OpenAI is embedding external safety researchers from Apollo.io within its own facilities for a four-month period. Additionally, Goodfire, METR, and Transluce will conduct evaluations on OpenAI's training processes, dangerous behaviors, and model releases. However, the access and disclosure capabilities for these external teams will be restricted. AI
IMPACT This initiative could lead to more robust AI safety protocols and better risk assessment for advanced models.
RANK_REASON This is a tool/process change by a company, not a core AI release or research.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →