.wikidata.org/wiki/Q92955","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS computer science and mathematics, Massachusetts Institute of Technology (1999)"],"canonical_papers":[],"first_verifiable_year":1999,"notable_systems":["Ximian / GNOME (Bonobo, Evolution)","Mono / Xamarin","GitHub Copilot (shipped as CEO)","GitHub Codespaces","nat.dev","Vesuvius Challenge (instigator and funder)"],"citations":1,"h_index":1,"patents":0,"dossier_notes":"OpenAlex found nothing for him, which is correct — he has no research corpus. The Semantic Scholar match (id 41234023) is a partial homonym merge, which I verified by retrieving the paper list: the Virgil's Eclogues classics paper belongs to a different Nat Friedman (the sculptor/mathematician of math-art conferences), while the GNOME-era entries are genuinely his; the headline '5 papers / 1 citation' is near-zero either way. The PubMed block (1134 raw hits, refined to 97, homonym_risk true) is dominated by unrelated medical researchers named Friedman N/NP/NR and was excluded entirely. The dossier's timeline is entirely null; his first verifiable year as a builder is 1999. The dossier omits the roles most relevant to this index — Copilot shipping under his tenure, the Vesuvius Challenge, AI Grant/NFDG, the Midjourney advisory role and the Meta Superintelligence Labs position — none of which, verified, amounts to personal research.","adjudication_notes":["foundations — SPLIT, final 6 (pass_1 5, pass_2 7). A small dispute. The verified fact is an MIT BS in computer science and mathematics (1999) and nothing beyond it: no thesis, no graduate training, no publication in optimization or statistical learning. That is real undergraduate quantitative training, which places him just below the 8-12 'strong graduate training' band in both passes' reasoning; 6 splits a difference neither pass can support more precisely.","transformers_lm — pass_1 closer, final 4 (pass_1 3, pass_2 6). I retrieved his complete Semantic Scholar paper list and it contains zero AI papers; pass_2's own rationale concedes 'no personal contribution to Codex's training, architecture or evaluation is verifiable'. Shipping Copilot as CEO is product leadership, which the rubric explicitly excludes from research credit, so the 3-7 'manages builders, no personal record' band is correct and 6 sits too high within it.","vector_embeddings — pass_1 right, final 2 vs pass_2's 4. Neither pass cited any embedding, retrieval or vector-search work by him, and his verified publication list contains none.","hands_on_engineering — PASS_2 closer, final 11 (pass_1 6, pass_2 12). Pass_1's 6 understates a verified twenty-year record of personally building shipped systems software — Bonobo/GNOME infrastructure, Evolution, Mono and Xamarin — and pass_2 correctly identifies him as a genuine author of the 1999 Bonobo work rather than merely a manager. It is not AI engineering, however, which keeps it below the 13-17 band.","industry_impact — SPLIT, final 11 (pass_1 10, pass_2 12). Both passes credit the same verified fact — GitHub Copilot, the first mass-deployed LLM coding product, shipped under his tenure as CEO — and differ only on weight. Xamarin and GitHub are real orgs he led, but only Copilot has these systems at its core, and the rubric's exclusion of fundraising and convening removes AI Grant, NFDG and the Vesuvius sponsorship from the calculation entirely.","penalties — both passes agree at 0/0 and both are right: his career is a verifiable engineering record, not an AI company founded on family money, and no citable evidence of paid coverage or purchased reach was found."],"years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"code","years":"2020-2021","evidence":"Deployed OpenAI Codex as GitHub Copilot while CEO of GitHub — application of a vendor model, not personal code-LM modeling"}],"technical_founder_roles":["Ximian — co-founder (technical, Mono/GNOME) — 1999-2003 (outside AI/LM)","Xamarin — co-founder & CEO (Mono) — 2011-2016 (outside AI/LM)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":5},"confidence":0.86,"duration_ms":60443},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":6},"confidence":0.83,"duration_ms":57774}],"source":"community","profile_url":"https://en.wikipedia.org/wiki/Nat_Friedman","wikipedia_title":"Nat Friedman","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"emin-gun-sirer","name":"Emin Gun Sirer","title":"Co-founder & CEO","company":"Ava Labs (Avalanche)","sector":"crypto","rank":68,"rubric_version":4,"dimensions":{"foundations":9,"vector_embeddings":3,"transformers_lm":1,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":14,"industry_impact":10,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":25,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":25,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.82,"rationale":"Sirer earned a PhD in Computer Science and Engineering from the University of Washington (2002, advisor Brian Bershad) and was an associate professor at Cornell University before founding Ava Labs. His genuinely deep, personally authored research record (OpenAlex: 156 works, 11,707 citations, h-index 49, unambiguous topical match to distributed systems/P2P/blockchain) spans operating systems (SPIN, 1995, 955 citations), peer-to-peer systems (Meridian, 2005), and blockchain/consensus, including the influential 'Majority Is Not Enough: Bitcoin Mining Is Vulnerable' (2014, 1,409 citations) and 'On Scaling Decentralized Blockchains' (2016, 1,189 citations). This is strong, PhD-level, personally-built systems and distributed-computing research with real academic depth and a long track record (first verifiable year 1994, 32 years active), but it sits in distributed systems, consensus, and networking -- not in the mathematics of embeddings, representation learning, or the attention/transformer/LM lineage the rubric weights most heavily. No AI/ML-specific publications were found. He designed the Avalanche consensus protocol underlying Ava Labs' blockchain, a real, personally-led technical system, supporting solid hands-on-engineering and foundations scores but low scores on the two core-AI dimensions.\n\nSirer's deep, personally-authored research record (h-index 49, 11,707 citations) sits entirely in operating systems (SPIN 1995), peer-to-peer systems (Meridian, KARMA), and blockchain consensus (selfish-mining 2014, On Scaling Decentralized Blockchains 2016, the Avalanche consensus protocol) — none of it touches the vector-space / distributional-semantics / attention / transformer / embedding lineage that today's frontier language models descend from, so no verifiable frontier_founder position exists. He has zero years of language-modeling work (no n-gram/neural LM, LSI/vector-space text, seq2seq, transformer, embedding or LLM papers, patents or shipped systems in the dossier or Google Scholar), so lm_domain_depth and lm_domain_breadth are near-absent with no distinct LM domains to list. He IS a genuine scientific/technical founder — he personally designed the Avalanche consensus protocol that Ava Labs (co-founded 2019, ~7 years) is built on and serves as founder-CEO setting technical direction — but that company's core is blockchain consensus, not language-modeling/AI systems, which places him in the 'technical founder outside this field' band. Counting one company and ~7 years as a technical founder outside the language-modeling field yields a mid-single-digit scientific_founder score.","evidence":[{"claim":"PhD Computer Science and Engineering, University of Washington (2002), advisor Brian Bershad; associate professor at Cornell University","source_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-authored 'Majority Is Not Enough: Bitcoin Mining Is Vulnerable' with Ittay Eyal (2014), an influential blockchain-security paper","source_url":"https://doi.org/10.1007/978-3-662-45472-5_28","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Designed the Avalanche consensus protocol and co-founded/CEO of Ava Labs","source_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD Computer Science and Engineering, University of Washington 2002, advisor Brian N. Bershad; Princeton undergraduate; associate professor at Cornell and former IC3 co-director; SPIN OS, HyperDex, KARMA (2003), selfish-mining paper, Avalanche consensus; founded Ava Labs 2019","source_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar (DaDmjMMAAAAJ), Cornell University: ~21,740 citations, h-index 58, i10 111; research focus operating systems, distributed systems, networking and blockchain - no machine learning, neural network, embedding or language-model papers among top works","source_url":"https://scholar.google.com/citations?user=DaDmjMMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Majority Is Not Enough: Bitcoin Mining Is Vulnerable' (Eyal & Sirer), Financial Cryptography 2014, ~1,409 citations","source_url":"https://doi.org/10.1007/978-3-662-45472-5_28","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sirer developed the Avalanche Consensus protocol and is CEO and co-founder of Ava Labs; former Cornell associate professor and IC3 co-director","source_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex/Google Scholar record is operating systems, P2P and blockchain consensus (SPIN, Meridian, selfish-mining, scaling blockchains) with no machine-learning, embedding or language-model works","source_url":"https://scholar.google.com/citations?user=DaDmjMMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Majority Is Not Enough: Bitcoin Mining Is Vulnerable' (Eyal & Sirer, Financial Cryptography 2014) — a blockchain-security, not language-modeling, contribution","source_url":"https://doi.org/10.1007/978-3-662-45472-5_28","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Designed the Avalanche consensus protocol underlying the Avalanche blockchain and is CEO/co-founder of Ava Labs (founded 2019); former Cornell associate professor and IC3 co-director — technical founder who authored the company's core science, but in distributed-systems/blockchain, not AI/LM","source_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Top works are 'Majority Is Not Enough: Bitcoin Mining Is Vulnerable' (2014), 'On Scaling Decentralized Blockchains' (2016), and SPIN OS (1995) — no embedding, representation-learning, neural-LM, seq2seq or transformer work in the record, confirming no frontier-model lineage and no language-modeling","source_url":"https://doi.org/10.1007/978-3-662-45472-5_28","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Google Scholar profile (DaDmjMMAAAAJ) research areas are operating systems, distributed systems, networking and blockchain — zero machine-learning/NLP/language-model publications","source_url":"https://scholar.google.com/citations?user=DaDmjMMAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science and Engineering, University of Washington (2002, advisor Brian Bershad)"],"canonical_papers":["Extensibility, Safety and Performance in the SPIN Operating System (1995)","Majority Is Not Enough: Bitcoin Mining Is Vulnerable (2014)","On Scaling Decentralized Blockchains (2016)"],"first_verifiable_year":1994,"notable_systems":["SPIN OS","KARMA (P2P currency, 2003)","HyperDex","Avalanche consensus protocol"],"citations":11707,"h_index":49,"patents":0,"dossier_notes":"OpenAlex match is flagged ambiguous (3 candidates) but the topics (Peer-to-Peer Network Technologies, Distributed systems, Blockchain Technology) and top works (SPIN, Bitcoin selfish-mining paper) are fully consistent with the known Sirer/Cornell/Ava Labs identity -- no evidence of homonym contamination. His deep, decades-long record is in distributed systems and blockchain consensus, not AI/ML/embeddings/transformers, which the rubric explicitly does not credit toward the core dimensions.","years_language_modeling":0,"years_as_technical_founder":7,"technical_founder_companies":1,"technical_founder_roles":["Ava Labs — co-founder & CEO, designer of Avalanche consensus protocol — 2019-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.9,"duration_ms":42296},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.9,"duration_ms":45763}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Emin_G%C3%BCn_Sirer","wikipedia_title":"Emin Gün Sirer","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"harry-grieve","name":"Harry Grieve","title":"Co-founder & CEO","company":"Gensyn","sector":"crypto","rank":69,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":3,"transformers_lm":7,"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":3,"hands_on_engineering":7,"industry_impact":8,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":25,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":25,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.55,"rationale":"The dossier's Wikipedia (chess grandmaster) and OpenAlex (University of Bath organometallic chemistry researcher) matches are both clearly the wrong person and are disregarded. However, one OpenAlex-listed 'top_work' — the arXiv paper 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (2509.08721, 2025) — was independently verified via direct arXiv fetch to be a genuine Gensyn systems paper, and Harry Grieve is confirmed as one of its 15 listed co-authors (alongside other Gensyn engineers such as Jeffrey Amico and Matthew J. Wright), with the paper reporting tests on a decentralized RL training network of thousands of Gensyn community nodes. No personal lead-authorship, individually-authored papers, specific contribution statement, university degree, or pre-Gensyn technical role could be verified for Grieve via direct web fetch (search tools were unavailable this session; Gensyn's site has no team/about page and LinkedIn/Crunchbase pages returned errors). Gensyn itself has a substantial, real technical research output (RL post-training, MoE routing, pipeline-parallelism security) rather than being vaporware, which supports a moderate industry_impact and hands_on_engineering score for building/leading the org, but with no verified personal research record Grieve individually scores low on foundations and vector_embeddings per rubric guidance to score lower when unsure. transformers_lm and hands_on_engineering get partial credit strictly for the one verified co-authorship credit on a real transformer/LM post-training systems paper produced by an org he leads, not for demonstrated personal depth.\n\nGrieve is co-founder (and per Crunchbase/getprog.ai a co-founder/CTO) of Gensyn since 2020, a decentralized machine-learning COMPUTE protocol — infrastructure for renting GPUs and coordinating training, not a language-modeling method that frontier GPT/Claude/Gemini-class models build on; there is no verifiable pre-2013 vector-space/distributional-semantics lineage work, so frontier_founder sits at the 'applies/builds adjacent infra' floor. His only verified language-modeling artifact is a 1-of-15 co-author credit on Gensyn's 2025 RL post-training paper (SAPO, arXiv:2509.08721), giving well under a year of personal, hands-on LM record in a single domain (natural-language LLM post-training) — placing lm_domain_depth and lm_domain_breadth in the lowest band; his pre-Gensyn ML work (Director of Data Research at Cytora on commercial-insurance data, an earlier skincare-recommender startup) is applied ML, not language modeling. As a founder he has ~6 years (2020-2026) in a technical/CTO role at one company whose core is AI/ML systems, with a background in applied ML and economics (MPA Brown, MA Economics/Finance Aberdeen) rather than the mathematics/optimization foundations of the field, supporting a mid-band scientific_founder score but not higher given his personal research output is a single shared authorship. Scored conservatively where the CTO title and personal core-authorship are only weakly corroborated by self-reported profiles.","evidence":[{"claim":"Harry Grieve is confirmed (via direct arXiv fetch) as one of 15 co-authors on 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing', a Gensyn paper on collective RL experience sharing tested on a decentralized network of thousands of Gensyn community nodes.","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The same paper's PDF confirms the 15-author list including Harry Grieve, and cites Gensyn's own prior work (gensyn2025genrl, gensyn2025rlswarm), confirming this is a genuine Gensyn-authored systems paper, not a homonym or unrelated work.","source_url":"https://arxiv.org/pdf/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn maintains an active research publication list (RL post-training, mixture-of-experts routing, pipeline-parallelism security, prediction-market mechanism design), indicating the company does genuine technical AI/ML infrastructure work rather than being purely business/marketing-driven.","source_url":"https://www.gensyn.ai/research","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn's product line (Delphi information markets, RL-Swarm, AXL peer-to-peer AI communication, CodeAssist) confirms the company's core business is decentralized ML training/inference infrastructure, consistent with the dossier's company description.","source_url":"https://www.gensyn.ai/news","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author (sixth of fifteen) of 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (arXiv:2509.08721, 10 September 2025), introducing SAPO, decentralized RL post-training via shared rollouts across heterogeneous nodes","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Gensyn litepaper (February 2022) is authored by 'the Gensyn team' and derives its verification approach from Jia et al. (2021) probabilistic proof-of-learning, Zheng et al. (2021) graph-based pinpoint protocol and Truebit-style incentive games, rather than original learning results","source_url":"https://docs.gensyn.ai/litepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"An arXiv author search for 'Grieve_H' returns zero results, indicating no independent arXiv publication record under that name","source_url":"http://export.arxiv.org/api/query?search_query=au:%22Grieve_H%22&start=0&max_results=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn's public product and research line (Delphi, CodeAssist, REE, AXL peer-to-peer communication) is company output listed on its own site, with no individual authorship attributed to Grieve","source_url":"https://www.gensyn.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn is a decentralized machine-learning compute protocol (GPU marketplace / training coordination infrastructure), not a frontier language-model lab.","source_url":"https://grieve.eth.link/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Grieve's sole verified language-modeling artifact is a 1-of-15 co-authorship on Gensyn's 2025 RL post-training paper (SAPO / collective RL experience sharing).","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Grieve confirmed as one of 15 co-authors on the 2025 Gensyn paper 'Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing' (SAPO), the only verifiable language-modeling artifact tied to him.","source_url":"https://arxiv.org/abs/2509.08721","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gensyn's own materials describe a decentralized machine-learning compute protocol/training network, not a frontier-model architecture or training objective that frontier labs build on.","source_url":"https://docs.gensyn.ai/litepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2025,"notable_systems":["Gensyn RL-Swarm / collective RL post-training network (co-author credit only, not verified as personal lead work)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Wikipedia match (English chess grandmaster, b. 2001) is clearly the wrong person — different domain entirely, disregarded. OpenAlex match (University of Bath researcher with 15 works on organometallic nickel NHC complexes and X-ray crystal structures, h-index 1) is clearly the wrong person — a chemistry researcher, not the Gensyn founder; disregarded except for one flagged top_work. That flagged top_work, arXiv 2509.08721 'Sharing is Caring', WAS independently verified as a real Gensyn paper with the real Harry Grieve (Gensyn CEO) as a listed co-author (1 of 15) — this single work should be attributed to the real person, not the OpenAlex chemistry identity it was bundled under. PubMed sample entries (veterinary pharmacovigilance and companion-animal disease surveillance research by a Canadian/UK 'Grieve H', plus a 1965 German-language public health law paper and a 2003 psycholinguistics paper) are all different homonyms unrelated to either the Gensyn founder or the chemistry researcher; disregarded. Could not verify Grieve's university, degree, or pre-Gensyn employment history via available tools this session (WebSearch budget was exhausted; Gensyn's own site has no team/about/founders page; LinkedIn, Crunchbase, and Tracxn fetches all failed with 403/404/402 errors) — scored conservatively low on personal research dimensions per 'if unsure, score lower.'","sector_note":"crypto","years_language_modeling":1,"years_as_technical_founder":6,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2025","evidence":"1-of-15 co-author, Gensyn SAPO RL post-training paper (arXiv:2509.08721); decentralized LLM RL post-training network"}],"frontier_lineage":["No verifiable foundational contribution to the frontier stack; Gensyn provides decentralized training/compute infrastructure adjacent to, not underneath, frontier models"],"technical_founder_roles":["Gensyn — Co-founder / CTO — 2020-2026 (~6 years)","Cytora — Director of Data Research (non-founder, applied ML on insurance data)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":9},"confidence":0.5,"duration_ms":90549},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":7},"confidence":0.6,"duration_ms":94737}],"source":"seeded","profile_url":null,"wikipedia_title":"Harry Grieve","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"tarun-chitra","name":"Tarun Chitra","title":"Founder & CEO","company":"Gauntlet","sector":"crypto","rank":70,"rubric_version":4,"dimensions":{"foundations":12,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":10,"industry_impact":8,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":25,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":25,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.75,"rationale":"Chitra has a genuine, verifiable quantitative-research record: B.A. Mathematics and B.S. Applied Engineering Physics from Cornell, prior quant/scientific-programming roles at D.E. Shaw Research and Vatic Labs (HFT), and a Google Scholar profile showing h-index 23 (1,882 citations) built almost entirely on DeFi mechanism-design and convex-optimization papers (Constant Function Market Makers / price oracles, Uniswap market analysis, MEV, intent-based markets). This is real applied mathematics and statistical/optimization work — strong foundations-dimension evidence — but it is financial/mechanism-design engineering, not the vector-embeddings or transformer/language-model lineage the rubric scores; no papers on embeddings, attention, or language modeling were found. Gauntlet applies simulation and optimization to DeFi risk parameters, not AI model-building. His technical depth is real but off-target for the core-AI dimensions.\n\nChitra's verifiable research record is entirely DeFi mechanism design and convex optimization — constant-function market makers, price oracles, optimal routing, staking/lending equilibria — with no vector-space, distributional-semantics, embeddings, attention, transformer or language-modeling work anywhere in OpenAlex, Semantic Scholar or his Google Scholar profile, so there is no position in the lineage that today's frontier language models descend from (frontier_founder ~2) and zero verifiable language-modeling years across any domain (lm_domain_depth/breadth ~2). His quant/HFT and DeFi financial work is market-microstructure and mechanism design, not 'financial language modeling' (no filings/news/time-series-as-language sequence models), so it does not earn a financial LM domain. He is, however, a genuine hands-on technical/scientific founder: he personally authored the CFMM/optimization papers and drives the technical direction of Gauntlet (founded 2018, ~8 years, one company), a simulation/optimization firm — but its core is DeFi risk parameters, not language-modeling/core-AI systems, so under the anchor 'a technical founder outside this field' he sits at the top of the 3-7 band. Company count is one.","evidence":[{"claim":"Google Scholar profile: h-index 23, 1,882 citations, top papers are AMM/DeFi mechanism-design and convex optimization (no ML/embeddings/LM papers)","source_url":"https://scholar.google.com/citations?user=_48EkToAAAAJ&hl=en","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"B.A. Mathematics and B.S. Applied Engineering Physics, Cornell University; prior roles at D.E. Shaw Research and Vatic Labs","source_url":"https://www.clay.com/dossier/gauntlet-ceo","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 6285600, affiliation Gauntlet: 44 papers, 1,241 citations, h-index 17; works include 'Improved Price Oracles: Constant Function Market Makers' (2020, 263 citations), 'An analysis of Uniswap markets' (2019, 250), 'Optimal Routing for Constant Function Market Makers' (2022), 'T","source_url":"https://api.semanticscholar.org/graph/v1/author/6285600?fields=name,affiliations,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author with Guillermo Angeris, Hsien-Tang Kao, Rei Chiang and Charlie Noyes of 'An analysis of Uniswap markets' (2019), which formally analyses constant product markets and validates stability via agent-based simulation","source_url":"https://arxiv.org/abs/1911.03380","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chief Executive Officer of Gauntlet; previously held positions in quantitative R&D at Vatic HFT and D.E. Shaw","source_url":"https://www.gauntlet.xyz/our-team","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5082358070: 49 works, 406 citations, h-index 11, topics blockchain technology applications and security, economic theories and models, financial markets, auction theory — no machine-learning or NLP topics","source_url":"https://api.openalex.org/authors/A5082358070","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 6285600 (affiliation Gauntlet): 44 papers, h-index 17, top works are 'Improved Price Oracles: Constant Function Market Makers', 'Optimal Routing for Constant Function Market Makers' — DeFi mechanism design, no embeddings/transformers/LM papers","source_url":"https://api.semanticscholar.org/graph/v1/author/6285600?fields=name,affiliations,paperCount,citationCount,hIndex,papers.title","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex A5082358070 topics are blockchain applications, economic theories, financial markets, auction theory — no machine-learning or NLP topics, confirming no language-modeling lineage","source_url":"https://api.openalex.org/authors/A5082358070","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chitra is founder-CEO of Gauntlet and personally authors its core CFMM/risk-simulation research (e.g. 'An analysis of Uniswap markets', 2019), making him a hands-on technical founder — but of a DeFi-optimization company, not an AI/LM-core company","source_url":"https://arxiv.org/abs/1911.03380","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chitra co-authored 'An Analysis of Uniswap Markets' (2019) and 'Improved Price Oracles: Constant Function Market Makers' (2020) — DeFi mechanism-design/optimization, no language-modeling or embedding content","source_url":"https://arxiv.org/abs/1911.03380","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chitra is founder and CEO of Gauntlet, whose core is DeFi risk simulation and optimization (not AI/language models); he personally authored the CFMM/oracle research it runs on","source_url":"https://www.gauntlet.xyz/our-team","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 6285600 (affiliation Gauntlet): 44 papers on AMM/DeFi/optimization; no NLP, embedding or transformer works","source_url":"https://api.semanticscholar.org/graph/v1/author/6285600","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["B.A. Mathematics, Cornell University","B.S. Applied Engineering Physics, Cornell University"],"canonical_papers":["Improved Price Oracles: Constant Function Market Makers (2020)","An Analysis of Uniswap Markets (2021)"],"first_verifiable_year":2020,"notable_systems":["Gauntlet risk-simulation platform"],"citations":1882,"h_index":23,"patents":0,"dossier_notes":"Dossier's PubMed block (24 results, gynecology/hematology papers by Indian medical authors 'Chitra T'/'Chitra TV') is a clear homonym unrelated to this person — discard entirely. OpenAlex flags 'ambiguous: true, candidates: 7' and undercounts citations (406) vs. Google Scholar's verified 1,882/h-index 23 — Google Scholar figure used instead. Dossier's earliest_year 1991/1982 (from OpenAlex/PubMed) is also homonym noise; true first verifiable year in his own record is ~2018-2020 (D.E. Shaw Research era papers, then Gauntlet founding 2018).","years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":1,"technical_founder_roles":["Gauntlet — Founder & CEO (authors core CFMM/optimization research and sets technical direction) — 2018-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.83,"duration_ms":52430},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.82,"duration_ms":59258}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"vincent-weisser","name":"Vincent Weisser","title":"Co-founder & CEO","company":"Prime Intellect","sector":"crypto","rank":71,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":3,"transformers_lm":7,"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"hands_on_engineering":8,"industry_impact":8,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":25,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":25,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.71,"rationale":"Vincent Weisser has no traditional CS/math degree: his education is a Product degree from CODE University of Applied Sciences plus an AI Safety Fundamentals course, not a PhD or research-track program. His prior ventures (Molecule GmbH, VitaDAO, Bio.xyz, dex.blue) are in biotech-DeFi and decentralized-exchange infrastructure, unrelated to core AI research. However, at Prime Intellect (co-founded with Johannes Hagemann) he is a listed contributor on real technical output — the dossier's OpenAlex/Semantic Scholar records show him as a co-author on the INTELLECT-3 technical report (arXiv 2512.16144, 2025), and Prime Intellect has shipped genuine decentralized/distributed LLM pretraining systems (the INTELLECT model series), which is hands-on engineering leadership of real training infrastructure rather than pure business role. This is a moderate case: some hands-on technical credibility via team-authored technical reports and building real distributed-training infra, but no personal foundational research record, so scored low-to-mid rather than high.\n\nWeisser co-founded Prime Intellect (~2023-24) and is a listed co-author only on the INTELLECT-3 team technical report (23 authors, 2025); he is not an author on INTELLECT-1 or INTELLECT-2, and Wikidata records his occupation as 'businessperson' with no research role — the technical science and core code (prime-rl, TOPLOC, SHARDCAST decentralized-training stack) are authored by Johannes Hagemann, Sami Jaghouar and the engineering team, not by him. Prime Intellect's decentralized/distributed pretraining is a training-orchestration approach, not a building block that frontier GPT/Claude/Gemini/Llama systems descend from, and Weisser has no verifiable pre-2013 vector-space or distributional-semantics lineage work, so frontier_founder sits low. His verifiable language-modeling record is confined to natural-language LLM training at Prime Intellect for roughly one to two years — the PubMed neuroimaging/multiple-sclerosis matches are an unrelated homonym, and his prior ventures (Molecule, VitaDAO, Bio.xyz) were biotech-funding DAOs, not biological language modeling, so both lm_domain_depth (<3 years) and lm_domain_breadth (a single natural-language domain) fall in the 3-7 band. He operates as a business founder-CEO whose company's science is executed by technical co-founders, which is the 3-7 scientific_founder band rather than a technical/scientific-founder score.","evidence":[{"claim":"Co-founder & CEO of Prime Intellect, building decentralized compute/training infrastructure for open AI models","source_url":"https://www.vincentweisser.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prior roles: Co-Initiator at Bio.xyz, Molecule GmbH, VitaDAO, founding member of dex.blue","source_url":"https://theorg.com/org/prime-intellect/org-chart/vincent-weisser","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed co-author, INTELLECT-3: Technical Report, arXiv 2512.16144 (2025)","source_url":"https://doi.org/10.48550/arxiv.2512.16144","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-3 Technical Report (106B-parameter MoE, prime-rl asynchronous RL framework) lists Vincent Weisser among 23 authors from the Prime Intellect Team","source_url":"https://arxiv.org/abs/2512.16144","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning — author list is Sami Jaghouar, Justus Mattern, Jack Min Ong, Jannik Straube, Manveer Basra, Aaron Pazdera, Kushal Thaman, Matthew Di Ferrante, Felix Gabriel, Fares Obeid, Kemal Erdem, Michael Keiblinger and","source_url":"https://arxiv.org/abs/2505.07291","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-1 Technical Report — authors are Sami Jaghouar, Jack Min Ong, Manveer Basra, Fares Obeid, Jannik Straube, Michael Keiblinger, Elie Bakouch, Lucas Atkins, Maziyar Panahi, Charles Goddard, Max Ryabinin and Johannes Hagemann; Weisser is not an author","source_url":"https://arxiv.org/abs/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prime Intellect's INTELLECT-2 introduces prime-rl, TOPLOC verifiable inference and SHARDCAST weight distribution for permissionless decentralized RL training","source_url":"https://www.primeintellect.ai/blog/intellect-2","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q126287602 lists Weisser's occupation as 'businessperson' and employer as Molecule GmbH, with no research role recorded","source_url":"https://www.wikidata.org/wiki/Q126287602","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub account vincentweisser has 0 public repositories","source_url":"https://github.com/vincentweisser","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q126287602 lists occupation 'businessperson' and employer Molecule GmbH, with no research role","source_url":"https://www.wikidata.org/wiki/Q126287602","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-1 Technical Report authors are Jaghouar, Ong, Basra, Obeid, Straube, Keiblinger, Bakouch, Atkins, Panahi, Goddard, Ryabinin and Hagemann — Weisser is not an author","source_url":"https://arxiv.org/abs/2412.01152","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-2 (prime-rl, TOPLOC, SHARDCAST decentralized RL training) author list does not include Weisser","source_url":"https://arxiv.org/abs/2505.07291","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"INTELLECT-3 Technical Report lists Vincent Weisser among ~23 Prime Intellect Team authors","source_url":"https://arxiv.org/abs/2512.16144","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder & CEO of Prime Intellect; prior roles at Bio.xyz, Molecule, VitaDAO, dex.blue (biotech-DeFi / DEX, not language modeling)","source_url":"https://theorg.com/org/prime-intellect/org-chart/vincent-weisser","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Weisser is listed among 23 Prime Intellect Team authors on INTELLECT-3: Technical Report (arXiv 2512.16144, 2025); he is not an author of INTELLECT-1 (arXiv 2412.01152) or INTELLECT-2 (arXiv 2505.07291)","source_url":"https://arxiv.org/abs/2512.16144","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prime Intellect's decentralized-training contributions (prime-rl, TOPLOC verifiable inference, SHARDCAST weight distribution) are the company's technical output, not a named building block of frontier models","source_url":"https://www.primeintellect.ai/blog/intellect-2","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q126287602 records Weisser's occupation as 'businessperson' and employer as Molecule GmbH, with no research role; GitHub vincentweisser has 0 public repositories","source_url":"https://www.wikidata.org/wiki/Q126287602","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Prior roles are biotech-DeFi and DEX infrastructure (Co-Initiator at Bio.xyz, Molecule GmbH, VitaDAO, founding member of dex.blue), not language-modeling domains","source_url":"https://theorg.com/org/prime-intellect/org-chart/vincent-weisser","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Product degree, CODE University of Applied Sciences","AI Safety Fundamentals course, AI Safety Fundamentals (non-degree)"],"canonical_papers":["INTELLECT-3: Technical Report (2025, team co-author)"],"first_verifiable_year":2025,"notable_systems":["Prime Intellect distributed training platform","INTELLECT-1/2/3 model series"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex record is explicitly flagged ambiguous (2 candidates) and shows only 1 work (INTELLECT-3), 0 citations; PubMed's 8 matches (PTSD neuroimaging, multiple sclerosis driving studies) are unrelated homonyms from a different Vincent Weisser and should not be attributed to this person. No prior AI research career before Prime Intellect (founded ~2023-24) was found; first_verifiable_year of 2003 in the dossier's timeline comes from the unrelated PubMed homonym and is not reliable — true first verifiable AI-relevant year is ~2024-2025.","years_language_modeling":2,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"natural_language","years":"2024-2025","evidence":"Prime Intellect decentralized LLM pretraining (INTELLECT series); co-author on INTELLECT-3 team report only"}],"frontier_lineage":["No verifiable personal contribution to the frontier-model foundation; Prime Intellect's decentralized-training orchestration is not a building block frontier GPT/Claude/Gemini/Llama systems descend from"],"technical_founder_roles":["Prime Intellect — co-founder & CEO (business/vision role; science led by co-founder Johannes Hagemann and team) — 2023-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":4},"confidence":0.7,"duration_ms":47331},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":5,"lm_domain_breadth":3,"scientific_founder":4},"confidence":0.74,"duration_ms":58242}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"vitalik-buterin","name":"Vitalik Buterin","title":"Co-founder","company":"Ethereum","sector":"crypto","rank":72,"rubric_version":4,"dimensions":{"foundations":10,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":12,"industry_impact":8,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":25,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":25,"tier":"informed_operator","tier_label":"Informed Operator","confidence":0.8,"rationale":"Buterin never completed a degree (he left the University of Waterloo after taking a Thiel Fellowship, though he did work as an undergraduate research assistant for cryptographer Ian Goldberg), so foundations is scored on verifiable self-taught depth in cryptography, distributed consensus, and mechanism design rather than credentials — this is real mathematical/algorithmic work (he authored the Ethereum whitepaper and Yellow Paper-adjacent design at 19-20) but is cryptography/game-theory/distributed-systems, not the linear-algebra/statistical-learning/ML core this rubric targets. OpenAlex confirms a genuine, non-trivial academic record (22 works, h-index 14, 789 citations) including peer-reviewed papers like 'A Flexible Design for Funding Public Goods' (Management Science, 2019) and 'Combining GHOST and Casper' — real, citable technical output, though entirely in blockchain consensus/cryptoeconomics, not vector embeddings or transformer/LM research. He personally designed and built Ethereum's protocol (hands-on engineering of a system that today runs at global scale), which supports high hands_on_engineering credit even though the system itself is not an AI system. No evidence found of him authoring or leading AI/ML research; his industry impact is enormous in crypto but not in the AI-systems sense this index measures, so transformers_lm and vector_embeddings score near the floor.\n\nButerin's entire verifiable technical record is in blockchain consensus, cryptoeconomics, mechanism design and applied cryptography (Ethereum whitepaper 2013, 'Combining GHOST and Casper' 2020, 'Aggregatable Subvector Commitments' 2020, 'A Flexible Design for Funding Public Goods' 2019) — none of it is in the vector-space/distributional-semantics/attention/transformer lineage that today's frontier language models descend from, and there are no pre-2013 term-vector or LSI patents or shipped LM systems, so frontier_founder sits near the floor. He has zero verifiable years of hands-on language-modeling work (statistical/neural LMs, embeddings, seq2seq, transformers, LLM pretraining/alignment), placing lm_domain_depth in the 0-2 band, and no distinct LM domains with a modeling record (his blockchain-energy PubMed co-authorship is commentary, not language modeling), placing lm_domain_breadth at the floor. He IS, however, a genuine scientific/technical founder — he personally authored the core research, protocol design and code that Ethereum runs on from 2013 onward (~11+ years, one company) — but that company's core is a blockchain, not language-modeling/AI systems, so under the anchors this is a technical founder OUTSIDE this field, scoring in the 3-7 band rather than higher. His in-field (language-modeling) technical-founder years and company count are therefore zero.","evidence":[{"claim":"Wrote the original Ethereum whitepaper in 2013 at age 19 and launched the network in 2015; worked as an undergraduate research assistant for cryptographer Ian Goldberg at University of Waterloo before leaving via Thiel Fellowship","source_url":"https://en.wikipedia.org/wiki/Vitalik_Buterin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Bitcoin Magazine in 2011 before designing and building Ethereum's smart-contract platform","source_url":"https://btcdirect.eu/en-eu/who-is-vitalik-buterin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attended University of Waterloo and was a research assistant for cryptographer Ian Goldberg; dropped out in 2014 after a $100,000 Thiel Fellowship; described Ethereum in a white paper in November 2013; his documented AI involvement is existential-risk philanthropy and commentary, with no machine-lea","source_url":"https://en.wikipedia.org/wiki/Vitalik_Buterin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author record (Ethereum Foundation, unambiguous single-candidate match): 22 works, 789 citations, h-index 14, topics blockchain, cryptography and data security, game theory and auction theory","source_url":"https://api.openalex.org/authors/A5069172917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Aggregatable Subvector Commitments for Stateless Cryptocurrencies' (SCN 2020), a vector-commitment cryptography paper","source_url":"https://doi.org/10.1007/978-3-030-57990-6_3","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Combining GHOST and Casper' (2020), specifying Ethereum's proof-of-stake fork-choice and finality gadget","source_url":"https://arxiv.org/abs/2003.03052","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wrote the Ethereum whitepaper in 2013 and deployed the protocol in 2015; his documented work is blockchain/cryptography/cryptoeconomics with no machine-learning or language-modeling authorship","source_url":"https://en.wikipedia.org/wiki/Vitalik_Buterin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author record (Ethereum Foundation, unambiguous): 22 works, 789 citations, h-index 14, topics blockchain, cryptography and data security, game theory, auction theory — none in embeddings, seq2seq or transformers","source_url":"https://api.openalex.org/authors/A5069172917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Aggregatable Subvector Commitments for Stateless Cryptocurrencies' is a cryptographic vector-commitment scheme, not a vector-embedding/representation-learning method — no frontier-model lineage","source_url":"https://doi.org/10.1007/978-3-030-57990-6_3","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Buterin authored the Ethereum whitepaper (Nov 2013) at age 19 and personally set and executed the protocol's technical/research direction, operating as its de facto chief scientist — a real technical founder, but of a blockchain protocol, not a language-modeling system","source_url":"https://en.wikipedia.org/wiki/Vitalik_Buterin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex record (22 works, h-index 14, topics blockchain, cryptography, game theory, auction theory) contains no NLP/language-model/embedding/transformer work","source_url":"https://api.openalex.org/authors/A5069172917","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["University of Waterloo (attended, did not complete degree)"],"canonical_papers":["Ethereum: A Next-Generation Smart Contract and Decentralized Application Platform (2013/2014 whitepaper)","A Flexible Design for Funding Public Goods (2019)","Combining GHOST and Casper (2020)"],"first_verifiable_year":2011,"notable_systems":["Ethereum protocol"],"citations":789,"h_index":14,"patents":0,"dossier_notes":"The dossier's single PubMed hit ('Confronting the Carbon-Footprint Challenge of Blockchain', 2023) is plausibly a genuine co-authorship (topic matches his known interest in blockchain energy use) but is flagged homonym_risk:false with low confidence given it is a single-paper match; treated as weak-corroborating evidence only, not weighted heavily. OpenAlex identity match is confident (ambiguous:false, candidates:1, Ethereum Foundation affiliation) and consistent with Wikipedia/Wikidata. Buterin's record is real and technical but is cryptography/cryptoeconomics, not core AI (linear algebra/embeddings/transformers) — scored per the rubric's explicit instruction that fame and blockchain success are not evidence of AI depth.