Show HN: ProofForge, AI agents whose proofs have to compile in Lean
Show HN: ProofForge, AI agents whose proofs have to compile in Lean Sanexxxx777 / ProofForge Public Notifications You must be signed in to change notification settings Fork 0 Star 0 Branches Tags Open more actions menu Latest commit History 5 Commits 5 Commits Folders and files Name Name Last commit message Last commit date proofs proofs LICENSE LICENSE README.md README.md Repository files navigation ProofForge An AI-agent pipeline that produces machine-verified Lean 4 / Mathlib proofs. Agents decompose a problem, prove the pieces, and formalize them in Lean.
This ProductUpdate is relevant to the technology intelligence record because it involves Google DeepMind, Google, OpenAI. The source article should remain the factual reference for follow-up coverage.
- Sanexxxx777 / ProofForge Public Notifications You must be signed in to change notification settings Fork 0 Star 0 Branches Tags Open more actions menu Latest commit History 5 Commits 5 Commits Folders and files Name Name Last commit message Last commit date proofs proofs LICENSE LICENSE README.md README.md Repository files navigation ProofForge An AI-agent pipeline that produces machine-verified Lean 4 / Mathlib proofs.
- Agents decompose a problem, prove the pieces, and formalize them in Lean.
- The Lean kernel rechecks every step: #print axioms shows only the standard axioms, and native_decide is never used.
- A wrong proof does not compile — so "the AI proved it" is not something you take on trust; the checker either accepts it or it doesn't.
- Contributions to Google DeepMind's formal-conjectures Six pull requests merged into google-deepmind/formal-conjectures : PR Problem What it contributes #4245 Erdős #1084 f₁(n) = n − 1 for unit-distance configurations on a line — the upper bound proved.
- #4244 Erdős #1052 The 5th unitary perfect number, 146361946186458562560000 (24 digits), via multiplicativity of the unitary divisor-sum σ* .