OpenAI chief scientist shares rare pretraining numbers, says algorithmic research may automate before alignment by 2027

sjgadler · x · 2026-11-23

OpenAI chief scientist Jakub Pachocki published "Perplexity and Intelligence", an unusually transparent release sharing concrete pretraining progress numbers and arguing for shared safety bars based on perplexity and other measurements, with further progress gated on our ability to understand AI behavior.

Commenters highlight a key claim: algorithmic pretraining research is verifiable and likely to be automated by 2027 — before alignment/monitoring research. The essay also argues that AI participation in pretraining algorithmic research itself may drive a step change, beyond the usual "full recursive self-improvement" framing.

Original post →

More from Models

Models channel →