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Ethereum — founder / core protocol designer & author, 2013-present (~11+ yrs) — technical founder OUTSIDE the language-modeling/AI field; not counted in in-field years"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.86,"duration_ms":46143},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.9,"duration_ms":48655}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Vitalik_Buterin","wikipedia_title":"Vitalik Buterin","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"david-minarsch","name":"David Minarsch","title":"Co-founder & CEO","company":"Valory (Olas)","sector":"crypto","rank":73,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":2,"transformers_lm":4,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":10,"industry_impact":6,"scientific_founder":8},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":24,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":24,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.78,"rationale":"Minarsch holds a PhD in Applied Game Theory from the University of Cambridge and has a modest but genuine peer-reviewed record (OpenAlex: 10 works, 62 citations, h-index 5; Semantic Scholar corroborates 9 papers, 64 citations, h-index 5) spanning conflict/network game theory, auction theory, and multi-agent systems applied to blockchains and supply chains. This is legitimate graduate-level quantitative training (game theory, optimization over strategic interactions) adjacent to but not squarely inside the rubric's core (linear algebra/matrix methods/statistical learning); his multi-agent-systems papers ('Autonomous Economic Agents as a Second Layer Technology for Blockchains,' 'Implementation of Autonomous Supply Chains... Multi-Agent Approach') are about coordinating decentralized software agents on blockchains, not about neural sequence models, embeddings, or attention/transformer architectures, so transformers_lm and vector_embeddings credit stays low. He is the co-founder/CEO of Valory, which built the Open Autonomy framework and Olas protocol for on-chain autonomous agents — real, personally-led engineering of a production multi-agent system — supporting a moderate hands_on_engineering score, but the systems are agent-orchestration/DLT infrastructure rather than AI models themselves, so industry_impact (measured by core-AI-system leadership, citations, patents) is comparatively modest given the small citation base.\n\nMinarsch's verifiable record begins in 2016 (game theory, auction theory, multi-agent systems) and contains no work in the frontier-model lineage — no pre-2013 distributional-semantics/vector-space/LSI precursors, no attention/transformer/embedding/tokenizer/optimizer/scaling/alignment authorship — so frontier_founder sits at the 'applies frontier models' floor: Olas/Mech Marketplace coordinates and hires AI agents built on others' models but contributes nothing the frontier stack descends from. He has zero verifiable years of hands-on language-modeling research or systems work (his papers are on economic agents on blockchains, supply-chain digital twins, and network conflict — not statistical/neural LMs, seq2seq, or vector-space text models), so lm_domain_depth and lm_domain_breadth are both at the 0-2 'nothing verifiable' band with no LM domains to list. He is, however, a genuine scientific/technical founder: co-founder and CEO of Valory (founded 2021, ~5 years), where he personally authored the Autonomous Economic Agent framework papers and led building the Open Autonomy/Olas stack — one company, ~5 verifiable years as a hands-on technical founder, placing scientific_founder in the 8-12 band (bottom, since the domain is multi-agent DLT orchestration adjacent to AI rather than core language modeling, and co-founder David Galindo carries the cryptography).","evidence":[{"claim":"Semantic Scholar author record (David E. N. Minarsch): 9 papers, 64 citations, h-index 5","source_url":"https://www.semanticscholar.org/author/David-E.-N.-Minarsch/3438886","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Valory team page: David Minarsch is co-founder and CEO of Valory, holds a PhD in Applied Game Theory from the University of Cambridge, and led the team that built 'the first framework for developing MAS in the DLT space'","source_url":"https://valory.xyz","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Valory; holds a PhD in Applied Game Theory from Cambridge University; led the team that built the first framework for developing multi-agent systems in the DLT space; Valory builds Pearl, the Mech Marketplace and the Olas Stack; co-founder David Galindo is the cryptographer CTO","source_url":"https://www.valory.xyz/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Olas coordinates autonomous AI agents that trade, influence and predict on behalf of owners, with the Mech Marketplace as an agent-hiring platform and Pearl as an agent app store; founded 2021","source_url":"https://olas.network/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of 'The Strategy of Conquest', Journal of Economic Theory (2020)","source_url":"https://doi.org/10.1016/j.jet.2020.105161","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Autonomous Economic Agents as a Second Layer Technology for Blockchains: Framework Introduction and Use-Case Demonstration' (CVCBT 2020)","source_url":"https://doi.org/10.1109/cvcbt50464.2020.00007","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Valory (founded 2021); led the team that built the first framework for developing multi-agent systems in the DLT space; co-founder David Galindo is the cryptographer CTO","source_url":"https://www.valory.xyz/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"David Minarsch is co-founder and CEO of Valory, holds a PhD in Applied Game Theory from Cambridge, and led the team that built the first MAS framework in the DLT space; Valory builds Pearl, the Mech Marketplace and the Olas Stack, with co-founder David Galindo as cryptographer CTO","source_url":"https://www.valory.xyz/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD, Applied Game Theory, University of Cambridge"],"canonical_papers":["The strategy of conquest (Journal of Economic Theory, 2020)","Autonomous Economic Agents as a Second Layer Technology for Blockchains (2020)","Implementation of Autonomous Supply Chains for Digital Twinning: a Multi-Agent Approach (2023)","Autonomous Economic Agent Framework (2022)","Dynamic Conflict on a Network (2016)"],"first_verifiable_year":2016,"notable_systems":["Olas / Open Autonomy protocol (multi-agent DLT framework, Valory)"],"citations":62,"h_index":5,"patents":0,"dossier_notes":"OpenAlex match is not ambiguous (candidates:2, name_exact:true) and is corroborated independently by Semantic Scholar (near-identical citation/h-index figures) and by Valory's own team page confirming the Cambridge PhD in Applied Game Theory, so identity confidence is high. Multi-agent systems and game theory are adjacent to AI (used in RL and mechanism design) but this record shows no linear algebra/matrix-methods thesis work, no vector embeddings, and no transformer/attention/LM authorship — scored as strong-but-adjacent graduate training plus real personal engineering leadership of a DLT multi-agent product, not core AI research.","years_language_modeling":0,"years_as_technical_founder":5,"technical_founder_companies":1,"technical_founder_roles":["Valory (Olas) — co-founder & CEO, author of the Autonomous Economic Agent framework and lead of the Open Autonomy stack — 2021-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":8},"confidence":0.72,"duration_ms":55967},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.75,"duration_ms":49023}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"eli-ben-sasson","name":"Eli Ben-Sasson","title":"Co-founder, President & CEO","company":"StarkWare Industries","sector":"crypto","rank":74,"rubric_version":4,"dimensions":{"foundations":14,"vector_embeddings":1,"transformers_lm":1,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":11,"industry_impact":7,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":24,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":24,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.9,"rationale":"The two passes agree entirely on the facts and disagree on how to apply the rubric, so this is adjudicated on rubric interpretation rather than on new evidence. Ben-Sasson is a distinguished theoretical computer scientist: PhD Hebrew University 2001 under Avi Wigderson, postdocs at Harvard and MIT, Technion faculty 2005-2020 rising to Full Professor, author of 'Short proofs are narrow — resolution made simple' (JACM 2001), 'SNARKs for C' (2013), Zerocash (2014) and, as first author, the original STARK paper 'Scalable, transparent, and post-quantum secure computational integrity' (ePrint 2018/046) that introduced FRI. He personally invented and built the proof systems that StarkEx and Starknet run on. The question is what that earns under a rubric that measures depth in the core of AI specifically. Pass 1 scored foundations 19 and industry_impact 17 on the strength of the mathematics and the company; pass 2 scored 12 and 6, reasoning that the rubric names the mathematics of statistical learning and that StarkWare's products are blockchain scaling, not AI. Pass 2 has the better reading, but goes too far in one direction as pass 1 does in the other. The rubric's foundations dimension names linear algebra, matrix methods, optimization and statistical learning; his work is proof complexity, error-correcting codes, algebraic complexity and Reed-Solomon proximity testing — first-principles mathematics of genuine depth in the adjacent-but-different tradition, which warrants a strong score but not the 18-20 'authored canonical work the field builds on' band, because the field in question here is AI and his canonical work is not in it. Industry impact likewise cannot sit at 17: that anchor requires orgs or products whose CORE is these systems, and StarkWare's core is validity rollups. Both lineage dimensions are at the floor: neither pass found, and I could not find, any publication, system or patent by him touching embeddings, attention, pretraining or neural networks.\n\nBen-Sasson's canonical work — proof complexity ('Short proofs are narrow', JACM 2001), zk-SNARKs/SNARKs for C (2013), Zerocash (2014), and the founding STARK/FRI paper (ePrint 2018/046) — is in cryptography, coding theory and computational-integrity proofs, none of which sits in the attention→transformer→language-model lineage or the pre-2013 vector-space/distributional-semantics precursor lineage that word2vec-era embeddings descend from; his frontier_founder position is therefore at the floor. There are zero verifiable years of language-modeling work (statistical/neural LMs, embeddings, seq2seq, transformers) across any of his 166 indexed works, and zero distinct language-modeling domains — the PubMed hits are homonym contamination, not his. He is, however, a genuine scientific/technical founder: he personally invented the STARK/FRI proof system that StarkWare's StarkEx and Starknet run on and co-founded the company in 2018 as its Chief Scientist/CEO — but that company's core is validity rollups for blockchain scaling, not language-modeling or AI systems, so under the anchor this is a deep technical founder OUTSIDE this field (3-7 band). Roughly 8 years founding one such company, in a technical-founder role, but out of scope for the AI/LM field this index measures.","evidence":[{"claim":"PhD theoretical computer science, Hebrew University 2001 under Avi Wigderson; postdocs at Harvard and MIT; Technion faculty 2005-2020 (Full Professor 2015); invented STARKs and FRI in 2018; co-founded StarkWare 2018, CEO from February 2024; no machine-learning or AI work documented in his biography","source_url":"https://en.wikipedia.org/wiki/Eli_Ben-Sasson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Scalable, transparent, and post-quantum secure computational integrity' (IACR ePrint 2018/046) with Bentov, Horesh and Riabzev — the founding STARK/FRI paper","source_url":"https://eprint.iacr.org/2018/046","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records the doctorate from Hebrew University with advisor Avi Wigderson and employers Technion and StarkWare","source_url":"https://www.wikidata.org/wiki/Q102301988","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD 2001, Hebrew University of Jerusalem, advisor Avi Wigderson; Technion faculty 2005 (Senior Lecturer) -> 2010 (Associate Professor) -> 2015 (Full Professor), departed 2020; co-founded StarkWare Industries in 2018 with Uri Kolodny, Michael Riabzev, and Alessandro Chiesa; became CEO/president Feb 2","source_url":"https://en.wikipedia.org/wiki/Eli_Ben-Sasson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of the original zk-STARK paper 'Scalable, transparent, and post-quantum secure computational integrity' (IACR ePrint 2018/046), with Iddo Bentov, Yinon Horesh, and Michael Riabzev -- the founding STARK/FRI protocol paper","source_url":"https://eprint.iacr.org/2018/046","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as Co-Founder and CEO of StarkWare on the company's official team/about page","source_url":"https://starkware.co/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records doctorate from Hebrew University with advisor Avi Wigderson, employers Technion and StarkWare, and Google Scholar ID M93Auk4AAAAJ","source_url":"https://www.wikidata.org/wiki/Q102301988","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of StarkWare, invented STARKs; first author of the founding STARK/FRI paper 'Scalable, transparent, and post-quantum secure computational integrity' (ePrint 2018/046) — cryptographic proof systems, not language modeling","source_url":"https://eprint.iacr.org/2018/046","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikipedia and Wikidata document his work as zero-knowledge proofs, proof complexity and cryptography (PhD under Avi Wigderson); no machine-learning, embedding, or language-modeling work in his record","source_url":"https://en.wikipedia.org/wiki/Eli_Ben-Sasson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"StarkWare's product core is validity/ZK rollups for blockchain scaling (StarkEx, Starknet) — the company Ben-Sasson technically founded","source_url":"https://starkware.co/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of StarkWare (2018); founding scientist of Zcash; first author of the founding STARK/FRI paper — technical founder who authored the company's core cryptographic research, but the field is zero-knowledge proofs / validity rollups, not AI or language modeling","source_url":"https://en.wikipedia.org/wiki/Eli_Ben-Sasson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Original zk-STARK paper 'Scalable, transparent, and post-quantum secure computational integrity' (IACR ePrint 2018/046) — proof-system cryptography, not any embeddings/attention/transformer/LM lineage work","source_url":"https://eprint.iacr.org/2018/046","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as Co-Founder and CEO on StarkWare's official team page; company core is validity rollups for blockchain scaling","source_url":"https://starkware.co/about-us/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Theoretical Computer Science, Hebrew University of Jerusalem, 2001 (advisor Avi Wigderson)","Postdoctoral research, Harvard University and MIT"],"canonical_papers":["Short proofs are narrow — resolution made simple (JACM, 2001)","SNARKs for C: Verifying Program Executions Succinctly and in Zero Knowledge (2013)","Zerocash: Decentralized Anonymous Payments from Bitcoin (2014)","Scalable, transparent, and post-quantum secure computational integrity (2018) — STARKs/FRI","Aurora: Transparent Succinct Arguments for R1CS (2019)"],"first_verifiable_year":1999,"notable_systems":["STARK proof system and FRI protocol","StarkEx / Starknet validity rollups","Zcash (founding scientist)"],"citations":8751,"h_index":41,"patents":0,"dossier_notes":"Both passes independently identified the PubMed block as homonym contamination despite homonym_risk=false: its items (a 1987 amiodarone-thrombocytopenia case report, a 1993 chromosome-deletion paper, developmental-milestone papers with A. Ben-Sasson) are not his. That error propagates into the timeline's claim of earliest year 1987 / 39 years active; his real record begins about 1999-2001.","adjudication_notes":["foundations (19 vs 12): pass 2's reading is right but its score is low; settled at 14. The verified fact both passes share is that his entire indexed corpus is proof complexity, coding theory and cryptography with no machine-learning topic. The rubric's 18-20 anchor is 'authored canonical work the field builds on' — he did, but not in this field, so the canonical-AI band is unavailable to him; the genuine first-principles mathematical depth (PhD under Wigderson, JACM, algebraic proof systems) still places him well into the PhD-level band.","industry_impact (17 vs 6): pass 2 was right; settled at 7. The anchor requires organizations or products whose CORE is these systems. StarkWare's core is validity rollups for blockchain scaling — verified from his own biography and the STARK paper — not AI systems. Pass 1's 17 credits a real but out-of-scope company.","hands_on_engineering (14 vs 9): split at 11. He genuinely designed and built the STARK/FRI stack that Starknet runs on, which pass 2 under-credits, but the rubric's dimension is about AI systems, models, or the hardware/infra under them, which this is not.","vector_embeddings (1 vs 2) and transformers_lm (0 vs 2): immaterial; both at 1. Neither pass could cite any ML artifact and neither could I."],"years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":1,"technical_founder_roles":["StarkWare Industries — co-founder & Chief Scientist/CEO (invented STARK/FRI, the core tech) — 2018-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.9,"duration_ms":53416},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.9,"duration_ms":50549}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Eli_Ben-Sasson","wikipedia_title":"Eli Ben-Sasson","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"juan-benet","name":"Juan Benet","title":"Founder & CEO","company":"Protocol Labs (IPFS, Filecoin)","sector":"crypto","rank":75,"rubric_version":4,"dimensions":{"foundations":7,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":14,"industry_impact":10,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":24,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":24,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.74,"rationale":"Juan Benet earned a BS and MS in Computer Science from Stanford University (2010, 2012), then founded Protocol Labs in 2014 and personally authored the IPFS whitepaper ('IPFS - Content Addressed, Versioned, P2P File System', arXiv:1407.3561), combining distributed-hash-table, content-addressing, and Merkle-DAG ideas — genuine, hands-on distributed-systems engineering and a real authored technical paper, plus he went on to build Filecoin. This demonstrates strong systems-engineering foundations (algorithms, cryptographic hashing, distributed protocols) and personal hands-on building of production infrastructure used at scale. However, none of this work is in the core-AI lineage the rubric measures — no linear algebra/optimization/statistical-learning research, no vector-embeddings or representation-learning work, and no transformer/language-model research or engineering. IPFS/Filecoin are content-addressed storage/distributed-systems protocols, not AI systems; any current 'AI' framing at Protocol Labs is business positioning rather than Benet's personal authored research. Scored with real hands-on-engineering credit for the IPFS whitepaper and system-building but very low on the AI-specific dimensions per rubric instructions to not reward adjacent-but-not-core technical work as if it were core AI depth.\n\nBenet's verifiable technical record is distributed-systems / content-addressed storage — the IPFS whitepaper (arXiv:1407.3561, 2014), Filecoin, libp2p and IPLD — none of which sits in the attention→transformer→language-model lineage or the pre-2013 vector-space / distributional-semantics precursor lineage that word2vec-era embeddings descend from, so frontier_founder is near-zero (a storage/CDN substrate is not a block frontier LMs are built on, and the OpenAlex/PubMed hits are a Spanish-novelist homonym, not him). He has zero verifiable hands-on language-modeling years (statistical/neural LMs, LSI, seq2seq, transformers) and zero distinct language-modeling domains — biological, financial, code, etc. — so lm_domain_depth and lm_domain_breadth are floored. He IS a genuine scientific/technical founder — he personally authored the core protocol papers and code Protocol Labs (founded May 2014, ~12 years) runs on, and Filecoin — but that founder work is OUTSIDE the language-modeling field the rubric scores, which caps scientific_founder in the 3-7 'technical founder outside this field' band. Any 'AI' framing at Protocol Labs (BitRobot, hosting Prime Intellect training runs) is org/business positioning, not Benet's personal authored language-modeling research.","evidence":[{"claim":"Founded Protocol Labs in May 2014; author of the IPFS whitepaper 'IPFS - Content Addressed, Versioned, P2P File System' (arXiv:1407.3561, 2014)","source_url":"https://arxiv.org/abs/1407.3561","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'IPFS - Content Addressed, Versioned, P2P File System', arXiv:1407.3561, sole author Juan Benet, submitted 14 July 2014; describes content-addressed block storage with a Merkle DAG, a distributed hashtable and a self-certifying namespace","source_url":"https://arxiv.org/abs/1407.3561","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub @jbenet, Protocol Labs: creator and lead developer of IPFS (23.1k stars), Kubo the Go IPFS implementation (17.1k stars), js-ipfs (7.4k stars) and go-libp2p (6.9k stars); 286 repositories","source_url":"https://github.com/jbenet","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Protocol Labs was founded in May 2014 by Juan Benet, participated in Y Combinator S14; it built IPFS, Filecoin (mainnet October 2020), libp2p, Multiformats and IPLD; its network also includes AI efforts such as BitRobot Network (2025) and Prime Intellect's INTELLECT-2 decentralized 32B RL training r","source_url":"https://pl.xyz/about/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"IPFS was created by Juan Benet, who later founded Protocol Labs in May 2014; alpha released February 2015","source_url":"https://en.wikipedia.org/wiki/InterPlanetary_File_System","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Protocol Labs and Juan Benet are listed as the original authors of Filecoin","source_url":"https://en.wikipedia.org/wiki/Filecoin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Benet is sole author of the IPFS whitepaper 'IPFS - Content Addressed, Versioned, P2P File System' (2014) — a content-addressed distributed file system, not an AI/LM system","source_url":"https://arxiv.org/abs/1407.3561","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Benet founded Protocol Labs in May 2014 and is listed as original author of Filecoin — a distributed storage protocol","source_url":"https://en.wikipedia.org/wiki/Filecoin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub @jbenet: creator/lead developer of IPFS, Kubo, js-ipfs and go-libp2p — distributed-systems engineering, no language-modeling repositories","source_url":"https://github.com/jbenet","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sole author of 'IPFS - Content Addressed, Versioned, P2P File System' (arXiv:1407.3561, 2014) — a distributed-storage protocol, not AI/LM work","source_url":"https://arxiv.org/abs/1407.3561","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Protocol Labs in May 2014 and created/led IPFS, Filecoin, libp2p as technical founder","source_url":"https://en.wikipedia.org/wiki/InterPlanetary_File_System","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Benet and Protocol Labs are the original authors of Filecoin (mainnet Oct 2020)","source_url":"https://en.wikipedia.org/wiki/Filecoin","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, Stanford University (2010)","MS Computer Science, Stanford University (2012)"],"canonical_papers":["IPFS - Content Addressed, Versioned, P2P File System (2014, arXiv:1407.3561)"],"first_verifiable_year":2014,"notable_systems":["IPFS (InterPlanetary File System)","Filecoin","Protocol Labs research org"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"CRITICAL: the dossier's Wikipedia/Wikidata AND OpenAlex matches are ALL the wrong person — a Spanish novelist/dramatist Juan Benet (1927-1993), completely unrelated. The OpenAlex 'top works' (Cuentos completos, Volverás a Región, Spanish literary criticism) belong to the novelist, not the Protocol Labs founder. The real Juan Benet has no discoverable OpenAlex/Semantic Scholar/PubMed academic record — his sole verifiable technical output is the self-published IPFS whitepaper (posted to arXiv but not peer-reviewed) plus the Filecoin protocol design. This is real, hands-on distributed-systems engineering but is orthogonal to the AI/ML lineage the rubric scores; treat all dossier bibliometric fields (citations, h-index, timeline) as invalid/homonym-contaminated and disregarded here.","years_language_modeling":0,"years_as_technical_founder":12,"technical_founder_companies":1,"technical_founder_roles":["Protocol Labs — founder / technical founder (authored IPFS & Filecoin core protocol) — 2014-2026 (outside language modeling)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.86,"duration_ms":38864},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.86,"duration_ms":45004}],"source":"seeded","profile_url":null,"wikipedia_title":"Juan Benet","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"gavin-wood","name":"Gavin Wood","title":"Founder","company":"Polkadot, Parity Technologies","sector":"crypto","rank":76,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":4,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":11,"industry_impact":5,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":23,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":23,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.76,"rationale":"Gavin Wood holds an MEng in Computer Systems and Software Engineering (2002) and a PhD from the University of York (2005), thesis 'Content-based visualization to aid common navigation of musical audio' — genuine graduate-level technical training, but in music-information-retrieval/signal-processing visualization, not the linear-algebra/optimization/statistical-learning core the rubric weights, and not in vector embeddings or the attention/transformer lineage. His major verified technical achievements are in blockchain systems engineering: he wrote most of the original Ethereum client code, authored the Ethereum Yellow Paper formally specifying the EVM, invented the Solidity language, and founded Polkadot/Kusama — substantial hands-on systems-engineering and protocol-design work, but this is cryptographic/distributed-systems engineering, not AI/ML research or infrastructure. No evidence was found of any authored work, patents, or shipped systems touching vector embeddings, transformers, or language modeling. hands_on_engineering is scored solidly for his genuine protocol/systems-building record; the AI-specific dimensions are scored near floor.\n\nWood's entire verifiable technical record — the Ethereum Yellow Paper/EVM, the Solidity language, Parity, Polkadot and Kusama — is cryptographic distributed-systems and protocol engineering, none of which sits anywhere in the attention→transformer→language-model lineage or the pre-2013 vector-space/distributional-semantics precursor lineage that word2vec-era embeddings descend from; frontier models build on none of his work, so frontier_founder is near floor. He has zero verifiable years and zero papers, patents or shipped systems in language modeling of any kind (his 2005 PhD was content-based *visualization* of musical audio — MIR/signal-processing, not a sequence/language model), so lm_domain_depth and lm_domain_breadth are at floor with no qualifying domains. He is, by contrast, a genuine hands-on scientific/technical founder — he personally wrote the core code and specifications his companies run on across ~13 years (Ethereum 2013–16, then Parity Technologies and the Web3 Foundation/Polkadot/Kusama, 2015–present) — but that is a technical-founder record entirely OUTSIDE the language-modeling field, which the rubric caps in the 3–7 band. Scored at the top of that band (7) to reflect multiple companies and a decade-plus of real technical founding, discounted because none of it is in this field.","evidence":[{"claim":"MEng Computer Systems and Software Engineering (2002) and PhD (2005), University of York, thesis on content-based visualization for musical audio navigation","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wrote most of the code for the first version of Ethereum, served as CTO of the Ethereum Foundation, authored the Ethereum Yellow Paper defining the EVM, credited with inventing Solidity","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Ethereum: A Secure Decentralised Generalised Transaction Ledger' co-authored work, 5309 citations per OpenAlex, is blockchain protocol design, not AI research","source_url":"https://doi.org/10.48550/arxiv.2005.13456","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MEng in Computer Systems and Software Engineering, University of York (2002); PhD 2005, thesis 'Content-based visualization to aid common navigation of musical audio'; research scientist at Microsoft; authored the Ethereum Yellow Paper specifying the EVM; proposed Solidity; first CTO of the Ethereum","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ethereum: A Secure Decentralised Generalised Transaction Ledger (the Yellow Paper) is credited to Gavin Wood with ~5,309 citations in OpenAlex","source_url":"https://ethereum.github.io/yellowpaper/paper.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Overview of Polkadot and its Design Considerations (arXiv 2005.13456) — Wood co-author","source_url":"https://arxiv.org/abs/2005.13456","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q43379767 records a PhD, education at University of York, and occupations computer scientist, researcher, software developer; notable works Ethereum and Polkadot","source_url":"https://www.wikidata.org/wiki/Q43379767","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wood wrote most of the first Ethereum client, authored the Ethereum Yellow Paper specifying the EVM, invented Solidity, and founded Parity Technologies and Polkadot/Kusama — all blockchain/distributed-systems work, none in AI or language modeling","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His OpenAlex/verified corpus contains only blockchain protocol papers (Ethereum Yellow Paper, Polkadot design overview); no embeddings, transformer, or language-modeling work exists in the record","source_url":"https://arxiv.org/abs/2005.13456","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD (University of York, 2005) thesis was 'Content-based visualization to aid common navigation of musical audio' — music-information-retrieval/signal processing, not a language or sequence model","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wood authored the Ethereum Yellow Paper formally specifying the EVM, wrote most of the first Ethereum client, and created the Solidity language — blockchain/distributed-systems engineering, not AI/LM lineage work","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wood co-founded Ethereum, founded Parity Technologies and the Web3 Foundation, and created Polkadot and Kusama, personally driving the technical/scientific direction — a technical-founder record, but in blockchain rather than AI/language modeling","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His 2005 University of York PhD thesis, 'Content-based visualization to aid common navigation of musical audio,' is signal-processing/music-information-retrieval visualization, not language or sequence modeling","source_url":"https://en.wikipedia.org/wiki/Gavin_Wood","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["MEng Computer Systems and Software Engineering, University of York (2002)","PhD, University of York (2005), thesis on content-based musical-audio visualization"],"canonical_papers":["Ethereum: A Secure Decentralised Generalised Transaction Ledger (Yellow Paper)","Overview of Polkadot and its Design Considerations (2020)"],"first_verifiable_year":2011,"notable_systems":["Ethereum (original client + EVM spec)","Solidity language","Polkadot","Kusama"],"citations":5339,"h_index":3,"patents":0,"dossier_notes":"Per the task brief's own guidance, this dossier's OpenAlex match is doubtful: works include 'Guidance for health care worker surveys in humanitarian contexts in LMICs' and topics like 'Disaster Response and Management' alongside genuinely his Ethereum/Polkadot papers, suggesting a mixed/homonym-polluted record; only the Ethereum and Polkadot papers were counted as verified. The dossier's PubMed refined sample (79 rows, nephrology/oncology/cardiology papers from UK-affiliated 'G. Wood') is entirely homonym noise from medical researchers and was excluded from scoring. Semantic Scholar match (192 papers, h-index 28, 8 candidates, ambiguous) was not used given the ambiguity flag; no independent verification of his true citation/h-index for CS work was completed this pass beyond the Ethereum paper itself.","years_language_modeling":0,"years_as_technical_founder":13,"technical_founder_companies":2,"technical_founder_roles":["Ethereum — co-founder & first CTO (author of Yellow Paper/Solidity) — 2013-2016 (outside LM field)","Parity Technologies / Web3 Foundation (Polkadot, Kusama) — founder & lead architect — 2015-present (outside LM field)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.85,"duration_ms":47489},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":7},"confidence":0.85,"duration_ms":50360}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Gavin_Wood","wikipedia_title":"Gavin Wood","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ari-juels","name":"Ari Juels","title":"Weill Family Foundation Professor, Cornell Tech; Co-director, IC3; Chief Scientist, Chainlink Labs","company":"Chainlink Labs","sector":"crypto","rank":77,"rubric_version":4,"dimensions":{"foundations":11,"vector_embeddings":2,"transformers_lm":3,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":7,"industry_impact":10,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":22,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":22,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.88,"rationale":"Juels is a distinguished researcher whose entire canonical corpus is applied cryptography and security, not the AI lineage this index measures. His 1996 UC Berkeley PhD was advised by Alistair Sinclair, a randomized-algorithms and Markov-chain theorist, and his canonical work — the fuzzy commitment scheme (1999), the fuzzy vault (2003), Proofs of Retrievability (CCS 2007), and later Town Crier/DECO/MEV — rests on real probability, coding theory and information-theoretic construction. That is genuine mathematical rigor and earns foundations at the top of the strong-graduate-training band, but it is not linear algebra, optimization or statistical learning for neural networks, so it does not reach the PhD-level-in-the-core anchor. I verified his single substantive machine-learning publication directly: 'Stealing Machine Learning Models via Prediction APIs' (Tramer, Zhang, Juels, Reiter, Ristenpart, USENIX Security 2016, arXiv:1609.02943), which extracts logistic-regression, neural-network and decision-tree models through black-box prediction APIs — an important security result about ML systems, not a contribution to embeddings, attention, pretraining or scaling. His own site lists his interests as blockchain technologies, AI security, applied cryptography and privacy, with no ML, embedding or language-model work; no authored work in the lineage exists. Industry impact is real but earned in security and oracle organizations (RSA Labs Chief Scientist, IC3 co-founder, Chainlink Labs), whose core is cryptography and distributed systems rather than language modeling, so it is credited on adjacency, not on AI systems the field runs on.\n\nJuels has no verifiable position in the language-modeling lineage that today's frontier models descend from: his canonical corpus is applied cryptography (fuzzy commitment/vault, Proofs of Retrievability), RFID security, and blockchain oracles, with no pre-2013 vector-space / distributional-semantics / LSI / term-relationship work and no attention/transformer/embedding/pretraining/alignment contribution — his one ML-adjacent paper ('Stealing Machine Learning Models via Prediction APIs', USENIX 2016) is an adversarial-ML security result, not foundational LM work, so frontier_founder is a 2. He has zero verifiable hands-on years in language modeling of any kind (natural-language, biological, financial, code, etc.), so lm_domain_depth and lm_domain_breadth are both 1. He is, however, a genuine scientific/technical founder — co-founder of Town Crier (acquired by Chainlink), co-director of IC3, and co-author of the 2017 Chainlink white paper as Chief Scientist of Chainlink Labs — but every one of those companies' core is cryptography, oracles and distributed systems, NOT language modeling, which places scientific_founder in the 3-7 'technical founder outside this field' band at 5. Roughly 8 years (2017–present) operating as a technical founder, across one to two such companies, all outside the AI/LM field.","evidence":[{"claim":"PhD in computer science, UC Berkeley 1996, doctoral advisor Alistair Sinclair; Weill Professor at Cornell Tech and co-director of IC3","source_url":"https://www.wikidata.org/wiki/Q102320479","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'Stealing Machine Learning Models via Prediction APIs' (Tramer, Zhang, Juels, Reiter, Ristenpart, USENIX Security 2016) — extraction attacks against logistic regression, neural networks and decision trees via black-box APIs; a security result about ML systems","source_url":"https://arxiv.org/abs/1609.02943","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His own faculty page states his research areas are blockchain technologies, AI security, applied cryptography and privacy — no machine-learning, embedding or language-model research is listed","source_url":"https://www.arijuels.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cornell Tech biography lists blockchains, cryptocurrency, smart contracts, applied cryptography and user authentication as his research areas","source_url":"https://www.tech.cornell.edu/people/ari-juels/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD 1996, UC Berkeley, advisor Alistair Sinclair; currently Weill Family Foundation Professor at Cornell Tech and co-director of IC3","source_url":"https://en.wikipedia.org/wiki/Ari_Juels","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cornell Tech faculty bio: research areas are blockchains, cryptocurrency, smart contracts, applied cryptography, user authentication, and privacy; previously Chief Scientist and Director of RSA Laboratories","source_url":"https://www.tech.cornell.edu/people/ari-juels/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD UC Berkeley 1996; 17 years at RSA Security, Chief Scientist from 2007; Weill Professor at Cornell Tech; co-director of IC3; Chief Scientist at Chainlink Labs and co-author of the 2017 Chainlink white paper; canonical work includes fuzzy commitment, fuzzy vault, PORs, client puzzles, proof-of-wor","source_url":"https://en.wikipedia.org/wiki/Ari_Juels","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder/co-director of IC3, Chief Scientist at Chainlink Labs and co-author of the 2017 Chainlink white paper; previously Chief Scientist of RSA Laboratories — technical leadership of cryptography/oracle organizations, not language-modeling labs","source_url":"https://en.wikipedia.org/wiki/Ari_Juels","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Town Crier, an authenticated-data-feed oracle system Juels co-built at Cornell, became a startup acquired by Chainlink — a crypto/oracle system, not an AI or language-modeling company","source_url":"https://www.tech.cornell.edu/people/ari-juels/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His only ML-related publication, 'Stealing Machine Learning Models via Prediction APIs' (2016), is a black-box model-extraction security result, not authorship of embeddings, attention, transformers, pretraining or scaling","source_url":"https://arxiv.org/abs/1609.02943","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'Stealing Machine Learning Models via Prediction APIs' (Tramer, Zhang, Juels, Reiter, Ristenpart, USENIX Security 2016) is his only substantive ML paper — a security/extraction result, not embedding, attention or LM authorship","source_url":"https://arxiv.org/abs/1609.02943","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"arijuels.com and Cornell Tech list his research areas as blockchains, cryptocurrency, smart contracts, applied cryptography, user authentication and privacy — no language-modeling, embedding or vector-space work","source_url":"https://www.tech.cornell.edu/people/ari-juels/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-director/co-founder of IC3 (academic initiative) and Chief Scientist and co-author of the 2017 Chainlink white paper — technical leadership of crypto/oracle organizations, not an LM company","source_url":"https://en.wikipedia.org/wiki/Ari_Juels","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science, University of California, Berkeley (1996), advisor Alistair Sinclair"],"canonical_papers":["A Fuzzy Commitment Scheme (CCS 1999)","A Fuzzy Vault Scheme (2003)","PORs: Proofs of Retrievability for Large Files (CCS 2007)","RFID security and privacy: a research survey (2006)","Stealing Machine Learning Models via Prediction APIs (USENIX Security 2016)","On Scaling Decentralized Blockchains (2016)"],"first_verifiable_year":1996,"notable_systems":["Proofs of Retrievability","Town Crier oracle","DECO","IC3","Chainlink (Chief Scientist)"],"citations":27079,"h_index":73,"patents":0,"dossier_notes":"OpenAlex match A5029249775 is flagged ambiguous (10 candidates) and its affiliation list (Beth Israel Deaconess Medical Center, NBER, Wisconsin-Madison, Johns Hopkins, Peking University) mixes in homonyms and co-author institutions; only the Cornell affiliation is his. The listed top works are genuinely his, so the citation and h-index totals are plausible but should be read as an upper bound. Semantic Scholar lookup failed entirely. The PubMed VeriChip-cloning result (2006) is genuinely his and is RFID security work, not biomedical ML. Patents were skipped by the collector, not verified as zero.","adjudication_notes":["foundations — SPLIT, final 11 between pass_1's 10 and pass_2's 14. Pass_2 is right that a Berkeley PhD under Alistair Sinclair plus information-theoretic constructions (fuzzy commitment/vault, PORs) is genuine graduate-level mathematical rigor, so pass_1's 10 slightly understates it; but pass_2's 14 reaches the 'PhD-level work in the core' anchor, and the verified record contains no linear-algebra, optimization or statistical-learning work — the rigor is in probability, coding theory and cryptography. 11 holds the strong-graduate-training band without crediting core-AI training he does not have.","vector_embeddings — pass_1 essentially right (1 vs pass_2's 3); final 2. Neither pass cited any embedding, retrieval or vector-search work, and arijuels.com lists none. Fuzzy vaults operate on unordered point sets for biometrics, which is not representation learning; a 2 reflects only incidental adjacency, not authored work.","transformers_lm — pass_1 right on kind, final 3. Both passes cited the same single item and I verified it directly: 'Stealing Machine Learning Models via Prediction APIs' (arXiv:1609.02943) attacks logistic regression, neural networks and decision trees through prediction APIs. It is an adversarial-ML security paper, not authorship of seq2seq, attention, pretraining or scaling work, so pass_2's 4 is marginally generous and pass_1's 2 marginally stingy.","hands_on_engineering — pass_1 closer, final 7 vs pass_2's 10. Town Crier and DECO are real systems he co-built, which justifies moving above pass_1's 6, but they are oracle/cryptography systems; there is no verifiable evidence he built models, training infrastructure, or embedding/retrieval systems, so the 8-12 'senior engineering adjacent to the core' band is not earned on AI grounds.","industry_impact — pass_1 right, final 10 vs pass_2's 9. The dispute is small; IC3 co-direction, RSA Labs Chief Scientist and Chainlink Labs Chief Scientist are genuine technical leadership of research organizations, but their core is cryptography and distributed systems, not language modeling, so this is adjacency credit rather than leadership of a lab that produced canonical AI work."],"years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":1,"technical_founder_roles":["Chainlink Labs — Chief Scientist / white-paper co-author — 2017-present (crypto/oracle, outside LM)","Town Crier — co-founder (oracle system, acquired by Chainlink) — outside LM","IC3 — co-founder/co-director — cryptography research initiative, outside LM"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.83,"duration_ms":51609},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.86,"duration_ms":49744}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Ari_Juels","wikipedia_title":"Ari Juels","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"alex-blania","name":"Alex Blania","title":"Co-founder & CEO","company":"Tools for Humanity / World (Worldcoin)","sector":"crypto","rank":78,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":8,"industry_impact":6,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":21,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":21,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.75,"rationale":"Blania holds a dual bachelor's in physics and industrial engineering from the University of Erlangen-Nuremberg (FAU) and pursued a physics master's, doing thesis research at the Max Planck Institute for the Science of Light and finishing at Caltech's Institute for Quantum Information and Matter. He co-authored one real, verifiable paper — 'Deep learning of spatial densities in inhomogeneous correlated quantum systems' (arXiv 2211.09050, 2022) with Sandro Herbig, Fabian Dechent, Evert van Nieuwenburg, and Florian Marquardt (a well-known quantum-machine-learning group leader) — which applies convolutional neural networks to predict physical observables in quantum many-body systems; this is genuine hands-on deep-learning work and gives some graduate-level statistical-learning/optimization credibility, but it is a single co-authored paper (likely thesis-derived, 3 citations) in computational physics, not language modeling, embeddings, or transformer research. He has no found publication, patent, or system in vector embeddings or transformer/LM research, consistent with his training being physics/quantum-computing rather than NLP. His hands-on engineering record is strongest as a builder of the Worldcoin/World Orb biometric hardware and World ID/World App systems since 2020 — real, personally-led systems engineering, but the core technology (iris-biometric hardware, blockchain identity) is not core language-modeling/transformer work, so industry_impact is scored on the basis of building real, complex technical systems rather than on funding raised or market cap.