OpenAI capabilities researcher Dan Selsam makes public statement on AI risk
connoraxiotes · x · 2026-09-15
Dan Selsam, a current OpenAI capabilities researcher since 2022 who helped pioneer chain-of-thought optimization, has published a personal statement on AI risk via former colleague Daniel Kokotajlo.
- Selsam has 15+ years in AI: probabilistic programming at MIT, early Lean theorem prover development at Microsoft Research, and early neural network reasoning work in his Stanford PhD.
- Commenters note he and fellow pretraining researcher Jacob Coxon are the people actively making models more powerful — not EA-aligned safety advocates — and recent events have genuinely freaked them out.
- The statement is being read as a notable signal of shifting sentiment inside frontier labs.
More from AGI Musings
- Musk proposes rival AI labs cross-test each other's frontier models before release — WhatTheLJW · 2026-09-15
- CoLLAs 2026 tutorial: make continual learning a design requirement, not an afterthought — apsarathchandar · 2026-09-15
- When does driving tech flip from complement to substitute? Fradkin's pushback on pro-worker AI — soumitrashukla9 · 2026-09-15
- Polymarket gives just 21% odds OpenAI declares AGI by end of 2026 — Polymarket · 2026-09-15
- XPeng's IRON humanoid walks off its own assembly line at a factory already 80% automated — cen6wkf · 2026-09-15
- LeCun amplifies: Anxious intellectuals fueled GMO and nuclear power panics, at deadly cost — ylecun · 2026-09-15