For over 150 years, the Riemann hypothesis has been the mathematical equivalent of that one puzzle you keep meaning to solve but never quite get around to - except the stakes are a $1 million bounty and eternal glory, not just bragging rights at a dinner party. The hypothesis, which concerns the distribution of prime numbers, remains unproven, and that million bucks is still sitting there, unclaimed, like a prize nobody can quite reach.

Enter Anthropic, with an unreleased AI model that, according to an announcement on Monday, managed to make significant progress on the problem. Specifically, it increased the lower bound of solutions for which the hypothesis holds true. Not a full proof, mind you - but hey, Rome wasn't built in a day, and neither is a millennium-old math problem.

What's particularly delightful is the method. An Anthropic staff member with no significant mathematical training simply prompted the model to "take a real stab" at proving the hypothesis, then left it to its own devices for a day and a half. The model, in a spectacular display of overachievement, tested 650 different ideas, coordinated across 60 subagents, and burned through 31 million output tokens. That's a lot of tokens - enough to make even the most verbose AI blush.

The paper's footnote reads like a corporate team-building exercise: out of the 60 subagents, two developed the key mathematical ideas, 13 contributed to those ideas, 30 tried and failed to come up with new ones, 13 served as validators, and the last two helped write the initial paper. It's like a math department where everyone has a role, even if that role is just checking the work and trying not to break anything.

Anthropic's in-house mathematicians confirmed the result, and it was formalized using the open-source proof assistant Lean, because why not add a little formal verification to the mix? This isn't the first time AI has flexed its mathematical muscles. Over the past year, AI models have solved several Erdős problems, and OpenAI's internal "Astra" model recently proved 10 major results. Anthropic also earlier disproved the longstanding Jacobian conjecture, which is like saying, "Oh, that thing you thought was true? Yeah, it's not."

All this progress has sparked both excitement and hand-wringing in the mathematical community. In June, a group of prominent mathematicians signed a declaration expressing concern that AI could undermine the field's core values - specifically, the expectation that proofs should be attributable to specific authors who take credit and responsibility. Because apparently, the idea of a proof with no one to blame when it's wrong is just too much to bear.

But not everyone is clutching their pearls. Fields Medal winner Timothy Gowers, in a blog post responding to the declaration, mused that a world where theorems aren't tied to mathematicians might not be any more problematic than the fact that stars aren't named after astronomers. Touché, Dr. Gowers. Touché. So while the Riemann hypothesis remains unsolved, at least we have the comforting thought that if an AI ever cracks it, we won't have to figure out who to congratulate - we can just thank the 60 subagents, their 31 million tokens, and call it a day.