\n\nBlania has NO verifiable position in the language-modeling lineage that today's frontier models descend from: his single co-authored paper (arXiv:2211.09050, 2022) applies CNNs to predict densities in quantum many-body systems — computational physics, not distributional semantics, embeddings, attention, transformers, tokenizers, optimizers, scaling or alignment — so frontier_founder is at the floor. He has zero verifiable years in language modeling of any kind (natural-language, biological, financial, code, etc.); the quantum-physics deep-learning work is not language modeling, so lm_domain_depth and lm_domain_breadth score in the 0-2 band with no qualifying domains. He IS a genuine technical/scientific founder-CEO — a physics/quantum-ML master's-level researcher who left Caltech's IQIM in 2019 to co-found World/Tools for Humanity (~5-6 years) and personally drove the technical direction of the Orb iris-biometric hardware, plus co-founded Merge Labs (2025) — but that founding work is biometric hardware and blockchain identity, a technical field OUTSIDE core language modeling, which places scientific_founder in the 3-7 'technical founder outside this field' band. Counting World/Tools for Humanity as one clearly technical-founder company (~5-6 years) and Merge Labs as a very recent second, none of it is the language-modeling/frontier-AI core this rubric measures.","evidence":[{"claim":"Blania holds degrees in physics and industrial engineering from the University of Erlangen-Nuremberg (FAU) and pursued a physics master's, doing thesis work at Max Planck Institute for the Science of Light and Caltech's Institute for Quantum Information and Matter","source_url":"https://www.fau.eu/2024/11/news/fau-alumni-mystory-alex-blania-ceo-tools-for-humanity/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-authored 'Deep learning of spatial densities in inhomogeneous correlated quantum systems' (arXiv:2211.09050, 2022) with Sandro Herbig, Fabian Dechent, Evert van Nieuwenburg, and Florian Marquardt, applying CNNs to predict observables in correlated quantum many-body systems","source_url":"https://arxiv.org/abs/2211.09050","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Is CEO of Tools for Humanity and co-founder/CEO of World (Worldcoin), which built the Orb iris-biometric verification hardware (development began 2020, manufacturing established in Erlangen, field testing by 2021), World ID, and World App","source_url":"https://en.wikipedia.org/wiki/Alex_Blania","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Tools for Humanity has raised over $300 million at approximately a $3 billion valuation from established investors (a16z, Blockchain Capital, others), not family/friends money","source_url":"https://en.wikipedia.org/wiki/Alex_Blania","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dual bachelor's in physics and industrial engineering from University of Erlangen-Nuremberg; master's research in quantum computing and AI at Caltech's Institute for Quantum Information and Matter; left Caltech in October 2019 after Sam Altman and Max Novendstern approached him, to found World and T","source_url":"https://en.wikipedia.org/wiki/Alex_Blania","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"First author of 'Deep learning of spatial densities in inhomogeneous correlated quantum systems' (arXiv:2211.09050, submitted 16 November 2022) with Sandro Herbig, Fabian Dechent, Evert van Nieuwenburg and Florian Marquardt — CNNs trained on random potentials to predict densities in 1D and 2D lattic","source_url":"https://arxiv.org/abs/2211.09050","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q135216343 records him as a German entrepreneur born 1993, occupation computer scientist, with no academic identifiers (no ORCID, no Google Scholar id).","source_url":"https://www.wikidata.org/wiki/Q135216343","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Blania left Caltech's Institute for Quantum Information and Matter in 2019 to co-found World and Tools for Humanity, where he is CEO and set the technical direction of the Orb iris-biometric verification hardware, World ID, and World App; also co-founded Merge Labs (2025), a brain-computer-interface","source_url":"https://en.wikipedia.org/wiki/Alex_Blania","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His only verifiable publication, 'Deep learning of spatial densities in inhomogeneous correlated quantum systems' (arXiv:2211.09050, 2022), applies convolutional neural networks to quantum many-body physics — not language modeling, embeddings, or transformers","source_url":"https://arxiv.org/abs/2211.09050","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dual bachelor's in physics and industrial engineering from FAU Erlangen-Nuremberg and physics master's research at Max Planck Institute for the Science of Light / Caltech, establishing a physics/quantum-ML rather than NLP background","source_url":"https://www.fau.eu/2024/11/news/fau-alumni-mystory-alex-blania-ceo-tools-for-humanity/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Blania's only verifiable publication is a 2022 CNN-based quantum many-body physics paper, with no transformer/LM, embedding, or pre-2013 vector-space lineage work","source_url":"https://arxiv.org/abs/2211.09050","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Blania co-founded Tools for Humanity and World (Worldcoin) after leaving Caltech in October 2019, serving as CEO and driving the Orb biometric hardware and World ID/World App — a technical founder role in biometrics/blockchain identity, not language modeling","source_url":"https://en.wikipedia.org/wiki/Alex_Blania","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records Blania as a German entrepreneur (born 1993), occupation computer scientist, with no academic identifiers (no ORCID/Scholar), consistent with no language-modeling research record","source_url":"https://www.wikidata.org/wiki/Q135216343","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Dual Bachelor's, Physics and Industrial Engineering, University of Erlangen-Nuremberg (FAU)","Physics Master's research, Max Planck Institute for the Science of Light / Caltech IQIM (deep learning applied to quantum many-body systems)"],"canonical_papers":[],"first_verifiable_year":2022,"notable_systems":["Worldcoin/World Orb biometric verification hardware","World ID decentralized identity protocol","World App self-custodial wallet"],"citations":3,"h_index":1,"patents":0,"dossier_notes":"Dossier's OpenAlex/Semantic Scholar match (1 work, arXiv 2211.09050) is confirmed correct on independent verification — the paper's co-author Florian Marquardt is a real, established quantum-ML physicist at Max Planck Institute for the Science of Light, consistent with Blania's stated master's thesis affiliation, and the PubMed search correctly returned zero results (no homonym confusion there). No evidence found of any personal transformer/LM or vector-embedding research or publications, consistent with the operator's framing that his core technical background is physics/quantum computing and biometric hardware, not core language-modeling research — scored transformers_lm and vector_embeddings at the floor (1) as directed. No credible source found describing Tools for Humanity/Worldcoin as funded by family or inherited wealth with no technical record; funding is from named institutional VCs, so capital_without_competence is not applied per brief's guidance to score 0 absent a citable source.","years_language_modeling":0,"years_as_technical_founder":6,"technical_founder_companies":2,"technical_founder_roles":["Tools for Humanity / World (Worldcoin) — co-founder & CEO, drove Orb biometric-hardware and World ID technical direction — 2019/2020–present","Merge Labs — co-founder (brain-computer interface) — 2025–present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":6},"confidence":0.82,"duration_ms":59536},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.82,"duration_ms":59242}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Alex_Blania","wikipedia_title":"Alex Blania","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"mark-zuckerberg","name":"Mark Zuckerberg","title":"Founder, Chairman & CEO","company":"Meta Platforms","sector":"general","rank":79,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":3,"transformers_lm":4,"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":6,"industry_impact":12,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":21,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":21,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.81,"rationale":"Zuckerberg attended Harvard studying psychology and computer science but dropped out in his second year (2004) without completing a degree, and has no graduate training, thesis, or authored research in linear algebra, optimization, statistical learning, embeddings, or the attention/transformer lineage. His youth programming (ZuckNet, the Synapse Media Player using a basic recommendation heuristic, ~2002) shows early general coding aptitude but predates and is unrelated to the vector-space/embeddings/transformer research lineage the rubric asks about, and is not itself ML research. He is listed as a co-founder of FAIR (Facebook AI Research, 2013) alongside Yann LeCun, Rob Fergus, and Serkan Piantino, but every source describes his role as organizational/funding leadership — LeCun directed FAIR's actual research program, and Zuckerberg is not credited as a contributor to FAIR's canonical outputs (fastText, PyTorch, Llama). His OpenAlex entry (1,219 works, h-index 15) is a clear name-collision aggregate — the top-cited 'works' are business/patent-law/operations-research papers (SIAM Journal on Optimization, LP relaxation) by other people named Mark Zuckerberg, not this individual, and Semantic Scholar's clean 2-paper/0-citation match is the more trustworthy proxy for his actual personal authorship record, which is essentially nil. Per the rubric, industry_impact still credits him meaningfully for having founded and funded Meta, whose org (under researchers he hired and funded, not personally led technically) produced PyTorch and the Llama model family, which the field runs on — but foundations/embeddings/transformers_lm dimensions stay low because he has no personal research record, exactly the rubric's stated case for 'famous CEO with no personal research record scores LOW' on those axes.\n\nZuckerberg has no personal work in the frontier lineage: no pre-2013 vector-space/distributional-semantics papers or patents, and no authored contribution to attention, transformers, embeddings, optimizers, tokenizers, pretraining objectives, scaling laws or alignment — the frontier stack Meta ships (PyTorch, fastText, Llama) is the output of researchers he hired and funded (Yann LeCun directed FAIR), so his own record earns only the 'applies/funds, no foundational contribution' band. He has zero verifiable hands-on years in language modeling specifically — no LM papers, no LM patents, no LM systems he personally built — so years_language_modeling is 0 and lm_domain_depth sits at the bottom band; his youth Synapse Media Player (a music-recommendation heuristic, ~2002) is not language modeling and is a single unrelated hobbyist artifact, not a hands-on modeling domain. He founded Facebook/Meta and did write its original product code, but as a founder-CEO of an AI company whose AI science and engineering are done by others he fits the 3-7 scientific_founder band, not the technical/scientific-founder tier that personally authors the core research, code or patents the company's AI runs on — years_as_technical_founder in this field is 0 and technical_founder_companies is 0.","evidence":[{"claim":"Attended Harvard 2002-2004 studying psychology and computer science; dropped out without completing a degree","source_url":"https://en.wikipedia.org/wiki/Mark_Zuckerberg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder of Facebook AI Research (FAIR) in 2013 alongside Yann LeCun, Rob Fergus, Serkan Piantino; LeCun directed FAIR's research 2013-2018, not Zuckerberg","source_url":"https://en.wikipedia.org/wiki/Meta_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Meta AI/FAIR produced fastText (2016), PyTorch (2017), and Llama (2023) as organizational outputs, attributed to the research org rather than to Zuckerberg personally","source_url":"https://en.wikipedia.org/wiki/Meta_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Studied psychology and computer science at Harvard from 2002, dropped out in his second year; built ZuckNet, Synapse Media Player (machine learning for listening preferences), CourseMatch and Facemash; Wikipedia documents no AI research, papers or technical contributions to ML, Llama or FAIR","source_url":"https://en.wikipedia.org/wiki/Mark_Zuckerberg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The SIAM Journal on Optimization paper 'Subset Algebra Lift Operators for 0-1 Integer Programming' (2004) attributed to a Mark Zuckerberg is by D. Bienstock and M. Zuckerberg (DBLP key journals/siamjo/BienstockZ04), an integer-programming researcher, not the Meta CEO","source_url":"https://api.semanticscholar.org/graph/v1/paper/DOI:10.1137/S1052623402420346?fields=title,year,authors,venue,externalIds","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record for the exact name 'Mark Zuckerberg' shows 2 papers and 0 citations, h-index 0","source_url":"https://api.semanticscholar.org/graph/v1/author/2075634992?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists his occupation as programmer, entrepreneur, computer scientist and CEO, education at Harvard (psychology, computer science) from 2002, employer Meta Platforms from 2004; no doctoral degree or advisor is recorded","source_url":"https://www.wikidata.org/wiki/Q36215","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"FAIR was co-founded in 2013 with Yann LeCun directing the research program; Zuckerberg's role was organizational/funding, and FAIR's canonical outputs (fastText, PyTorch, Llama) are attributed to the research org, not to him personally","source_url":"https://en.wikipedia.org/wiki/Meta_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar's exact-name author record shows 2 papers, 0 citations, h-index 0 — essentially no personal authorship in any research lineage, including language modeling","source_url":"https://api.semanticscholar.org/graph/v1/author/2075634992?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attended Harvard 2002-2004 (psychology and computer science), dropped out with no degree, no thesis or graduate training; Wikipedia documents no AI/ML research or technical contributions to Llama or FAIR","source_url":"https://en.wikipedia.org/wiki/Mark_Zuckerberg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"FAIR (2013) was co-founded organizationally with Yann LeCun directing the research program; Meta AI's fastText/PyTorch/Llama are org outputs, not Zuckerberg's personal authorship","source_url":"https://en.wikipedia.org/wiki/Meta_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record for 'Mark Zuckerberg' shows 2 papers, 0 citations, h-index 0 — no personal language-modeling research record","source_url":"https://api.semanticscholar.org/graph/v1/author/2075634992?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attended Harvard 2002-2004 (psychology and computer science), dropped out without a degree; no graduate training, thesis, patents, or authored ML research","source_url":"https://en.wikipedia.org/wiki/Mark_Zuckerberg","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Harvard College, psychology and computer science, 2002-2004 (no degree completed)"],"canonical_papers":[],"first_verifiable_year":2002,"notable_systems":["Facebook/Meta Platforms (founder, funder)","FAIR / Meta AI (co-founder, non-technical role)","PyTorch (organizational output, not personal authorship)","Llama (organizational output, not personal authorship)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex block (works_count 1219, cited_by 891, h_index 15, affiliations BHP/Melbourne/Columbia/Israel-Meta) is almost certainly a mixed pool of multiple people named Mark Zuckerberg (the SIAM Journal on Optimization / LP-relaxation / integer-programming papers are operations-research work unrelated to this individual). The clean Semantic Scholar match (2 papers, 0 citations, h-index 0) is far more credible and was weighted accordingly. PubMed match (1 result, 2024 Nepal encephalitis metagenomics paper) is also an unrelated homonym and was excluded entirely.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"frontier_lineage":["No personal foundational contribution; Meta org (funded/led non-technically by him) produced PyTorch, fastText and Llama, but authored by hired researchers, not Zuckerberg"],"technical_founder_roles":["Meta Platforms — founder/CEO, product code originally, AI science done by others — 2004-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":4},"confidence":0.82,"duration_ms":46106},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":5},"confidence":0.83,"duration_ms":52406}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Mark_Zuckerberg","wikipedia_title":"Mark Zuckerberg","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"arvind-krishna","name":"Arvind Krishna","title":"Chairman & CEO","company":"IBM","sector":"general","rank":80,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":2,"transformers_lm":3,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":6,"industry_impact":10,"scientific_founder":2},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":20,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":20,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.77,"rationale":"Krishna holds a PhD in electrical engineering from the University of Illinois Urbana-Champaign (1991), with graduate research reported as being in distributed systems and data management — genuine doctoral-level technical training, but not in the AI/ML/transformer lineage this rubric targets, giving modest foundations credit. He joined IBM's Thomas J. Watson Research Center in 1990 and spent ~18 years in research roles (database servers, security software) before moving into executive leadership; no evidence was found of personally authored papers, patents, or code in linear algebra/optimization, embeddings, or transformer/LM research specifically — his later Director of IBM Research and CEO roles (2015-2020-present) are organizational leadership over research divisions (including Watson-era AI), not personal authorship of core-AI work. As CEO, he architected the Red Hat acquisition and has directed IBM's AI/cloud/quantum strategy, which is real industry impact on AI-adjacent business lines, but the rubric explicitly excludes 'manages builders, no personal record' from higher scores, so industry_impact reflects organizational leadership of a major tech company rather than lab leadership that personally produced canonical AI work.\n\nKrishna's verifiable technical record — a UIUC PhD (1991) and ~18 years at IBM's Watson Research Center — is in databases, distributed systems and security software, none of which sits in the vector-space / distributional-semantics / attention-transformer lineage that today's frontier language models descend from; no pre-2013 embeddings/LSI/term-relationship work or later transformer/scaling/alignment contribution is attributable to him (the AI-topic OpenAlex record A5071248780 is a homonym electronic-packaging engineer). He has zero verifiable hands-on years in language modeling of any kind — his LLM connection (watsonx / Granite) is CEO-level organizational stewardship since 2020, not personally authored modeling work — so no distinct LM domain can be counted. He is a career IBM employee and executive, not the founder of any company: he founded zero companies and has zero years operating as a scientific/technical founder. As CEO he applies and directs frontier-class AI at IBM but contributes nothing foundational to the frontier stack himself, placing him in the 'applies frontier models, no foundational contribution' band.","evidence":[{"claim":"PhD Electrical Engineering, University of Illinois Urbana-Champaign, 1991; BTech Electrical Engineering, IIT Kanpur, 1985.","source_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Joined IBM's Thomas J. Watson Research Center in 1990, PhD research reported in distributed systems and data management, spent 18 years at Watson Research through 2009 in technical/software roles (database servers, security software).","source_url":"https://grainger.illinois.edu/alumni/distinguished/Arvind-Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Named Senior Vice President and Director of IBM Research in January 2015; CEO of IBM since April 2020, chairman since January 2021; principal architect of the Red Hat acquisition.","source_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BTech in electrical engineering, IIT Kanpur (1985); PhD in electrical engineering, University of Illinois Urbana-Champaign (1991); joined IBM Thomas J. Watson Research Center 1990 and stayed 18 years; SVP IBM Research 2015; SVP Cloud and Cognitive Software; chairman and CEO of IBM from April 2020; c","source_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education at IIT Kanpur (BTech electrical engineering, 1980-1985) and University of Illinois Urbana-Champaign (MS and PhD in electrical engineering, 1985-1991), occupation chief executive officer from 2020, employer IBM","source_url":"https://www.wikidata.org/wiki/Q56276330","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No AI-lineage authorship is retrievable: the OpenAlex author record matched to this name (A5071248780, 24 works, 66 citations, h-index 4) consists of electronic-packaging, solder-void and polymer-fracture papers with affiliations at Aptiv, Qualcomm UK, PES University and Ohio State, which do not cor","source_url":"https://api.openalex.org/authors/A5071248780","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Krishna began at IBM in 1990 at the Thomas J. Watson Research Center in technical/research roles (databases, security software) before executive leadership; CEO since April 2020 — a career IBM executive, not a company founder, with no personal authorship in the vector-space/embedding/transformer lan","source_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Krishna's career is entirely at IBM (joined Watson Research 1990, SVP 2015, CEO 2020, chairman 2021); he is not a company founder, and his research roles preceded modern AI.","source_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The only AI/ML-topic OpenAlex author record matched to the name is an electronic-packaging/solder-void homonym (Aptiv, Qualcomm UK, Ohio State), not the IBM CEO — no LM-lineage authorship is retrievable.","source_url":"https://api.openalex.org/authors/A5071248780","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Electrical Engineering, University of Illinois Urbana-Champaign (1991)","MS Electrical Engineering, University of Illinois Urbana-Champaign","BTech Electrical Engineering, IIT Kanpur (1985)"],"canonical_papers":[],"first_verifiable_year":1990,"notable_systems":["IBM Cloud & Cognitive Software (executive oversight)","Red Hat acquisition","IBM Research (director)"],"citations":66,"h_index":4,"patents":0,"dossier_notes":"Dossier's OpenAlex block (24 works, h-index 4, topics 'Electronic Packaging and Soldering Technologies,' '3D IC and TSV technologies,' affiliations Aptiv/Qualcomm UK/PES University/Ohio State) is a clear homonym — an electronics-packaging engineer, not the IBM CEO with a distributed-systems/database PhD — disregarded in scoring. The dossier's PubMed block (853 raw hits, homonym_risk flagged, titles/years mostly null) is unusable noise and was disregarded. No independently verifiable personal papers, patents, or code from this Arvind Krishna in core AI/ML (foundations math, embeddings, or transformers) were found via this pass's searches; his technical record predates and is adjacent to modern AI (databases, security software), and his AI-era impact is executive/strategic rather than personally authored.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":1},"confidence":0.86,"duration_ms":47665},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":2},"confidence":0.82,"duration_ms":52566}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Arvind_Krishna","wikipedia_title":"Arvind Krishna","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"daniel-gross","name":"Daniel Gross","title":"Co-founder (former, June 2024-July 2025); investor/operator","company":"Safe Superintelligence Inc. (formerly); Meta Superintelligence Labs (from July 2025)","sector":"general","rank":81,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":5,"transformers_lm":2,"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":3,"hands_on_engineering":6,"industry_impact":6,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":20,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":20,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.69,"rationale":"Gross has no degree, thesis, or authored paper in linear algebra, optimization, or statistical learning that any source corroborates (Wikidata's only 'educated_at' entry is a pre-military prep academy in Israel, not a university; no evidence of a CS/ML PhD or dropout-from-Stanford narrative could be verified). His one concrete technical artifact is Greplin/Cue (founded 2010, launched at 19), a product letting users search across email, social media and cloud-storage accounts from one place with 'predictive search features' added in 2012 - this is consumer full-text/cross-account search and light personalization, not documented vector-space, embedding, or LM research, so it earns modest vector_embeddings/hands_on_engineering credit for shipping a real search system, not for research depth. After Apple acquired Cue in 2013 (Apple then shut Cue down), Gross became 'a director focused on machine learning' at Apple - a leadership/management title with no corroborated description of him personally building models or infrastructure. His subsequent record (YC partner running the YC AI program 2017+, prolific angel investor in Uber/GitHub/Figma/Perplexity/CoreWeave, co-deploying the Andromeda Cluster of 2,512 H100 GPUs for startups with Nat Friedman) is investing/infrastructure-provisioning, which the rubric explicitly excludes from credit. He co-founded SSI with Ilya Sutskever and Daniel Levy in June 2024 but left after about a year (July 2025) for Meta Superintelligence Labs; no source found describes his specific day-to-day role at SSI as research/engineering versus operations, fundraising, or recruiting, so industry_impact credit reflects only being an early co-founder of a lab whose stated mission is building safe superintelligence, not personal technical leadership of its research. Semantic Scholar's 17-paper/h-index-5 match and the PubMed entries (yeast transcription/genome biology by 'Gross DS', viral immunology by 'Gross DA') are unrelated homonyms with no overlap in co-authors, venue, or subject matter and are excluded entirely.\n\nNothing of Gross's own authorship sits in the frontier-model lineage: Greplin/Cue (2010–2013) was consumer cross-account full-text search with light 'predictive' personalization, not documented vector-space, distributional-semantics, embedding, attention or LM work, and no paper, patent or shipped model in that lineage is verifiable (arXiv cs.LG returns zero, OpenAlex found nothing, the Semantic Scholar/PubMed hits are confirmed homonyms), so he scores as someone who invests in and provisions compute for frontier labs rather than a foundational contributor. His verifiable language-modeling record is at most the ~3 years building Cue's search product (2010–2013), a single natural-language-text domain — no biological, financial, code, chemistry or other LM domain with a hands-on record exists. As technical founder in this field he has one company, Cue, and even that is consumer search adjacent to the core rather than language-modeling science; at SSI (co-founded June 2024, left ~July 2025) the science is Ilya Sutskever's and no source describes Gross's day-to-day role as research/engineering versus fundraising, recruiting and operations, so he does not earn the scientific-founder dimension. Apple 'director focused on machine learning' and the YC AI program are leadership/management titles with no corroborated personal model-building.","evidence":[{"claim":"Gross launched Greplin (later renamed Cue) in 2010 at age 19; it let users search online accounts (social media, email, cloud storage) from one place, with predictive search features added in 2012; Apple acquired Cue in 2013 for a reported $40-60M and shut it down shortly after.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"After the acquisition, Gross joined Apple as a director focused on machine learning; in 2017 he joined Y Combinator as a partner and created the 'YC AI' program; he is described as a notable technology investor (Uber, Instacart, Figma, GitHub, Airtable, Rippling, CoreWeave, Character.ai, Perplexity","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gross and Nat Friedman deployed the Andromeda Cluster, a supercomputer cluster of 2,512 H100 GPUs for startup use.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"In June 2024 Gross co-founded Safe Superintelligence Inc. with Ilya Sutskever and Daniel Levy; in July 2025 Gross left SSI to join Meta Superintelligence Labs.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Safe Superintelligence Inc.'s stated mission is 'Building safe superintelligence (SSI) is the most important technical problem of our time' and it is presented as the company's sole focus, though the SSI website itself lists no founder biographical detail confirming Gross's specific technical role.","source_url":"https://ssi.inc","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No university degrees documented; co-founded Greplin 2010 (rebranded Cue 2012, a unified search product with predictive search), acquired by Apple 2013; became a director focused on machine learning at Apple; joined Y Combinator as partner 2017 and created the YC AI program; co-founded Safe Superint","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"An arXiv author search for 'Daniel Gross' restricted to cs.LG returns zero results","source_url":"http://export.arxiv.org/api/query?search_query=au:%22Daniel_Gross%22+AND+cat:cs.LG&start=0&max_results=20","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records occupation 'businessperson', education limited to Bnei David Mechina, employer Cue, and notable work Cue — no academic degree, affiliation or research output","source_url":"https://www.wikidata.org/wiki/Q19364797","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gross launched Greplin/Cue in 2010, a unified cross-account search product with predictive-search features; Apple acquired Cue in 2013 and shut it down; he later became an Apple ML director, a YC partner, an investor, and co-founded SSI in June 2024 before leaving for Meta in July 2025.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Greplin/Cue (2010, launched at 19) was a unified cross-account search product (email, social media, cloud storage) with predictive search added in 2012; Apple acquired Cue in 2013 and shut it down — no vector-space or LM research described.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Gross co-founded SSI with Ilya Sutskever and Daniel Levy in June 2024 and left for Meta Superintelligence Labs in July 2025; SSI's stated technical/research leadership is Sutskever's, with no source documenting Gross authoring its research or code.","source_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"An arXiv cs.LG author search for 'Daniel Gross' returns zero results; Wikidata lists occupation 'businessperson', education only 'Bnei David Mechina', with no academic degree or research output.","source_url":"https://www.wikidata.org/wiki/Q19364797","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["No verified university degree; Wikidata lists only 'Bnei David Mechina', an Israeli pre-military preparatory academy, as 'educated_at' - not a CS/ML program. Commonly repeated claims of a Stanford or other college dropout could not be corroborated by any source fetched in this pass; treat as unverified."],"canonical_papers":[],"first_verifiable_year":2010,"notable_systems":["Greplin/Cue (cross-account personal search, 2010-2013)","Andromeda Cluster (2,512 H100 GPU cluster, with Nat Friedman)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"OpenAlex found nothing (correctly - no academic profile exists). The Semantic Scholar match (17 papers, h-index 5, 'name_exact': true) is almost certainly a homonym: no corroborating co-author, venue, or subject-matter overlap with Daniel Gross the entrepreneur was found, and nothing in his verified biography suggests authored ML papers; excluded from scoring. All 9 refined PubMed hits are confirmed homonyms - 'Gross DS' (yeast transcription/genome biology, Mol Cell/J Biol Chem/Gene, consistent single research group unrelated to AI) and 'Gross DA' (viral immunology/gene therapy, Sci Rep/J Virol) - clearly different people; excluded entirely. Wikidata's education field is thin/misleading in isolation (lists only a prep academy) and does not itself establish whether Gross attended or dropped out of a university; this pass could not verify the widely-repeated 'college dropout' claim from a primary/reliable source and reports it as unverified rather than assuming it. No profile_page or self_claims data was available in the dossier to cross-check.","years_language_modeling":3,"years_as_technical_founder":3,"technical_founder_companies":1,"lm_domains":[{"domain":"natural_language","years":"2010-2013","evidence":"Greplin/Cue cross-account full-text search with predictive-search personalization; not documented embedding/LM research, single domain"}],"frontier_lineage":["No verifiable contribution; Cue was consumer cross-account search, not vector-space/embedding/attention/transformer lineage. Investor and compute provisioner (Andromeda Cluster with Nat Friedman), not a foundational builder."],"technical_founder_roles":["Cue (Greplin) — founder-CEO of consumer search product — 2010-2013","Safe Superintelligence Inc. — co-founder, non-research role (science led by Ilya Sutskever) — 2024-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":4},"confidence":0.8,"duration_ms":54316},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":4,"lm_domain_breadth":3,"scientific_founder":6},"confidence":0.72,"duration_ms":58447}],"source":"community","profile_url":"https://en.wikipedia.org/wiki/Daniel_Gross_(entrepreneur)","wikipedia_title":"Daniel Gross (businessman)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"emad-mostaque","name":"Emad Mostaque","title":"Founder (former Co-founder & CEO, Stability AI)","company":"Stability AI / Intelligent Internet","sector":"general","rank":82,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":2,"transformers_lm":3,"frontier_founder":4,"lm_domain_depth":3,"lm_domain_breadth":4,"hands_on_engineering":4,"industry_impact":8,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":20,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":20,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.79,"rationale":"Mostaque holds an MA in mathematics and computer science from Oxford but his career before Stability AI was hedge-fund management (crude oil trading) and geopolitical consulting, with no personal record of AI/ML research, code, or publications prior to 2022 — the OpenFold paper (Nature Methods, 2024; bioRxiv preprint 2022) lists him among 28 authors from Stability AI/Columbia/OpenFold-consortium, but no author-contributions detail surfaces any technical (modeling/coding/math) role for him specifically, consistent with his documented pattern as a funder/resource-provider rather than a hands-on researcher. Stable Diffusion itself was developed by Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz (CompVis/LMU Munich, building on their earlier latent-diffusion research), with Stability AI providing compute and organizational backing — Mostaque was not a co-inventor of the diffusion/transformer techniques involved. A June 2023 investigative report (30+ sources) found he had misrepresented his educational background and overstated his personal involvement in Stable Diffusion's development, and a former co-founder sued him alleging fraud in a stock buyback; these findings directly bear on the reliability of his own self-claims and support scoring his research/engineering dimensions on verified record only, which is thin. Industry_impact reflects that he did found and lead (2020-2024) the organization that funded and released Stable Diffusion, a genuinely significant open-weights system, even though his personal technical contribution to its science is not established.\n\nNothing of Mostaque's OWN authorship sits in the frontier language-model lineage: Stable Diffusion's latent-diffusion architecture was authored by Rombach, Blattmann, Esser and Lorenz (CompVis/LMU), with Stability AI supplying compute and funding, and it is an image model rather than a language-modeling building block; there is no pre-2013 vector-space/distributional-semantics work, and no named block (attention, tokenizer, optimizer, scaling result) traces to him. His only verifiable personal research artifact is co-authorship (among 28, Stability AI affiliation, technical role unspecified) on OpenFold, a protein-structure/biological-sequence model — a thin, single-domain hands-on record dated 2022–2024 with essentially zero personal natural-language modeling years. As a founder he operated as the business/vision CEO of Stability AI (2019/2020–2024) and later Intelligent Internet/Schelling AI, but Forbes (June 2023, 30+ sources) documented he overstated his personal technical involvement, so he does not qualify as a scientific/technical founder who authored the core code, papers or patents — that work was done by employed researchers. Counting only what sources support: 0 verifiable years as a technical founder and 0 companies founded in that role, one thin biological language-modeling domain via co-authorship.","evidence":[{"claim":"MA mathematics and computer science, University of Oxford; pre-Stability AI career was hedge fund management (crude oil) and geopolitical consulting; no documented AI research background before founding Stability AI in 2019/2020","source_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of 28 authors (affiliation: Stability AI) on the OpenFold paper (Nature Methods, 2024) and its 2022 bioRxiv preprint, alongside the actual OpenFold/AlphaFold-retraining research team; no contribution statement in the fetched source specifies his individual technical role","source_url":"https://www.biorxiv.org/content/10.1101/2022.11.20.517210v1","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"June 2023 investigation citing 30+ sources (investors, former employees) found Mostaque misled investors and the public about his educational background, an AWS partnership, and the extent of his personal involvement in developing Stable Diffusion","source_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stable Diffusion was developed by Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz (university researchers), with Stability AI providing computational resources rather than the core research","source_url":"https://en.wikipedia.org/wiki/Stability_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder Cyrus Hodes sued Mostaque in July 2023 alleging he was fraudulently induced to sell his 15% Stability AI stake for $100 (across two 2021-2022 transactions) shortly before a $1B-valuation raise made it worth ~$150M; Mostaque stepped down as CEO March 23, 2024","source_url":"https://en.wikipedia.org/wiki/Stability_AI","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Holds an MA in mathematics and computer science from Oxford; Forbes reported June 2023, citing over 30 sources, that he misled investors and the public about his educational background and misrepresented his involvement in developing Stable Diffusion, and made unsubstantiated claims of partnerships","source_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Stable Diffusion's latent-diffusion architecture was developed by Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz of the CompVis group at LMU Munich with Runway, trained on LAION-5B data; Stability AI's role was compute (256 A100 GPUs, ~150,000 GPU-hours), funding, employing the r","source_url":"https://en.wikipedia.org/wiki/Stable_Diffusion","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of 28 authors on the OpenFold preprint with affiliation 'Stability AI'; the paper is a trainable reimplementation of AlphaFold2 for protein structure prediction","source_url":"https://www.biorxiv.org/content/10.1101/2022.11.20.517210v1","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education at the University of Oxford and CEO of Stability AI 2019-2024; no doctorate, thesis or advisor is recorded","source_url":"https://www.wikidata.org/wiki/Q114049362","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Mostaque's only indexed research output is co-authorship (among 28 authors, Stability AI affiliation) on OpenFold, a trainable AlphaFold2 reimplementation for protein structure — a biological-sequence model; no individual technical contribution is stated","source_url":"https://doi.org/10.1038/s41592-024-02272-z","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Forbes (June 2023, 30+ sources) reported Mostaque misled investors about his educational background and overstated his personal involvement in developing Stable Diffusion; his pre-Stability career was hedge-fund management and consulting with no AI research record","source_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education at Oxford and CEO of Stability AI 2019–2024; no doctorate, thesis, advisor, or technical-founder role is recorded","source_url":"https://www.wikidata.org/wiki/Q114049362","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Mostaque is listed as one of 28 authors on OpenFold (Nature Methods 2024; bioRxiv 2022), a trainable AlphaFold2 reimplementation for protein structure, under a 'Stability AI' affiliation with no individual technical-contribution statement","source_url":"https://doi.org/10.1038/s41592-024-02272-z","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Forbes (June 2023), citing 30+ sources, reported Mostaque misled investors and the public about his background and overstated his personal involvement in developing Stable Diffusion; he founded and was CEO of Stability AI 2019–2024 with no recorded doctorate, thesis or advisor","source_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["MA Mathematics and Computer Science, University of Oxford"],"canonical_papers":["OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization (Nature Methods, 2024) — co-author among 28, technical role not established"],"first_verifiable_year":2022,"notable_systems":["Stable Diffusion (funded/organized, not personally authored)","Stability AI (founder/CEO 2020-2024)"],"citations":497,"h_index":2,"patents":0,"dossier_notes":"OpenAlex flags ambiguous match (candidates:2) but the works found (OpenFold, Stability AI affiliation) are plausibly this person and not a homonym, so citation/h-index figures are used as-is but noted as reflecting large-team co-authorship rather than a personal research record — his h-index of 2 on 2 papers is not indicative of an independent research program. Could not find a citable source documenting paid/sponsored coverage or family-wealth funding of Stability AI (his funding appears to have come from hedge-fund earnings and later venture capital, not family money), so bought_popularity and capital_without_competence penalties are reported as 0 per the rubric's 'no source = discard/0' rule, though the Forbes-reported pattern of overstated technical claims is noted as a genuine credibility concern in the rationale.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"lm_domains":[{"domain":"biological","years":"2022-2024","evidence":"co-author (1 of 28, Stability AI affiliation) on OpenFold protein-structure model, Nature Methods 2024 / bioRxiv 2022; individual technical role not established"}],"frontier_lineage":["None established — Stable Diffusion is an image model he funded/organized but did not author; no named language-modeling block or pre-2013 vector-space precursor traces to him"],"technical_founder_roles":["Stability AI — founder & CEO (business/vision, science done by employed researchers) — 2019/2020–2024","Intelligent Internet / Schelling AI — founder — 2024–"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":4,"lm_domain_depth":3,"lm_domain_breadth":4,"scientific_founder":4},"confidence":0.82,"duration_ms":61483},{"pass":"pass_2","dimensions":{"frontier_founder":4,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":4},"confidence":0.8,"duration_ms":59811}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Emad_Mostaque","wikipedia_title":"Emad Mostaque","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sam-altman","name":"Sam Altman","title":"Co-founder & CEO","company":"OpenAI","sector":"general","rank":83,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":1,"transformers_lm":4,"frontier_founder":4,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":5,"industry_impact":13,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":20,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":20,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.92,"rationale":"Altman is the clearest case in this batch of the pattern the rubric explicitly separates out: enormous influence over the field with no personal record in its core. He studied computer science at Stanford for two years and left in 2005 without a degree, so there is no thesis, no graduate training in linear algebra, optimization or statistical learning, and no verifiable coursework record. He has authored no research in the lineage: his only OpenAlex-indexed item is the GPT-5 System Card (2025), an institutional document listing OpenAI staff, and he appears on the GPT-4 Technical Report the same way — corporate authorship convention across hundreds of names, not a technical contribution to embeddings, attention, pretraining or scaling. No first-author paper, no research-lead role, and no AI/ML patent as inventor were found in any source; the patents on his record are Loopt-era location-based social networking. His hands-on engineering is not zero because he wrote part of the original Loopt codebase as a working founder-engineer from 2005, but that is mobile social software, not AI systems, models or the infrastructure under them, and the engineering at Loopt was led by co-founder Nick Sivo. Industry impact is the one genuinely high dimension, credited strictly on the permitted ground that he co-founded and leads the laboratory that produced GPT-3, InstructGPT/RLHF and GPT-4 — canonical work the field builds on — and that he set its research direction and compute strategy. His fundraising, the Microsoft deal, ChatGPT's user numbers, his investor record and his public profile are excluded entirely; stripped of those, his personal technical record in the core of AI is thin.