Axiom Math wants to keep AI slop out of mathematics
As AI floods math with unchecked proofs, Axiom Math is betting on verification. The startup’s AI agents just formally verified the Stacks Project, one of the field’s biggest libraries.
TL;DR
- OpenAI has released numerous manuscripts claiming AI solutions to math problems, raising concerns about the accuracy of AI-generated proofs.
- Axiom Math aims to build a verified foundation for AI agents to use for writing software and conducting science.
- The company's AI agents have formally verified the Stacks Project, a large and highly regarded collection of mathematical works.
- Formal verification involves rewriting mathematical proofs in a programming language like Lean, with a proof assistant checking the code line by line.
- This process ensures the logical soundness of mathematical theorems and definitions, similar to how engineers verify chip designs.
- Axiom Math's AI agents translated approximately 7,654 pages of the Stacks Project into about 16 million lines of computer-checked code.
- The verification process for Stacks was complex due to the interconnected nature of its theorems and definitions.
- Axiom Math trained its own AI models to be proficient in mathematics and used a parallel processing approach where agents identified their own tasks.
- Formalizing algebraic geometry, a significant area previously lacking in formalized proofs, is a key achievement of Axiom's work.
- Axiom believes that verified results serve as building blocks for future proofs, enabling AI to create more reliable software and advance scientific discovery.