The AI (164news.com's guide to ai) Revolution: Proving Fermat’s Last Theorem in 11 Days
A Mathematical Marvel and AI’s Game-Changing Ability
A mathematician funded to formalize Fermat’s Last Theorem has had his work upstaged by AI. Dozens of Claude agents crafted 13 million lines of Lean code in 11 days, achieving what took a human five years.
In a remarkable feat, Anthropic published a computer-checked proof of Fermat’s Last Theorem, a conjecture proposed by Pierre de Fermat in 1637. The proof, involving 30,300 intermediate theorems, was generated by a swarm of AI agents, as detailed in their research post.
Kevin Buzzard, a mathematician at Imperial College London, led a community effort to formalize the theorem since 2024, funded by the EPSRC. He noted that while the AI’s proof adds nothing new to mathematics, it demonstrates the potential of AI in formalizing complex research at an unprecedented pace.
The Cost and Complexity
Anthropic’s achievement raises questions about the cost and effort involved. They estimate that the proof required six billion output tokens from an internal model comparable to Claude Fable 5.1. At $50 per million tokens, this equates to $300,000, though Anthropic notes this is an illustration rather than a precise cost.
Overcoming Challenges
The project faced initial setbacks when agents lost track of the project’s state, but these were overcome by adopting Prove2Me, an open platform designed to manage theorem statements and speed up compilation. This platform, combined with a multi-agent harness, enabled the AI to complete the proof in under two weeks.
In conclusion, this event underscores the transformative potential of AI in mathematics and research, as well as the evolving role of technology in advancing human knowledge.