\n\nNothing in the transformer→language-model lineage is attributable to Altman's own hand: no attention/transformer method, no embedding, optimizer, tokenizer, pretraining objective, scaling result, dataset or alignment technique bears his authorship, and there are no pre-2013 vector-space/distributional-semantics patents or shipped systems to place him in the pre-word2vec foundation — his only indexed items (GPT-4 Technical Report, GPT-5 System Card) are organizational bylines across hundreds of names, so frontier_founder sits in the 'applies/leads, no foundational contribution' band. His verifiable hands-on language-modeling record is zero years: he set OpenAI's direction and compute strategy as CEO but authored none of its LM research or systems, so lm_domain_depth is at the 0-2 anchor and there are no distinct LM domains with a personal modeling record, leaving lm_domain_breadth at the single-'none' anchor. As a founder he is a business founder-CEO — Loopt's engineering was led by co-founder Nick Sivo, and OpenAI's science was built by Ilya Sutskever, Greg Brockman and the research staff — so he is a founder/CEO of an AI company whose science was done by others (scientific_founder 3-7 band), not a technical/scientific founder who wrote the core research, code or patents. I therefore report zero verifiable years as a technical founder and zero companies founded in that role.","evidence":[{"claim":"Attended Stanford University for two years studying computer science and left in 2005 without earning a degree","source_url":"https://en.wikipedia.org/wiki/Sam_Altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records his occupations as businessperson, programmer, entrepreneur and chief executive officer, educated at Stanford in computer science, with no doctoral or research affiliation","source_url":"https://www.wikidata.org/wiki/Q7407093","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of many co-authors on the GPT-4 Technical Report (2023), an organizational authorship credit across hundreds of names rather than a personal research contribution","source_url":"https://arxiv.org/abs/2303.08774","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dropped out of Stanford University after two years studying computer science; no degree completed","source_url":"https://finance.yahoo.com/technology/ai/articles/sam-altman-dropped-stanford-2-181500595.html","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Loopt (2005); technical/engineering work led by co-founder Nick Sivo, Altman in CEO/business role","source_url":"https://interestingengineering.com/culture/who-is-sam-altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"President of Y Combinator 2014-2019 (investor/accelerator leadership, not research)","source_url":"https://techcrunch.com/2019/03/08/y-combinator-president-sam-altman-is-stepping-down-amid-a-series-of-changes-at-the-accelerator","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Attended Stanford University for two years studying computer science and dropped out without earning a bachelor's degree in 2005; co-founded Loopt at 19; joined Y Combinator 2011 and became president 2014; co-founded OpenAI in 2015 and has been CEO since 2019; described in executive and entrepreneur","source_url":"https://en.wikipedia.org/wiki/Sam_Altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Altman is CEO of OpenAI since 2019 and co-founded it in 2015; described as entrepreneur/investor, not a research scientist — the technical/scientific work is led by others (e.g. Ilya Sutskever as chief scientist)","source_url":"https://en.wikipedia.org/wiki/Sam_Altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Loopt (2005) in a CEO/business role; engineering led by co-founder Nick Sivo — not language-modeling work","source_url":"https://interestingengineering.com/culture/who-is-sam-altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Listed as one of many organizational co-authors on the GPT-4 Technical Report (2023), an authorship convention across hundreds of names rather than a personal contribution to attention, pretraining or scaling","source_url":"https://arxiv.org/abs/2303.08774","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Altman co-founded OpenAI in 2015 and has been CEO since 2019; his role is organizational leadership and direction-setting, not authorship of the models' architecture or training methods","source_url":"https://en.wikipedia.org/wiki/Sam_Altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His only research-indexed items are the GPT-4 Technical Report and GPT-5 System Card, institutional documents listing OpenAI staff — corporate authorship convention, not a personal contribution to attention, embeddings, pretraining or scaling","source_url":"https://arxiv.org/abs/2303.08774","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"At Loopt (co-founded 2005) the technical/engineering work was led by co-founder Nick Sivo, with Altman in the CEO/business role — he has not operated as the scientific/technical founder authoring the core code, papers or patents","source_url":"https://interestingengineering.com/culture/who-is-sam-altman","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Stanford University, computer science, 2003-2005, left without a degree"],"canonical_papers":[],"first_verifiable_year":2005,"notable_systems":["Loopt (co-founder; wrote part of the original codebase)","OpenAI (co-founder and CEO; organizational leadership of GPT-3/GPT-4, not personal authorship)"],"citations":18,"h_index":1,"patents":2,"dossier_notes":"Several dossier fields are unusable. (1) first_verifiable_year 1995 is wrong: it derives from the PubMed query whose 16 'refined' results are all homonyms (coral-reef ecology, dreissenid mussels, histochemistry, typhoid) — Altman was ten years old in 1995. The correct first verifiable year is 2005. (2) The Semantic Scholar match 'S. Altman' with 31 papers and 31,001 citations is flagged name_exact false with 3 candidates and is a different person or a merged record; it must not be credited to him, and neither pass did. (3) The OpenAlex record is flagged ambiguous (4 candidates) and its only substantive item is an institutional system card. (4) The Wikidata employer field contains an unresolved QID (Q2616400).","adjudication_notes":["industry_impact — SPLIT, final 13 (pass_1 12, pass_2 14). Both passes agree on the permitted ground: he co-founded and leads the lab that produced GPT-3, InstructGPT/RLHF and GPT-4, canonical work the field builds on, and set its research direction. The rubric credits 'leadership of labs that produced canonical work' explicitly, which supports a high score; but it withholds the 18-20 anchor for principal builders of systems the field runs on, and he is not a builder of them, so 13 sits at the top of the 'production systems built and led personally' band without crossing into authorship credit.","transformers_lm — SPLIT, final 4 (pass_1 3, pass_2 5). Neither pass found any authored contribution; the dispute is only how much a system-card byline is worth. The verified fact is that his appearances on the GPT-4 Technical Report and GPT-5 System Card are corporate authorship convention across hundreds of names — real proximity to the work, no personal contribution to attention, pretraining or scaling. 4 keeps him inside the 3-7 'manages builders, no personal record' band, which is what the evidence shows.","foundations — SPLIT, final 3 (pass_1 2, pass_2 4). Two years of undergraduate computer science with no degree, no thesis and no verifiable coursework in the field's mathematics. Pass_2's 4 credits more than the record shows; pass_1's 2 is the near-'nothing verifiable' anchor, which two years of CS study slightly exceeds.","hands_on_engineering — SPLIT, final 5 (pass_1 2, pass_2 7). Pass_2 is right that he wrote part of the original Loopt codebase as a working founder-engineer, so pass_1's 2 is too harsh; but pass_1 is right that the Loopt engineering was led by co-founder Nick Sivo and that none of it is AI systems, models or their infrastructure, so 7 overstates. 5 is mid-band for 'uses the tools, manages builders'.","vector_embeddings — pass_1 right, final 1 vs pass_2's 2. No embedding, retrieval or vector-search work of any kind is attributable to him; this is the 0-2 anchor.","penalties — both passes agree at 0/0 and both are right. This is the profile shape the capital_without_competence penalty targets, but the rubric requires cited evidence of family, friends or inherited money funding the AI company, and OpenAI was founded on named external donor commitments; asserting the penalty without a citable source would be discarded by the merge step and is not permitted on rumour."],"years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["OpenAI — co-founder & CEO (business/executive, not technical founder) — 2015-present","Loopt — co-founder & CEO (business role, engineering led by Nick Sivo) — 2005-2012"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.86,"duration_ms":49948},{"pass":"pass_2","dimensions":{"frontier_founder":5,"lm_domain_depth":3,"lm_domain_breadth":2,"scientific_founder":3},"confidence":0.85,"duration_ms":68318}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Sam_Altman","wikipedia_title":"Sam Altman","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"anna-kazlauskas","name":"Anna Kazlauskas","title":"Co-founder & CEO","company":"Vana (Open Data Labs)","sector":"crypto","rank":84,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":3,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":7,"industry_impact":6,"scientific_founder":7},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":19,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":19,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.64,"rationale":"Kazlauskas is not a co-founder of Gensyn (the task context's suggestion appears mistaken) — she is the co-founder/CEO of Vana, a data-ownership/DePIN protocol for AI training data, and this is corroborated across multiple independent sources (MIT News, Crunchbase, podcast profiles). She studied computer science and economics at MIT (no evidence of a completed degree or thesis found — sources describe her as a student/dropout rather than a graduate) and ran a YC-backed (W18) fintech/document-automation ML startup, Iambiq, and later worked as an early engineer at the Celo Foundation adapting the Celo blockchain for mobile. No peer-reviewed papers, patents, or canonical AI research were found under her name; the sole OpenAlex-listed work is a 2026 SSRN research-agenda essay ('The Economics of AI Training Data') with zero citations, which is a position paper, not primary technical research. Her verifiable technical record is real but shallow — production engineering on a blockchain (Celo) and a small ML-for-documents startup — rather than personal contributions to embeddings, attention, or transformer/LM research; Vana's core protocol design (data attestation, proof-of-contribution) is infrastructure/tokenomics work adjacent to AI data pipelines rather than model research itself.\n\nNo verifiable contribution sits in the lineage today's frontier models descend from: Kazlauskas has no authored attention/transformer/embedding/optimizer/dataset work, and Vana is a data-ownership/DePIN protocol (encrypted storage, on-chain permissions, proof-of-contribution) that is infrastructure adjacent to AI training data, not a foundational modeling block — frontier_founder scores near the floor. Her only indexed work is a 2026 SSRN research-agenda essay with zero citations; there is no verifiable language-modeling research or shipped LM/vector-space system under her name, so years_language_modeling is effectively 0 and lm_domain_depth/breadth score at the bottom (Iambiq's 'document automation' is not a confirmed LM record, and applying data pipelines to model training is not modeling in a domain). She is a genuine technical founder-CEO with a real engineering background (early engineer at Celo, YC-backed Iambiq, founder of Vana ~2021), giving roughly 6 years across two companies she co-founded — which anchors scientific_founder in the 8-12 band — but no papers or patents establish that she personally authored the core science, and Vana's core is not language modeling, holding it to the low end. Overall: real entrepreneurial/engineering record, no frontier-lineage or language-modeling depth.","evidence":[{"claim":"Anna Kazlauskas is co-founder and CEO of Vana (Open Data Labs), a protocol for user-owned AI training data — not affiliated with Gensyn","source_url":"https://news.mit.edu/2025/vana-lets-users-own-piece-ai-models-trained-on-their-data-0403","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Before Vana she was an early engineer at the Celo Foundation, working on adapting the Celo blockchain for mobile devices","source_url":"https://nocap.blog/founder/anna-kazlauskas/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MIT class of 2019; joined the MIT Bitcoin Club in 2015; met co-founder Art Abal in the Media Lab class 'Emergent Ventures' taught by Ramesh Raskar, who still advises Vana on AI research; worked at Celo before founding Vana; Vana uses data DAOs so users pool exported personal data and receive proport","source_url":"https://news.mit.edu/2025/vana-lets-users-own-piece-ai-models-trained-on-their-data-0403","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Vana's documentation describes it as an open protocol for private, user-owned data with encrypted storage, on-chain permission management and a local Personal Server; DataDAOs are an optional application layer, not the core.","source_url":"https://docs.vana.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kazlauskas is co-founder/CEO of Vana, a protocol for user-owned AI training data (data attestation, proof-of-contribution) — infrastructure adjacent to AI data, not model research","source_url":"https://news.mit.edu/2025/vana-lets-users-own-piece-ai-models-trained-on-their-data-0403","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Before Vana she was an early engineer at the Celo Foundation and ran YC-backed (W18) ML document-automation startup Iambiq — a verifiable but non-LM technical-founder track record","source_url":"https://nocap.blog/founder/anna-kazlauskas/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Vana's own documentation describes it as an open protocol for private user-owned data with encrypted storage and on-chain permissions — no language-modeling or transformer research","source_url":"https://docs.vana.org/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kazlauskas is co-founder/CEO of Vana, an open protocol for user-owned AI training data (encrypted storage, on-chain permissions) — infrastructure adjacent to AI, not model research","source_url":"https://news.mit.edu/2025/vana-lets-users-own-piece-ai-models-trained-on-their-data-0403","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Before Vana she ran Iambiq (YC W18), an ML document-automation startup, and was an early engineer at the Celo Foundation adapting the Celo blockchain for mobile","source_url":"https://nocap.blog/founder/anna-kazlauskas/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Computer Science and Economics, MIT (degree completion not independently confirmed — multiple sources describe her as leaving MIT to found companies)"],"canonical_papers":[],"first_verifiable_year":2018,"notable_systems":["Vana protocol (user-owned data / AI training data DPO network)","Iambiq (YC W18, ML document automation)","early engineering contributions to Celo blockchain"],"citations":2,"h_index":1,"patents":0,"dossier_notes":"The task context suggesting she might be a Gensyn co-founder is NOT supported by any source found — she is consistently identified across Crunchbase, MIT News, podcasts, and her own site as Vana/Open Data Labs founder, which matches the dossier's own 'company: Vana' field; Gensyn's actual co-founders are Ben Fielding and Harry Grieve. The dossier's OpenAlex match (1 work, the 2026 SSRN piece) appears correctly matched (non-ambiguous, single candidate) but is thin — a research-agenda essay, not original technical research. The PubMed 'Kazlauskas A' results (233 rows, diabetic retinopathy / bearing degradation / quantum imaging papers) are unrelated homonyms (different field entirely, likely Andrius Kazlauskas or similar biomedical/engineering researchers) and were correctly not counted. Overall her identity is confidently established but her technical AI research depth is low and mostly self-reported/entrepreneurial rather than published; scored with moderate-low confidence due to thin corroborating detail on the extent of her personal coding/engineering contribution at Celo and Iambiq.","years_language_modeling":0,"years_as_technical_founder":6,"technical_founder_companies":2,"technical_founder_roles":["Iambiq — co-founder (ML document automation, YC W18) — 2018","Vana / Open Data Labs — co-founder & CEO — ~2021-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":8},"confidence":0.6,"duration_ms":53367},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":3,"lm_domain_breadth":3,"scientific_founder":6},"confidence":0.6,"duration_ms":52616}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"dominic-williams","name":"Dominic Williams","title":"Founder & Chief Scientist","company":"DFINITY (Internet Computer)","sector":"crypto","rank":85,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":1,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":11,"industry_impact":6,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":19,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":19,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.84,"rationale":"Williams has a verifiable and substantial technical record entirely in distributed systems and cryptography, with nothing anywhere in the AI lineage. I confirmed the authorship of his principal paper directly against the IACR ePrint archive: 'Internet Computer Consensus' (ePrint 2021/632, with Jan Camenisch, Manu Drijvers, Timo Hanke, Yvonne-Anne Pignolet and Victor Shoup) introduces the ICC family of Byzantine fault-tolerant atomic-broadcast protocols under partial synchrony, with probabilistic leader rotation and optimistic responsiveness. He is also a co-author of the earlier 'DFINITY Technology Overview Series, Consensus System' (2018). Co-authoring a consensus protocol alongside cryptographers of Shoup's and Camenisch's standing is real mathematical competence in probability, cryptography and distributed algorithms, which supports a foundations score adjacent to graduate training — but it is not the linear algebra, optimization or statistical learning the rubric names, and none of it touches representation learning. He founded DFINITY in 2016 and is credibly the principal architect of the Internet Computer's chain-key design, which is why hands-on engineering is his highest dimension. I found no paper, model or system authored by him in vector-space models, embeddings, retrieval, attention, transformers, pretraining, scaling or alignment. The Internet Computer now markets 'Caffeine AI' and an AI app builder, but that is a platform consuming third-party models, and marketing positioning is not evidence under this rubric.\n\nWilliams's entire verifiable technical output — the ICC Byzantine fault-tolerant consensus family (IACR ePrint 2021/632, PODC 2022) and the 2018 DFINITY Consensus System overview — sits in distributed systems and cryptography, not in the attention/transformer/embedding lineage; nothing of his is a building block today's frontier language models descend from, and there is no pre-2013 vector-space, LSI, distributional-semantics or term-relationship work under his name, so frontier_founder is floored just above zero. He has zero verifiable years in language modeling of any kind — no n-gram, neural-LM, seq2seq, transformer, retrieval or embedding record — and Caffeine AI is a platform consuming third-party models, not modeling work, so both lm_domain_depth and lm_domain_breadth reflect an absent record (0 LM years, 0 LM domains). He is, however, a genuine hands-on technical founder: he founded DFINITY in 2016 (~10 years, still active) and personally set and executed the Internet Computer's chain-key/consensus technical direction, with earlier founder roles at Smartdrivez and the MMO Fight My Monster — but that founding work is in distributed systems/crypto, entirely OUTSIDE language modeling, which is the 3-7 'technical founder outside this field' band. scientific_founder is scored at the top of that band for a real, multi-company, decade-long technical-founder record that simply is not in this field.","evidence":[{"claim":"Co-author of 'Internet Computer Consensus' (Camenisch, Drijvers, Hanke, Pignolet, Shoup, Williams, 2021), introducing the ICC family of leader-based Byzantine fault-tolerant consensus protocols assuming partial synchrony, with probabilistic leader rotation and optimistic responsiveness","source_url":"https://eprint.iacr.org/2021/632","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 2115924821 lists the 'DFINITY Technology Overview Series, Consensus System' (2018, with T. Hanke and M. Movahedi) among his papers","source_url":"https://api.semanticscholar.org/graph/v1/author/2115924821?fields=name,paperCount,citationCount","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata describes him as an 'Erlang programmer', software developer and entrepreneur, with no academic degree, research affiliation or doctoral record recorded","source_url":"https://www.wikidata.org/wiki/Q115266892","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BSc Computer Science, King's College London (1992-1995); pre-blockchain career founding Smartdrivez and the MMO game Fight My Monster","source_url":"https://usethebitcoin.com/crypto-personalities/all-you-need-to-know-about-dominic-williams-the-co-founder-of-dfinity/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata describes him as an 'Erlang programmer' / software developer and entrepreneur, consistent with a distributed-systems (not AI research) background","source_url":"https://www.wikidata.org/wiki/Q115266892","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-author of 'DFINITY Technology Overview Series, Consensus System' (2018, 319 citations, with T. Hanke and M. Movahedi); Semantic Scholar author 2115924821 records 5 papers and 319 citations","source_url":"https://api.semanticscholar.org/graph/v1/author/2115924821?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year,papers.citationCount,papers.authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-authored 'Internet Computer Consensus' (Camenisch, Drijvers, Hanke, Pignolet, Shoup, Williams), a Byzantine consensus protocol — a distributed-systems/cryptography contribution with no place in the transformer/LM lineage","source_url":"https://eprint.iacr.org/2021/632","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him as an 'Erlang programmer', software developer and entrepreneur with no research affiliation or doctoral record — consistent with a distributed-systems technical founder, not an AI/LM researcher","source_url":"https://www.wikidata.org/wiki/Q115266892","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BSc Computer Science, King's College London (reported; not confirmed from a primary institutional source)"],"canonical_papers":["Internet Computer Consensus (IACR ePrint 2021/632; PODC 2022)","DFINITY Technology Overview Series, Consensus System (2018)"],"first_verifiable_year":2018,"notable_systems":["Internet Computer / chain-key cryptography platform","ICC consensus protocol family","Caffeine AI (product built on third-party models)"],"citations":319,"h_index":2,"patents":0,"dossier_notes":"The dossier's OpenAlex block is overwhelmingly the WRONG PERSON. A5000266943 (94 works, 152 citations) carries topics 'Memory, Trauma, and Commemoration', 'German History and Society' and 'Italian Fascism and Post-war Society' with affiliations at Leeds, Huddersfield, Liverpool and Northumbria — a UK humanities scholar of Holocaust studies whose works ('Matters of Testimony: Interpreting the Scrolls of Auschwitz', a Slayer 'Angel of Death' article) sit in the list beside the single genuine entry, 'Internet Computer Consensus'. Its earliest_year of 1971 produces the dossier's first_verifiable_year 1971 and years_active 55, which are meaningless; his real first verifiable publication is 2018. The Semantic Scholar profile is also mixed, with only the two DFINITY consensus entries genuinely his. The PubMed block (14,796 raw hits refined to 97, all titles and years null) is unusable noise from an overly broad affiliation query and must be discarded entirely.","adjudication_notes":["foundations — SPLIT, final 6 (pass_1 2, pass_2 8). Pass_2 is right that pass_1's 2 is untenable: I verified against IACR ePrint 2021/632 that he co-authored a Byzantine-consensus protocol with Victor Shoup and Jan Camenisch, which is not something achievable without genuine mathematical competence in probability and cryptography. But pass_2's 8 places him in the 'strong graduate training' band on evidence that is entirely cryptography and distributed algorithms, with no verifiable degree beyond a reported BSc and nothing in linear algebra, optimization or statistical learning — the mathematics this dimension names. 6 credits demonstrated rigor without asserting core-AI training.","hands_on_engineering — PASS_2 right, final 11 (pass_1 6, pass_2 12). He founded DFINITY in 2016 and is credibly the principal architect of the Internet Computer's chain-key design, a large production distributed system — the 8-12 band for senior engineering, which pass_1's 6 understates. It is not AI infrastructure, so it does not reach 13-17.","transformers_lm — pass_1 right, final 2 (pass_1 1, pass_2 3). Both passes searched and neither found any authored work in attention, transformers, pretraining or alignment. The only AI-adjacent item is Caffeine AI, a product consuming third-party models, and its 'self-writing internet' marketing language is explicitly not evidence.","vector_embeddings — pass_1 right, final 1 (pass_1 0, pass_2 2). No embedding, retrieval or vector-search work exists under his name; 1 rather than 0 only because a verifiable technical career sits behind the absence.","industry_impact — SPLIT, final 6 (pass_1 4, pass_2 7). DFINITY is a real R&D organization with a substantial research team and 319 citations attach to the consensus work, which exceeds pass_1's 4; but the org's core is consensus and cryptography rather than language models, so it earns no credit as a lab that produced canonical AI work.","dossier homonym — BOTH PASSES CORRECT and both flagged it: the OpenAlex record is a Holocaust-studies humanities scholar, and the dossier's first_verifiable_year of 1971 (years_active 55) is an artifact of that merge, not his record. Neither pass counted it, and the adjudication uses 2018."],"years_language_modeling":0,"years_as_technical_founder":10,"technical_founder_companies":3,"technical_founder_roles":["DFINITY / Internet Computer — founder & chief architect (chain-key & consensus) — 2016-present","Fight My Monster (MMO) — founder — pre-2016","Smartdrivez — founder — pre-2016"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.87,"duration_ms":56037},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.82,"duration_ms":62022}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sam-blackshear","name":"Sam Blackshear","title":"Co-founder & CTO (former)","company":"Mysten Labs (Sui)","sector":"crypto","rank":86,"rubric_version":4,"dimensions":{"foundations":8,"vector_embeddings":1,"transformers_lm":2,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":10,"industry_impact":6,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":19,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":19,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.86,"rationale":"Sam Blackshear holds a PhD in Programming Languages from the University of Colorado Boulder and a BA in CS and Philosophy from Williams College; his entire publication record (26 OpenAlex works, h-index 11, topics: software testing/debugging, formal methods in verification, malware detection) is in static analysis, program verification and compositional race detection (e.g. RacerD, PACM PL 2018; Thresher 2013), not in machine learning, statistical learning theory, or optimization for learning. At Meta he was a principal engineer on the Infer/RacerD static-analysis team and then created the Move smart-contract programming language for the Diem/Libra project, later co-founding Mysten Labs and building the Sui blockchain (2021); this is deep, personal, PhD-level systems/languages engineering, but it is compiler/VM/verification infrastructure, not AI model or embedding/representation-learning infrastructure. No evidence in the dossier or in verification searches of any personal research, code, or shipped system involving embeddings, attention, transformers, pretraining, or neural network training. Mysten Labs/Sui's core product is blockchain L1 infrastructure, not an AI/LM system, so industry_impact is capped by the rubric's 'org's CORE is these systems' requirement despite his strong citations/engineering leadership. Per WebSearch, he has since left Mysten Labs to join Anthropic for defensive security research (2026) -- a security, not ML-research, role -- which does not change the AI-core scoring.\n\nBlackshear's verifiable record is entirely in programming languages, static analysis and program verification (RacerD, Thresher, Verification modulo versions) and in the Move smart-contract language / Sui blockchain — none of it is in the attention→transformer→language-model lineage, in vector-space/distributional-semantics text representation, or in any optimizer/tokenizer/embedding/pretraining/alignment block that frontier models descend from, so frontier_founder is near-zero (his 2026 move to Anthropic is a defensive-security role, not foundational LM research). He has zero verifiable years in statistical or neural language modeling; designing a compiled smart-contract language (Move) is language DESIGN/compilation, not statistical/neural language MODELING, so lm_domain_depth and lm_domain_breadth are both at the floor with no qualifying domain. He is, however, a genuine scientific/technical founder: he co-founded Mysten Labs in September 2021 and personally authored the Move language and core research the Sui L1 runs on (~4 years, one company) — but this is a technical founder OUTSIDE the AI/language-modeling field, which the anchor caps in the 3-7 band. Net: one real technical-founder role, but no language-modeling lineage, depth, breadth, or frontier-model foundation.","evidence":[{"claim":"PhD in Programming Languages, University of Colorado Boulder; BA Computer Science and Philosophy, Williams College; ~6 years at Meta as Principal Engineer leading development of Move before co-founding Mysten Labs","source_url":"https://www.sui.io/blog/move-origins-sam-blackshear","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored 'RacerD: compositional static race detection' (PACM PL / OOPSLA 2018), a compositional static analysis tool for Java built on Facebook's Infer framework -- program analysis/verification, not ML","source_url":"https://research.facebook.com/publications/racerd-compositional-static-race-detection/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar profile corroborates: 17 papers, 665 citations, h-index 13, consistent with the OpenAlex program-analysis/formal-methods focus","source_url":"https://www.semanticscholar.org/author/Sam-Blackshear/1748060","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Mysten Labs (Sept 2021) and built the Sui Layer-1 blockchain, using a variant of Move; recently departed Mysten Labs to join Anthropic for defensive security research","source_url":"https://www.tradingview.com/news/coinpedia:3d1cb34b4094b:0-move-creator-sam-blackshear-leaves-mysten-labs-what-next-for-sui/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"arXiv record shows six papers: Resources: A Safe Language Abstraction for Money (2020, with Dill, Qadeer, Barrett, Mitchell), Robust Safety for Move (2021), The Move Borrow Checker (2022), Sui Lutris (2023), Generating Move Smart Contracts based on Concepts (2024), and Collaboration is all you need:","source_url":"http://export.arxiv.org/api/query?search_query=au:%22Blackshear%22&start=0&max_results=30&sortBy=submittedDate&sortOrder=ascending","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Creator of the Move programming language at Meta's Diem project; co-founded Mysten Labs September 2021 with Evan Cheng, Adeniyi Abiodun, George Danezis and Kostas Chalkias; no AI or machine-learning systems documented for Sui or Mysten Labs","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar records 17 papers, 665 citations, h-index 13 under an exact name match","source_url":"https://www.semanticscholar.org/author/1748060","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Blackshear created the Move programming language at Meta's Diem project and co-founded Mysten Labs (Sept 2021), building the Sui Layer-1 blockchain on a Move variant — blockchain/PL infrastructure, not AI/LM systems","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Publication record (RacerD PACM PL 2018, Thresher OOPSLA 2013, Verification modulo versions PLDI 2014) is entirely static analysis / formal verification — no embeddings, attention, transformers, or neural language modeling","source_url":"https://research.facebook.com/publications/racerd-compositional-static-race-detection/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar profile: 17 papers, h-index 13, all in program analysis / formal methods, confirming no language-modeling lineage work","source_url":"https://www.semanticscholar.org/author/Sam-Blackshear/1748060","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Blackshear co-founded Mysten Labs (Sept 2021) as creator of the Move language and architect of Sui, personally authoring the core language/VM — a technical founder, but of a blockchain L1, not a language-modeling system","source_url":"https://en.wikipedia.org/wiki/Sui_(blockchain_platform)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His entire authored corpus (RacerD OOPSLA 2018, Thresher 2013, Move safety/borrow-checker papers 2020-2022, Sui Lutris 2023) is in program analysis, verification and blockchain — none in embeddings, transformers or language modeling, so there is no frontier-model lineage","source_url":"https://research.facebook.com/publications/racerd-compositional-static-race-detection/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD, Programming Languages, University of Colorado Boulder","BA, Computer Science and Philosophy, Williams College"],"canonical_papers":["RacerD: compositional static race detection (PACM PL / OOPSLA 2018)","Thresher (OOPSLA 2013)","Verification modulo versions (PLDI 2014)","Almost-correct specifications (PLDI 2013)"],"first_verifiable_year":2010,"notable_systems":["Infer/RacerD static analysis (Facebook/Meta)","Move programming language (Diem/Libra, then Sui)","Sui blockchain (Mysten Labs)"],"citations":398,"h_index":11,"patents":0,"dossier_notes":"OpenAlex/Semantic Scholar match is correct and well-supported (co-author/venue/affiliation consistent with known Facebook/Meta program-analysis work; corroborated independently via web search of PhD institution, Meta role, and Move/Sui history). The PubMed hits (Blackshear SG, 1959-1960 breast cancer surgery papers) are an unrelated homonym and were excluded. This is a genuine PL/formal-methods/systems researcher, not an AI/ML researcher -- his domain is adjacent CS theory (program verification, static analysis) rather than the rubric's target lineage (linear algebra/optimization/statistical learning, embeddings, transformers). Scored honestly low on the three AI-specific dimensions despite strong general CS credentials, per brief instructions not to inflate for accomplished non-AI CS people.","years_language_modeling":0,"years_as_technical_founder":4,"technical_founder_companies":1,"technical_founder_roles":["Mysten Labs — co-founder & creator of Move / Sui core tech — 2021-2025"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.9,"duration_ms":55831},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.92,"duration_ms":42664}],"source":"seeded","profile_url":null,"wikipedia_title":"Sui (blockchain platform)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"satya-nadella","name":"Satya Nadella","title":"Chairman & CEO","company":"Microsoft","sector":"general","rank":87,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":2,"transformers_lm":4,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"hands_on_engineering":5,"industry_impact":12,"scientific_founder":2},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":19,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":19,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.89,"rationale":"Nadella holds a BE in Electrical Engineering (Manipal), an MS in Computer Science (University of Wisconsin-Milwaukee), and an MBA (Chicago Booth) — solid technical education but no PhD, no publications, no AI/ML research record, and no canonical papers of any kind found in Wikidata, OpenAlex, Semantic Scholar, or web search. His entire post-1992 career at Microsoft is executive/managerial: Bing, Server & Tools, Cloud & Enterprise/Azure, then CEO since 2014, where he has directed (not personally authored) Microsoft's massive OpenAI partnership and Copilot rollout. The dossier's OpenAlex 'publications' (Shaping the Fourth Industrial Revolution, Navigating Digital Transformation, Hit Refresh) are business/leadership commentary and a memoir, not technical AI research, and per rubric these do not count as core-AI depth. Patents attributed to him (34, per patent-analytics sites) are typical of a senior tech executive named as co-inventor on business/product patents, not evidence of personal hands-on model-building. Industry impact is scored moderately for directing one of the largest AI deployments in the industry (Azure OpenAI Service, Copilot) even though this is managerial rather than a personal research/engineering record — per rubric, a famous CEO with no personal technical record scores low on the research dimensions specifically.\n\nNothing in Nadella's own record enters the frontier-model lineage: no attention/transformer/embedding/optimizer/scaling/alignment work, no pre-2013 vector-space, distributional-semantics or term-relationship-network papers, patents or shipped systems authored by him — his connection to GPT-class models is executive sponsorship of Microsoft's OpenAI partnership and Copilot, which the rubric explicitly excludes as branding/deployment, not foundation. He has zero verifiable hands-on years in language modeling of any kind: OpenAlex shows only business/leadership commentary (Hit Refresh, 'Shaping the Fourth Industrial Revolution'), and the PubMed 'Nadella S' hits are unrelated medical-research homonyms (spinal-cord/CAR-T papers with disjoint co-authors), so no LM domain — natural-language, biological, financial or otherwise — is supported. He is not a founder of any company (Microsoft was founded in 1975 by Gates and Allen; he joined in 1992 and became CEO in 2014), and never operated as a founder-CTO/Chief-Scientist authoring core research or code, so the scientific-founder record is empty. Accordingly all four dimensions sit at the 0-2 'nothing verifiable' floor.","evidence":[{"claim":"MS in Computer Science, University of Wisconsin-Milwaukee; BE Electrical Engineering, Manipal Institute of Technology; MBA, University of Chicago Booth School of Business","source_url":"https://en.wikipedia.org/wiki/Satya_Nadella","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Joined Microsoft in 1992; held engineering-leadership (not individual-contributor research) roles across Bing, Server & Tools, and Cloud & Enterprise/Azure before becoming CEO in 2014","source_url":"https://www.ebsco.com/research-starters/biography/satya-nadella/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Patent portfolio (~34 patents) is business/systems-oriented, not AI-model research; no AI/ML papers found on Google Scholar, arXiv, or DBLP under his name","source_url":"https://insights.greyb.com/satya-nadella-patents/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BE Electrical Engineering, Manipal Institute of Technology 1988; MS Computer Science, University of Wisconsin-Milwaukee 1990; MBA University of Chicago Booth 1997; member of technology staff at Sun Microsystems before joining Microsoft in 1992; president of Server & Tools Division 2011-2014, then EV","source_url":"https://en.wikipedia.org/wiki/Satya_Nadella","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5067894343 lists only 4 works, 232 citations, h-index 3, all business/strategy items (Hit Refresh, Navigating Digital Transformation, Shaping the Fourth Industrial Revolution, Global maxima through local action); topics are Big Data and Business Intelligence and Business Strategies","source_url":"https://api.openalex.org/authors/A5067894343","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q7426870 records education at Manipal Institute of Technology (Bachelor of Engineering), University of Wisconsin-Milwaukee (MS, computer science) and Booth School of Business (MBA), employers Sun Microsystems and Microsoft from 1992; no doctoral advisor and no Google Scholar ID are recorded","source_url":"https://www.wikidata.org/wiki/Q7426870","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Nadella joined Microsoft in 1992 and became CEO in 2014; he is a business executive, not a founder, with no doctoral advisor, Google Scholar ID or authored AI/ML research recorded","source_url":"https://www.wikidata.org/wiki/Q7426870","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5067894343 lists only 4 works (Hit Refresh memoir and business-strategy essays), h-index 3, topics 'Big Data and Business Intelligence' / 'Business Strategies' — no language-modeling or transformer research","source_url":"https://api.openalex.org/authors/A5067894343","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Microsoft was founded by Bill Gates and Paul Allen in 1975, not by Nadella; his roles (Bing, Server & Tools, Cloud & Enterprise/Azure, CEO) are managerial leadership","source_url":"https://en.wikipedia.org/wiki/Satya_Nadella","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career is executive/managerial at Microsoft since 1992 through CEO (2014); no doctoral advisor, no Google Scholar ID, no founded company recorded","source_url":"https://www.wikidata.org/wiki/Q7426870","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5067894343 lists only 4 works (memoir + business-strategy essays), 232 citations, h-index 3, topics 'Big Data and Business Intelligence' / 'Business Strategies' — no LM or AI-research authorship","source_url":"https://api.openalex.org/authors/A5067894343","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Education is BE (Manipal), MS CS (Wisconsin-Milwaukee), MBA (Booth); no PhD and no research/lineage record","source_url":"https://en.wikipedia.org/wiki/Satya_Nadella","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BE Electrical Engineering, Manipal Institute of Technology","MS Computer Science, University of Wisconsin-Milwaukee","MBA, University of Chicago Booth School of Business"],"canonical_papers":[],"first_verifiable_year":2017,"notable_systems":["Microsoft Azure (executive leadership, not personal engineering)","Azure OpenAI Service / Copilot (executive sponsorship)"],"citations":232,"h_index":3,"patents":34,"dossier_notes":"Dossier's OpenAlex/Semantic Scholar/PubMed matches all appear correctly identified but reflect business-book/commentary output (Hit Refresh, Fourth Industrial Revolution essays) and unrelated medical-research homonyms (spinal cord injury, CAR T cell papers under 'Nadella S'), not personal AI/ML research — none of this is canonical AI work and should not be weighted as such. No evidence found of Nadella personally authoring any AI/ML technical paper, patent claim, or open-source system.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":1},"confidence":0.9,"duration_ms":48719},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":2},"confidence":0.9,"duration_ms":44564}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Satya_Nadella","wikipedia_title":"Satya Nadella","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sandeep-nailwal","name":"Sandeep Nailwal","title":"Co-founder, Polygon; co-founder, Sentient","company":"Polygon / Sentient","sector":"crypto","rank":88,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":2,"transformers_lm":4,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":7,"industry_impact":5,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":18,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":18,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.83,"rationale":"Nailwal is a blockchain engineer and founder, not an AI researcher. His verifiable technical record is Ethereum scaling infrastructure: he co-founded Matic Network, later Polygon, in 2017 with Jaynti Kanani, Anurag Arjun and Mihailo Bjelic, all described in the sources only as software engineers, with no biographical or educational detail confirmable from any primary source. His single indexed publication is 'OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution' (arXiv:2411.03887, 2024), and I verified the author list directly: he is ninth of twelve, on a paper whose senior names are Sewoong Oh, Himanshu Tyagi and Pramod Viswanath. I also confirmed the paper's subject matter from the abstract — it introduces a primitive for cryptographically enforced usage authorization of freely distributed models, with security definitions for model-extraction and permission-forgery resistance implemented via fingerprinting and crypto-economic enforcement. That is security and mechanism design about distributing models, not work on the models themselves: no architecture, training, embedding or language-modelling contribution. The paper has zero citations, it is his only one, and it dates to 2024, giving him two years in this lineage. Sentient, which he co-founded, does release open models, but nothing in the verifiable record shows him personally designing, training or authoring language-model work, so the research dimensions sit in the 3-7 'uses the tools, manages builders' band, with engineering credited slightly higher for real protocol-level building at Polygon that lies outside the AI core.