Two Lean groups verify Claude-assisted proof of the Riemann hypothesis
Anthropic mathematician Levent Alpöge posted his claim a day after the company's Nasdaq debut, an Italian who worked twelve years without AI released an incomplete proof the same day
Images
Alpöge said he solved the math problem using the Anthropic AI tool Claude a day after the company went public on the Nasdaq stock market index. Photograph: Brendan McDermid/Reuters
theguardian.com
Two organizations announced Saturday that they have independently verified a mathematician's claim that the AI system Claude helped him prove the Riemann hypothesis. The Lean Focused Research Organization and the Mathlib Initiative — both built around the Lean programming language — wrote that they have "full confidence" in the claim and congratulated its author.
The author is Levent Alpöge, a Turkish American mathematician employed at Anthropic, whose note appeared on his personal website on Friday — one day after the San Francisco company went public on Nasdaq. He wrote that Claude had helped him develop the solution over about six weeks ("Claude flew, and danced, and sparkled near that finish line") and linked the proof as Lean code, a formal language a computer can check line by line. That format is what allowed verification within a day rather than a refereeing cycle measured in months.
Alpöge knows how the timing reads. "The precise timing is horrible. If I could have, I'd have finished this all much earlier," he wrote of the post-listing date. The Riemann hypothesis, proposed by Bernhard Riemann in 1859, concerns how prime numbers are distributed along the number line; the Clay Mathematics Institute selected it as one of seven Millennium prize problems, each carrying a million dollars.
The claim lands in a crowded season of AI-and-mathematics announcements. In September, Anthropic's competitor OpenAI said a non-public version of its tools had helped solve part of a different Millennium problem — a version outsiders cannot inspect. Some mathematicians have said they are uneasy about well-capitalized private companies spending millions of dollars on AI aimed at the field's hardest problems. And on the same Friday that Alpöge posted, the New York Times reported, an independent mathematician in Italy, Alessandro Masullo, released an incomplete proof of the same hypothesis.
Masullo has been working on his proof for nearly twelve years. He did not use AI, because he does not fully trust the system.