Clay Institute says review of Riemann claim will take time

Levent Alpöge, a member of Anthropic’s technical staff, made the announcement Friday in a four-page explanation posted on his own website. He wrote that thanks to his work with Claude he had found a proof of the Riemann Hypothesis, the conjecture that has been open since 1859.

Writing about the collaboration, Alpöge said he and the system had covered enormous ground together in recent months and that this time the machine went further than he did as the work approached its conclusion. “Claude y yo lo hemos pasado muy bien juntos con un abanico tan amplio en los últimos meses, pero esta vez Claude me superó con creces cerca de la línea de meta,” he wrote. “No, Claude voló, y bailó, y brilló cerca de esa línea de meta.”

Formulated in 1859, the Riemann Hypothesis concerns the distribution of prime numbers — those divisible only by one and by themselves — as they grow toward infinity.

According to Alpöge, he made the finding on Sunday night. He messaged Anthropic chief executive Dario Amodei to ask about “mi capacidad para anunciarlo,” his ability to announce it, because the San Francisco-based company was about to collect headlines with its stock market debut. One day after the company’s initial public offering — described in the report as the highest-valued in history — Alpöge disclosed that he had solved a problem that had proved elusive for more than 150 years.

Anthropic did not respond to an email from The Associated Press seeking comment on its employee’s claim.

Alpöge published his proof of the Riemann Hypothesis — and Claude’s — in 17 million lines of Lean code, a programming language that allows proofs of mathematical theorems to be written as code. Traditionally, findings buried in such code would be published in a formal paper, said Alex Kontorovich, a distinguished professor and chair of the mathematics department at Rutgers University.

“Por lo general, cuando alguien te entrega un artículo, es también una especie de código para una demostración y tienes que leer el artículo, seguirlo por tu cuenta, determinar si crees cada una de las líneas de ese artículo,” Kontorovich said. “Y cuando terminas, deberías tener una comprensión profunda de por qué algo es cierto.”

He added that Claude followed that code and that the demonstration is true. “¿Significa eso que yo lo entiendo? Todavía no. Pero la computadora es extraordinariamente selectiva, mucho más que yo, y así que si puedes satisfacer a la computadora, hay una probabilidad muy, muy grande de que también me satisfagas a mí,” he said. “Pero tengo que hacer el trabajo.”

Hours after the announcement, the Riemann Hypothesis still appeared as unsolved on the website of the Clay Mathematics Institute. The institute selected the hypothesis in 2000 as one of its seven Millennium Prize Problems and offered a $1 million reward for a solution.

The evaluation process for determining whether the problem has actually been resolved takes time, said Martin Bridson, the institute’s president. “Este es un anuncio de importancia histórica, que encierra la promesa de una transformación dramática en la percepción de lo que es posible en la lucha por ampliar la comprensión humana de las matemáticas,” he said in a statement. “Pero el camino desde el anuncio de hoy hasta un lugar de comprensión real será largo y difícil.”

Alpöge said he is presenting the world with a complete argument before having mastered a vast landscape surrounding its mathematical foundations. He said more than a quarter of a million pages emerged from his work with artificial intelligence.

In September, OpenAI said its technology was key to a similar advance on Navier-Stokes, the equation that governs the flow of fluids. “Cayó Navier-Stokes y ahora la Hipótesis de Riemann,” Kontorovich said. He described a kind of out-of-body experience in which he pictures himself saying “Sí, la Hipótesis de Riemann ha sido resuelta,” and trying to work out what the words leaving his mouth mean. “Hay una especie de experiencia extracorporal en la que me veo a mí mismo diciendo ‘Sí, la Hipótesis de Riemann ha sido resuelta’, y tratando de comprender qué demonios son estas palabras que están saliendo de mi boca,” he said.

It is the second time in recent months that a company in the sector, or someone linked to one, has claimed to have solved a famous unsolved mathematical problem.

Anthropic’s Wall Street debut this week raised a record 81 billion dollars at a valuation of one and a half trillion dollars. Amodei has projected an optimistic vision of AI’s future in which the technology drives significant advances in mathematical and scientific research, while warning about the potential risks it poses.

The claim extends a sequence MSI has followed in AI mathematics: in September, MSI reported that OpenAI had published what it described as a solution to the Navier-Stokes Millennium Prize Problem, the equation governing fluid flow that Kontorovich on Friday paired with the Riemann claim. What has changed is the company at the center of the announcement — a rival’s staff mathematician now says Claude carried the work — while what remains unresolved is the distance Bridson described between an announcement and real understanding.