\n\nNothing in Nailwal's verifiable record places his own work in the lineage today's frontier models descend from: his single indexed paper, OML (arXiv:2411.03887, 2024), is 9th-of-12 authorship on a cryptographic model-distribution / licensing primitive (fingerprinting + crypto-economic usage authorization), not attention, embeddings, pretraining, scaling or any pre-2013 vector-space precursor — so frontier_founder sits at the 'applies/associates with frontier models, no foundational contribution' band. He has no verifiable years of hands-on language-modeling work (the 2024 paper is security/mechanism design about distributing models, not modeling them), so lm_domain_depth is near the floor and years_language_modeling is 0. No distinct language-modeling domain (natural-language, biological, financial, code, etc.) has a verifiable hands-on modeling record in his name, so lm_domain_breadth is at the floor. He IS a genuine technical/engineering co-founder — Polygon/Matic Network from 2017 (~9 years of real shipped protocol engineering) and later Sentient — but Polygon's core is Ethereum-scaling blockchain, not these systems, and Sentient's science is carried by academic co-authors (Sewoong Oh, Himanshu Tyagi, Pramod Viswanath), so scientific_founder lands in the 3-7 'technical founder outside this field / AI company whose science is done by others' band rather than higher.","evidence":[{"claim":"'OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution', arXiv:2411.03887 (2024); Nailwal is ninth of twelve authors, after Zerui Cheng, Edoardo Contente, Ben Finch, Oleg Golev, Jonathan Hayase, Andrew Miller, Niusha Moshrefi and Anshul Nasery, and before Sewoong O","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The OML paper addresses cryptographically enforced usage authorization for locally executed models — security definitions for model-extraction and permission-forgery resistance via AI-native fingerprinting and crypto-economic enforcement — not model training or architecture","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Polygon was founded as Matic Network in 2017 by Jaynti Kanani, Sandeep Nailwal, Mihailo Bjelic and Anurag Arjun, described as software engineers; it is an Ethereum-compatible proof-of-stake scaling platform","source_url":"https://en.wikipedia.org/wiki/Polygon_(blockchain)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5114645762: 1 work, 0 citations, h-index 0, earliest year 2024","source_url":"https://api.openalex.org/authors/A5114645762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sandeep Nailwal is one of 12 co-authors on 'OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution' (arXiv, 2024), alongside academics including Sewoong Oh and Andrew Miller.","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sandeep Nailwal co-founded Polygon (originally Matic Network) in 2017 with Jaynti Kanani, Anurag Arjun, and Mihailo Bjelic; the founders are described generically as software engineers, with no AI/ML focus — Polygon's core is Ethereum-compatible blockchain scaling.","source_url":"https://en.wikipedia.org/wiki/Polygon_(blockchain)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sentient Labs (the organization behind the OML paper) is a real open-source AI reasoning research lab publishing at NeurIPS, ICML, and COLM, and OML is one of its shipped systems.","source_url":"https://sentient.xyz","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OML: A Primitive for Reconciling Open Access with Owner Control in AI Model Distribution, arXiv 2411.03887 (2024); Nailwal is ninth of twelve authors, paper led by Cheng, Contente, ... Oh, Tyagi, Viswanath","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Polygon founded as Matic Network in 2017 by Jaynti Kanani, Sandeep Nailwal, Mihailo Bjelic and Anurag Arjun, described as software engineers; no educational or biographical detail given","source_url":"https://en.wikipedia.org/wiki/Polygon_(blockchain)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OML (arXiv:2411.03887, 2024) is Nailwal's only indexed paper — 9th of 12 authors, a cryptographic usage-authorization/fingerprinting primitive for model distribution, not language-model architecture, training, embeddings or scaling","source_url":"https://arxiv.org/abs/2411.03887","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5114645762: 1 work, 0 citations, h-index 0, earliest year 2024 — no language-modeling or vector-space publication record","source_url":"https://api.openalex.org/authors/A5114645762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Polygon (originally Matic Network) was co-founded in 2017 by Nailwal, Jaynti Kanani, Anurag Arjun and Mihailo Bjelic as software engineers; its core is Ethereum-compatible proof-of-stake scaling, outside the AI/language-modeling core","source_url":"https://en.wikipedia.org/wiki/Polygon_(blockchain)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Polygon (originally Matic Network) was co-founded in 2017 by Jaynti Kanani, Sandeep Nailwal, Mihailo Bjelic and Anurag Arjun as an Ethereum-compatible proof-of-stake scaling platform — blockchain infrastructure with no AI core","source_url":"https://en.wikipedia.org/wiki/Polygon_(blockchain)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5114645762 (Sandeep Nailwal): 1 work, 0 citations, h-index 0, earliest year 2024 — no publication record before 2024 and none in vector-space/distributional-semantics","source_url":"https://api.openalex.org/authors/A5114645762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2024,"notable_systems":["Polygon / Matic Network (co-founder, 2017)","Sentient (co-founder)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier has no person-level Wikipedia or Wikidata record for Nailwal at all: the 'wikipedia' block resolves to the Polygon (blockchain) article and the 'wikidata' block to Q111446062, which is the Polygon protocol entity (label 'Polygon', github 0xPolygon), not a human. Those blocks must not be read as biography. The OpenAlex match is correct and is not a homonym, but is a single 2024 paper, so the dossier's timeline (first_verifiable_year 2024, years_active 2) reflects only his publication record and understates roughly nine years of blockchain engineering from 2017 — engineering that is real but lies outside the AI core this index measures, so it does not lift the research dimensions. His education could not be verified from any source consulted and is left empty rather than asserted. Patents were skipped by the collector, not verified as zero.","adjudication_notes":["transformers_lm — pass_1 closer on kind, final 4 (pass_1 3, pass_2 5). Both passes identified the same single item; I verified its author list and subject matter directly. The deciding facts are that he is ninth of twelve authors and that the paper is about cryptographic usage authorization and fingerprinting for model distribution, not about the models — so no authored contribution to attention, pretraining or scaling exists. Both scores sit in the correct band; 4 reflects genuine co-authorship on an AI-adjacent paper without crediting language-modelling work he did not do.","foundations — pass_1 closer, final 3 (pass_1 1, pass_2 4). No degree, thesis or publication in linear algebra, optimization or statistical learning is verifiable, and no primary source establishes his education at all. Pass_2's 4 asserts more than the record shows; pass_1's 1 is the 'nothing verifiable' anchor, which nine years of shipped protocol engineering slightly exceeds.","vector_embeddings — near-agreement, final 2 (pass_1 1, pass_2 3). Neither pass found any embedding, retrieval or vector-search work; the OML paper's 'AI-native fingerprinting' touches model representations only incidentally and is a security mechanism.","hands_on_engineering — PASS_2 right, final 7 (pass_1 3, pass_2 8). Pass_1's 3 understates verifiable protocol-level building at Polygon from 2017, which is real shipped engineering; but it is Ethereum scaling infrastructure with no AI core, so it cannot reach the 8-12 band that requires engineering adjacent to the core of AI. This is the one dimension where the disagreement was pass_1 missing engineering rather than pass_2 inflating research.","industry_impact — near-agreement, final 5 (pass_1 4, pass_2 6). Sentient is a real lab releasing open models, but no primary source shows he leads its research rather than providing business leadership; his own citation record is zero; and Polygon, his major org, has no AI core. Polygon's token market cap and fundraising are explicitly excluded.","penalties — both passes report 0/0 and both are right, for the reason pass_1 articulated best: the profile shape (an 'AI' venture founded on crypto-derived personal wealth with one mid-list co-authored paper as the entire personal technical record) is a genuine concern, but no citable source documenting family, friends or personal-wealth funding was found, and the brief forbids asserting a penalty on rumour. Pass_2 adds the correct mitigating fact that Polygon was built and shipped before Sentient, so a technical record does precede the AI venture."],"years_language_modeling":0,"years_as_technical_founder":9,"technical_founder_companies":2,"technical_founder_roles":["Polygon / Matic Network — technical co-founder (blockchain scaling, outside the LM field) — 2017-2026","Sentient — co-founder of AI lab whose language-model science is led by academic co-authors — 2023-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":6},"confidence":0.8,"duration_ms":49657},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":3,"lm_domain_breadth":2,"scientific_founder":5},"confidence":0.78,"duration_ms":55185}],"source":"seeded","profile_url":null,"wikipedia_title":"Polygon (blockchain)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"anatoly-yakovenko","name":"Anatoly Yakovenko","title":"Co-Founder & CEO","company":"Solana Labs","sector":"crypto","rank":89,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":1,"transformers_lm":1,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":10,"industry_impact":6,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":17,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":17,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.77,"rationale":"Yakovenko holds a BS in computer science from UIUC and spent roughly a decade at Qualcomm as an engineer on wireless and distributed systems, followed by roles at Mesosphere and Dropbox — a real, verifiable, hands-on distributed-systems engineering record. He personally authored the Solana whitepaper and the 'Proof of History' technical paper (2017-2018), and Solana's consensus design is his own engineering work, not a business-only role, which supports hands_on_engineering and foundations (distributed-systems math/algorithms) scores. However none of this touches the rubric's core: there is no vector-embeddings work, no seq2seq/attention/transformer/language-model authorship or training, and no OpenAlex/arXiv record in the AI research lineage — Solana is a blockchain consensus protocol, not an AI system. industry_impact reflects a real, large engineering organization (Solana Labs) but its product is not core-AI infrastructure per the rubric's definition, so it is scored moderately for general technical leadership rather than AI industry impact.\n\nNothing in Yakovenko's verifiable record enters the language-model lineage that frontier systems (GPT/Claude/Gemini/Llama) descend from: no attention/transformer/embedding/optimizer/tokenizer/pretraining/alignment work, and no pre-2013 vector-space, distributional-semantics or term-relationship-network patents or shipped systems — his authorship (the Solana whitepaper and the Proof of History consensus paper, 2017-2018) is blockchain consensus, not language modeling, so frontier_founder scores at floor. He has zero verifiable years in any language-modeling domain — natural-language text, biological, financial, code or otherwise — so both lm_domain_depth and lm_domain_breadth score at floor. He IS, however, a genuine scientific/technical founder: he personally authored the core research and consensus design that Solana Labs runs on and set its technical direction from 2017 to the present (~9 years, one company) — but that founder role is in distributed-systems/blockchain, entirely OUTSIDE the AI/language-modeling field, which places scientific_founder in the 3-7 'technical founder outside this field' band. No sources place him inside the AI research or language-modeling lineage.","evidence":[{"claim":"BS in computer science, University of Illinois Urbana-Champaign; immigrated from Ukraine as a child","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Spent over a decade at Qualcomm as an engineer on wireless and distributed systems before later roles at Mesosphere and Dropbox","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Authored 'Solana: A new architecture for a high performance blockchain' and the 'Proof of History: A Clock for Blockchain' whitepaper, and co-founded Solana Labs starting 2017","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar match (A. Yakovenko, 20 papers, 279 citations) is name_exact:false with 3 candidates — homonym risk not independently resolved to this Yakovenko; not relied on for AI-lineage claims","source_url":"https://www.semanticscholar.org/author/114558484","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS in computer science, University of Illinois Urbana-Champaign; more than a decade at Qualcomm as an engineer on wireless and distributed systems; engineering roles at Mesosphere and Dropbox from 2016; developed Proof of History in 2017 and co-founded Solana Labs; authored 'Solana: A new architectu","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 114558484 lists 'Solana: A new architecture for a high performance blockchain v0.8' (2018) as his sole computing paper; the remaining ~19 works under that id are 1980-2011 Russian-language metallurgy papers (blast-furnace stoves, lime kilns, steel-teeming ladles) by a differe","source_url":"https://api.semanticscholar.org/graph/v1/author/114558484/papers?fields=title,year,venue,authors","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata identifies him only as 'Co-Founder - Solana', with no recorded doctorate, advisor, research occupation or publication identifiers","source_url":"https://www.wikidata.org/wiki/Q115947586","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Solana Labs (2017) and personally authored 'Proof of History: A Clock for Blockchain' and the Solana architecture whitepaper — a distributed-systems/blockchain-consensus technical-founder role, not AI/language-modeling","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him only as 'Co-Founder - Solana' with no doctorate, research occupation or publication identifiers; no AI/LM lineage work is attributable","source_url":"https://www.wikidata.org/wiki/Q115947586","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Yakovenko authored 'Solana: A new architecture for a high performance blockchain' and the 'Proof of History' whitepaper and co-founded Solana Labs (2017) — a blockchain consensus protocol, not an AI/language-model system","source_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records him only as 'Co-Founder - Solana' with no research occupation, doctorate, advisor or publication identifiers; no AI/ML/language-modeling affiliation exists","source_url":"https://www.wikidata.org/wiki/Q115947586","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 114558484 lists only the Solana architecture paper as his computing work (the remaining ~19 works are unrelated Russian-language metallurgy papers by a homonym); no embeddings/transformer/LM authorship in any source","source_url":"https://www.semanticscholar.org/author/114558484","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, University of Illinois Urbana-Champaign"],"canonical_papers":[],"first_verifiable_year":2017,"notable_systems":["Solana blockchain (Proof of History consensus)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex block shows not found; Semantic Scholar match (h-index 1, name_exact:false, 3 candidates) is unresolved homonym risk and not usable for AI-lineage attribution. Wikipedia/Wikidata identity match is solid and consistent (Solana co-founder). No AI/ML-specific technical record (embeddings, transformers, LM training) was found for this person in any source — his verifiable expertise is distributed systems and blockchain consensus, adjacent to but outside the rubric's AI core.","years_language_modeling":0,"years_as_technical_founder":9,"technical_founder_companies":1,"technical_founder_roles":["Solana Labs — co-founder / lead architect (authored Proof of History consensus, sets technical direction) — 2017-present (blockchain, outside the AI/LM field)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.9,"duration_ms":42041},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.9,"duration_ms":42885}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Anatoly_Yakovenko","wikipedia_title":"Anatoly Yakovenko","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"nikil-viswanathan","name":"Nikil Viswanathan","title":"Co-founder & CEO","company":"Alchemy","sector":"crypto","rank":90,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":9,"industry_impact":7,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":17,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":17,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.72,"rationale":"Viswanathan holds a BS and MS in Computer Science from Stanford focused on distributed systems and computer networks — real, verifiable technical education, though at the master's (not PhD) level and not in AI/ML specifically. He worked as an engineer on core products at Google, Microsoft and Facebook before co-founding Down To Lunch (a consumer social app) and then Alchemy, a blockchain infrastructure company (node/API infrastructure, not an AI/ML company). No evidence was found of him authoring papers, patents, or systems in vector embeddings, transformers, or language modeling; Alchemy's core product is Web3 developer infrastructure (RPC nodes, indexing) rather than AI systems, so hands_on_engineering and industry_impact are scored for legitimate infrastructure-engineering leadership rather than AI-specific depth. He is a real hands-on technical co-founder (not merely a business-side CEO), which supports moderate hands_on_engineering credit, but there is no verifiable connection between his work and the core-AI lineage this rubric measures.\n\nNo verifiable contribution places Viswanathan in the frontier-model lineage: no vector-space, distributional-semantics, embedding, attention, transformer, optimizer, tokenizer, dataset or alignment work appears under his name in any primary source (the Semantic Scholar 'N. Viswanathan' and PubMed 'Viswanathan N' hits — linguistics, MOF photocatalysis, chemistry — are homonyms the dossier itself flags), so frontier_founder is 1. He has zero verifiable years in language modeling of any kind — his Stanford MS is in distributed systems/networks and Alchemy is Web3 node/RPC/indexing infrastructure, not any LM system — so lm_domain_depth is 1 and lm_domain_breadth is 1 with no qualifying domains. He is, however, a genuine hands-on technical co-founder (Down To Lunch ~2015, then Alchemy 2017–present, ~9–11 years), not a business-only CEO — but that founding is entirely outside language modeling, which the scientific_founder anchor caps at 3–7 ('a technical founder outside this field'), so 5. No in-field (language-modeling) company was founded.","evidence":[{"claim":"BS and MS in Computer Science from Stanford University, focused on distributed systems and computer networks","source_url":"https://www.clay.com/dossier/alchemy-ceo","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founder and CEO of Alchemy, a blockchain developer platform (Web3 node/API infrastructure) used by JPMorgan, Robinhood, Visa, Stripe, Polymarket","source_url":"https://www.alchemy.com/company","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records Nikil Viswanathan as an American software engineer, born 6 October 1987 in Chicago, educated at Stanford University and Stanford University School of Engineering, co-founder and CEO of Alchemy, notable work 'Down To Lunch'","source_url":"https://www.wikidata.org/wiki/Q30069857","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Wikipedia article titled 'Alchemy (company)' is an unrelated American film distributor that filed for bankruptcy in 2016, confirming there is no Wikipedia record of his blockchain company or of any technical/AI work by him","source_url":"https://en.wikipedia.org/wiki/Alchemy_(company)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alchemy's Semantic Scholar name match ('N. Viswanathan', 7 papers, 6 citations, h-index 1) is a non-exact match and does not correspond to him; no AI-lineage publication is attributable","source_url":"https://api.semanticscholar.org/graph/v1/author/40791257?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alchemy is a blockchain/Web3 developer platform (node/API/RPC infrastructure), not an AI or language-modeling company; Viswanathan is co-founder and CEO","source_url":"https://www.wikidata.org/wiki/Q30069857","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records Viswanathan as an American software engineer, notable work 'Down To Lunch' (a consumer social app), educated at Stanford — no AI/ML or language-modeling research attributed","source_url":"https://www.wikidata.org/wiki/Q30069857","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The only Semantic Scholar match ('N. Viswanathan', 7 papers, 6 citations, h-index 1) is a non-exact match and PubMed results are flagged homonym_risk:true — no AI-lineage or LM publication is attributable to him","source_url":"https://api.semanticscholar.org/graph/v1/author/40791257?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists Nikil Viswanathan as an American software engineer, educated at Stanford, with notable work 'Down To Lunch' — no AI/language-modeling research, patents or papers recorded","source_url":"https://www.wikidata.org/wiki/Q30069857","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alchemy is a Web3 developer platform providing blockchain node/RPC/API and indexing infrastructure (not AI/language-modeling systems), co-founded and led by Viswanathan","source_url":"https://www.alchemy.com/company","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar 'N. Viswanathan' (7 papers, h-index 1, name_exact:false) and PubMed 'Viswanathan N' hits are homonyms unrelated to the Alchemy founder; no frontier-lineage or LM publication is attributable","source_url":"https://api.semanticscholar.org/graph/v1/author/40791257?fields=name,paperCount,citationCount,hIndex","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Stanford University, BS Computer Science","Stanford University, MS Computer Science"],"canonical_papers":[],"first_verifiable_year":2013,"notable_systems":["Alchemy Web3 developer platform","Down To Lunch (prior consumer app)"],"citations":6,"h_index":1,"patents":0,"dossier_notes":"Dossier's Semantic Scholar match (name_exact:false, id 40791257, 7 papers/6 citations) and the PubMed sample (mandarin tone sandhi linguistics, MOF photocatalysis, chemistry papers under 'Viswanathan N') are homonym noise unrelated to the Alchemy founder, correctly flagged homonym_risk:true in the dossier. No Wikipedia page exists; Wikidata Q30069857 ('Down To Lunch' notable work) correctly identifies him. His technical education is real but there is no verifiable AI/ML research record — Alchemy is blockchain infrastructure, not an AI company, despite general crypto-industry visibility.","years_language_modeling":0,"years_as_technical_founder":11,"technical_founder_companies":0,"technical_founder_roles":["Down To Lunch — technical co-founder (consumer social app, outside language modeling) — ~2015-2017","Alchemy — co-founder & CEO / hands-on technical founder (Web3 node/API infrastructure, outside language modeling) — 2017-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.9,"duration_ms":47135},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.83,"duration_ms":49065}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"paolo-ardoino","name":"Paolo Ardoino","title":"CEO, Tether / CTO, Bitfinex","company":"Tether","sector":"crypto","rank":91,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":8,"industry_impact":4,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":17,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":17,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.81,"rationale":"Paolo Ardoino holds a Computer Science degree from the University of Genoa (2008) and has a genuine hands-on software engineering background — early work as a researcher on a military high-availability/self-recovering-networks/cryptography project, then joining Bitfinex in 2014 as a senior developer optimizing its matching engine, rising to CTO. This shows real systems/infrastructure engineering competence, but none of it is in the core AI lineage: no publications, degrees, or shipped systems in linear algebra/optimization/statistical learning, vector embeddings, or transformer/language-model research were found in Wikipedia, Wikidata, OpenAlex, PubMed, or web search. His current role leading Tether (stablecoin infrastructure) and Bitfinex is financial-exchange and crypto-infrastructure engineering, not an AI company whose core is these systems. hands_on_engineering is scored moderately for genuine matching-engine/distributed-systems work; all AI-specific dimensions are scored near floor since there is no verifiable AI research or engineering record.\n\nNo verifiable contribution sits anywhere in the lineage today's frontier language models descend from: there is no OpenAlex/PubMed/Semantic Scholar record, no patents, and no pre-2013 vector-space, distributional-semantics or term-relationship-network work — the Tether AI Research offline-translation models (Sept 2026) are organizational output, not personally authored by Ardoino, so frontier_founder and lm_domain_depth score at floor (0 verifiable language-modeling years). His verifiable hands-on record is trading/matching-engine and stablecoin/distributed-systems infrastructure (Bitfinex from 2014, CTO 2016; Tether CEO from Dec 2023), which is not language modeling in any domain, so lm_domain_breadth is at floor with zero distinct LM domains. He does operate as a genuine technical founder/executive — he founded Fincluster (2013) and personally set and wrote core technical direction (matching engine) at Bitfinex — but that is a technical founder OUTSIDE this field (crypto/fintech infrastructure), which caps scientific_founder in the 3-7 band. Counting one company he technically founded and technical-executive years all outside language modeling, scientific_founder is scored 5.","evidence":[{"claim":"Computer Science degree, University of Genoa, graduated 2008","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Joined Bitfinex in 2014 as Senior Software Developer optimizing the matching engine; promoted to CTO in 2016; CEO of Tether since December 2023","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex, PubMed, or Semantic Scholar record found for Paolo Ardoino","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Bachelor's degree in computer science from the University of Genoa; worked as a researcher there on cybersecurity and cryptography; developed trading algorithms at a hedge fund; founded Fincluster in 2013; joined Bitfinex 2014 as software engineer, CTO from 2016 working on the trading engine and bac","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q126537585 records only University of Genoa as education, with no doctorate and no research occupation listed","source_url":"https://www.wikidata.org/wiki/Q126537585","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PubMed search 'Ardoino P[Author]' returns 0 results; no OpenAlex or Semantic Scholar author record exists for him","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Ardoino+P%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Tether AI Research released open-source offline translation model families (TranslatePsy-AfriSLM, TranslatePsy-EuroNano) in September 2026 — organisational output, not personally authored by Ardoino","source_url":"https://tether.io/news/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub account PaoloArdoino shows a single public repository","source_url":"https://github.com/paoloardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ardoino has no OpenAlex, Semantic Scholar or PubMed author record and no patents; 'Ardoino P[Author]' returns 0 PubMed results","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Ardoino+P%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"CS degree University of Genoa; founded Fincluster (2013); joined Bitfinex 2014 as software engineer, CTO from 2016 working on the trading engine; CEO of Tether since Dec 2023 — infrastructure, not AI/LM research","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q126537585 lists only University of Genoa as education, no doctorate and no research occupation","source_url":"https://www.wikidata.org/wiki/Q126537585","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Fincluster in 2013; joined Bitfinex 2014 as software engineer, CTO from 2016 working on the trading/matching engine; CEO of Tether since December 2023 — a technical founder/CTO record in crypto/fintech, not AI/LM","source_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex, Semantic Scholar or PubMed author record and no patents exist for Paolo Ardoino — nothing in the frontier-model or language-modeling lineage","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Ardoino+P%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Computer Science degree, University of Genoa (2008)"],"canonical_papers":[],"first_verifiable_year":2008,"notable_systems":["Bitfinex matching engine","Tether (USDT) infrastructure"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier correctly shows no OpenAlex/PubMed/Semantic Scholar record for this person; no homonym issue. Tether/Bitfinex are not AI companies and no evidence surfaced of any personal AI/ML work, so this profile is scored per the 'senior engineering adjacent to core, not the core itself' anchor at best, and lower where there is zero AI-specific evidence (vector_embeddings, transformers_lm).","years_language_modeling":0,"years_as_technical_founder":13,"technical_founder_companies":1,"technical_founder_roles":["Fincluster — founder / technical lead (fintech, outside LM field) — 2013","Bitfinex — CTO (matching-engine / crypto infra, outside LM field) — 2016-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":5},"confidence":0.8,"duration_ms":53143},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":5},"confidence":0.85,"duration_ms":49957}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Paolo_Ardoino","wikipedia_title":"Paolo Ardoino","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sundar-pichai","name":"Sundar Pichai","title":"CEO of Alphabet Inc. and Google","company":"Alphabet & Google","sector":"general","rank":92,"rubric_version":4,"dimensions":{"foundations":5,"vector_embeddings":2,"transformers_lm":4,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"hands_on_engineering":4,"industry_impact":11,"scientific_founder":1},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":17,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":17,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.89,"rationale":"Pichai holds a B.Tech in metallurgical engineering from IIT Kharagpur, an MS in materials science and engineering from Stanford, and an MBA from Wharton — engineering-adjacent graduate training but not in CS, ML, or applied mathematics, and no doctoral or thesis-level research record of any kind. His Google career (Toolbar, Chrome, Chrome OS, Google Drive, Gmail/Maps oversight, Android, then CEO) is documented as product management and executive leadership, not individual-contributor engineering or research; no personally authored papers, no patents, and no Google Scholar profile were found. The dossier's OpenAlex match (4 works, all 2019-2021 congressional-testimony reprints and a Fox Business interview) is not scientific authorship and is excluded from scoring; the PubMed 'Pichai S' records (dental/orthodontic finite-element papers, hepatitis seroprevalence, orthopedic surgery, all India-affiliated, 2012-2025) are a clear homonym — a different person entirely — and are also excluded. Industry_impact is scored moderately: as CEO he leads the organization that ships Gemini, TensorFlow, and Search/Ads infrastructure and that employed the original Transformer authors and Google DeepMind, but this credits organizational leadership of a company whose core is these systems, not personal authorship of the underlying research (Attention Is All You Need, BERT, etc. were built by named researchers, not Pichai). All core research dimensions (foundations, vector_embeddings, transformers_lm, hands_on_engineering) reflect the complete absence of a personal, verifiable technical/research record per the rubric's explicit instruction that fame and company branding do not count.\n\nPichai has no personally authored method, architecture, dataset, optimizer, tokenizer, embedding, or training/inference system that today's frontier language models descend from — the transformer, BERT, TensorFlow and Gemini foundations were authored by named Google researchers, not by him, and CEO oversight of the org that produced them is not a verifiable position in the technical lineage (frontier_founder=2). He has zero verifiable hands-on years in language modeling of any kind: his degrees are in metallurgical/materials engineering and business, and his Google career is documented product management and executive leadership, not IC research or systems work in statistical/neural LMs, vector-space models, or embeddings (years_language_modeling=0, lm_domain_depth=2). He has no hands-on record in any distinct language-modeling domain — natural-language, biological, financial, code or otherwise (lm_domain_breadth=1). He is a hired executive, not the scientific/technical founder of any company: he founded no company and authored none of the core research, code or patents the systems run on, so scientific_founder is effectively nil (years_as_technical_founder=0, companies=0). The dossier's OpenAlex works are congressional-testimony reprints and a press interview, and the PubMed 'Pichai S' records are an unrelated Indian medical/dental homonym — both excluded.","evidence":[{"claim":"B.Tech in metallurgical engineering, IIT Kharagpur; MS materials science and engineering, Stanford University; MBA, Wharton School (Siebel Scholar, Palmer Scholar)","source_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata confirms MS from Stanford (materials science field), MBA from Wharton (business management field), employer Google from 2004, CEO of Google 2015 and Alphabet 2019","source_url":"https://www.wikidata.org/wiki/Q3503829","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career at Google was product management and leadership across Chrome, Chrome OS, Google Drive, Gmail/Maps oversight, and Android, not individual-contributor engineering or research; no research papers, patents, or hands-on AI/ML engineering documented","source_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PubMed 'Pichai S[Author]' records (17 results, 2012-2025) are dental/orthodontic, hepatitis, and orthopedic-surgery papers by India-affiliated co-authors (Vetriselvan A, Peddu R, Bose VC, etc.) — a different person, not Google's Sundar Pichai","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Pichai+S%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"B.Tech metallurgical engineering IIT Kharagpur (1993), MS materials science and engineering Stanford, MBA Wharton; engineering/product roles at Applied Materials then McKinsey; joined Google 2004 as a product manager over Chrome, ChromeOS, Drive, Gmail, Maps; added Android 2013; CEO of Google 2015 a","source_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records education limited to Stanford (MS, materials science) and Wharton (MBA), with employers Google (from 2004) and Alphabet (from 2019) and no doctorate or academic affiliation","source_url":"https://www.wikidata.org/wiki/Q3503829","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The transformer paper that anchors this lineage was authored by Google Brain and Google Research staff (Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, Polosukhin), not by Pichai","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Education limited to Stanford MS (materials science) and Wharton MBA; occupation executive/CEO; employer Google from 2004, Alphabet from 2019 — no doctorate, no CS/ML research affiliation, no founded company","source_url":"https://www.wikidata.org/wiki/Q3503829","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Career at Google is product management and executive leadership (Toolbar, Chrome, Chrome OS, Drive, Android, then CEO); no authored research papers, patents, or hands-on AI/ML engineering documented","source_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The transformer paper anchoring this lineage was authored by Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser and Polosukhin — not by Pichai","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Pichai joined Google in 2004 as a product manager and rose through product/executive roles (Chrome, ChromeOS, Android) to CEO of Google (2015) and Alphabet (2019); he is not a company founder and has no authored research papers or patents","source_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Transformer architecture underpinning today's frontier language models was authored by Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser and Polosukhin — not by Pichai","source_url":"https://arxiv.org/abs/1706.03762","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records his education as an MS in materials science (Stanford) and an MBA (Wharton), with employers Google (2004) and Alphabet (2019) and no doctorate, academic affiliation, or founder role","source_url":"https://www.wikidata.org/wiki/Q3503829","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["B.Tech, Metallurgical Engineering, IIT Kharagpur","MS, Materials Science and Engineering, Stanford University","MBA, The Wharton School, University of Pennsylvania (Siebel Scholar, Palmer Scholar)"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["Google Chrome (product leadership, not engineering)","Chrome OS (product leadership)","Android (executive oversight from 2013)","Google/Alphabet CEO — oversees Gemini, TensorFlow, DeepMind, Search/Ads infra as organizational leader, not personal author"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's OpenAlex match (works_count 4, e.g. 'Heads of Facebook, Twitter, and Google Testify on Content Moderation') is clearly not research — these are congressional-testimony reprints and a Fox Business press piece, a homonym/false-match artifact against name-matching, not authored scientific work; excluded from scoring. The dossier's PubMed sample (17 results under 'Pichai S[Author]': dental/orthodontic finite-element-analysis papers, hepatitis seroprevalence, orthopedic surgery papers, all India-affiliated co-authors 2012-2025) is a different person — an Indian medical/dental researcher homonym — not Google's Sundar Pichai; excluded from scoring. No patents or Google Scholar profile were found for the actual Sundar Pichai in any source checked.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":1},"confidence":0.93,"duration_ms":45238},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":1},"confidence":0.95,"duration_ms":34327}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Sundar_Pichai","wikipedia_title":"Sundar Pichai","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"ahmad-shadid","name":"Ahmad Shadid","title":"Founder & CEO, O.XYZ; founder and former CEO, io.net","company":"O.XYZ","sector":"crypto","rank":93,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":7,"industry_impact":5,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":16,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":16,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.5,"rationale":"I could not verify any personal research record in the AI lineage for this person, and the two academic identities the dossier supplies are both homonyms, so the research dimensions are scored near the bottom on absence of evidence rather than on disproof. The dossier's OpenAlex profile belongs to 'Ahmad Jamal Shadid', a University of Ottawa master's student whose 2008 thesis is on load balancing for real-time HLA/RTI distributed simulation - verified directly from the uOttawa repository record (doi 10.20381/ruor-19046) - and whose five works from 2007-2008 are on DEVS, Petri nets and discrete-event simulation, not machine learning; the PubMed cluster is a third, medical, Shadid. What is verifiable about the subject is organisational and infrastructural rather than scientific: io.net, which he founded, is a decentralised GPU compute platform whose own documentation describes its technical basis as adopting the open-source Ray library to distribute workloads across heterogeneous GPUs, and O.XYZ is a follow-on venture. Assembling and operating distributed GPU capacity is real systems work adjacent to the training stack, which is why hands_on_engineering is scored at the 'uses the tools / senior engineering adjacent to the core' band rather than lower, but it is compute brokerage, not model design: I found no paper, patent, preprint, public repository or shipped model by him on embeddings, retrieval, attention, transformers, pretraining, scaling or alignment. Industry impact is low on this rubric's terms because the rubric excludes fundraising, token market capitalisation and 'AI company' branding as evidence, and what remains - an infrastructure marketplace - has ML systems as its customers rather than its core. I report both penalties as zero because I have no citable source for pay-for-play, purchased reach or family funding; that is an absence of evidence, not a clearance, and my overall confidence is correspondingly low.\n\nNo verifiable contribution by Ahmad Shadid sits in the frontier-model lineage: the OpenAlex/Semantic Scholar/PubMed records are all confirmed homonyms (a uOttawa distributed-simulation student, a 'K. Shadid', and a medical-genetics cluster), and his own venture io.net is a decentralised GPU-compute marketplace built on the open-source Ray library — inference/training capacity that others' models run on, not an authored architecture, optimizer, tokenizer, embedding or alignment method that GPT/Claude/Gemini-class systems descend from, and no pre-2013 vector-space/distributional-semantics patent or shipped system is attributable to him. His years of hands-on language-modeling work — statistical/neural LMs, seq2seq, transformers, pretraining or vector-space text systems — are zero on the verifiable record, so lm_domain_depth sits in the 0-2 nothing-verifiable band. Likewise no distinct language-modeling domain (natural-language, biological, financial, code, etc.) has a hands-on modeling record tied to him, so breadth is at the floor. He is the founder/CEO of io.net and the follow-on O.XYZ/ORGN, but the science and engineering rest on existing distributed-computing tooling and there is no paper, patent, preprint or repository showing he personally authored the core — placing him in the 3-7 'founder of an AI company whose science was done by others' band, not a technical/scientific founder, so years_as_technical_founder and technical_founder_companies are reported as 0.","evidence":[{"claim":"The dossier's OpenAlex author A5069023772 is a University of Ottawa researcher with 5 works (2007-2008) on distributed simulation, DEVS, HLA/RTI load balancing and Petri nets - topics with no machine-learning content","source_url":"https://api.openalex.org/authors/A5069023772","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"That OpenAlex identity's thesis (doi 10.20381/ruor-19046) is by 'Ahmad Jamal Shadid', a 2008 University of Ottawa master's thesis - establishing the OpenAlex cluster as an academic homonym, not the io.net/O.XYZ founder","source_url":"http://ruor.uottawa.ca/handle/10393/28023","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"io.net's own documentation describes its technical basis as adopting the open-source Ray library to distribute AI workloads across GPUs, cutting infrastructure build time, motivated by the cost of NVIDIA A100 capacity - i.e. a compute-aggregation platform built on existing distributed-computing tool","source_url":"https://io.net/docs","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"O.XYZ's site now redirects to orgn.com, which presents a 'confidential agentic stack' product for defense and regulated teams and names no individual founders","source_url":"https://orgn.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"io.net, founded by Ahmad Shadid, is a decentralized GPU-compute platform whose own documentation describes adopting the open-source Ray library to distribute AI workloads across heterogeneous GPUs — compute aggregation built on an existing distributed-computing tool, not a model, architecture or tra","source_url":"https://io.net/docs","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The OpenAlex identity A5069023772 attached to the dossier is 'Ahmad Jamal Shadid', a 2008 University of Ottawa master's thesis on HLA/RTI distributed-simulation load balancing — a homonym with no machine-learning or language-modeling content, so it supplies no LM years for the subject","source_url":"http://ruor.uottawa.ca/handle/10393/28023","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The OpenAlex identity A5069023772 attributed in the dossier is 'Ahmad Jamal Shadid', a 2008 University of Ottawa master's student working on HLA/RTI distributed-simulation load balancing (thesis doi 10.20381/ruor-19046) — a homonym with no machine-learning or language-modeling content, so no LM year","source_url":"https://api.openalex.org/authors/A5069023772","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"O.XYZ redirects to orgn.com, presenting a 'confidential agentic stack' product and naming no individual as its technical/scientific founder, consistent with a business-founder rather than a core-code/paper author role","source_url":"https://orgn.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["io.net decentralised GPU compute network (as founder)","O.XYZ / ORGN (as founder)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Both scholarly blocks are WRONG PEOPLE. (1) OpenAlex A5069023772 is 'Ahmad Jamal Shadid', a University of Ottawa master's student (thesis 2008, doi 10.20381/ruor-19046, verified at ruor.uottawa.ca/handle/10393/28023) working on HLA/RTI distributed-simulation load balancing - the dossier's first_verifiable_year of 2007 and years_active of 19 come entirely from this homonym and are invalid for the subject. (2) The Semantic Scholar entry is filed under 'K. Shadid', name_exact false, and is a different person again. (3) The PubMed cluster (92 hits: complement-system genetics, immune-checkpoint cardiotoxicity, dermatology) is a third set of homonyms. No Wikipedia or Wikidata entry exists. I also could not independently confirm the subject's own education or technical background from any primary source: the current O.XYZ domain redirects to orgn.com and no longer names him, and the press coverage of his io.net tenure was unreachable from this session (archive.org is blocked and the outlet URLs I attempted returned 404/403). Scores therefore rest on absence of a verifiable record, and confidence is set low to reflect that.