Within hours of an Anthropic mathematician announcing an AI-assisted solution, a second Riemann hypothesis proof landed online, and this one has no AI in it. Alessandro Masullo, a postdoctoral researcher at Stanford, posted his paper on Friday and said he did the work himself over 12 years.
The result is a human-versus-machine contest over the most famous open problem in mathematics. Neither paper had been formally verified by Friday afternoon.
A paper released earlier than planned
Masullo told readers he was about a year from releasing his results when Levent Alpöge’s announcement arrived. Alpöge, who works at Anthropic, said he reached his solution by working with Claude for weeks.
Stanford professor Ravi Vakil, president of the American Mathematical Society, explained the plan in a public post. The hope was to bring Masullo to Stanford for a year to polish the writing, then announce the result after showing it to enough people. With the other claim public, it seemed best to put the current draft out.
Masullo said the same in his paper: he was advised to release the version he had, and it may contain errors because it was rushed. Vakil stressed that the work is “all human, with no AI.”
“We are still here”
The introduction goes beyond mathematics. Masullo addressed mathematicians who feel discouraged by the pace and tone of the current climate. “We are still here,” he wrote. “We are very much alive. We are not going anywhere.”
The message follows months of AI companies announcing high-profile solutions to long-standing problems. OpenAI has said it solved the Navier-Stokes equations, which belong to the same Millennium Prize category as the Riemann hypothesis. Each of those problems carries a $1 million prize.
What the hypothesis predicts
The Riemann hypothesis describes how often prime numbers appear among the other numbers. Cornell mathematician Steven Strogatz called it the “White Whale” of mathematics. His reaction to the double announcement on X was brief: “The plot thickens!”
Sergei Gukov, a math professor at Caltech, called a solution a “HUGE deal” and said the problem had been the biggest in the field. He spoke of the problem in the past tense. Asked about the two claims, he added that things were now more complicated.
How the two papers differ
Setting the two Riemann hypothesis proof attempts side by side shows how far apart they are.
- Masullo’s proof is far shorter than Alpöge’s.
- Masullo’s text is written for human readers, and it carries no formal verification yet.
- Alpöge’s solution comes as a Lean file, which software can check, but it comes without an explanation people can follow.
Lean is an open-source language that expresses mathematics as code. Participants in a public Lean discussion have since confirmed that Alpöge’s file compiles, which makes his claim easier to test. A human-written paper has no such shortcut. Experts have to read it line by line, and without a Lean version it is too early to say whether it holds.
Why the credibility question matters
Mathematicians are now racing to examine both solutions, using their own judgment and AI tools. One camp expects the formal file to win on correctness. The other values a proof people can read and learn from.
The stakes go past one prize. If the AI-assisted route is right, mathematicians will have to decide what credit and trust mean when a model does much of the discovery. If the short human proof is right, it would show that decades of focused individual work still competes with the newest tools.
Masullo, his Stanford partner Brian Conrad, and Alpöge did not respond to requests for comment on Friday.
What happens next
Expect weeks of checking. Specialists will test Masullo’s draft for gaps, and some will try to express it in Lean so a compiler can judge it too. Both papers may also be right, which would give the field two very different routes to the same Riemann hypothesis proof.
Until the reviews come in, the outcome is open, and the community is watching a contest between a rushed draft and a machine-checked file.
Source: Business Insider