Skip to main content

AI Proves Fermat's Last Theorem in 11 Days—A Milestone for Math

In a stunning display of artificial intelligence's growing prowess, Anthropic's AI model, Claude, has accomplished what many thought would take years: it formalized the proof of Fermat's Last Theorem in a mere 11 days. But before you imagine the AI conjuring up a brand-new mathematical proof, let's clarify what actually happened.

Fermat's Last Theorem, a problem that stumped mathematicians for over three centuries, was finally proven by British mathematician Andrew Wiles in 1995. His proof, spanning 129 pages, was a monumental achievement that required months of meticulous manual verification. Now, Claude has taken that existing proof and translated it into a language that computers can understand and verify—a process known as formalization.

What Does Formalization Mean?

Think of it this way: when mathematicians write proofs, they often skip steps that seem obvious to them. But computers, being literal-minded, need every single logical step spelled out. That's where Lean comes in—a proof assistant that checks each step of a proof to ensure it's logically sound. Formalizing a proof means converting it into a format that Lean can verify, which is a painstakingly detailed process.

Anthropic had previously estimated that formalizing Fermat's Last Theorem could take several years. So, when Claude did it in under two weeks, the math community took notice.

The AI's Impressive Feat

Claude didn't work alone. It operated through a multi-agent system, collaborating on defining concepts, proving intermediate theorems, and deriving complex propositions. The project, initiated by Anthropic researcher Tianyi Peng, used a platform called Prove2Me, which organizes theorems and their dependencies in a way that allows multiple AI agents to work in parallel.

Over the course of 11 days, Claude generated approximately 13 million lines of Lean code and proved around 30,300 theorems. Of those, about 29,500 were incorporated into the final proof. The entire proof was checked by Lean, relying on only three standard axioms.

A Collaborative Effort

Human researchers provided high-level guidance, such as deciding which mathematical objects to prioritize, but the nitty-gritty proof work was left to Claude. The final proof followed a simplified version of Wiles' proof by Darmon, Diamond, and Taylor, making it more accessible for formalization.

Why This Matters

This achievement isn't about AI replacing mathematicians. Instead, it showcases AI's potential to handle the tedious, time-consuming task of formalization. If this technology can be extended to more modern mathematical results, it could drastically reduce the manual effort required to verify new proofs. As AI continues to generate mathematical insights, providing both human-readable and formalized versions of proofs might become standard practice.

Kevin Buzzard, a mathematician who reviewed the proof, called it an important step forward in AI-assisted formalization. The complete Lean proof is now publicly available on GitHub for anyone to explore.

The Road Ahead

While this is a remarkable milestone, it's just the beginning. The ability to formalize proofs automatically could accelerate mathematical research, allowing mathematicians to build on verified results with confidence. And who knows? Maybe one day, AI will not just formalize proofs but also discover new ones, pushing the boundaries of human knowledge even further.

Key Points

  • Claude formalized Fermat's Last Theorem in 11 days, a task previously estimated to take years.
  • The AI generated 13 million lines of Lean code and proved over 30,000 theorems.
  • This was a multi-agent effort, with human researchers providing high-level guidance.
  • The proof is computer-verified, ensuring its logical correctness.
  • This breakthrough could transform mathematical verification, making it faster and more reliable.

Image

As we look to the future, one thing is clear: the collaboration between human intuition and AI's tireless precision is opening new frontiers in mathematics. And it's happening faster than anyone expected.