VCG
A bombshell claim had the math world buzzing this weekend. Levent Alpöge, a mathematician at artificial intelligence company Anthropic, announced that he had worked with the company's AI model Claude to produce a proof of the 167-year-old Riemann hypothesis.
"This is a great contribution. You could say it has shaken the world and marked the beginning of a new era in mathematics," Fields Medalist Shing-Tung Yau, dean of Tsinghua University's Qiuzhen College, said on Saturday evening.
Proposed in 1859 by German mathematician Bernhard Riemann, the hypothesis concerns the distribution of prime numbers. Primes, such as 2, 3, 5 and 7, are natural numbers greater than 1 that are divisible only by 1 and themselves. The hypothesis points to an underlying mathematical pattern in their seemingly random distribution along the number line. It is one of the seven Millennium Prize Problems, each carrying a $1 million prize for a correct solution. A proof would have far-reaching implications across mathematics and could also have implications in fields such as cryptography and physics.
Levent Alpöge, a mathematician at Anthropic. /Archives of the Mathematisches Forschungsinstitut Oberwolfach
Alpöge made the announcement on Friday, releasing more than 17 million lines of code written in Lean, a programming language in which a computer checks every step of a proof. He said Claude had also generated a written draft running to more than 250,000 pages, which he held back because it was too unwieldy to read. Alpöge said he did not yet fully understand all the mathematics in the argument.
A day later, the Lean and Mathlib communities said in a joint statement that Alpöge's Riemann hypothesis proof had passed community verification.
The work has sparked debate among mathematicians worldwide over what an AI-assisted proof would mean for the field and how it might change the way research is done.
"The page count doesn't matter. What matters is readability and no one can read 250,000 pages," Fields Medalist Deng Yu, a professor of mathematics at the University of Chicago, said on Saturday. He said a more readable version might emerge soon and that he would reserve judgment until then.
Fields Medalist Shing-Tung Yau speaks on stage, Beijing, March 28, 2025. /VCG
"AI will become an indispensable part of our research," Yau said. "As for human ingenuity, I think machines still have a long way to go." He said AI was far better at solving problems than at posing them. In his view, asking questions takes more creativity and the key is knowing which questions matter.
"If you know a theorem is true, you can use it. If you understand why it's true, you can try applying the same method to problems we still haven't solved," said Artan Sheshmani, a researcher at the Beijing Institute of Mathematical Sciences and Applications. He said he wanted to understand what mathematical principle made the Riemann hypothesis true, assuming the proof holds up, and where else that principle might apply.
Alpöge's account also offered a glimpse of Claude's role in shaping the argument. He said a simpler mathematical problem had led him to explore a new approach. Later, Claude suggested how many times to repeat a key mathematical operation. Sheshmani saw that as a creative contribution. AI can surprise experts with an idea and explore where it leads, he said, while Lean checks the logic. Mathematicians still have to decide what is worth studying, interpret the findings and take responsibility for the conclusions they make public.
The logo of Anthropic. /VCG
Anthropic began trading on Nasdaq on Thursday, a day before Alpöge's announcement. The timing prompted speculation about whether the release was intended to generate publicity. Alpöge himself described the timing as "terrible."
CHOOSE YOUR LANGUAGE
互联网新闻信息许可证10120180008
Disinformation report hotline: 010-85061466