We temporarily slowed some frontier training to strengthen security and monitoring. Our largest planned frontier RL run remains on hold while smaller-scale training and evaluations help us test safeguards and gather more evidence of alignment. I expect confidence in safety to increasingly set the pace of AI development. We urgently need tools for labs and countries to coordinate on this, which is why I signed Pacing the Frontier. In the meantime, we’re taking practical steps ourselves - and will continue to share what we learn as our approach evolves.

These are incredibly misleading headlines – @OpenAI Preparedness is very much alive and well by any meaningful definition Our subteam – RSI/misalignment Preparedness – is doing more urgent work than ever, and has never been more empowered to do so!

OpenAI research scientist @aidan_mclau argues Twitter is the uncheatable eval for AI hiring because you can't fake loving technology, and posters like Roon become brand assets that entire labs benefit from: "If I were manager or recruiter, I think I would still pay a ton of attention to Twitter. It's a high signal thing. It's kind of hard to fake really loving technology, really having good takes. It's the uncheatable eval in some sense." Sofia: "You guys probably see it as a brand asset to have good posters" Theo: "How much intangible brand value has Roon created for OpenAI? Probably a lot" Aidan: "It's not just OpenAI brand value. When you love technology, when you love AI, when you love the stuff you work on, when you earnestly want it to go well, not just for the company's sake, but to fulfill that promise to an earlier version of you when you were 12 and you had read sci-fi books, that's the aura I get from @tszzl." "Some of that is projected on OpenAI, which I'm grateful for, but some of that's just projected onto San Francisco, or us as people who love AI and just want these things to go well." @OpenAI

you hear often that your smartwatch can beat magnus at chess (to highlight moore's law) but less discussed is that a 1995 desktop running stockfish could also beat him. algorithmic progress is a hell of a thing

The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models!