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":4},"confidence":0.5,"duration_ms":68494},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.6,"duration_ms":54370}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"elon-musk","name":"Elon Musk","title":"Founder & CEO, xAI; CEO, Tesla and SpaceX","company":"xAI, Tesla, SpaceX","sector":"general","rank":94,"rubric_version":4,"dimensions":{"foundations":3,"vector_embeddings":1,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"hands_on_engineering":5,"industry_impact":10,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":16,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":16,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.9,"rationale":"Musk holds a BA in physics and a BS in economics from the University of Pennsylvania (1997) and never enrolled in the Stanford graduate programme he was admitted to in 1995, so there is no thesis, graduate coursework or publication record in linear algebra, optimization or statistical learning behind him. His entire indexed scholarly output is a single 2019 Neuralink white paper, 'An Integrated Brain-Machine Interface Platform With Thousands of Channels' (JMIR, doi 10.2196/16194) — electrode arrays and signal acquisition hardware, published under Neuralink's collective authorship convention for its founder-CEO, and neuroscience hardware rather than machine learning in any case; his OpenAlex profile's five works are duplicate versions of that one document plus two 2025 Zenodo preprints of uncertain provenance that neither pass could attribute to him. I found no paper, preprint, patent as inventor or public code by him on embeddings, attention, transformers, pretraining, scaling or alignment; on this rubric's spine his personal record is effectively empty. His association with the lineage is as founder and funder — OpenAI co-founder in 2015, departing the board in 2018, and xAI founder and CEO from 2023 — with no documented hands-on research role at either; Grok was built by researchers and engineers hired from Google, DeepMind and OpenAI. Hands-on engineering earns a little more than the research axes but not much: his documented personal coding is Zip2-era web software in the 1990s, and his technical role at Tesla and SpaceX is documented as detailed engineering direction rather than personally designing the Dojo training stack or the FSD networks. Industry impact is genuine on the rubric's terms because Tesla's autonomy programme and xAI's Grok have machine learning at their core and he built those organizations, but it is organizational impact with no citation, inventor-patent or canonical-paper record underneath it; his net worth, fame and 'AI company' branding are excluded entirely.\n\nNothing in the language-model lineage that today's frontier systems descend from — attention, transformers, word2vec/GloVe-class embeddings, tokenizers, optimizers, scaling laws, RLHF, or any pre-2013 vector-space / distributional-semantics / term-relationship patent or shipped system — is attributable to Musk; his sole substantive indexed paper is a 2019 Neuralink brain-machine-interface hardware white paper, and no paper, preprint, inventor-patent or public code by him on embeddings, attention, transformers, pretraining or alignment exists, so his frontier position is founder/funder only (OpenAI 2015–2018 board, xAI from 2023), not authorship. His verifiable hands-on years in language modeling specifically are effectively zero: the timeline's first verifiable year (2019) is Neuralink neuroscience hardware, and founding xAI in 2023 is organizational, not personal LM research. There are no distinct language-modeling domains with a hands-on record of his own (biological, financial, code, natural-language text or otherwise) — the Neuralink work is neural-signal decoding hardware, not sequence/language modeling. As a scientific/technical founder in this field he scores in the 'AI company whose science and engineering were done by others' band: xAI's Grok was built by researchers and engineers hired from Google, DeepMind and OpenAI, and the prior assessment found his role documented as engineering direction rather than personally authoring the models, training stack or FSD networ","evidence":[{"claim":"BA in physics and BS in economics, University of Pennsylvania (1997); admitted to a Stanford materials-science graduate programme in 1995 but never enrolled; co-founded OpenAI in December 2015 as funder and board member and departed the board in 2018; founded xAI in 2023","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"His entire OpenAlex record (A5026992422) is 5 works with h-index 3, comprising duplicate versions of the 2019 Neuralink brain-machine-interface paper plus two 2025 Zenodo preprints with 0 citations each","source_url":"https://api.openalex.org/authors/A5026992422","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"'An Integrated Brain-Machine Interface Platform With Thousands of Channels', J Med Internet Res 2019;21(10):e16194 — his sole substantive indexed publication, describing electrode and signal-acquisition hardware under Neuralink's collective authorship","source_url":"https://doi.org/10.2196/16194","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record shows 2 papers and 0 citations, confirming no independent research corpus","source_url":"https://www.semanticscholar.org/author/Elon-Musk/2064796567","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BA Physics, BS Economics, University of Pennsylvania (1997); no completed graduate degree (admitted to Stanford materials science PhD 1995, did not enroll)","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded OpenAI in Dec 2015 as funder/board member, pledged $1B but donated far less, departed board 2018; no documented hands-on research role","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded xAI (2023) as CEO; built Grok by hiring researchers/engineers from Google and OpenAI rather than personally engineering the models","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BA physics and BS economics, University of Pennsylvania 1997; accepted to Stanford materials-science graduate programme but did not enrol; co-founded OpenAI 2015 and left its board 2018; launched xAI July 2023; father contributed 10% of a later Zip2 funding round","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Entire OpenAlex record for Elon Musk is 5 works / h-index 3, all versions of the 2019 Neuralink brain-machine-interface paper plus two 2025 Zenodo items; topics are brain-computer interfaces and neural engineering, not machine learning","source_url":"https://api.openalex.org/authors/A5026992422","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Entire OpenAlex record (A5026992422) is 5 works / h-index 3, reducing to duplicate versions of the 2019 Neuralink brain-machine-interface paper plus two unattributable 2025 Zenodo preprints — topics are brain-computer interfaces and neural engineering, not language modeling or machine learning","source_url":"https://api.openalex.org/authors/A5026992422","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Musk co-founded OpenAI in Dec 2015 as funder/board member and left the board in 2018, and founded xAI in 2023 as CEO; no documented hands-on research role, and Grok was built by hired researchers/engineers","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Sole substantive indexed publication is 'An Integrated Brain-Machine Interface Platform With Thousands of Channels', J Med Internet Res 2019;21(10):e16194 — electrode and signal-acquisition hardware under Neuralink collective authorship, not language modeling","source_url":"https://doi.org/10.2196/16194","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Musk co-founded OpenAI in Dec 2015 as funder/board member and departed the board in 2018; founded and leads xAI from 2023 — organizational roles, with no documented hands-on research contribution to embeddings, transformers, pretraining, scaling or alignment","source_url":"https://en.wikipedia.org/wiki/Elon_Musk","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Entire OpenAlex record (A5026992422) is 5 works reducing to the 2019 Neuralink brain-machine-interface paper plus two unattributable 2025 Zenodo preprints; topics are brain-computer interfaces and neural engineering, not language modeling","source_url":"https://api.openalex.org/authors/A5026992422","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"xAI's Grok was built by researchers and engineers recruited from Google DeepMind and OpenAI, not personally engineered or authored by Musk","source_url":"https://en.wikipedia.org/wiki/XAI_(company)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BA Physics, University of Pennsylvania, 1997","BS Economics, Wharton School, University of Pennsylvania, 1997","Admitted to Stanford materials-science PhD programme; did not enrol"],"canonical_papers":[],"first_verifiable_year":2019,"notable_systems":["Neuralink N1 brain-machine interface (white-paper byline)","Tesla Autopilot/FSD and Dojo (as CEO, not as personal author)","xAI Grok (as founder/CEO)"],"citations":1288,"h_index":3,"patents":0,"dossier_notes":"The dossier's PubMed block is a complete HOMONYM despite homonym_risk being reported as false: 'Musk E' on the SARS-CoV-2 antibody and Fc-effector-function papers belongs to an immunology group (Fischinger, Kaplonek, Bartsch), and the dossier's earliest_pubmed_year of 2019 comes from that person, coinciding with the Neuralink year only by accident. The OpenAlex match is flagged ambiguous (5 candidates) and its five 'works' reduce to one Neuralink white paper plus two unattributable 2025 Zenodo preprints, so the 1,288 citations rest on a single collective-authorship document. Wikidata lists 'programmer' and 'inventor' as occupations, which is not evidence of AI research. Patents were skipped by the collector, not verified absent.","adjudication_notes":["foundations — SPLIT, final 3 (pass_1 2, pass_2 5). The verified record is an undergraduate physics and economics degree and a graduate programme he never enrolled in — no thesis, no coursework, no publication in the field's mathematics. Pass_2's 5 credits an unearned quantitative reputation; pass_1's 2 sits at the 'nothing verifiable' anchor, which an actual physics BA slightly exceeds.","transformers_lm — pass_1 right, final 2 (pass_1 1, pass_2 4). Both passes searched independently and both concluded, in their own words, that no paper, preprint, patent or public code by him on embeddings, attention, transformers, pretraining, scaling or alignment exists. Pass_2 then scored 4, which its own finding does not support: founding and funding xAI is organizational, and the rubric assigns that to industry_impact, not to this dimension. 2 is the 0-2 'nothing verifiable' anchor the evidence actually shows.","hands_on_engineering — SPLIT, final 5 (pass_1 3, pass_2 8). Pass_2 is right that documented 1990s Zip2 coding and sustained detailed engineering direction at Tesla and SpaceX exceed pass_1's 3; but pass_2's own rationale concedes the record shows 'engineering direction rather than personally designing the Dojo training stack or FSD networks', and 8 enters the band reserved for senior engineering adjacent to the AI core, which that concession rules out. 5 is mid 'manages builders'.","industry_impact — SPLIT, final 10 (pass_1 8, pass_2 12). Tesla's autonomy programme and xAI's Grok have machine learning at their core and he built both organizations, which is the rubric's permitted org-building ground and exceeds pass_1's 8. But he has no citations as an author, no patents as inventor and no canonical paper beneath that impact, and OpenAI's canonical output postdates his 2018 board departure and was led by others, so pass_2's 12 — level with executives who lead labs producing canonical work — is too high.","vector_embeddings — pass_1 right, final 1 (pass_1 1, pass_2 2). Nothing in embeddings, retrieval or vector search is attributable to him.","penalties — both passes agree at 0/0 and both are right. Pass_2 examined the closest candidate fact, that his father contributed to a later Zip2 funding round, and correctly found it fails the test: it is a small, contested, long-ago contribution to a non-AI company, and the rubric requires cited evidence of family money funding an AI company with no technical record. No citable evidence of paid coverage or purchased reach was found by either pass."],"years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["xAI — founder & CEO (2023–), core research authored by hired researchers, not by him","OpenAI — co-founder/funder & board member (2015–2018), no hands-on research role"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.83,"duration_ms":64991},{"pass":"pass_2","dimensions":{"frontier_founder":3,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.83,"duration_ms":41444}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Elon_Musk","wikipedia_title":"Elon Musk","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"yu-hu","name":"Yu Hu","title":"Founder & CEO","company":"Kaito AI","sector":"crypto","rank":95,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":4,"transformers_lm":3,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":6,"industry_impact":6,"scientific_founder":3},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":16,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":16,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.45,"rationale":"The dossier's Wikipedia/Wikidata block ('Yu (Jeffrey) Hu', Purdue/Georgia Tech business professor, PhD MIT Sloan under Erik Brynjolfsson) is a homonym and is NOT the Kaito AI founder — the real Yu Hu of Kaito AI studied at the University of Cambridge, then worked as an investment banking analyst at Deutsche Bank and a portfolio manager at Citadel before founding Kaito AI, an AI-powered search/analytics engine for digital assets. No verifiable degree in computer science, mathematics, or a technical PhD was found for him; his background is finance, not AI research. Kaito AI is a real, shipped product (crypto-market search/analytics using LLMs), so there is some hands-on-engineering/industry-impact credit for building and leading an AI-product company, but no evidence of Yu Hu personally authoring papers, code, or research in vector embeddings, transformers, or LM training — the rubric requires personal, verifiable technical depth, which is not established here. Scored low across all core-research dimensions, with modest industry-impact credit for founding and running a functioning AI product company.\n\nNothing of Yu Hu's own authorship sits in the lineage that frontier language models descend from — no pre-2013 vector-space/distributional-semantics patents or systems, no attention/transformer/embedding/optimizer/tokenizer/alignment contribution, and no papers or code of his in the record (the dossier's papers are homonym noise for a Purdue business professor and unrelated 'Yu Hu' authors). His verifiable background is finance — University of Cambridge, an investment-banking analyst role at Deutsche Bank, and portfolio management at Citadel — not language-modeling research or engineering, so there are zero verifiable hands-on years authoring or training LMs in any single domain, let alone across biological/financial/code/text domains. Kaito AI (founded ~2022) is a real shipped LLM-powered crypto search/analytics product, but applying vendor models to the crypto-attention domain is not language modeling in that domain, and there is no evidence Hu personally wrote the core models, retrieval stack, or research the company runs on. He is thus a founder-CEO of an AI company whose science and engineering were done by technical staff — the 3-7 band for scientific_founder — with one company across roughly three to four years, not a verifiable technical founder who authored the core code/patents/papers.","evidence":[{"claim":"Kaito AI founder Yu Hu studied at the University of Cambridge, previously an Investment Banking Analyst at Deutsche Bank and managed a $500M portfolio at Citadel","source_url":"https://x.com/Param_eth/status/1965085813236002964","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founder/CEO of Kaito AI, an AI-powered search engine for digital assets, backed by Dragonfly, Sequoia and Jane Street","source_url":"https://www.linkedin.com/in/yuhu9277/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Wikipedia/Wikidata entry titled 'Yu (Jeffrey) Hu' (Q15109655) describes a Purdue University Daniels School of Business professor and MIT Digital Fellow, a different person from the Kaito AI founder","source_url":"https://en.wikipedia.org/wiki/Yu_(Jeffrey)_Hu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kaito's own site describes it as 'The intelligence and financial markets platform for the attention economy', offering Mindshare Arena, Aura profiles, Capital Launchpad and trading rewards; it names no founders and details no model or retrieval architecture","source_url":"https://www.kaito.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kaito's documentation frames the company around 'InfoFi' and the attention economy rather than any described AI/search technology stack, and contains no founder biography","source_url":"https://docs.kaito.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Wikipedia article 'Yu (Jeffrey) Hu' used by the dossier is a Purdue University business-school professor (Accenture Chair, Daniels School of Business, INFORMS Distinguished Fellow, MIT IDE Digital Fellow) with no mention of Kaito AI, crypto or startup founding — a different person","source_url":"https://en.wikipedia.org/wiki/Yu_(Jeffrey)_Hu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kaito AI founder Yu Hu studied at the University of Cambridge and previously worked as an investment-banking analyst at Deutsche Bank and a portfolio manager at Citadel before founding Kaito — a finance, not AI-research, background","source_url":"https://www.linkedin.com/in/yuhu9277/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Yu Hu is founder/CEO of Kaito AI, an AI-powered search/analytics platform for digital assets; the company markets itself around 'InfoFi'/the attention economy and names no founder as the author of its model or retrieval architecture","source_url":"https://www.kaito.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Wikipedia/Wikidata 'Yu (Jeffrey) Hu' (Q15109655) is a Purdue Daniels School of Business professor, a different person, so the dossier's degrees, papers and citations do not attach to the Kaito founder","source_url":"https://en.wikipedia.org/wiki/Yu_(Jeffrey)_Hu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Kaito's own site and docs describe an 'InfoFi'/attention-economy intelligence and financial-markets platform (Mindshare, Aura, Capital Launchpad) with no founder-authored model or retrieval architecture disclosed","source_url":"https://www.kaito.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's 'Yu (Jeffrey) Hu' (Q15109655) is a Purdue Daniels School of Business professor, a different person from the Kaito AI founder","source_url":"https://en.wikipedia.org/wiki/Yu_(Jeffrey)_Hu","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["University of Cambridge (undergraduate)"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["Kaito AI search/analytics platform for digital assets"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"CRITICAL: the dossier's entire Wikipedia and Wikidata block is a homonym mismatch — 'Yu (Jeffrey) Hu' (Q15109655) is a Purdue/Georgia Tech information-systems business professor (PhD MIT Sloan, advisor Erik Brynjolfsson), NOT the Kaito AI founder, who has a finance background (Cambridge, Deutsche Bank, Citadel) with no found CS/ML degree. The OpenAlex block (COVID-19 clinical papers, semiconductor FETs, robotics obstacle-avoidance — all under generic 'Yu Hu') is also unrelated homonym noise; the dossier itself does not flag this mismatch, so it is reported here per the brief's instruction. All dimension scores are based on the verified Kaito-AI Yu Hu, not the professor in the dossier's programmatic data.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Kaito AI — founder & CEO (business/finance founder, ~2022–2026) — science and engineering done by technical staff, not verifiably authored by Hu"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.72,"duration_ms":45217},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":3},"confidence":0.72,"duration_ms":51143}],"source":"seeded","profile_url":null,"wikipedia_title":"Yu (Jeffrey) Hu","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"hayden-adams","name":"Hayden Adams","title":"Founder & CEO","company":"Uniswap Labs","sector":"crypto","rank":96,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":10,"industry_impact":6,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":15,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":15,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.76,"rationale":"Adams holds a BS in mechanical engineering from Stony Brook University (2016), not a computer science, mathematics, or AI-related degree, and has no verifiable AI/ML research record — his canonical contribution (the Uniswap constant-product automated market maker, x*y=k, and the v3 concentrated-liquidity design) is a real, personally-authored and personally-built piece of financial/algorithmic engineering, but it is decentralized-exchange mechanism design, not vector embeddings, attention, transformers, or language modeling. This supports real hands_on_engineering credit (he taught himself Solidity after being laid off from Siemens and single-handedly built and shipped Uniswap v1-v3, a system that today processes billions in volume) and some industry_impact as the builder of foundational DeFi infrastructure, but the rubric explicitly scores AI-core dimensions, and there is no verifiable evidence he has authored or built anything in that space. All three AI-core dimensions (foundations beyond generic engineering, vector_embeddings, transformers_lm) score at or near the floor.\n\nNothing in Adams's verifiable record touches the frontier-model lineage: his canonical contributions are the Uniswap constant-product AMM (x*y=k) and v3 concentrated liquidity — decentralized-exchange mechanism design in Solidity, with no vector-space, distributional-semantics, attention, transformer, embedding, optimizer, tokenizer or pretraining work, pre- or post-2013, so frontier_founder floors. He has zero verifiable years in language modeling of any kind (natural-language text, biological, financial, code, chemistry, legal or otherwise) — his BS is in mechanical engineering and his entire hands-on history is DeFi smart contracts — so lm_domain_depth and lm_domain_breadth also floor with no qualifying domains. He IS, however, a genuine technical founder: laid off from Siemens, he taught himself Solidity and personally wrote and shipped Uniswap v1–v3 and co-authored the whitepapers, running as the technical founder of Uniswap Labs from 2018 to now (~8 years, one company). Because that founding record is entirely outside the AI/language-modeling field this rubric measures, it lands in the 'technical founder outside this field' band rather than the AI-core bands.","evidence":[{"claim":"Author of the Uniswap v3 Core whitepaper (March 2021), introducing concentrated liquidity","source_url":"https://app.uniswap.org/whitepaper-v3.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Adams was a former mechanical engineer at Siemens, launched Uniswap in November 2018 inspired by a Vitalik Buterin blog post, and is credited as co-author of the Uniswap v3 whitepaper (March 2021) with Noah Zinsmeister; no AI or machine learning is mentioned","source_url":"https://en.wikipedia.org/wiki/Uniswap","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uniswap is a decentralized exchange protocol on Ethereum using smart contracts and liquidity pools (constant-product rule), i.e. its core technology is not AI systems","source_url":"https://www.wikidata.org/wiki/Q104438477","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uniswap v3 core is a Solidity smart-contract codebase (5.0k stars, BUSL-1.1/GPL-2.0 licensed) with no machine-learning components","source_url":"https://github.com/Uniswap/v3-core","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Adams, a former mechanical engineer at Siemens, taught himself Solidity and personally built and launched Uniswap (Nov 2018), and is credited author of the v3 whitepaper (March 2021) with no AI/ML component","source_url":"https://en.wikipedia.org/wiki/Uniswap","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uniswap v3 core is a Solidity smart-contract codebase with no machine-learning or language-modeling components","source_url":"https://github.com/Uniswap/v3-core","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Adams personally taught himself Solidity after leaving Siemens and single-handedly built and launched Uniswap (Nov 2018), and is credited co-author of the Uniswap v3 whitepaper — no AI/ML work in any source","source_url":"https://en.wikipedia.org/wiki/Uniswap","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Uniswap v3-core is a Solidity smart-contract codebase authored under his direction with no machine-learning components","source_url":"https://github.com/Uniswap/v3-core","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Author of the Uniswap v3 Core whitepaper (March 2021) introducing concentrated liquidity — AMM mechanism design, not language-model lineage","source_url":"https://app.uniswap.org/whitepaper-v3.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Stony Brook University, BS Mechanical Engineering (2016)"],"canonical_papers":["Uniswap v3 Core (2021, Adams et al.)"],"first_verifiable_year":2017,"notable_systems":["Uniswap v1/v2/v3 automated market maker protocol"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex is not found and Semantic Scholar match (2 papers, 0 citations) plausibly matches the Uniswap v3 whitepaper listing rather than peer-reviewed research; the dossier correctly has no first_verifiable_year populated. No AI/ML background found in any source — his entire verifiable technical record is DeFi/AMM mechanism design, mechanical engineering, and Solidity smart-contract development, unrelated to the core-AI lineage this rubric measures.","years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":1,"technical_founder_roles":["Uniswap Labs — founder/CEO, personally authored core Solidity contracts and whitepapers — 2018-2026"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.95,"duration_ms":43010},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.9,"duration_ms":47956}],"source":"seeded","profile_url":null,"wikipedia_title":"Uniswap","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jeff-yan","name":"Jeff Yan","title":"Founder","company":"Hyperliquid","sector":"crypto","rank":97,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":10,"industry_impact":4,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":15,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":15,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.55,"rationale":"This dossier's programmatic matches (Wikidata Q6175301 'Jeff Yang' the Harvard journalist/businessman, the OpenAlex/Semantic Scholar CAPTCHA-security author, and the PubMed hits) are all different people and must be disregarded. The real Jeff Yan of Hyperliquid studied mathematics and computer science at Harvard College (2013-2017), was an International Physics Olympiad medalist, then worked as a low-latency algorithm/market-making developer at Hudson River Trading before founding Chameleon Trading and Hyperliquid, a custom L1 perpetuals DEX. This is strong quantitative/systems foundations and personal, hands-on trading-infrastructure engineering, but there is no verifiable paper, patent, thesis, or public code repository establishing any personal record in vector embeddings, attention, or language modeling — his technical output is exchange/market-microstructure engineering, not AI research. industry_impact is scored for building and leading the engineering of a real, technically substantial trading system (Hyperliquid), not for its market cap or fame.\n\nNone of Jeff Yan's verifiable work sits in the language-model lineage: there is no distributional-semantics/vector-space/attention/transformer paper, patent, embedding method, tokenizer, optimizer, dataset, benchmark, or pretraining/alignment contribution that today's frontier models build on — his technical output is low-latency market-making and a custom L1 exchange (HyperBFT consensus, on-chain order book), so frontier_founder, lm_domain_depth and lm_domain_breadth are all near-zero with zero verifiable language-modeling years and zero language-modeling domains. He IS, however, a genuine scientific/technical founder — a founder-CEO who personally set and executed the technical direction and wrote core code, self-funding and building Hyperliquid 'from first principles' after co-founding the market-making firm Chameleon Trading — but that role is entirely OUTSIDE this field (trading infrastructure, not AI/LM), which the rubric places in the 3-7 'technical founder outside this field' band. Across ~2 companies over roughly 6 years (Chameleon ~2020, Hyperliquid launched 2023) he is a strong technical founder, but not of an AI/language-modeling company. The dossier's OpenAlex/Semantic-Scholar/PubMed/Wikidata matches are all different people (a CAPTCHA-security researcher, journalist Jeff Yang, biomedical 'Yan J' homonyms) and contribute nothing to these dimensions.","evidence":[{"claim":"Jeff Yan studied mathematics and computer science at Harvard College, 2013-2017, and was a physics olympiad medalist","source_url":"https://www.datawallet.com/crypto/who-is-jeff-yan-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"After Harvard, Yan worked at Hudson River Trading as an algorithm developer in low-latency equities market-making before founding Chameleon Trading and then Hyperliquid","source_url":"https://colossus.com/article/beyond-the-sky-jeffrey-yan-hyperliquid/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hyperliquid is a self-funded, custom Layer-1 decentralized perpetuals exchange founded by Jeff Yan and co-founder iliensinc, with no outside VC funding","source_url":"https://hyperliquidguide.com/ecosystem/who-created-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hyperliquid is a layer-1 blockchain 'written and optimized from first principles' using HyperBFT, a custom consensus algorithm inspired by HotStuff; HyperCore runs fully on-chain perpetual and spot order books with one-block finality at 200k orders/second; HyperEVM adds general smart contracts.","source_url":"https://hyperliquid.gitbook.io/hyperliquid-docs/about-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The official hyperliquid-dex/node repository (Apache-2.0) contains node binaries and validator documentation and attributes no named individuals or founders.","source_url":"https://github.com/hyperliquid-dex/node","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata has no entity for Hyperliquid's Jeff Yan — a search for 'Jeff Yan' returns only Jeff Yang (a Taiwanese-American writer), a researcher named Jeff Yang, a Chinese urologist and Jeff D Yanosky.","source_url":"https://www.wikidata.org/w/index.php?search=Jeff+Yan","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The academic 'Jeff Yan' in the dossier is a security researcher whose corpus is CAPTCHA-breaking, password guessing, acoustic side channels and image forensics (Newcastle/Strathclyde/Linkoping) — a different person.","source_url":"https://www.semanticscholar.org/author/1704945","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jeff Yan personally set and executed Hyperliquid's technical direction, self-funded and built from first principles after Harvard math/CS and HRT low-latency market-making, co-founding Chameleon Trading first","source_url":"https://colossus.com/article/beyond-the-sky-jeffrey-yan-hyperliquid/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hyperliquid is a layer-1 blockchain written 'from first principles' using HyperBFT consensus with on-chain perpetual/spot order books — an exchange/trading system, not an AI or language-modeling system","source_url":"https://hyperliquid.gitbook.io/hyperliquid-docs/about-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hyperliquid is a self-funded custom L1 perpetuals DEX founded by Jeff Yan and co-founder iliensinc with no outside VC funding","source_url":"https://hyperliquidguide.com/ecosystem/who-created-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jeff Yan personally worked as a low-latency algorithm developer at Hudson River Trading, then founded Chameleon Trading (crypto market-making) before Hyperliquid, setting and executing the technical direction himself","source_url":"https://colossus.com/article/beyond-the-sky-jeffrey-yan-hyperliquid/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hyperliquid is a self-funded custom Layer-1 perpetuals DEX founded by Jeff Yan, built and optimized 'from first principles' with the HyperBFT consensus algorithm and fully on-chain HyperCore order books — a trading system, not an AI/language-modeling system","source_url":"https://hyperliquid.gitbook.io/hyperliquid-docs/about-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Yan studied mathematics and computer science at Harvard College (2013-2017) with no graduate degree and no known academic publication record in AI, embeddings, or language modeling","source_url":"https://www.datawallet.com/crypto/who-is-jeff-yan-hyperliquid","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Harvard College — mathematics and computer science, 2013-2017 (no advanced/graduate degree found)"],"canonical_papers":[],"first_verifiable_year":2017,"notable_systems":["Hyperliquid (custom L1 perpetuals exchange)","Chameleon Trading (crypto market-making firm)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier is built around the wrong Jeff Yan(g) throughout: Wikidata Q6175301 is Jeff Yang, a Taiwanese-American writer/journalist/TV producer educated at Harvard College/Saint Ann's School, not the Hyperliquid founder (shared first name + surname spelling collision, and Wikidata also mislabels him 'Jeff Yang' not 'Jeff Yan'). The OpenAlex (id A5100864308, h-index 22) and Semantic Scholar (id 1704945, h-index 24) matches are a CAPTCHA/password-security researcher (affiliations: Linkoping, Xidian, CUHK, Strathclyde, Cambridge) — an entirely different person the brief specifically warned about. The PubMed hits (glioblastoma, mitochondrial DNA, liver cancer papers) are unrelated biomedical authors sharing the common surname+initial 'Yan J'. None of these programmatic sources should be used for scoring; all figures above come from independently verified web sources about the actual Hyperliquid founder, who has no known academic publication record at all.","years_language_modeling":0,"years_as_technical_founder":6,"technical_founder_companies":2,"technical_founder_roles":["Chameleon Trading — co-founder / technical (crypto market-making) — ~2020","Hyperliquid — co-founder & CEO, personally built L1/consensus/order-book (trading infrastructure, outside AI/LM) — 2022/2023-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":7},"confidence":0.8,"duration_ms":48976},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.8,"duration_ms":54379}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"niraj-pant","name":"Niraj Pant","title":"Co-founder","company":"Ritual","sector":"crypto","rank":98,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"hands_on_engineering":4,"industry_impact":6,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":15,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":15,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.57,"rationale":"Niraj Pant studied Computer Science at the University of Illinois Urbana-Champaign, including some privacy research in the school's Decentralized Systems Lab as an undergraduate, but dropped out at 19 after about a month to intern at Polychain Capital, where he spent roughly six years as a General Partner investing in crypto/AI infrastructure startups (Offchain Labs, EigenLayer, Polymarket, dYdX, Compound). He then co-founded Ritual, a decentralized execution layer for AI/compute, in 2023. This is a genuine crypto-investing and infrastructure-entrepreneurship record, but there is no evidence of a completed degree, published research, authored papers, patents, or personally built AI/ML systems — his technical depth is that of an investor/allocator who backed AI-adjacent crypto infrastructure, not a researcher or engineer with a first-principles technical record in the core-AI lineage (math foundations, embeddings, transformers). Per rubric, this scores low on the research dimensions; industry_impact reflects Ritual's stated mission (decentralizing AI compute/inference) and his investing track record, but with low confidence given the dossier's OpenAlex match is a wrong-person homonym and little independent technical verification exists.\n\nNo verifiable contribution by Niraj Pant to the frontier-model lineage exists: he has no pre-2013 vector-space / distributional-semantics patents or systems, no attention/transformer/embedding/optimizer/tokenizer/scaling/alignment work, and Ritual's own research output (verifiable inference, speculative decoding, watermarking, execution-aware consensus) is authored by Arka Pal, Rahul Thomas, Micah Goldblum, Maryam Bahrani and Naveen Durvasula, not by him. He has zero verifiable hands-on years in language modeling of any kind — his record is a ~6-year Polychain GP investing career (Offchain Labs, EigenLayer, Polymarket) followed by co-founding Ritual in 2023, which is capital-allocation and company-building, not personal LM research or systems work. No distinct language-modeling domain (biological, financial, code, natural-language text, etc.) shows a hands-on record authored by him. He co-founded exactly one AI-infrastructure company (~3 years) but as a former investor whose scientific/engineering direction is carried by technical co-founders and researchers, placing him in the 'founder of an AI company whose science was done by others' band rather than the technical-founder band.","evidence":[{"claim":"Attended University of Illinois Urbana-Champaign for Computer Science, conducted privacy research at the school's Decentralized Systems Lab, but dropped out after ~1 month of a Polychain internship at age 19","source_url":"https://siebelschool.illinois.edu/about/awards/alumni-awards/alumni-awards-past-recipients/81999","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Spent roughly six years as a General Partner at Polychain Capital, leading investment rounds in Offchain Labs, EigenLayer, and Polymarket, among 30+ companies","source_url":"https://fortune.com/crypto/2023/11/08/two-former-polychain-partners-fundraise-25-million-ritual-decentralize-ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Ritual, described as 'the decentralized execution layer for AI', raising $25M in 2023","source_url":"https://fortune.com/crypto/2023/11/08/two-former-polychain-partners-fundraise-25-million-ritual-decentralize-ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual lists Niraj Pant among its team alongside Akilesh Potti and Arka Pal; the company builds decentralized AI infrastructure and an SDK for integrating AI into decentralized applications, emphasising censorship resistance, privacy and verifiable computation","source_url":"https://ritual.net/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual's research/blog output on verifiable and privacy-preserving LLM inference, speculative decoding and sampling, model watermarking and execution-aware consensus is authored by Arka Pal, Rahul Thomas, Micah Goldblum, Maryam Bahrani and Naveen Durvasula; Niraj Pant does not appear as an author on","source_url":"https://ritual.net/blog","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual describes itself as a lab for autonomous intelligence with research frontiers in AI, mechanism design, systems and cryptography, and publishes a whitepaper on delegated execution and attestation rather than named academic papers","source_url":"https://ritual.net/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The OpenAlex author record A5078600276 attached to this slug (3 works, 220 citations, Whittier College) consists of TeV blazar VLBA observation papers in The Astrophysical Journal, i.e. a different person","source_url":"https://doi.org/10.1088/0004-637x/723/2/1150","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual's verifiable/private LLM-inference, speculative-decoding, watermarking and consensus research is authored by Arka Pal, Rahul Thomas, Micah Goldblum, Maryam Bahrani and Naveen Durvasula; Niraj Pant is not listed as an author","source_url":"https://ritual.net/blog","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Niraj Pant co-founded Ritual in 2023 after ~6 years as a Polychain Capital General Partner, framed as a former-investor founding team decentralizing AI compute","source_url":"https://fortune.com/crypto/2023/11/08/two-former-polychain-partners-fundraise-25-million-ritual-decentralize-ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No genuine academic publication or patent record exists for the Ritual co-founder; the dossier's OpenAlex match (TeV blazar VLBA astrophysics, Whittier College) is a wrong-person homonym","source_url":"https://doi.org/10.1088/0004-637x/723/2/1150","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual's technical/research output (verifiable LLM inference, speculative decoding, watermarking, execution-aware consensus) is authored by Arka Pal, Rahul Thomas, Micah Goldblum, Maryam Bahrani and Naveen Durvasula; Niraj Pant does not appear as an author","source_url":"https://ritual.net/blog","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Pant spent roughly six years as a General Partner at Polychain Capital (crypto investing) before co-founding Ritual in 2023 with $25M raised — an investor/allocator path, not a language-modeling research or engineering record","source_url":"https://fortune.com/crypto/2023/11/08/two-former-polychain-partners-fundraise-25-million-ritual-decentralize-ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Ritual describes itself as a decentralized execution layer for AI with named technical team members (Akilesh Potti, Arka Pal); the science/engineering is led by technical co-founders","source_url":"https://ritual.net/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended (did not complete) Computer Science, University of Illinois Urbana-Champaign"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["Ritual (decentralized AI execution/compute layer, as co-founder/investor-turned-founder, not sole technical architect)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"CRITICAL: the dossier's OpenAlex match (id A5078600276, works on TeV blazar jets / VLBA radio astronomy at Whittier College) is the WRONG homonym — a completely unrelated astrophysics researcher. No genuine academic publication record for the Ritual co-founder was found in any source checked; his verifiable record is investment/career history (Polychain GP), not research or engineering. Dossier's timeline first_verifiable_year (2006, from the astrophysics homonym) should be disregarded entirely.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Ritual — co-founder (business/investing lead, science carried by co-founders) — 2023-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.72,"duration_ms":47687},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":2,"scientific_founder":4},"confidence":0.72,"duration_ms":49843}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"alex-gluchowski","name":"Alex Gluchowski","title":"Co-founder & CEO","company":"Matter Labs (ZKsync)","sector":"crypto","rank":99,"rubric_version":4,"dimensions":{"foundations":6,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":8,"industry_impact":4,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":14,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":14,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.55,"rationale":"Gluchowski is publicly known as the founder of Matter Labs, the company behind the zkSync zero-knowledge rollup on Ethereum — a cryptography/zero-knowledge-proof systems company, not a vector-embeddings, LSA, or transformer/language-modeling company. This session's dossier for him returned no usable Wikipedia, Wikidata, OpenAlex, or Semantic Scholar match, and its PubMed block (40 hits under 'Gluchowski A') is entirely unrelated homonyms (UK physiotherapy/exercise-science researchers, Polish food-science researchers) with no connection to this person. WebSearch capacity was exhausted mid-session and WebFetch attempts against Matter Labs' own site and LinkedIn did not return usable biographical content, so his education and pre-crypto career could not be independently verified in this pass; per the brief's instruction to score lower and disclose when unsure, scores here reflect only what is reliably attributable (building and leading a real cryptographic-systems engineering company) with no verified personal record in the core AI/LM lineage this rubric measures. Zero-knowledge cryptography is adjacent mathematical/systems engineering but is not itself linear algebra/optimization-for-ML, embeddings, or LM research, so foundations and hands_on_engineering are scored modestly for general technical company-building rather than AI-specific depth, and vector_embeddings/transformers_lm are scored at zero for lack of any verifiable authored work in those areas.