OpenAI Proposes Perplexity-Based Shared Safety Bars for Pretraining Progress

dhadfieldmenell · x · 2026-11-23

OpenAI's merettm published "Perplexity and Intelligence", arguing for common pretraining limits based on perplexity and other measurements, with progress beyond them gated on understanding AI behavior. Commenters hope other labs mirror this transparency to enable quantifiable pacing standards.

Related event: OpenAI Publishes Pretraining Perplexity Data, Proposes Industry Safety Threshold(41 posts)→

Original post →

More from AGI Musings

AGI Musings channel →