Levent Alpöge, a mathematician who works at Anthropic, has posted a statement on his own website that makes a large claim: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." He says the same method also proves a wider version, the generalized Riemann hypothesis, and that with more time he expects it to prove a third, the extended Riemann hypothesis. The Riemann hypothesis is one of the Clay Mathematics Institute's Millennium Prize Problems, which carry a million-dollar prize.
The question goes back to a paper Riemann wrote in 1859. It concerns the zeta function, a formula whose zeros, the points where it equals zero, together pin down how the prime numbers are spread out. Plotted as points, the zeros that matter all fall on one vertical line, the hypothesis says. Many results in mathematics assume it is true, Anthropic noted in August. A correct proof would make those results settled facts.
Alpöge has not published a write-up that people can read. He has posted the proof only as computer code, split into 14 download files and written in Lean, a programming language in which a computer can check every step of a proof. He calls the argument "incredibly long at the moment" and says he chose not to post the huge, unreadable document a full write-up would produce now. He says he will work to understand the argument better and write it up properly.
He describes Claude, Anthropic's AI model, as doing much of the final work. He says he steered the two of them toward the right approach, and that "this time Claude far outpaced me near the finish line." He also writes that he has not yet mastered much of the mathematics around the argument, that the parts Claude worked out are less familiar to him, and that his own summary of those parts "might make zero sense." He expects shorter proofs to follow fairly quickly.
Alpöge says the computer version is finished and rests on no unproven assumptions, but no one outside has yet reported whether Lean accepts the code. Even if it does, experts still have to check that the statement typed into the computer really is Riemann's hypothesis, since a computer check covers only the problem as it was typed in. "Let's see if it compiles first," Kevin Buzzard, a mathematician at Imperial College London, said, according to a news report, adding that it "might be a while" before the right experts are found to explain the argument. The Clay institute considers a solution only after it appears in a qualifying publication, two years pass, and mathematicians generally accept it. So even if the proof holds, any prize decision is at least two years away.
In August, Anthropic said an unreleased version of Claude had raised the share of the zeros that matter proven to lie on the line from 41.6% to 67.2%. Alpöge was one of two Anthropic mathematicians who examined that result. At the time, the company said it did not expect those techniques to lead to a proof of the hypothesis.
Alpöge also writes that "the precise timing is horrible." He says the work came together late on a Sunday night, and the timeline in his statement dates the final step to Nov. 16. He says he asked Anthropic's leadership about announcing it, and that "there was a constraint to respect." A news report notes that his statement came out the day after Anthropic's shares began trading on Nov. 19. He does not say what the constraint was, and he calls a cynical reading of the timing false.