\n\nNothing in Gluchowski's verifiable record touches the language-modeling lineage that today's frontier models build on: his authored code (the 'bellman' zkSNARK library and 'plasma_winter' PoC in Rust) and Matter Labs' shipped systems (zkSync Lite, ZK Stack, Airbender RISC-V prover) are zero-knowledge cryptography, with no attention/transformer, embedding, distributional-semantics, tokenizer, optimizer, scaling or alignment component in the frontier stack — no pre-2013 vector-space or LSI precursor work either, so frontier_lineage is empty and frontier_founder scores at the floor. There are zero verifiable years of language-modeling work (the PubMed hits under 'Gluchowski A' are unrelated exercise-science and food-science homonyms), so lm_domain_depth and lm_domain_breadth sit at the bottom band with no LM domains. He IS, however, a genuine hands-on technical founder — he personally authored core cryptographic code that Matter Labs (founded ~2018) runs on — but that founder role is entirely OUTSIDE this field, which the anchors cap at 3-7; hence scientific_founder = 5 for one company as a real technical founder in ZK cryptography, not language modeling.","evidence":[{"claim":"GitHub user gluk64 ('Alex G'), organisation Matter Labs, author of 'plasma_winter' (PoC of Plasma with account model and zkSNARK-proven state transitions, Rust) and 'bellman' (zkSNARK library with Ethereum BN256 support, Rust); 56 public repositories, no machine-learning or language-model projects","source_url":"https://github.com/gluk64","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alex Gluchowski is featured as 'CEO, Matter Labs'; ZKsync builds a network of chains secured by zero-knowledge cryptography, the ZK Stack open-source framework, Prividium for institutional privacy, and Airbender, a RISC-V prover for subsecond block proofs","source_url":"https://zksync.io/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The zkSync repository maintained by Matter Labs implements zkSync Lite, a ZK Rollup scaling and privacy engine for Ethereum using zero-knowledge proofs for transaction validation","source_url":"https://github.com/matter-labs/zksync","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub user gluk64 (Matter Labs), author of 'bellman' (zkSNARK library, Rust) and 'plasma_winter' (zkSNARK-proven state transitions, Rust); 56 repos, no ML/LM projects — technical-founder code base is cryptography, not the LM lineage","source_url":"https://github.com/gluk64","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alex Gluchowski is CEO of Matter Labs / ZKsync, a network secured by zero-knowledge cryptography (ZK Stack, Prividium, Airbender prover) — the company's core is ZK proving, not embeddings/transformers/language modeling","source_url":"https://zksync.io/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"GitHub user gluk64 (Matter Labs) authored 'bellman' (zkSNARK library) and 'plasma_winter' (zkSNARK-proven state transitions) in Rust; 56 repos, no ML/LM projects — his built record is ZK cryptography, not language modeling","source_url":"https://github.com/gluk64","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Alex Gluchowski is CEO of Matter Labs; ZKsync builds chains secured by zero-knowledge cryptography (ZK Stack, Airbender RISC-V prover) — a cryptographic-systems company he co-founded and technically leads, whose core is not language modeling","source_url":"https://zksync.io/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["zkSync / ZK Stack (Matter Labs)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier has essentially no usable programmatic identity data: Wikipedia, Wikidata, OpenAlex, and Semantic Scholar all show found:false, and the PubMed block (40 refined hits, homonym_risk marked false but this is incorrect — every sampled result is a UK exercise-science/physiotherapy researcher or a Polish food-science researcher, not the Matter Labs founder). This session's WebSearch quota was exhausted before additional searches on Gluchowski's education/pre-crypto background could be run, and WebFetch against matterlabs.com/about and his LinkedIn returned no usable content (404/blocked). Confidence is accordingly low; a re-run with search capacity should verify his educational background and whether he has any cryptography/math publication record before scores are finalized.","years_language_modeling":0,"years_as_technical_founder":8,"technical_founder_companies":1,"technical_founder_roles":["Matter Labs (ZKsync) — founder/CEO, authored core ZK cryptography code (bellman, plasma_winter) — ~2018-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.6,"duration_ms":46083},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.62,"duration_ms":54921}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"sergey-nazarov","name":"Sergey Nazarov","title":"Co-founder & CEO","company":"Chainlink Labs","sector":"crypto","rank":100,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":1,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"hands_on_engineering":8,"industry_impact":6,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":14,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":14,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.73,"rationale":"Nazarov holds a 2007 NYU bachelor's degree in philosophy and management (no math/CS/ML graduate training), then built a sequence of early startups (ExistLocal, QED Capital, CryptaMail/Secure Asset Exchange) before co-founding SmartContract.com in 2014, which became Chainlink. He co-authored the 2017 Chainlink whitepaper 'A Decentralized Oracle Network' together with Steve Ellis and Cornell professor Ari Juels, and has personally led Chainlink's technical roadmap (oracle network design, Proof-of-Reserve, CCIP) for nearly a decade — this is real, sustained hands-on systems-building and technical leadership, credited under hands_on_engineering/industry_impact. However, Chainlink's core technology is decentralized-oracle/distributed-systems and applied cryptography engineering, not language modeling, vector embeddings, or transformer architectures; no publication, patent, or project ties him to that lineage, so those dimensions score at floor per the brief's explicit guidance on Chainlink. foundations is scored low-but-nonzero for the applied cryptographic/distributed-systems design work embedded in the whitepaper, not for math/ML theory.\n\nNazarov's foundational contribution is the decentralized-oracle network (Chainlink whitepaper 2017, co-authored with Steve Ellis and Ari Juels) — applied cryptography and distributed systems, with no position in the attention→transformer→language-model lineage and no verifiable pre-2013 vector-space, distributional-semantics, LSI or term-relationship-network patents or shipped systems, so frontier_founder is at floor. There is no verifiable language-modeling record of any kind (no LM papers, patents, or shipped statistical/neural/vector-space text systems), so lm_domain_depth and lm_domain_breadth are near floor at 0 years and 0 distinct LM domains. He IS a genuine, sustained technical/scientific founder — he co-authored the core whitepaper and has personally driven Chainlink's technical roadmap (Proof-of-Reserve, CCIP) across ~12 years and prior ventures (CryptaMail, Secure Asset Exchange, SmartContract.com) — but that founder-technologist work sits entirely OUTSIDE the AI/language-modeling field, which the scientific_founder anchor caps at the 3-7 'technical founder outside this field' band. Scored 5 for a decade-plus of real technical-founder execution in a non-LM domain.","evidence":[{"claim":"Nazarov graduated NYU in 2007 with a bachelor's degree in philosophy and management; no graduate STEM degree found","source_url":"https://en.wikipedia.org/wiki/Sergey_Nazarov_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded SmartContract.com with Steve Ellis in 2014; co-authored the Chainlink whitepaper 'A Decentralized Oracle Network' with Steve Ellis and Ari Juels, published 2017","source_url":"https://research.chain.link/whitepaper-v1.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"As CEO of Chainlink Labs, Nazarov has led development of Proof-of-Reserve and the Cross-Chain Interoperability Protocol (CCIP), and joined the CFTC's Innovation Advisory Committee in Feb 2026","source_url":"https://en.wikipedia.org/wiki/Sergey_Nazarov_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's Wikipedia/Wikidata match (Q130442938, Ukrainian political strategist and former MP from Odesa) is a namesake, not the Chainlink co-founder","source_url":"https://en.wikipedia.org/wiki/Sergey_Nazarov","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Verified from the whitepaper's own title page: 'ChainLink: A Decentralized Oracle Network — Steve Ellis, Ari Juels, and Sergey Nazarov, 4 September 2017 (v1.0)', on oracle connectivity, on-chain data aggregation, off-chain consensus and reputation/security monitoring.","source_url":"https://research.chain.link/whitepaper-v1.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Born 1986; graduated NYU 2007 with a bachelor's degree in philosophy and management; founded ExistLocal (2009), QED Capital (2011), CryptaMail and Secure Asset Exchange (2014); co-founded SmartContract.com with Steve Ellis in 2014; no AI work mentioned.","source_url":"https://en.wikipedia.org/wiki/Sergey_Nazarov_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chainlink was created in 2017 by Sergey Nazarov and Steve Ellis, who co-authored the whitepaper with Cornell professor Ari Juels; it is a decentralized blockchain oracle network bridging on-chain contracts to off-chain data.","source_url":"https://en.wikipedia.org/wiki/Chainlink_(blockchain_oracle)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chainlink's published research corpus (Chainlink 1.0/2.0, Town Crier, Mixicles, DECO, OCR3, Confidential Compute) is cryptography and oracle-network design; none of it is machine-learning research.","source_url":"https://chain.link/whitepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chainlink whitepaper 'A Decentralized Oracle Network' (2017, Ellis/Juels/Nazarov) concerns oracle connectivity, on-chain aggregation and off-chain consensus — cryptography and distributed systems, not any transformer/embedding/LM component","source_url":"https://research.chain.link/whitepaper-v1.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chainlink's research corpus (Chainlink 1.0/2.0, Town Crier, Mixicles, DECO, OCR3) is oracle-network and cryptography research with no machine-learning or language-modeling work","source_url":"https://chain.link/whitepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded SmartContract.com with Steve Ellis in 2014 (which became Chainlink) after founding ExistLocal, QED Capital, CryptaMail and Secure Asset Exchange — a decade-plus technical-founder record entirely in crypto, none in AI/LM","source_url":"https://en.wikipedia.org/wiki/Sergey_Nazarov_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Nazarov co-authored 'ChainLink: A Decentralized Oracle Network' (2017) on oracle connectivity, on-chain aggregation and off-chain consensus — cryptography/distributed-systems, not language modeling","source_url":"https://research.chain.link/whitepaper-v1.pdf","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Chainlink's research corpus (Chainlink 1.0/2.0, Town Crier, DECO, Mixicles, OCR3) is applied cryptography and oracle-network design, with no machine-learning or language-modeling work","source_url":"https://chain.link/whitepaper","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded SmartContract.com with Steve Ellis in 2014 (later Chainlink) and has led its technical roadmap as CEO/founder for roughly a decade","source_url":"https://en.wikipedia.org/wiki/Chainlink_(blockchain_oracle)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["New York University — B.A. philosophy and management, 2007"],"canonical_papers":["ChainLink: A Decentralized Oracle Network (whitepaper, 2017, with Steve Ellis and Ari Juels)"],"first_verifiable_year":2014,"notable_systems":["Chainlink (decentralized oracle network)","SmartContract.com"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's Wikipedia and Wikidata entries (Q130442938, 'Ukrainian political strategist, former member of parliament') are for a completely different, unrelated Sergey Nazarov, not the Chainlink co-founder — a clear homonym mismatch the assessor should disregard. OpenAlex found nothing (correctly, since the real Nazarov has no academic publication record). Semantic Scholar returned a match (id 153516042, h-index 8, 13 papers, 316 citations) but the brief flags high homonym risk for common Slavic surnames and I could not independently confirm this Semantic Scholar profile is the Chainlink Nazarov rather than an unrelated scientist named Sergey Nazarov (a fairly common Russian/Ukrainian name) — I did not use it for scoring given the risk of a false match, and note this as an open gap rather than crediting phantom foundations. No PubMed or patent records were verified as his.","years_language_modeling":0,"years_as_technical_founder":12,"technical_founder_companies":0,"technical_founder_roles":["Chainlink Labs / SmartContract.com — co-founder & CEO, technical roadmap owner — 2014-2026 (oracle networks, OUTSIDE AI/LM)","Secure Asset Exchange / CryptaMail — founder — 2014 (crypto, outside AI/LM)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.85,"duration_ms":54613},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"scientific_founder":7},"confidence":0.86,"duration_ms":44063}],"source":"seeded","profile_url":null,"wikipedia_title":"Sergey Nazarov","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"adam-back","name":"Adam Back","title":"Co-founder & CEO","company":"Blockstream","sector":"crypto","rank":101,"rubric_version":4,"dimensions":{"foundations":7,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"hands_on_engineering":7,"industry_impact":4,"scientific_founder":6},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":13,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":13,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.77,"rationale":"Back holds a PhD in computer science (distributed systems) from the University of Exeter (1995, dissertation on parallelization of general-purpose programs using optimistic techniques from parallel discrete-event simulation), giving him genuine PhD-level formal training in distributed algorithms -- adjacent to but not overlapping the rubric's core areas. He invented Hashcash in 1997, a proof-of-work scheme that became foundational to Bitcoin mining, and is the only person explicitly named in Satoshi Nakamoto's Bitcoin whitepaper, making him a real, personally-credited contributor to cryptographic systems engineering. He co-founded and is CEO of Blockstream (2014). None of this work touches vector embeddings, representation learning, or the transformer/language-model lineage, and no publications, patents, or systems in those areas were found; Semantic Scholar shows only 1 indexed paper (41 citations), and OpenAlex has no match. Scores reflect a legitimate cryptography/distributed-systems PhD and hands-on engineering record with essentially zero footprint in the specific AI-core dimensions the rubric measures.\n\nBack's canonical work is Hashcash (1997), a proof-of-work scheme, and his PhD is in distributed systems (parallel discrete-event simulation) — none of it is vector-space, distributional-semantics, attention, transformer, tokenizer, optimizer, embedding or pretraining work, so there is no verifiable lineage into today's frontier language models. He has no verifiable record in language modeling of any kind (natural-language text, biological, financial, code or otherwise) — zero years, zero domains. He is, however, a genuine technical founder: he personally authored the core cryptographic technology (Hashcash, credlib) his companies build on and co-founded/leads Blockstream (2014–present, ~11 years) as a hands-on cryptographer, plus prior engineering at Zero-Knowledge Systems — but this founder record is entirely in cryptography/Bitcoin infrastructure, i.e. a technical founder OUTSIDE the language-modeling field, which caps the scientific_founder score in the 3–7 band. His footprint in all four rubric-v4 dimensions is essentially the same as in the prior pass: strong technical founder, wrong field, no language-modeling lineage.","evidence":[{"claim":"PhD in computer science (distributed systems), University of Exeter, 1995","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Invented Hashcash (1997), the proof-of-work scheme later used in Bitcoin mining; the only person explicitly named in the Bitcoin whitepaper","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Blockstream in 2014, became CEO in 2016","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PhD in computer science (distributed systems), University of Exeter, 1995 thesis 'Parallelization of general purpose programs using optimistic techniques from parallel discrete event simulation', supervisor Stephen Turner; invented Hashcash 1997; implemented credlib; Zero-Knowledge Systems; co-found","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q348671 lists occupation 'cryptographer' and notable work 'Hashcash'; educated at University of Exeter; no AI-related occupation, affiliation or work","source_url":"https://www.wikidata.org/wiki/Q348671","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record for Adam Back: 1 paper, 41 citations, h-index 1 - no machine-learning corpus","source_url":"https://www.semanticscholar.org/author/Adam-Back/2058627663","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Invented Hashcash (1997), a proof-of-work scheme (Bitcoin mining lineage) — a cryptography contribution, not a language-modeling or frontier-AI building block","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Blockstream in 2014 and is its CEO; a Bitcoin infrastructure company, not a language-modeling company","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q348671 lists occupation 'cryptographer' and notable work 'Hashcash'; no AI/ML occupation, affiliation or work","source_url":"https://www.wikidata.org/wiki/Q348671","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Adam Back is a cryptographer who invented Hashcash (proof-of-work); occupation 'cryptographer', notable work 'Hashcash' — no AI/ML or language-modeling occupation or notable work listed","source_url":"https://www.wikidata.org/wiki/Q348671","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Blockstream (2014) and is its CEO; PhD (distributed systems) Exeter 1995; no publications or systems in embeddings, representation learning, or the transformer/LM lineage","source_url":"https://en.wikipedia.org/wiki/Adam_Back","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author record shows 1 indexed paper, 41 citations, h-index 1 — no machine-learning or language-modeling corpus","source_url":"https://www.semanticscholar.org/author/Adam-Back/2058627663","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["PhD Computer Science (distributed systems), University of Exeter, 1995"],"canonical_papers":[],"first_verifiable_year":1997,"notable_systems":["Hashcash","Blockstream / Liquid Network"],"citations":41,"h_index":1,"patents":0,"dossier_notes":"Dossier's OpenAlex block is found:false (no match) and PubMed flags homonym_risk with 355 hits and no filtering possible; neither was usable. Wikipedia/Wikidata confirm identity and PhD details cleanly. His genuine technical depth is in cryptography and distributed systems (proof-of-work, Bitcoin-adjacent protocol design), not AI/ML -- the rubric's core dimensions score near zero not because his record is weak in general, but because it is in an entirely different field.","years_language_modeling":0,"years_as_technical_founder":11,"technical_founder_companies":1,"technical_founder_roles":["Blockstream — co-founder & CEO (technical, cryptography) — 2014–present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.86,"duration_ms":52138},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"scientific_founder":5},"confidence":0.9,"duration_ms":45520}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Adam_Back","wikipedia_title":"Adam Back","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"andy-jassy","name":"Andy Jassy","title":"President & CEO","company":"Amazon","sector":"general","rank":102,"rubric_version":4,"dimensions":{"foundations":1,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":4,"industry_impact":8,"scientific_founder":3},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":13,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":13,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.88,"rationale":"Andy Jassy has no personal research, publication, or engineering record in AI: he holds an AB from Harvard and an MBA from Harvard Business School, joined Amazon in 1997 in a marketing role, and built his career in product management and business leadership. He founded and led AWS (2003-2021) as an executive, which today is critical infrastructure for AI training and inference (Bedrock, Trainium, SageMaker), but this is organizational/business leadership of a cloud platform, not personal authorship of AI research, models, or the math/embeddings/transformer lineage. There is no evidence in Wikipedia, Wikidata, OpenAlex, PubMed, or web search of any authored paper, patent, degree in a technical field, or hands-on coding/model-building record. industry_impact is scored moderately (not low) because AWS under his leadership became foundational infrastructure many AI labs run on, but foundations/vector_embeddings/transformers_lm are near-floor since he has no personal technical record in the core lineage — exactly the 'famous CEO with no personal research or engineering record scores low' anchor case for the core dimensions.\n\nJassy has no personal position in the frontier-model lineage: no authored method, architecture, dataset, optimizer, tokenizer, embedding, or pre-2013 distributional-semantics/vector-space patent or shipped system — AWS supplies infrastructure (Bedrock, Trainium, SageMaker) that AI labs rent, but that is organizational output, not his own contribution to the foundation, so frontier_founder floors at 2. There is zero verifiable hands-on language-modeling record (no papers, patents, or systems he personally built) across any era, so lm_domain_depth is ~0 years and lm_domain_breadth shows no distinct LM domain with a hands-on record — both near the floor. He founded and led AWS from 2003 and served as its CEO 2016–2021, but as a product/business executive whose science and engineering were done by the AWS technical staff — he did not author the core code, papers, or patents — which is exactly the anchor's 'founder/CEO of a company whose science was done by others,' scored 3 rather than the technical-founder band. He holds an AB (Government) from Harvard and an MBA from Harvard Business School with no technical degree, thesis, or publication.","evidence":[{"claim":"AB Harvard University, MBA Harvard Business School; joined Amazon 1997 in a marketing role","source_url":"https://en.wikipedia.org/wiki/Andy_Jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded and led AWS from inception in 2003, CEO of AWS 2016-2021, then Amazon CEO from July 2021","source_url":"https://www.businessbecause.com/news/mba-degree/7456/andy-jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex, PubMed, or Semantic Scholar record found for Andy Jassy","source_url":"https://en.wikipedia.org/wiki/Andy_Jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jassy holds a BA in Government from Harvard College and an MBA from Harvard Business School; joined Amazon in 1997 as a marketing manager; conceived AWS with Bezos in 2003, led the original 57-person team, AWS CEO 2016-2021, Amazon CEO from July 2021. No engineering credentials, research background","source_url":"https://en.wikipedia.org/wiki/Andy_Jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q41812531 lists his education as Harvard Business School (MBA) and Harvard University (BA) and his occupation solely as 'business executive'","source_url":"https://www.wikidata.org/wiki/Q41812531","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"PubMed search 'Jassy A[Author]' returns 0 results","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Jassy+A%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"AWS custom AI silicon (Trainium, Inferentia) is built by the AWS organisation Jassy founded and led","source_url":"https://aws.amazon.com/ai/machine-learning/trainium/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jassy founded and led AWS from its inception and was AWS CEO April 2016–July 2021, then Amazon CEO from July 2021; occupation listed solely as 'business executive'","source_url":"https://en.wikipedia.org/wiki/Andy_Jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex, Semantic Scholar, or PubMed authorship record exists for Andy Jassy; PubMed 'Jassy A[Author]' returns 0 results","source_url":"https://pubmed.ncbi.nlm.nih.gov/?term=Jassy+A%5BAuthor%5D","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q41812531 lists education as Harvard University (BA) and Harvard Business School (MBA), occupation 'business executive' — no technical/research field","source_url":"https://www.wikidata.org/wiki/Q41812531","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q41812531 lists occupation solely as 'business executive'; education is Harvard BA and HBS MBA — no technical/research role or degree","source_url":"https://www.wikidata.org/wiki/Q41812531","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Jassy founded and led AWS as its CEO (business/organizational leadership) from inception; no OpenAlex, Semantic Scholar, PubMed or patent record exists for him","source_url":"https://en.wikipedia.org/wiki/Andy_Jassy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["AB, Harvard University","MBA, Harvard Business School"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["Amazon Web Services (organizational leadership, not personal engineering)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier correctly shows no OpenAlex/PubMed/Semantic Scholar record — this is an accurate absence, not a coverage gap; no homonym issue to flag. No first_verifiable_year exists in the AI-research sense; his career start (Amazon 1997) is a business, not technical-research, timeline.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.95,"duration_ms":43226},{"pass":"pass_2","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.95,"duration_ms":38896}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Andy_Jassy","wikipedia_title":"Andy Jassy","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"changpeng-zhao","name":"Changpeng Zhao","title":"Founder","company":"Binance","sector":"crypto","rank":103,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":0,"transformers_lm":1,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":8,"industry_impact":3,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":13,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":13,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.81,"rationale":"Zhao holds a genuine Bachelor of Science in Computer Science from McGill University and built a real, verifiable software-engineering career prior to Binance: developing trade-order-matching software for a Tokyo Stock Exchange subcontractor, four years as a developer on Bloomberg Tradebook's futures trading software, and founding Fusion Systems (2005, Shanghai), which built high-frequency trading platforms, followed by technical/CTO roles at Blockchain.info and OKCoin. This is substantive hands-on systems/infrastructure engineering — low-latency matching engines and trading systems — earning credit on hands_on_engineering, but it is financial-trading-systems engineering, not AI/ML research, and no source found documents any personal AI/ML paper, patent, or authored model. Binance (founded 2017) is a cryptocurrency exchange; recent Binance-branded AI products (e.g. 'Binance Agent OS') are corporate initiatives of the company, not documented as Zhao's personal technical work. Per the rubric, Binance's market dominance, CZ's net worth, and media presence are excluded from scoring; on the specific core-AI dimensions (foundations math depth in the ML sense, embeddings, transformer/LM lineage, industry impact whose core is AI systems) his personal record is minimal to none.\n\nZhao's verifiable technical record is financial trading-systems engineering — trade-order-matching software for a Tokyo Stock Exchange subcontractor, Bloomberg Tradebook futures software, and Fusion Systems' high-frequency trading platforms — with no work anywhere in the AI language-modeling lineage: no vector-space/distributional-semantics or LSI patents, no embeddings, seq2seq, attention or transformer contributions, so his position in the foundation today's frontier models descend from is nil (OpenAlex not found; PubMed hits are surname/initial homonyms). He has zero verifiable years in language modeling — order-matching engines and HFT are not language modeling in any domain (natural-language, biological, financial-as-language, code, etc.), so no distinct LM domain qualifies. He did operate as a genuine technical founder of Fusion Systems (2005) building trading platforms he personally engineered, but that company is outside the AI/language-modeling field, and Binance is a crypto exchange whose AI-branded products are corporate initiatives, not his personal science — placing scientific_founder in the 'technical founder outside this field' band. Net worth, Binance's dominance and media presence are excluded per rubric.","evidence":[{"claim":"Changpeng Zhao earned a Bachelor of Science in Computer Science from McGill University in Montreal.","source_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Zhao developed trade-order-matching software for a Tokyo Stock Exchange subcontractor, worked four years as a developer on Bloomberg Tradebook's futures trading software, and founded Fusion Systems (2005, Shanghai) building high-frequency trading platforms, before serving as head of development at B","source_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Zhao founded Binance in July 2017; no AI/ML research publications or patents are documented for him, and Binance's recent AI-branded products (e.g. 'Binance Agent OS') are described as company initiatives, not personal technical work by Zhao.","source_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BSc Computer Science, McGill University; Tokyo Stock Exchange order-matching software, four years at Bloomberg Tradebook building futures trading software, founded Fusion Systems 2005 (high-frequency trading platforms), Blockchain.info head of development 2013, OKCoin CTO, co-founded Binance July 20","source_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q52714313: educated at McGill University School of Computer Science; occupation businessperson / chief executive officer; employer Binance; no research or academic properties recorded","source_url":"https://www.wikidata.org/wiki/Q52714313","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q52714313 records education at McGill University School of Computer Science and occupation businessperson/CEO, employer Binance — no research or academic properties.","source_url":"https://www.wikidata.org/wiki/Q52714313","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Zhao founded Fusion Systems in 2005 building high-frequency trading platforms, and earlier developed order-matching software for a Tokyo Stock Exchange subcontractor and futures trading software at Bloomberg Tradebook — financial trading systems, not AI/language modeling.","source_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q52714313 records Zhao's occupation as businessperson/CEO with employer Binance and no research or academic properties; OpenAlex returns no author record, consistent with no publication or language-modeling research history.","source_url":"https://www.wikidata.org/wiki/Q52714313","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Computer Science, McGill University"],"canonical_papers":[],"first_verifiable_year":2005,"notable_systems":["Fusion Systems (2005, high-frequency trading platforms)","Binance (2017, cryptocurrency exchange)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's OpenAlex block was not found (found=false), consistent with Zhao having no academic publication record. Zhao's 2023 guilty plea to a Bank Secrecy Act violation at Binance is a real, well-documented legal matter but does not fall under either defined penalty category (bought_popularity or capital_without_competence) and was not scored as a penalty; it is noted here only for context, not applied to dimension scores or penalties.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Fusion Systems — founder, HFT trading platforms — 2005+ (technical founder OUTSIDE the AI/language-modeling field)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.85,"duration_ms":49684},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":6},"confidence":0.85,"duration_ms":43497}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Changpeng_Zhao","wikipedia_title":"Changpeng Zhao","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"brian-armstrong","name":"Brian Armstrong","title":"Co-founder & CEO","company":"Coinbase","sector":"crypto","rank":104,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":1,"transformers_lm":1,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":6,"industry_impact":2,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":12,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":12,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.82,"rationale":"Armstrong holds a dual BA in economics/computer science (2005) and an MS in computer science (2006) from Rice University, giving him general CS graduate training but no evidence of coursework, thesis, or research in linear algebra/optimization/statistical learning specifically, or any ML specialization. His pre-Coinbase roles (developer at IBM, consultant at Deloitte, software engineer at Airbnb working on payments infrastructure) are standard generalist software engineering with zero AI/ML content. As a technical co-founder he personally built early Coinbase engineering infrastructure (a crypto exchange, not an AI/LM system), which supports a modest hands_on_engineering score for production systems built and led, but none of that work touches vector embeddings, attention, transformers, or language modeling. No papers, patents, or verifiable ML research record were found; the Semantic Scholar 35-paper/h-index-12 match in the dossier is an unverified name match (no affiliation/co-author overlap with Armstrong) and should be treated as a homonym, not credited. Coinbase's core product is a cryptocurrency exchange, not an AI/language-modeling system, so per the rubric's explicit exclusion industry_impact is scored low despite the company's scale, fundraising and fame; Coinbase's AI-agent-commerce initiatives (x402, AgentKit) could not be confirmed as Armstrong's personal technical authorship versus other Coinbase engineers.\n\nNo part of Armstrong's own work sits in the lineage today's frontier language models descend from: there is no verifiable pre-2013 distributional-semantics / vector-space / LSI work, and none of the later blocks (attention, transformers, embeddings, optimizers, tokenizers, pretraining objectives, scaling or alignment methods) — his record is a cryptocurrency exchange, so frontier_founder is 0-2. He has zero verifiable years in language modeling of any kind (statistical/neural LMs, vector-space text models, seq2seq, transformers): his CS degrees and pre-Coinbase roles (IBM, Deloitte, Airbnb payments) and Coinbase itself contain no LM research or systems, so lm_domain_depth and lm_domain_breadth both sit at 0-2 with zero domains. He is, however, a genuine hands-on technical co-founder — he personally built early Coinbase engineering infrastructure and co-founded NewLimit — but of companies OUTSIDE this field (a crypto exchange and an anti-aging biotech, neither an AI/language-modeling system), which places scientific_founder in the 3-7 'technical founder outside this field' band rather than higher. The Semantic Scholar 35-paper/h-index-12 match is an unverified homonym with no affiliation or co-author link and is not credited to him.","evidence":[{"claim":"Armstrong earned a dual BA in economics and computer science from Rice University in 2005, followed by an MS in computer science in 2006.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Before Coinbase, Armstrong worked as a developer at IBM, a consultant at Deloitte, and a software engineer at Airbnb (2011) on international payments systems.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Wikipedia biography contains no mention of AI, machine learning research, or patents associated with Armstrong's career.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Armstrong founded Coinbase in 2012 with Fred Ehrsam via Y Combinator; his $8.8B net worth is classified by Forbes as self-made (Self-Made Score 8/10), consistent with VC/YC funding rather than inherited wealth.","source_url":"https://www.forbes.com/profile/brian-armstrong/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records Armstrong's only formal 'educated_at' entry as Bellarmine College Preparatory (high school) and lists occupation including 'software engineer', with no ML/AI research affiliations.","source_url":"https://www.wikidata.org/wiki/Q64705310","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Dual BA economics and computer science, Rice University 2005; MS computer science 2006; developer at IBM, consultant at Deloitte, software engineer at Airbnb from 2011; co-founded Coinbase 2012; no research publications, patents or AI systems documented","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists occupation as chief executive officer / business executive / software engineer with employer Coinbase, and records no doctorate, academic affiliation or research output","source_url":"https://www.wikidata.org/wiki/Q64705310","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Armstrong co-founded Coinbase in 2012 as a technical co-founder/software engineer (Rice MS CS 2006); Wikipedia biography records no AI, machine-learning, or language-modeling research, papers or patents.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists occupation as CEO/business executive/software engineer with employer Coinbase and records no doctorate, academic affiliation, or research output in AI/language modeling.","source_url":"https://www.wikidata.org/wiki/Q64705310","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Coinbase is a cryptocurrency exchange platform Armstrong co-founded with Fred Ehrsam; its core is crypto trading/custody, not AI or language modeling.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Armstrong's Wikidata record lists occupation as CEO/business executive/software engineer with employer Coinbase and no doctorate, academic affiliation, or research output; no ML/AI or vector-space work.","source_url":"https://www.wikidata.org/wiki/Q64705310","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Armstrong is co-founder of NewLimit, an anti-aging startup — a second founder role, but outside language modeling.","source_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Rice University — BA Economics & Computer Science (2005)","Rice University — MS Computer Science (2006)","Bellarmine College Preparatory (high school, per Wikidata)"],"canonical_papers":[],"first_verifiable_year":2011,"notable_systems":["Coinbase (cryptocurrency exchange, co-founded 2012)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier omitted Rice University education entirely (Wikidata only listed his high school); confirmed via Wikipedia/Forbes. OpenAlex not found. Semantic Scholar match (35 papers, h-index 12, name_exact=false, 8 candidates, no affiliations) is almost certainly a homonym ('B. Armstrong' is a common name pattern) with no corroborating affiliation or co-author link to Coinbase/Rice/payments engineering — not credited in scoring. PubMed match explicitly flagged homonym_risk=true with 1651 results, not usable. No patents found. Coinbase funding was Y Combinator/VC-backed and Forbes classifies Armstrong's wealth as self-made, not family/inherited — capital_without_competence does not apply on current evidence, and Coinbase is a crypto exchange, not marketed as an 'AI company' per se, so the penalty's second prong also fails. Could not confirm or deny Armstrong's personal technical involvement in Coinbase's x402/AgentKit AI-agent-commerce initiatives (Coinbase blog page returned 403); treated as unconfirmed and not credited to him personally.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Coinbase — technical co-founder / CEO, 2012–present (crypto exchange, outside the LM field)","NewLimit — co-founder, anti-aging biotech (outside the LM field)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.86,"duration_ms":41904},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.9,"duration_ms":52845}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Brian_Armstrong_(businessman)","wikipedia_title":"Brian Armstrong (businessman)","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"humayun-sheikh","name":"Humayun Sheikh","title":"Founder & Chairman","company":"Fetch.ai","sector":"crypto","rank":105,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":1,"transformers_lm":2,"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"hands_on_engineering":3,"industry_impact":6,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":12,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":12,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.51,"rationale":"Humayun Sheikh is publicly known primarily as an early investor in DeepMind and as founder/chairman of Fetch.ai, a project combining distributed-ledger technology with autonomous 'AI agents.' No Wikipedia page exists for him, and this session's attempts to independently verify his educational background or any personal technical AI/ML record via multiple sources (Fetch.ai's own site, Crunchbase, LinkedIn, Bloomberg, Forbes) were blocked (403/404 errors) or returned no biographical content, so his formal education and prior technical career could not be confirmed from primary sources this session. Being an early investor in DeepMind demonstrates investment judgment and access, not personal research or engineering competence, and per the rubric does not count toward the AI-research dimensions. Fetch.ai markets itself heavily around 'AI agents' and has a working codebase and published technical papers as an organization, which supports a modest industry_impact score for having built a real (if crypto-token-centric) AI-agent platform, but no evidence was found that Sheikh personally authored research, wrote production ML code, or holds an AI/ML-relevant degree, so the personal-research and hands-on-engineering dimensions are scored near the 'nothing verifiable' floor rather than assumed.\n\nNo verifiable contribution of Sheikh's own to the frontier-model lineage exists: the OpenAlex and PubMed hits are confirmed homonyms (a Bangladesh materials-engineering academic and a US headache researcher), and there are no authored papers, patents, or shipped vector-space/embedding/transformer systems attributable to him, so frontier_founder sits at the floor. His only language-modeling claims — '20 years developing ML/AI algorithms for market trading and price prediction' — come from his own interview and are uncorroborated by any paper, patent or named system, so per the self-claim rule they score as absent: 0 verifiable language-modeling years and 0 verifiable LM domains. At Fetch.ai (co-founded Aug 2017) the technical/scientific roles are held by others — Toby Simpson (co-founder & CTO, AI/complex-systems background) and Thomas Hain (co-founder, a speech/language-processing academic) — while Sheikh is the CEO/founding-investor on the business side, which places scientific_founder in the 3-7 band (founder of an AI company whose science and engineering were done by others). His education is Electrical/Electronics/Communications engineering (NED University, 1995) plus a Cambridge Judge entrepreneurship program, with no AI/ML research record.","evidence":[{"claim":"No Wikipedia page exists for Humayun Sheikh (dossier wikipedia.found=false, wikidata.found=false), and this session's attempts to fetch Fetch.ai's own team/about pages, Crunchbase, Bloomberg, and Forbes profile pages for independent biographical verification all failed (404/403 errors) or returned n","source_url":"https://fetch.ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fetch.ai is publicly described as a project combining blockchain/token infrastructure with autonomous AI agents (ASI:One, Agentverse), with Sheikh as founder/chairman; his role is consistently described in available secondary characterizations as investor/chairman rather than as a technical architec","source_url":"https://fetch.ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fetch.ai builds AI platforms and services: ASI:One personal assistant, Agentverse agent marketplace, Fetch Business and the uAgents developer framework; the site carries no biography of the founder","source_url":"https://fetch.ai/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"A Wikidata search for 'Humayun Sheikh Fetch.ai' returns no matching entries — no person-level structured record exists","source_url":"https://www.wikidata.org/wiki/Special:Search?search=Humayun+Sheikh+Fetch.ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's OpenAlex match A5025548219 is 'Sheikh Muhammad Humayun Kabir', Chittagong University of Engineering & Technology, publishing on steel plastic-strain memory, aluminium composites and wet cooling towers — not the Fetch.ai founder","source_url":"https://api.openalex.org/authors/A5025548219","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The ASI Alliance unites Fetch.ai, SingularityNET and CUDOS; the site names no individual leaders or founder biographies","source_url":"https://superintelligence.io/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Fetch.ai was co-founded in August 2017 by Humayun Sheikh with Toby Simpson (co-founder & CTO, AI/complex-systems and artificial-life background) and Thomas Hain, i.e. the technical direction sits with the co-founders, not Sheikh, who is CEO and a founding DeepMind investor with a commodities-trading","source_url":"https://iq.wiki/wiki/humayun-sheikh","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Toby Simpson is Fetch.ai's co-founder and CTO, bringing the technical knowledge in artificial intelligence and complex systems to the organization.","source_url":"https://iq.wiki/wiki/toby-simpson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The dossier's OpenAlex match A5025548219 ('Sheikh Muhammad Humayun Kabir', Chittagong University of Engineering & Technology — steel/aluminium/cooling-tower papers) is a homonym mismatch, and no Wikipedia/Wikidata person record for the Fetch.ai founder exists, leaving no verifiable AI-research or pa","source_url":"https://api.openalex.org/authors/A5025548219","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Thomas Hain is Fetch.ai's Chief Science Officer and co-founder (Sheffield-based speech/language researcher), i.e. the technical/scientific founder rather than Sheikh.","source_url":"https://theorg.com/org/fetch-ai/org-chart/humayun-sheikh","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No Wikipedia or Wikidata person-record exists for Humayun Sheikh, and the OpenAlex author record is a confirmed homonym (Chittagong University materials/mechanical engineering), leaving no verifiable personal research or language-modeling publication record.","source_url":"https://www.wikidata.org/wiki/Special:Search?search=Humayun+Sheikh+Fetch.ai","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":[],"canonical_papers":[],"first_verifiable_year":2010,"notable_systems":["Fetch.ai (founder/chairman) — AI-agent + blockchain platform"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's OpenAlex block (affiliations: Chittagong University of Engineering & Technology, Bangladesh; papers on cooling towers, structural steel, aluminum metal matrix composites) is a confirmed HOMONYM MISMATCH — this is a mechanical/materials engineering academic in Bangladesh, not Fetch.ai's UK-based Humayun Sheikh, and was excluded entirely from scoring. No Wikipedia/Wikidata entry exists, and profile_page/self_claims were both empty in the dossier, leaving very little verifiable primary-source material; multiple independent web sources (Crunchbase, Bloomberg, Forbes, LinkedIn, Fetch.ai's own about/team pages) were attempted and all failed to return usable biographical content in this session (403/404/no-content errors), so his formal education and any personal technical record could not be confirmed either way. Given the near-total absence of verifiable primary-source evidence, scores and confidence are set conservatively low rather than assuming either a strong or weak record — this should be treated as a data-scarcity case, not a confirmed 'nothing exists' case.