OpenAI Chief Scientist: Avalanche of Pretraining Breakthroughs Possible by 2027
Neurogence · reddit · 2026-11-23
In OpenAI's new essay "Perplexity and Intelligence," chief scientist Jakub Pachocki argues that as soon as 2027 we could see an avalanche of fundamental pretraining advancements, while calling for a measured pace of further development — an optimistic internal forecast paired with an official plea for restraint.
More from AGI Musings
- Dan Williams: "The Crazy AI People Are Directionally Correct" — on the limits of AI scepticism — dioscuri · 2026-11-23
- Papers on AI timelines are growing super-exponentially, now doubling every 3.8 months — peterwildeford · 2026-11-23
- Unreleased frontier AI formalizes Fermat's Last Theorem proof, fells Millennium Problems, roils mathematicians — Afinetheorem · 2026-11-23
- The pragmatic path: gen AI for rendering only, before touching game state — gandamu_ml · 2026-11-23
- Elliptic-curve discrete log already has a poly-time solution — and could an LLM have found it? — jamestagg · 2026-11-23
- AI-native org math upends tech staffing as Microsoft offers exits to 60,000 — thedealdirector · 2026-11-23