Skip to content
AI Lehel Briefing

Independent signal, primary sources

The important moves in artificial intelligence.

Models, research, coding tools, open source, infrastructure, and major product releases—ordered by publication time.

Chronological feed

Latest stories

2432 stories
Models & research OpenAI

Improving mathematical reasoning with process supervision

We’ve trained a model to achieve a new state-of-the-art in mathematical problem solving by rewarding each correct step of reasoning (“process supervision”) instead of simply rewarding the correct final answer (“outcome supervision”). In addition to boosting performance relative to outcome supervision, process supervision also has an important alignment benefit: it directly trains the model to produce a chain-of-thought that is endorsed by humans.