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Fetch.ai — CEO / founding investor (business, not technical, founder; CTO is Toby Simpson) — 2017-present"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.66,"duration_ms":57539},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":2,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.55,"duration_ms":72260}],"source":"seeded","profile_url":null,"wikipedia_title":null,"scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"marc-benioff","name":"Marc Benioff","title":"Chair & CEO","company":"Salesforce","sector":"general","rank":106,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":2,"transformers_lm":2,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":4,"industry_impact":7,"scientific_founder":3},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":12,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":12,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.85,"rationale":"Marc Benioff has a BS in business administration from USC (1986) and spent 13 years at Oracle in sales, marketing, and product-development (business) roles before founding Salesforce in 1999 — no CS/math degree, no personal ML/AI research record. His OpenAlex-listed works (a clean, unambiguous identity match) are all business/trade books and essays: 'Behind the Cloud' (Salesforce origin story), 'Compassionate Capitalism', 'The Business of Changing the World', and a 2005 DTIC cyber-policy piece — none are technical AI, math, or engineering papers. The dossier's PubMed hits (10 results, 1948-1973, lung-cancer/cytology papers by 'Benioff MA') are a clear homonym from a different era and field and were excluded. He wrote assembly code for Macintosh as an intern in his youth, which is real but pre-professional, general-purpose (non-AI) programming exposure, not AI engineering. Salesforce under his leadership acquired MetaMind (Richard Socher's company, 2016) and built Einstein AI / Agentforce as AI-feature layers on a CRM/SaaS business whose core is not AI systems — Benioff is the business/CEO figure directing AI strategy and acquisitions, not a personal researcher or builder of these systems, so industry_impact is scored moderate (real AI product surface exists at scale, but the rubric requires the company's CORE to be these systems, which Salesforce's is not) while the four research/engineering dimensions are scored at 'nothing verifiable' to 'minimal' per the rubric's anchors.\n\nNothing in Benioff's verifiable record sits in the lineage that today's frontier language models descend from: no pre-2013 vector-space, distributional-semantics or term-relationship-network papers, patents or shipped systems, and no attention/transformer/embedding/optimizer/tokenizer/pretraining work authored or built by him — his OpenAlex 'works' are trade books (Behind the Cloud, Compassionate Capitalism) and the PubMed hits are a 1948–1973 homonym. He has zero verifiable hands-on years in language modeling of any kind (natural-language, biological, financial, code or otherwise); Salesforce's Einstein and Agentforce are AI feature layers on a CRM/SaaS platform, and its NLP capability came via the MetaMind/Richard Socher acquisition (2016), not from Benioff's own modeling work, so no LM domain can be credited to him. As a founder he is the business/CEO founder of Salesforce, whose engineering was built by technical co-founders Parker Harris, Dave Moellenhoff and Frank Dominguez, not the founder-CTO or Chief Scientist who authored the core code or patents — placing him at the low end of the 'founder/CEO of a company whose science and engineering were done by others' band. He holds a BS in business administration and no CS/math degree, no patents, and no authored technical papers, so years and companies as a technical founder in this field are zero.","evidence":[{"claim":"Marc Benioff earned a BS in business administration from USC (1986) and worked at Oracle for 13 years in sales, marketing, and product development roles before founding Salesforce","source_url":"https://en.wikipedia.org/wiki/Marc_Benioff","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Salesforce is the company Benioff co-founded, chairs, and leads as CEO; its AI products (Einstein, Agentforce) are features on a CRM/SaaS platform","source_url":"https://en.wikipedia.org/wiki/Marc_Benioff","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS in business administration, USC (1986); founded Liberty Software at 15 in 1979 writing Atari games; wrote assembly code for the Macintosh as an Apple programmer during USC; 13 years at Oracle in sales, marketing and product development; no patents, papers or AI research attributed to him","source_url":"https://en.wikipedia.org/wiki/Marc_Benioff","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5021953126: 9 works, 227 citations, top items are trade books (Behind the Cloud 2009, Compassionate Capitalism 2004, The Business of Changing the World 2006) — no research publications","source_url":"https://api.openalex.org/authors/A5021953126","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q1521908: educated at USC Marshall School of Business in business administration; occupation businessperson, CEO, Internet entrepreneur, philanthropist","source_url":"https://www.wikidata.org/wiki/Q1521908","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Salesforce was co-founded by Marc Benioff with Parker Harris, Dave Moellenhoff and Frank Dominguez, who built the software; Benioff came from a 13-year sales/marketing/product career at Oracle with a BS in business administration","source_url":"https://en.wikipedia.org/wiki/Marc_Benioff","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Salesforce's natural-language/AI capability was acquired via MetaMind (Richard Socher's company) in 2016 and productized as Einstein — the modeling was done by acquired researchers, not by Benioff","source_url":"https://en.wikipedia.org/wiki/Salesforce","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5021953126 (Marc Benioff, 9 works, 227 citations) lists only trade books and essays — no vector-space, embedding, transformer or language-modeling research","source_url":"https://api.openalex.org/authors/A5021953126","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Benioff holds a BS in business administration from USC (1986), worked 13 years at Oracle in sales/marketing/product roles, and co-founded Salesforce with technical co-founders who built the platform; no AI research, papers or patents are attributed to him","source_url":"https://en.wikipedia.org/wiki/Marc_Benioff","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex author A5021953126 lists 9 works that are trade books and essays (Behind the Cloud 2009, Compassionate Capitalism 2004), not technical AI/ML research","source_url":"https://api.openalex.org/authors/A5021953126","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Business Administration, University of Southern California (1986)"],"canonical_papers":[],"first_verifiable_year":1986,"notable_systems":["Salesforce CRM platform","Einstein AI (feature layer, not personally built)","Agentforce (feature layer, not personally built)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's PubMed block (10 results, term 'Benioff M[Author]', earliest_year 1948, homonym_risk: false) is actually a clear homonym failure — the papers are lung-cancer/cytology research from 1948-1973 by a 'Benioff MA' who cannot be the businessman born in 1964; the dossier's own homonym_risk flag is wrong here and should have been true. The OpenAlex match itself (candidates: 1) is correctly identified as Marc Benioff the businessman, but his 'works' are all trade books/essays, not research — this is a case where a clean OpenAlex match still yields zero technical AI evidence, exactly the profile the rubric describes as 'famous CEO with no personal research record.'","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.9,"duration_ms":46290},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.93,"duration_ms":46503}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Marc_Benioff","wikipedia_title":"Marc Benioff","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"michael-saylor","name":"Michael Saylor","title":"Executive Chairman & Co-founder","company":"Strategy (MicroStrategy)","sector":"crypto","rank":107,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":1,"transformers_lm":0,"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":4,"industry_impact":4,"scientific_founder":5},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":12,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":12,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.81,"rationale":"Saylor holds a BS from MIT double-majoring in aeronautics/astronautics and the history of science (via an Air Force ROTC scholarship) — real STEM training but not in mathematics, ML, or the core-AI lineage this rubric measures, and no graduate degree or thesis in the field was found. His career is entirely business intelligence and enterprise software: computer simulation modeling and market-forecasting models at The Federal Group and DuPont in the late 1980s, then co-founding MicroStrategy (1989) as a BI/data-analytics software company, and since 2020 pivoting the company (renamed Strategy) into a Bitcoin-treasury vehicle. No personally authored papers, patents in ML/AI (Wikipedia notes 31 patents as of 2016, but their subject matter was not verified as AI/ML-specific in this pass and should not be assumed so), Google Scholar/arXiv/DBLP presence, or built AI models/systems were found — Semantic Scholar's 3-paper/15-citation, h-index-2 record under his name could not be verified as genuinely his (name-only match, high homonym risk, no affiliation match) and is not counted as evidence. His public 'AI' commentary and MicroStrategy's descriptions of AI-adjacent BI tooling are commentary/product-marketing rather than personal research or engineering depth, consistent with the rubric's instruction to not credit commentary/books/talks as research.\n\nNo verifiable contribution sits anywhere in the lineage today's frontier language models descend from: MicroStrategy/Strategy is business-intelligence, OLAP and mobile software, not distributional-semantics, vector-space text modeling, LSI, neural LMs, attention or transformers, and no pre-2013 vector-space or term-relationship-network patents/systems by Saylor were found, so frontier_founder is near-zero. There are zero verifiable years of hands-on language-modeling work (papers, LM/vector-space patents, or shipped LM systems) — the Semantic Scholar hits are an unverified name-match and the PubMed 'Saylor MA' record is an unrelated geriatric-care homonym — so lm_domain_depth and lm_domain_breadth score at the floor with no qualifying domain. Saylor is a genuine technical/founding executive of one company (MicroStrategy, co-founded 1989), doing early computer-simulation and market-forecasting modeling, but that company's core is BI/analytics software, NOT language modeling or the AI core this rubric measures, which places scientific_founder in the 3-7 'technical founder outside this field' band. He founded/co-founded one such company and has no verifiable technical-founder years within the language-modeling field.","evidence":[{"claim":"BS from MIT, double major in aeronautical & astronautical engineering and history of science, via Air Force ROTC scholarship; no graduate degree found.","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Early career: computer simulation modeling at The Federal Group (1987), market-forecasting computer models as an internal consultant at DuPont (1988), before co-founding MicroStrategy (business intelligence software) in 1989 with Sanju Bansal.","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No mention of personal AI or machine learning research in his Wikipedia biography; technical work centered on business intelligence software and data mining applications; as of 2016 held 31 patents (subject matter not independently confirmed as AI-specific in this pass).","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS from MIT (1983-1987), double major in aeronautical & astronautical engineering and history of science; co-founded MicroStrategy in 1989 with Sanju Bansal; CEO until August 2022, then executive chairman focused on bitcoin strategy; granted 31 patents as of 2016; authored The Mobile Wave (2012); no","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MicroStrategy/Strategy builds business intelligence and mobile software — MicroStrategy Analytics, MicroStrategy Mobile and Usher (digital credentials); the article records no dedicated AI or machine-learning products, and Saylor's position is described as strategic and executive rather than technic","source_url":"https://en.wikipedia.org/wiki/MicroStrategy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata lists education at MIT Sloan (BS, aeronautics, history of science), occupations entrepreneur, philanthropist, businessperson, computer scientist, inventor and author, employer Strategy from 1989, notable work The Mobile Wave — no research field or academic affiliation","source_url":"https://www.wikidata.org/wiki/Q6833771","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Saylor co-founded MicroStrategy in 1989 (business-intelligence/analytics software); early career was computer-simulation modeling at The Federal Group and market-forecasting models at DuPont — technical founder work, but in BI/data analytics, not language modeling.","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MicroStrategy/Strategy's products are business-intelligence, mobile and cloud software (Analytics, Mobile, Usher) with no dedicated AI/language-modeling systems recorded; no OpenAlex publication identity exists for Saylor.","source_url":"https://en.wikipedia.org/wiki/MicroStrategy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Saylor did computer-simulation modeling at The Federal Group (1987) and market-forecasting models at DuPont (1988), then co-founded MicroStrategy (business-intelligence software) in 1989 — no AI/language-modeling research or products.","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"MicroStrategy/Strategy builds business intelligence, mobile software (MicroStrategy Analytics, Mobile, Usher) and now operates as a Bitcoin-treasury vehicle; the record shows no dedicated language-modeling or transformer/embedding work.","source_url":"https://en.wikipedia.org/wiki/MicroStrategy","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"OpenAlex found no publication identity; PubMed 'Saylor MA' hits are an unrelated geriatric-care/caregiving researcher (homonym), leaving no verifiable language-modeling research record.","source_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Aeronautics & Astronautics and History of Science (double major), MIT"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":[],"citations":0,"h_index":0,"patents":31,"dossier_notes":"Dossier's OpenAlex block is found:false (no academic-publication identity for this person, consistent with a pure business-executive career). The dossier's Semantic Scholar match (3 papers, 15 citations, h-index 2, name_exact:true but 8 candidates) could not be corroborated as genuinely this Michael Saylor via any independent source in this pass and was excluded from scoring rather than credited. The dossier's PubMed block (105 raw/10 refined hits, earliest_year 2014) is entirely unrelated homonym noise — a 'Saylor MA' co-author on geriatric-care and caregiving research papers, a different person — and was disregarded. No citable evidence of paid/sponsored AI coverage or of founding an AI-branded venture on family/inherited wealth was found in this pass, so both penalty fields are 0 per the brief's 'if you cannot cite it, report 0' rule; however, the concern is noted here: Saylor has no verifiable personal language-modeling or ML research record of any kind, and MicroStrategy/Strategy's public AI-adjacent messaging should not be read as technical depth.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":1,"technical_founder_roles":["MicroStrategy/Strategy — co-founder & CEO/exec chairman, 1989-present — business-intelligence software, outside the language-modeling field"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":2,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.82,"duration_ms":52825},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":5},"confidence":0.86,"duration_ms":48468}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Michael_J._Saylor","wikipedia_title":"Michael J. Saylor","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"charles-hoskinson","name":"Charles Hoskinson","title":"Founder & CEO","company":"Input Output (Cardano)","sector":"crypto","rank":108,"rubric_version":4,"dimensions":{"foundations":4,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":4,"industry_impact":4,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":10,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":10,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.74,"rationale":"Hoskinson studied analytic number theory at Metropolitan State University of Denver and began but did not complete doctoral coursework in mathematics at the University of Colorado Boulder — partial, incomplete graduate mathematics training with no thesis, no degree conferred, and no published papers found in OpenAlex, Semantic Scholar, or web search. He co-founded Ethereum in 2014 (departing before launch after a dispute over for-profit vs. non-profit structure) and later founded IOHK/Cardano, a proof-of-stake blockchain platform — real blockchain-protocol and cryptography-adjacent entrepreneurship, but this is smart-contract/consensus-protocol engineering, not core AI/ML work. No verifiable evidence exists of Hoskinson personally authoring AI/ML research, building embeddings or language-model systems, or holding any AI-related patent; his public commentary on AI (e.g., Cardano's occasional AI-oracle integrations) is business/product positioning rather than personal technical depth in the field this index measures. Scored near the floor on all AI-specific dimensions per rubric — an incomplete math degree plus blockchain-protocol entrepreneurship does not constitute verifiable AI depth.\n\nNothing in Hoskinson's verifiable record sits anywhere in the vector-space → embeddings → attention → transformer lineage that today's frontier language models descend from: no pre-2013 distributional-semantics/LSI/term-vector patents or shipped systems, no embeddings, attention, tokenizer, optimizer, scaling or alignment work — the OpenAlex miss is total and the Semantic Scholar (planetary-science bolide spherules) and PubMed (gut-microbiome) hits are confirmed homonyms. He has zero verifiable years in language modeling of any kind and zero distinct language-modeling domains (natural-language, biological, financial, code, etc.). As founder-CEO of IOHK and Cardano he sets direction for a proof-of-stake blockchain, but the peer-reviewed cryptography/consensus research is authored by IOG scientists (Aggelos Kiayias et al.) and Hoskinson does not appear as author on any of the 296 IOG library papers — so he is a technical founder outside this field rather than one personally authoring the core research, and he founded zero companies whose core is AI/language-model systems. He is scored near the floor on the three research dimensions and in the low 'technical founder outside this field' band for scientific_founder.","evidence":[{"claim":"Studied analytic number theory at Metropolitan State University of Denver, then began (but did not complete) a Master's/PhD in mathematics at University of Colorado Boulder","source_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded IOHK (Input Output Hong Kong, now Input Output) and the Cardano blockchain platform","source_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hoskinson attended Metropolitan State University of Denver and the University of Colorado Boulder to study mathematics; Wikipedia does not state that he completed a degree. He was one of five original Ethereum founders in late 2013 and was removed in 2014 after a dispute over commercial versus nonpr","source_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Input Output (IOG) research library contains 296 research papers on blockchain, proof of stake, smart contracts, game theory, cryptography and universal composition; Charles Hoskinson does not appear as an author on any of the listed papers","source_url":"https://www.iog.io/en/research/library/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q90574565 lists education at Metropolitan State University of Denver and University of Colorado Boulder, occupation 'mathematician, entrepreneur', and records no degree, no doctoral advisor and no Google Scholar ID","source_url":"https://www.wikidata.org/wiki/Q90574565","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Semantic Scholar record attached to this slug (C. Hoskinson, id 2235822893, 4 papers) consists of planetary-science papers on spherules recovered from the CNEOS 2014-01-08 bolide site, i.e. a different person","source_url":"https://api.semanticscholar.org/graph/v1/author/2235822893","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Hoskinson co-founded IOHK/Input Output and the Cardano blockchain platform and was a co-founder of Ethereum — all blockchain-protocol companies, none whose core is AI/language modeling","source_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"The Input Output (IOG) research library holds 296 papers on blockchain, proof of stake, cryptography and game theory, and Charles Hoskinson does not appear as an author on any of them","source_url":"https://www.iog.io/en/research/library/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q90574565 records education at Metropolitan State University of Denver and University of Colorado Boulder and occupation 'mathematician, entrepreneur', with no degree, no doctoral advisor, no Google Scholar ID and no language-modeling record","source_url":"https://www.wikidata.org/wiki/Q90574565","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"IOG research library holds 296 papers on consensus, cryptography and game theory; Hoskinson does not appear as author on any, so Cardano's science is authored by others","source_url":"https://www.iog.io/en/research/library/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q90574565 lists Hoskinson as mathematician/entrepreneur and founder of Cardano/IOHK, with no Google Scholar ID, no doctoral advisor and no degree recorded","source_url":"https://www.wikidata.org/wiki/Q90574565","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikipedia describes Hoskinson as a cryptocurrency entrepreneur and co-founder of IOHK, Cardano and (pre-launch) Ethereum — a blockchain, not AI/language-modeling, record","source_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar id 2235822893 (4 papers, planetary-science bolide spherules) and PubMed 'Hoskinson C' (gut-microbiome) are homonyms, not this person — no AI/LM authorship exists","source_url":"https://api.semanticscholar.org/graph/v1/author/2235822893","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended (did not complete degree) Metropolitan State University of Denver, analytic number theory","Attended (did not complete PhD) University of Colorado Boulder, mathematics"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["Ethereum (co-founder, departed pre-launch)","Cardano / IOHK"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex has no match at all; Semantic Scholar match (C. Hoskinson, 4 papers, 12 citations) is unverified/likely a different C. Hoskinson and was not relied on given no corroboration. PubMed sample (Saccharomycetes/gut-microbiome research under 'Hoskinson C') is an unrelated homonym and was disregarded. No completed degree and no AI/ML publication record found anywhere; this is a purely business/blockchain-entrepreneurship profile with respect to the AI-specific rubric.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["IOHK / Input Output — founder-CEO — 2015-present (blockchain, OUTSIDE the AI/language-modeling field; core research authored by IOG scientists)","Cardano — founder — 2015-present (proof-of-stake protocol, not an AI/LM system)"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.85,"duration_ms":49637},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.86,"duration_ms":50737}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Charles_Hoskinson","wikipedia_title":"Charles Hoskinson","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"erik-voorhees","name":"Erik Voorhees","title":"Founder","company":"Venice.ai / ShapeShift","sector":"crypto","rank":109,"rubric_version":4,"dimensions":{"foundations":1,"vector_embeddings":1,"transformers_lm":2,"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"hands_on_engineering":2,"industry_impact":4,"scientific_founder":3},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":9,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":9,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.79,"rationale":"Voorhees attended the University of Puget Sound (started 2003) with no computer-science, mathematics or ML degree on record, and his entire pre-AI career is in cryptocurrency: SatoshiDice (2012, sold for $11.5M, later settled SEC unregistered-securities charges), BitInstant (2012, marketing/communications role), and ShapeShift (founded 2014, a crypto exchange). No published papers, patents, degrees, or personal authored technical work in linear algebra, embeddings, or the transformer/LM lineage were found anywhere. His AI venture, Venice.ai (founded 2024), is a privacy-focused proxy/UI layer that explicitly does not train its own models — it routes user requests to existing open-source models via decentralized GPU providers — so founding it demonstrates product/privacy-infrastructure entrepreneurship, not personal AI research or model-building engineering. This is a clear case the rubric flags: a well-known crypto entrepreneur with essentially no personal core-AI technical record; industry_impact is scored only slightly above zero to acknowledge Venice.ai is a real, functioning product (routing/privacy infrastructure) rather than vaporware, while foundations/vector_embeddings/transformers_lm/hands_on_engineering remain at 'nothing verifiable.'\n\nThere is no verifiable position for Voorhees anywhere in the frontier-model lineage: no pre-2013 vector-space, distributional-semantics or term-relationship-network patents or shipped systems, no embeddings, attention, transformer, optimizer, tokenizer, pretraining, scaling or alignment contribution — OpenAlex and Semantic Scholar find nothing and the 66 PubMed hits are flagged homonyms, so frontier_founder is at the floor. His years in language modeling are zero: Venice.ai (2024) is an explicitly non-training privacy/proxy layer that routes users to existing open-source models over decentralized GPUs, which is application of others' models, not LM research or systems work, so lm_domain_depth and lm_domain_breadth score at 'nothing verifiable' with no distinct LM domains to list. On scientific_founder, he is a genuine serial founder (SatoshiDice 2012, ShapeShift 2014, Venice.ai 2024) and Venice.ai is a real AI-branded company, but the science and engineering are done by others (open-source model authors and his team) and no source shows him personally authoring core AI research, code or patents — this is exactly the 3-7 'CEO/founder of an AI company whose science was done by others' case, scored at the floor (3), with 0 verifiable years as a technical founder in this field.","evidence":[{"claim":"Erik Voorhees attended the University of Puget Sound starting 2003, founded SatoshiDice (2012, sold 2013 for $11.5M in bitcoin, later settled SEC charges), worked at BitInstant (2012), and founded ShapeShift (2014) — no CS/ML degree or technical research credentials are documented.","source_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Voorhees is listed as Founder of Venice AI, but Wikipedia's coverage of Venice AI provides no detail on founding date, technical role, or any AI/ML credentials for him.","source_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Venice AI (founded 2024, based in Sheridan, Wyoming) does not train its own models; it provides privacy-preserving access to existing open-source AI models via decentralized GPU compute providers.","source_url":"https://venice.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Education University of Puget Sound (enrolled 2003); career BitInstant Director of Marketing 2012, SatoshiDice co-founder 2012-2013, ShapeShift founder/CEO 2014; Wikipedia attributes no technical or engineering work to him","source_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Venice.ai founded 2024, Venice.ai Inc., Sheridan Wyoming; serves open-source models over decentralised GPU providers and does not train its own models","source_url":"https://venice.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q16886381 lists him as an American businessman and entrepreneur with no academic, research or technical properties recorded","source_url":"https://www.wikidata.org/wiki/Q16886381","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Venice.ai does not train its own models; it provides privacy-preserving access to existing open-source AI models via decentralized GPU compute providers.","source_url":"https://venice.ai/about","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikipedia describes Voorhees solely as a cryptocurrency entrepreneur (ShapeShift, SatoshiDice, BitInstant, Venice AI) with no technical, research or ML work attributed to him.","source_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Voorhees's documented career is cryptocurrency (BitInstant Director of Marketing 2012, SatoshiDice 2012, ShapeShift founder/CEO 2014, Venice AI founder 2024) with no CS/ML degree, papers or patents.","source_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q16886381 records him as an American businessman and entrepreneur with no academic, research or technical properties.","source_url":"https://www.wikidata.org/wiki/Q16886381","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["Attended University of Puget Sound (from 2003); no CS/ML degree documented"],"canonical_papers":[],"first_verifiable_year":2012,"notable_systems":["SatoshiDice (2012)","ShapeShift (2014)","Venice.ai (2024, privacy-focused access layer over open-source models, does not train models)"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"The dossier's wikidata date_of_birth field ('+1901-00-00') is clearly a data-quality error (Voorhees was born 1985) and was ignored. OpenAlex/Semantic Scholar found no record (correctly — he has no academic publication record), and the 66 PubMed hits under 'Voorhees E[Author]' are flagged homonym_risk=true and are almost certainly unrelated medical researchers named Voorhees, not this person (no AI/crypto-relevant affiliation match) — excluded entirely. Could not find a specific citable source substantiating a capital_without_competence claim (e.g. documented family/inherited funding specifically for Venice AI with proof of no technical competence), so that penalty is reported as 0 per the rubric's evidentiary bar, though the general profile (large crypto wealth funding an 'AI' branded company with no personal ML background) is the kind of pattern the rule targets.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.9,"duration_ms":44855},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":3},"confidence":0.82,"duration_ms":41342}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Erik_Voorhees","wikipedia_title":"Erik Voorhees","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"jeremy-allaire","name":"Jeremy Allaire","title":"Co-founder, Chairman & CEO","company":"Circle","sector":"crypto","rank":110,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"hands_on_engineering":5,"industry_impact":4,"scientific_founder":4},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":8,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":8,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.75,"rationale":"Allaire holds a BA in political science/philosophy with an economics concentration from Macalester College (1993) — no technical/mathematics degree. His career is a sequence of internet/fintech entrepreneurship: co-founded Allaire Corporation (creator of ColdFusion, a web application-development platform) in 1995, served as CTO of Macromedia after its acquisition of Allaire Corp working on Flash Player, founded video-platform company Brightcove in 2004, and founded stablecoin/crypto company Circle (issuer of USDC) in 2013. This is genuine, hands-on software/product engineering leadership across multiple companies, but none of it touches the core-AI lineage — no linear algebra/optimization/statistical-learning work, no embeddings, no transformer/LM research or engineering. No AI/ML papers, patents, or systems were found under his name in OpenAlex, Semantic Scholar, or web search. Per rubric, a successful, technically-literate software entrepreneur with zero personal AI/ML research or engineering record scores at the floor on all AI-specific dimensions regardless of his fintech/crypto business success.\n\nNothing in Allaire's verifiable record enters the frontier-model lineage: no vector-space/distributional-semantics/LSI patents, no embeddings, no attention/transformer/pretraining/alignment work — his two Semantic Scholar items are 2002 Macromedia Flash MX/Flash Communication Server white papers, not language modeling, so frontier_founder and both language-modeling dimensions score at the floor. He has zero verifiable years in language modeling (natural-language or any other type — biological, financial, code, etc.), so lm_domain_depth and lm_domain_breadth are 0 and lm_domains is empty. He IS a genuine hands-on technical founder — co-founded Allaire Corporation (ColdFusion) in 1995, was CTO of Macromedia building the MX platform, founded Brightcove (2004) and Circle/USDC (2013) — but every one of these is web/video/fintech, none an AI or language-modeling company, which places scientific_founder in the 3-7 'technical founder outside this field' band rather than higher. Years and company count as a technical founder OF LANGUAGE-MODELING systems are therefore zero.","evidence":[{"claim":"BA political science/philosophy (economics concentration), Macalester College, 1993","source_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Allaire Corporation (creator of ColdFusion) in 1995; served as CTO of Macromedia after its 2001 acquisition of Allaire Corp","source_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Founded Brightcove (online video platform) in 2004 and Circle (crypto/stablecoin company, issuer of USDC) in August 2013","source_url":"https://www.circle.com/leadership/jeremy-allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Graduated from Macalester College in 1993 with a degree in political science and philosophy with a concentration in economics; co-founded Allaire Corporation 1995 with brother J.J.; CTO of Macromedia after the 2001 acquisition, helped develop the MX platform; founded Brightcove 2004; launched Circle","source_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 144226940 ('J. Allaire', name_exact false) has 2 papers: 'Macromedia Flash MX: A next-generation rich client' (2002) and 'Flash Communication Server MX' (2002) — web/multimedia platform white papers, not machine-learning research","source_url":"https://api.semanticscholar.org/graph/v1/author/144226940","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q6181059 records education only at Macalester College, occupation 'businessperson, entrepreneur', employers Circle and Brightcove; no doctoral advisor, no thesis, no Google Scholar ID","source_url":"https://www.wikidata.org/wiki/Q6181059","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Allaire Corporation (ColdFusion) in 1995, CTO of Macromedia after 2001 acquisition, founded Brightcove 2004 and Circle 2013 — all web/video/fintech, no AI/LM company","source_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 144226940 'J. Allaire' has 2 papers, both 2002 Macromedia Flash MX white papers — no ML/embeddings/transformer/language-modeling work","source_url":"https://api.semanticscholar.org/graph/v1/author/144226940","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q6181059 records occupation 'businessperson, entrepreneur', education only Macalester College — no thesis, advisor, or scholarly AI/LM record","source_url":"https://www.wikidata.org/wiki/Q6181059","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Co-founded Allaire Corporation (creator of ColdFusion) in 1995 with brother JJ; became CTO of Macromedia after the 2001 acquisition and helped develop the Macromedia MX platform; founded Brightcove and Circle — all web/video/fintech, none AI/LM research","source_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q6181059 records education only at Macalester College and occupation 'businessperson, entrepreneur' with employers Circle and Brightcove — no thesis, doctoral advisor, or scholarly identifiers indicating LM/AI research","source_url":"https://www.wikidata.org/wiki/Q6181059","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Semantic Scholar author 144226940 ('J. Allaire', name_exact false) shows only two 2002 white papers on Flash MX / Flash Communication Server MX — multimedia platform docs, not distributional-semantics, embeddings, or language-model work","source_url":"https://api.semanticscholar.org/graph/v1/author/144226940","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BA Political Science/Philosophy (Economics concentration), Macalester College"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":["ColdFusion (Allaire Corporation)","Brightcove","Circle / USDC"],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier has no OpenAlex match; Semantic Scholar 'J. Allaire' match (2 papers, name_exact:false) is unverified and not relied on. PubMed hits are unrelated homonyms. No evidence anywhere of AI/ML technical work personally authored by Allaire; his degree is non-technical (poli-sci/philosophy) and his entire career is web/video/fintech product entrepreneurship, not AI research or engineering.","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0,"technical_founder_roles":["Allaire Corporation — co-founder (ColdFusion), 1995-2001 — NOT language modeling","Macromedia — CTO (MX platform), 2001-2004 — NOT language modeling","Brightcove — founder, 2004 — NOT language modeling","Circle — founder/CEO (USDC), 2013-present — NOT language modeling"]},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":0,"lm_domain_breadth":0,"scientific_founder":4},"confidence":0.85,"duration_ms":42823},{"pass":"pass_2","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":1,"scientific_founder":4},"confidence":0.9,"duration_ms":55171}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Jeremy_Allaire","wikipedia_title":"Jeremy Allaire","scored_at":"2026-09-14T09:48:33.896571+00:00"},{"slug":"tim-cook","name":"Tim Cook","title":"Executive Chairman (former CEO), Apple","company":"Apple","sector":"general","rank":111,"rubric_version":4,"dimensions":{"foundations":2,"vector_embeddings":0,"transformers_lm":0,"frontier_founder":0,"lm_domain_depth":0,"lm_domain_breadth":0,"hands_on_engineering":2,"industry_impact":4,"scientific_founder":0},"dimension_labels":{"foundations":"Mathematical Foundations","vector_embeddings":"Vector Embeddings","transformers_lm":"Transformer & LM Lineage","frontier_founder":"Frontier Founder","lm_domain_depth":"Deep Knowledge Domain Expert","lm_domain_breadth":"Cross-Domain Language Modeling","hands_on_engineering":"Hands-On Engineering","industry_impact":"Scientific & Industry Impact","scientific_founder":"Scientific & Technical Founder"},"weighted_score":4,"penalties":{"bought_popularity":0,"capital_without_competence":0},"penalty_evidence":[],"score":4,"tier":"narrative_only","tier_label":"Narrative Only","confidence":0.93,"rationale":"Cook's entire verifiable education and career record is in operations, supply chain, and business management, with no personal research, engineering, or coursework record in linear algebra, optimization, statistical learning, embeddings, or the transformer/LM lineage. He holds a BS in industrial engineering from Auburn (1982) and an MBA from Duke's Fuqua School of Business (1988), then spent his career at IBM, Intelligent Electronics, and Compaq in operations/supply-chain roles before joining Apple in 1998 as SVP of Worldwide Operations, becoming COO and then CEO in 2011. No dossier source (Wikipedia, Wikidata, OpenAlex, Semantic Scholar) shows any authored technical paper, patent, or built AI system personally attributable to him; OpenAlex has no match at all, and the Semantic Scholar record ('T. Cook', non-exact name match, 8 candidate homonyms) is not verifiably the same person. Apple has shipped AI-adjacent products (Siri, Apple Intelligence, on-device ML chips) during his tenure, but per the rubric this is organizational/product leadership of teams built and run by others, not personal technical depth, and his industry_impact score reflects leading the world's most valuable consumer-hardware company (which now ships AI features) rather than any personal AI research or engineering contribution — which is essentially nonexistent and should not be inflated by Apple's AI branding.\n\nNothing in Cook's verifiable record places any of his own work in the lineage that today's frontier language models descend from — no distributional-semantics, vector-space, embedding, attention, transformer, optimizer, tokenizer, dataset or alignment contribution, pre- or post-2013; his BS is in industrial engineering (Auburn, 1982) and his career is operations and supply chain. He has zero verifiable years of hands-on language-modeling work (research, patents or shipped LM/vector-space systems); the OpenAlex record is absent and the Semantic Scholar 'T. Cook' entry is a flagged non-exact homonym, so no year count is supportable. There are no distinct language-modeling domains with a hands-on record, so lm_domain_breadth is nil. Cook is not a technical or scientific founder of any company — he did not found Apple and joined in 1998 as an operations executive; he sets no personal technical research direction and authors none of the core code/papers/patents, so scientific_founder is essentially zero across zero companies.","evidence":[{"claim":"Cook holds a BS in industrial engineering from Auburn University (1982) and an MBA from Duke University's Fuqua School of Business (1988), where he was named a Fuqua Scholar (top 10% of class)","source_url":"https://www.biography.com/business-leaders/tim-cook","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cook was elected to the National Academy of Engineering in 2026 specifically for 'transformative leadership in the supply chain of consumer electronics' — i.e., for operations/supply-chain leadership, not AI/ML research","source_url":"https://eng.auburn.edu/news/2026/02/tim-cook-nae","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cook's career prior to Apple was in operations roles at IBM (1982-1994) and other companies; he joined Apple in 1998 as SVP of Worldwide Operations, became COO, and CEO in 2011","source_url":"https://en.wikipedia.org/wiki/Tim_Cook","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex author record found for Tim Cook (found: false) and the Semantic Scholar match ('T. Cook') is explicitly a non-exact name match with 8 other candidates, indicating no confirmed personal research publication record","source_url":"https://www.wikidata.org/wiki/Q265852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"BS industrial engineering, Auburn University (1982); MBA, Duke Fuqua (1988); twelve years at IBM's personal computer business ending as director of North American fulfilment; COO at Intelligent Electronics' reseller division; VP corporate materials at Compaq 1997; joined Apple 1998 as SVP worldwide","source_url":"https://en.wikipedia.org/wiki/Tim_Cook","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata records his education (Auburn BS industrial engineering, Duke MBA) and employers (IBM 1982-1994, Apple from 1998), with no academic or research identifiers such as ORCID or Google Scholar.","source_url":"https://www.wikidata.org/wiki/Q265852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Apple describes Apple Intelligence as built by its machine-learning and AI research organisation; Cook's role is as chief executive, not as an author or engineer on those systems.","source_url":"https://machinelearning.apple.com/","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Cook holds a BS in industrial engineering from Auburn (1982) and an MBA from Duke Fuqua (1988); his pre-Apple career was operations roles at IBM, Intelligent Electronics and Compaq, and he joined Apple in 1998 as SVP of Worldwide Operations, becoming COO then CEO in 2011 — no research, engineering-a","source_url":"https://en.wikipedia.org/wiki/Tim_Cook","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex author record exists for Tim Cook and Wikidata records no ORCID/Scholar research identifier, confirming no personal publication record in the language-modeling lineage.","source_url":"https://www.wikidata.org/wiki/Q265852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"No OpenAlex author record exists for Tim Cook and no patents were found; his degrees are BS industrial engineering (Auburn, 1982) and MBA (Duke Fuqua, 1988), with employers IBM (1982-1994) and Apple (from 1998) — an operations/management, not research, trajectory.","source_url":"https://en.wikipedia.org/wiki/Tim_Cook","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"},{"claim":"Wikidata Q265852 records education and employers with no research identifiers (ORCID/Google Scholar) and lists his role as CEO from 2011, not founder of any company.","source_url":"https://www.wikidata.org/wiki/Q265852","verified":true,"verified_at":"2026-09-14T09:48:33.896571+00:00"}],"metadata":{"education":["BS Industrial Engineering, Auburn University (1982)","MBA, Fuqua School of Business, Duke University (1988), Fuqua Scholar"],"canonical_papers":[],"first_verifiable_year":null,"notable_systems":[],"citations":0,"h_index":0,"patents":0,"dossier_notes":"Dossier's OpenAlex lookup returned found: false — correctly no fabricated research record was attached. The Semantic Scholar 'T. Cook' entry (id 2066982687, 3 papers, h-index 2) is flagged non-exact by the dossier itself (name_exact: false, 8 candidates) and could not be verified as this Tim Cook via any independent source; treated as not this person and excluded from scoring/evidence. The dossier's PubMed block is entirely unrelated homonym contamination (radiology AI registries, chemistry, physical therapy, cardiovascular studies by other 'Cook T' authors) and was disregarded — this is the expected, correct outcome for a CEO with no personal technical record, exactly as the task brief anticipated; no homonym mis-attribution risk found that would need correcting upward, only downward (i.e., no reason to credit any of the found records to him).","years_language_modeling":0,"years_as_technical_founder":0,"technical_founder_companies":0},"dossier_sources":{},"passes":[{"pass":"pass_1","dimensions":{"frontier_founder":1,"lm_domain_depth":1,"lm_domain_breadth":0,"scientific_founder":1},"confidence":0.96,"duration_ms":33352},{"pass":"pass_2","dimensions":{"frontier_founder":0,"lm_domain_depth":0,"lm_domain_breadth":0,"scientific_founder":0},"confidence":0.97,"duration_ms":39521}],"source":"seeded","profile_url":"https://en.wikipedia.org/wiki/Tim_Cook","wikipedia_title":"Tim Cook","scored_at":"2026-09-14T09:48:33.896571+00:00"}]}