Research

AI-generated text

Claimed AI-Assisted Riemann Proof Raises Questions About Mathematics

A source article claims that Anthropic employee Levent Alpöge used Claude to produce a formal proof of the Riemann hypothesis, but provides no human-readable demonstration.

Claimed AI-Assisted Riemann Proof Raises Questions About Mathematics

Anthropic employee Levent Alpöge published a PDF on November 20 claiming that he and the Claude artificial intelligence system had proved the Riemann hypothesis, according to the source article. It says he also released 17 million lines of Lean code that passed multiple independent checks, but did not publish a demonstration humans could follow. His attempt to produce one reportedly resulted in a text of more than 250,000 pages.

The source does not provide the details needed to verify those claims, and its account does not establish that the proof is valid or has been accepted by the mathematical community. The Riemann hypothesis concerns zeros of the zeta function and has long been among mathematics’ best-known unsolved problems.

Verification and understanding

The article builds its argument around the distinction between formal verification and human understanding. It reports that Maksym Radziwiłł, a mathematics professor at New York University, told Quanta Magazine that science fundamentally asks why something is true, a question the material did not answer. Andrew Granville, a professor at the Université de Montréal, told Nature that an intelligible mathematical exposition was needed. Brian Conrad, a professor at Stanford University, criticized the reported decision to announce the proof without fully understanding it, according to The New York Times.

The source also describes more measured or enthusiastic reactions. It says Fields Medalist Terence Tao expected the proof to take time to digest, while fellow Fields Medalist James Maynard told Nature that, once properly understood, it could rank among mathematics’ greatest achievements. Martin Bridson, president of the Clay Mathematics Institute, reportedly called the announcement historically significant. Steven Strogatz, a Cornell University mathematician, compared the human–AI partnership to a skilled rider on a fast horse.

The source’s accounts of these statements also require verification. It attributes to Alpöge the view that, although the development might feel like the end of an era, it could also mark a new beginning for science.

What a proof could change

According to the article, mathematicians have for decades assumed the Riemann hypothesis to be true in many contexts. An accepted proof would therefore allow some results that depend on it to dispense with that condition. The author argues, however, that the effect on the profession could be greater than the immediate practical benefit.

The source also claims that Alpöge, Tristan Buckmaster and Claude solved a harder form of the Navier–Stokes problem during the same week; it does not provide enough information to assess that result independently either. The author outlines several possible futures: AI might open new areas of mathematical research, or machines might increasingly take over tasks now performed by mathematicians.

In one possible role, human researchers would remain interpreters of machine-produced results. The article also raises the possibility that this role could become difficult if proofs grow too complex for people to follow.

Solving and posing problems

The author distinguishes solving known problems from formulating new questions. The article notes that conceiving the Riemann hypothesis was itself a major intellectual achievement and argues that current AI systems’ reasoning abilities do not automatically mean they are equally capable of identifying new areas of mathematics.

Quoting Alpöge, the source says he set the direction for combining two ideas and identified the appropriate mathematical setting, after which Claude rapidly outpaced him near the finish. The article suggests that the purported proof may have used extensive existing mathematics, but acknowledges that the unknown argument does not establish whether fundamentally new methods were needed. In an earlier remark quoted by the source, Tao suggested that the prospect of creating such new mathematics is one reason the Riemann hypothesis is considered so important.

Access and power

The article questions not only capability but also access. It claims that Alpöge used a Claude model unavailable to the public and contacted Anthropic leader Dario Amodei before making the announcement. The author draws from this a broader concern that access to frontier models and computing resources could concentrate opportunities for scientific discovery in the hands of a few companies. Those conclusions are opinions, not established facts.

Finally, the article compares a possible future for mathematics with computer chess: after a period of human–machine collaboration, machines might reach a point where people can no longer contribute meaningfully to their work. That scenario is speculative. The source’s central question remains what a machine-verifiable result means as human knowledge if researchers themselves cannot follow its reasoning.