Episode 77 · The Machine's Theorems
Seven Hundred and Twenty-Two Papers in a Day
The release in OpenAI's own words; attempted vs released numbers and the missing denominators; a guided tour of the repository; how Lean and Comparator checking works, with the actual challenge files for the plane-colouring, Thompson's group F and Unique Games families; the π Lean scope; the sign error (−2m ≠ 0) and the 7 October history entry; withdrawn is not false; the Advisory Group dispute; reactions from Tao, Scholze, Green, Agol, Wilton, Lozano-Robledo and Hales. Keywords: OpenAI math release, GitHub, Lean, Comparator, Mathlib, formalization, withdrawal, sign error, Hodge, Advisory Group, AHM, Tao, Math 2.0
Listen to the show on your preferred podcast platform.
See it work
Explore the working principle
Inside this episode
- Cold open: the parody and the repository
- What was released, in OpenAI's own words
- Attempted versus released: what the numbers do and do not say
- A guided tour of the repository
- Forty-two percent: what a formalization buys you
- How to read one entry: a worked example
- How Lean checking actually works
- Three Lean pages, up close
- The sign error, step by step
- Publish first, check later: is this new?
- Who was consulted: the Advisory Group dispute
- The community reacts, in their own words
- Why it matters, and the verification ledger
Production credits & AI disclosure
Scripts generated with Claude Opus. Synthetic narration generated with Edge TTS. Reviewed, listened to, and approved by Sankalp Thakur.