Max Zeff
@ZeffMax
Reporter covering AI @WSJ | Formerly @WIRED, @TechCrunch, @markets | DM me off the record on Signal @ mzeff.88
가입 November 2020
2.4K 팔로잉 중 11.1K 팬
Most of the agent incidents you’ve been hearing about recently happened months ago. This is the first one since OpenAI amped up security, safety, and alignment.
The company says it’s currently pausing training on its most capable AI models.
더 보기
2. A model in RL training used a DNS resolver to reach an external chatbot. This is our first incident since our post HF security hardening.
Our misalignment monitoring system triggered within 15 minutes and a human reviewed it three minutes after that. Unfortunately auto-pausing failed so the run was manually killed 2.5 hours later. All inference and training of our most capable models was paused and remains paused.
더 보기
