The Researchers

Core contributor on o1 and GPT-4o. Brown cognitive science PhD in vision and computational neuroscience; ex-Vicarious and ex-X, The Moonshot Factory, where he worked on robotics and computer vision.

OpenAI research and engineering contributor since 2021, including work on o1. Former cofounder and CTO of Storm8, the mobile games company behind more than 1B downloads and over $1B in lifetime revenue.

MTS at Anthropic. MIT CSAIL PhD on multi-agent coordination under limited communication, with earlier training in Stanford Symbolic Systems and CS. Previously founded Metamanagement through YC S20 and worked at Dropbox.

AI researcher at Anthropic in London. IOI 2015 silver medalist and HSE graduate. Previously a research engineer at Google DeepMind, where he worked on AlphaDev's faster sorting algorithms.

Anthropic interpretability researcher. MIT PhD under Dimitris Bertsimas, Max Tegmark, and Neel Nanda. Author of "Language models represent space and time" (ICLR 2024) and core contributor on Circuit Tracing / sparse probing / universal neurons. Reverse-engineers how language models compose primitives into higher-level circuits.

Leads Anthropic's Cognitive Oversight team — alignment auditing, honesty, oversight. Harvard math PhD → postdoc with David Bau on interpretability. Co-authored sparse feature circuits, alignment faking, and the WMDP benchmark.

Research engineer at DeepMind. Co-led AlphaProof (IMO silver-medal performance) and co-author on AlphaGo Zero, AlphaZero, and MuZero — basically every flagship RL paper of the last decade. École Centrale Paris engineering diploma + Stanford MS in financial math.

Conceived and led AlphaGeometry — his PhD passion project under NYU's He He that became a Nature paper and an Olympiad-level geometry solver. Vietnamese, four years building it from scratch at Google Brain → DeepMind.

Research scientist at Google DeepMind, incoming Assistant Professor at CMU. Berkeley PhD under Pieter Abbeel. Co-author of OpenLLaMA, RingAttention with Blockwise Transformers (near-infinite context), Large World Model (LWM), and Gemini 2.5.

Senior research scientist at Google DeepMind. Led the overall technical direction for Gemini competitive programming and ICPC 2025 (gold-medal level). Also involved in Gemini Deep Think IMO 2025 gold-medal effort. Post-training research on LaMDA and PaLM 2.

OpenAI research scientist. Led o1-mini and o3-mini development. Stanford PhD under Jure Leskovec on knowledge graphs / reasoning. Major contributor on GPT-4o, o1 system card, OGB benchmark.

Foundational contributor on o1-preview, o1, and Deep Research at OpenAI. Before that, he worked at Google Brain on PaLM and the Flan-T5/Flan-PaLM instruction-tuning lines. Recently moved to Meta Superintelligence Labs.

Co-author of RoBERTa, BART, RAG, and LLaMA 3. Edinburgh PhD on combining symbolic and distributed semantics. Best Paper at EMNLP 2016 and Best Resource at ACL 2017.

Now at Thinking Machines Lab. Previously led LLaMA pretraining at Meta GenAI. Co-author of RoBERTa, LLaMA 1/2/3, BART, OPT, and XLM-R.

Harvard PhD in interpretability (Wattenberg/Viégas/Pfister), funded by Kempner Institute Graduate Fellowship. Author of "Emergent World Representations" (Othello world-model paper, ICLR Oral) and "Inference-Time Intervention" (NeurIPS Spotlight, 1.1K cites). Joined Thinking Machines Lab in 2026 after 10 months at Meta.

Member of Technical Staff at TML. Spent 2.5 years on Microsoft's AI Superintelligence team pre-training the Phi family. Lead author of Samba (ICLR 2025) — simple hybrid state-space architecture — and Phi-4-mini Decoder-Hybrid-Decoder for efficient long-generation reasoning. UIUC PhD with Chengxiang Zhai.

Research lead on Cursor's Composer model series. Northeastern PRL (Arjun Guha) undergrad researcher who shipped MultiPL-E and SelfCodeAlign (NeurIPS) before finishing his degree. Ex-Scale AI, Roblox, Trail of Bits. Code-LLM training methodology specialist.

At Fireworks AI in Mountain View. Stanford PhD with ~4K citations on Scholar (h-index 20) — published-researcher signature in ML. Path through Calico Life Sciences → Waymo → Tesla → Ceramic.ai → Fireworks AI.

Co-founder and CEO of Radical Numerics. Lead author of Hyena Hierarchy (ICML) and co-author of Evo (Science), two important non-Transformer architecture lines from the past several years. Stanford PhD with Stefano Ermon and Chris Ré; founding scientist at Liquid AI.

Co-founder of Radical Numerics. Lead author of Evo (Science 2024) — first long-context 7B foundation model for biology that designs DNA/RNA/proteins zero-shot — and HyenaDNA (NeurIPS). Stanford PhD in Bioengineering co-advised by Steve Baccus (neurobiology) and Chris Ré. Gave a TED2025 talk: "How AI could generate new life forms."

VP of Kernels at Together AI and incoming Assistant Professor at UCSD. Co-inventor of FlashAttention, winner of the Stanford Open Source Software Prize, and co-author on Hyena, H3, FlashFFTConv, Monarch Mixer, and ThunderKittens. Stanford PhD with Chris Ré on efficient ML kernels.

Principal Research Manager at MSR Redmond's Physics of AGI group. First author on "Textbooks Are All You Need" (Phi-1) and central author on Phi-1.5, Phi-3, Phi-4, and Phi-4-reasoning. Her earlier theory work on implicit bias of gradient descent is widely cited.

Co-creator of TinyStories — the result of wondering "could a 4-year-old's vocabulary be enough to train a coherent LM?" while reading to his daughter. Lead/co-author on Phi-1.5, Phi-4. Background is pure mathematics (geometric functional analysis) — not the typical ML pedigree. Won the Erdős Prize in mathematics.

CEO and Co-Founder of Harmonic, the AI lab building mathematical superintelligence. CMU CS BS, Stanford PhD candidate with Stefano Ermon. Previously Co-founder/CTO of Helm.ai. Lead author of "Aristotle: IMO-level automated theorem proving" — Harmonic's Aristotle hit gold-medal level at IMO 2025 and recently produced a formal Lean proof for an Erdős problem.

Research engineer at Harmonic working on post-training and search algorithms. UMich PhD with Ambuj Tewari on principled uncertainty quantification, conformal prediction, and AI for Science. Princeton BA. Ex-Anthropic research fellow on robotics red-teaming, ex-Meta CV/graphics. Continuum Transformers + neural operators for PDEs in the recent pipeline.