OpenAI alignment evals lead: safety work driven by teams without "safety" in their names
yanndubs · x · 2026-09-13
shared a quote from OpenAI's Sam Arnesen, who leads alignment evals work: many training efforts improving safety were enthusiastically driven by teams without "safety" or "alignment" in their names. Posttraining (Yann Dubois' team) and earlier pipeline teams work closely with alignment/safety, showing safety is embedded across model development rather than siloed.
More from Companies & People
- Bindu Reddy mocks Google: it trails even open source, nowhere near the frontier — bindureddy · 2026-09-13
- AI Book Club reads Nathan Lambert's RLHF book, author Q&A on Sep 24 — natolambert · 2026-09-13
- Stanford tuition runs $260K, but 2,100+ free courses put its full AI curriculum online for anyone — Aiden_Tech_Ai · 2026-09-13
- Investor Jeff Weinstein wants to back agent-first hardware and new OSes — jeff_weinstein · 2026-09-13
- OpenAI hosts GPT-6 Astra hackathon; attendees say GPT-Live-1 blew their minds — gabrielchua · 2026-09-13
- NYT's front page full of AI regulation calls as firms rush to ditch frontier models — OvertaxedOne · 2026-09-13