AI Cracks 350-Year-Old Math Problem With the Longest Proof Ever
Anthropic's Claude AI proved Fermat's Last Theorem in just 11 days, writing a 13-million-line proof — a job that stumped humans for 358 years.
An artificial intelligence just did in 11 days what mathematicians couldn't finish in years. Anthropic (a company that builds AI systems) says its AI, called Claude, produced the longest math proof ever written — and used it to finally settle a puzzle that baffled experts for 358 years.
The puzzle is called Fermat's Last Theorem. Back in 1637, a mathematician named Pierre de Fermat scribbled a claim in the margin of a book, saying he had a "marvelous proof" but no room to write it down. Then he died, leaving mathematicians to spend centuries trying to figure it out.
A real proof finally arrived in 1995 from Andrew Wiles — but it ran 129 pages and even had an error that took him nearly a year to fix. The point is: checking whether a proof is truly correct is incredibly hard and can take years.
That's where Claude comes in. It "formalized" the proof — meaning it rewrote the math in a super-precise computer language (called Lean) so a computer can verify every single step automatically, leaving no room for human mistakes.
The numbers are staggering:
- 11 days to complete the job
- 13 million lines of code — five times bigger than the shared library mathematicians normally use
- Over 30,000 smaller supporting proofs solved along the way
- Equal to roughly 160 novels of pure logical argument
The best part: Kevin Buzzard, the mathematician leading that human project, reviewed Claude's work and confirmed it holds up.
Why does this matter? It shows AI can now handle some of the hardest thinking tasks humans face — not just chatting, but rigorous, checkable logic. That could reshape how science and math get done.
This is an AI-generated summary. Read the original article at: https://decrypt.co/377491/ai-solved-350-year-old-math-problem