OpenAI chief scientist Jakub Pachocki proposes shared training caps across major AI labs
AntDX316 · x · 2026-11-23
OpenAI chief scientist Jakub Pachocki is proposing that Google, Anthropic, and other major labs agree on when to stop training AI models, warning that AI could start accelerating its own development as early as 2027, with breakthroughs outpacing human understanding.
Key points:
- Labs would commit to shared training limits nobody can cross until safety monitoring improves;
- OpenAI is willing to reveal internal training scores (the newly published perplexity curve) to build trust;
- Commenters are skeptical: given US–China competition, the odds of frontier 'pacing' actually happening are near zero.
The first public call by a frontier lab for cross-institution training pacing — a landmark in AI governance debate.
Related event: OpenAI Chief Scientist Warns of 2027 AI Acceleration, Urges Training Limits(3 posts)→
More from Companies & People
- Mustafa Suleyman and Yuval Harari discuss AI, consciousness and humanity — krishnan · 2026-11-23
- Treasury Secretary blames bank outages on AI-driven security patches, likens them to a vaccine — rohanpaul_ai · 2026-11-23
- Josh Elman pushes back on 'Apple is dead' takes: mainstream AI products take tortoise work — rxwei · 2026-11-23
- Anthropic built a good doc editor inside Claude before Google built a good Claude integration in Docs — sjgadler · 2026-11-23
- Ant's pathfinding and mistakes for AI companies are 'a gift from god' to OpenAI, argues investor — eigenron · 2026-11-23
- Tesla's Christmas Eve 2008 rescue: Musk bet everything days before bankruptcy — elonmusk · 2026-11-23