Anthropic's Levent Alpöge claims Riemann hypothesis proof found with Claude

Anthropic's Levent Alpöge claims Riemann hypothesis proof found with Claude

Levent Alpöge, a mathematician who works at Anthropic, has published an essay announcing that he found a Lean-compilable solution to the Riemann hypothesis, working with Claude. TechCrunch's wording is hedged: he "appears to have solved" one of the most important problems in mathematics. The hypothesis dates back to 1859 and is among the Millennium Prize problems. It conjectures that there is a hidden structure underlying the appearance of prime numbers.

The timing is awkward, and Alpöge says so. This week his employer, Anthropic, became the first publicly traded frontier AI lab, per TechCrunch. "The precise timing is horrible," he wrote. In the essay he describes messaging Anthropic CEO Dario Amodei last Sunday night to ask how to announce the result.

Mathematicians have been working through the result in a chat on Zulip. Chris Birkbeck, a number theorist at the University of East Anglia, posted that the cohomology bound is proved by what looks like an automorphy-lifting argument, one that genuinely uses Galois representations attached to regular algebraic, polarized cuspidal automorphic representations of GL_n over CM fields, along with base change, Jacquet-Shalika, Clozel's purity, Bernstein-Zelevinsky, the main theorem of CM, Deligne-Ribet and more. Twenty readers reacted to that comment with a 🤯 emoji.

Lean is an open-source coding language used to express mathematical ideas as code, which allows proofs to be formally verified. Several participants in the Zulip chat, including MIT mathematician Andrew Sutherland, say they have confirmed that Alpöge's solution compiles. TechCrunch added an editor's note that the Lean proof has been checked by independent sources. It will be a while before the mathematics community has a handle on the proof itself, but most appear confident that the formalized Lean result will hold.

In the essay Alpöge appears to feel a bit guilty about releasing a proof without a truly human-intelligible explanation. But he has been obsessed with the problem since childhood, and writes that he "could not live" if he waited. He calls the essay a "stream-of-consciousness jaunt through the experience of finding all the pieces of the argument through incredible confusion with Claude", and says the model "flew, and danced, and sparkled near that finish line." TechCrunch notes that for the result to mean anything, humans still have to understand it and explain it to each other, which Alpöge has committed to pursue.

Alpöge and NYU mathematician Tristan Buckmaster also released another important proof, related to the Navier-Stokes equations. TechCrunch says problems like these helped kick off a race this summer between frontier labs and mathematicians to solve outstanding problems.

A second claim arrived at the same time. Ravi Vakil, a Stanford professor and the president of the American Mathematics Association, posted a proof of the General Riemann hypothesis written by Stanford postdoc Alessandro Masullo. Vakil noted that "it is all human, with no AI. The hope was to bring Alessandro to Stanford for a year to polish the writing and then announce it after running it by enough people. Now it seems best to put out the current draft." Masullo says the proof is the result of 12 years of work. It is not expressed in Lean, so TechCrunch says it is too early to tell whether it checks out, much less whether anyone understands it. In his introduction Masullo wrote that the work's message to discouraged mathematicians is: "we are still here. We are very much alive. We are not going anywhere."

Key facts

  • Levent Alpöge, a mathematician at Anthropic, published an essay announcing a Lean-compilable solution to the Riemann hypothesis, found with Claude.
  • Several Zulip chat participants, including MIT's Andrew Sutherland, say the solution compiles in Lean; TechCrunch's update says independent sources have checked the Lean proof.
  • Mathematicians say it will be a while before the community understands the proof itself, and Alpöge has committed to pursue a human-understandable explanation.
  • Separately, Ravi Vakil posted a proof of the General Riemann hypothesis by Stanford postdoc Alessandro Masullo, described as all human with no AI and the result of 12 years of work.
  • Masullo's proof is not expressed in Lean, so TechCrunch says it is too early to say whether it checks out.

Why it matters

The Riemann hypothesis dates back to 1859 and is among the Millennium Prize problems. It conjectures a hidden structure underlying the appearance of prime numbers. If the claimed result holds, TechCrunch says we are entering sublime-mysteries-of-the-universe territory. The story also pairs an AI-assisted route with an all-human one. TechCrunch connects it to a race that began this summer between frontier labs and mathematicians to solve outstanding problems, and notes that Alpöge and Buckmaster also released another important proof, related to the Navier-Stokes equations.

Who it affects

Number theorists and other mathematicians are the first audience: they are the ones working through the argument on Zulip. It also touches AI labs, since Alpöge works at Anthropic and credits Claude with finding the pieces of the argument. TechCrunch notes Anthropic became the first publicly traded frontier AI lab this week. Mathematicians who feel discouraged are addressed directly by Masullo, who wrote that his work is meant to tell them: we are still here.

How to use it

There is nothing to install or run here. This is a claimed result, and the practical way to follow it is to watch how the mathematics community responds. Lean, the open-source coding language involved, expresses mathematical ideas as code so that proofs can be formally verified, which is how participants in the Zulip chat checked that Alpöge's solution compiles. Alpöge has committed to pursuing a human-intelligible explanation, which is what would let others learn from the argument.

How solid is it

Mixed, and the two claims differ. For Alpöge's result, several Zulip participants, including MIT mathematician Andrew Sutherland, say they confirmed it compiles in Lean, and TechCrunch's editor's note says independent sources have checked the Lean proof. Most in the community appear confident the formalized result will hold, but it will be a while before anyone has a handle on the proof itself. TechCrunch's own wording is that he "appears to have solved" the problem. For Masullo's all-human proof, TechCrunch says that without being expressed in Lean it is too early to say whether it checks out. No official verification, journal submission, or Clay Institute Millennium Prize response is mentioned in the article.

Risks and caveats

A proof that compiles is not the same as a proof people understand. Alpöge appeared in his essay to feel a bit guilty about releasing it without a truly human-intelligible explanation, and TechCrunch says that for the action to mean anything, humans must be able to understand the result and explain it to each other. The article does not say how much of the proof Claude produced versus Alpöge. It also does not say whether the Lean proof covers the original Riemann hypothesis or a general version; it says only that Alpöge's solution compiles. Masullo's proof has not been checked. TechCrunch's account is also partly a description of Alpöge's essay and of Zulip chatter, not a review of the mathematics.

“we are still here. We are very much alive. We are not going anywhere.”

— Alessandro Masullo, in the introduction to his proof, as quoted by TechCrunch