Anthropic says unreleased AI model made progress on Riemann hypothesis

In short: Anthropic says an unreleased AI model made measurable progress on the Riemann hypothesis, a famous math problem that has gone unsolved for over 150 years.

What happened

Anthropic, the company behind the Claude chatbot, published research saying one of its unreleased models improved results related to the Riemann hypothesis. This hypothesis is a long-standing question tied to prime numbers, which are the building blocks of math in the same way atoms are building blocks of matter.

Anthropic did not claim it solved the problem. Instead, it said the model “significantly increased the lower bound” for cases where the hypothesis holds true. In plain terms, it pushed the verified range further, like checking more miles of a long bridge and confirming the structure still holds.

Anthropic also described how the work was done. A staff member without deep math training asked the model to try to prove the hypothesis and then let it run for about a day and a half. The system tried 650 ideas, split work across 60 sub-agents (smaller helper AIs working like a team), and used 31 million in total spending, according to the report.

Two in-house mathematicians at Anthropic confirmed the result. The work was also formalized with Lean, a “proof assistant,” which is software that checks each step of a math proof the way a strict editor checks every line.

Why it matters

The Riemann hypothesis has a $1 million prize for a complete proof, and it is still unclaimed. Even partial progress can be important because it suggests today’s AI tools may help researchers explore hard problems faster. At the same time, mathematicians are debating what it means for credit and responsibility when AI helps create proofs.

Source: TechCrunch AI