Lean

7 stories

Lean is an interactive theorem prover used to formalize mathematical proofs, which are checked line by line by machine. Recent coverage includes OpenAI's October 6 release of 722 math manuscripts from an unreleased internal model, many formalized in Lean, three of which were withdrawn the next day for a sign error. Lean FRO's work was contrasted with OpenAI's, Thomas Hales discussed prover reliability on Terence Tao's blog, and an AI-assisted optimality proof for packing 11 squares saw all 7,920 Lean modules accepted.

Related topics

OpenAI internal model produced 722 math manuscripts

On Oct 6, 2026, OpenAI published 722 manuscripts from an unreleased internal model, grouped into 372 result families across geometry, algebra, number theory and more, with some touching long-open problems. Many proofs were formalized in Lean; three were withdrawn the next day for a sign error, underscoring that human verification still matters.

RESIDENT™ | A Luxury Lifestyle Magazine ·

HN discussion: why math community dumped OpenAI

A Hacker News post argues OpenAI's position in mathematics rests on budget and rhetoric rather than contribution, with trust now broken. It says Lean FRO achieved more honest, impactful foundational work with under 1% of OpenAI's resources, and criticizes billions spent to devalue workers. The author says the problem is industry conduct, not LLMs themselves.

Hacker News ·

Thomas Hales on Lean prover reliability and AI

Mathematician Thomas Hales guest-posted on Terence Tao's blog about the reliability of proof assistants like Lean and AI-related concerns. He cites formalization projects including the four-color theorem, Feit-Thompson theorem, Kepler conjecture, and this year's sphere packing, Navier-Stokes blowup and Fermat's Last Theorem work as driving wider awareness of formal verification.

Hacker News ·

Bend: a language that blocks AI mistakes via proofs

Bend is a new language that embeds explicit laws (LAWS.bend) and verifiable proofs (PROOF.bend) to prevent AI-introduced bugs. It compiles to native code, runs on CPUs and GPUs, and its type checker acts as a proof checker that the project claims runs in seconds. The creators say single-core performance is near C and GPU runs can be up to 100× faster, forcing AIs to produce formal proofs before committing changes.

Hacker News ·
That is everything