A Tiny Tweet Just Blew Up a Century-Old Math Problem. Blame AI.

Imagine a math problem so stubborn, so universally accepted as probably true, that it's stumped the sharpest minds for over a century. Now imagine it getting unceremoniously debunked by a single, concise social media post. And not just any social media post, but one born from the digital brain of an AI.

That's precisely what happened when Levent Alpöge, a mathematician at AI company Anthropic, dropped a bomb on the Jacobian conjecture. His weapon? Anthropic's large language model, Claude Fable 5, which coughed up a counterexample so elegant, it fit into a tweet: "Hello there the Jacobian conjecture is false thanx."

The Conjecture That Wouldn't Quit

For the uninitiated, a conjecture is basically a mathematical hunch. Everyone thinks it's true, but nobody's managed to prove it or disprove it. The Jacobian conjecture, first floated in 1884, deals with polynomial functions — the kind that take numbers, shuffle them around, and spit out new ones, essentially moving points through space. The idea was that if a function moved points without folding or crushing the space (a property measured by its "Jacobian determinant"), it should be perfectly reversible, like an undo button for geometry.

Many brilliant mathematicians, including a Fields Medalist, tried to crack it. Proofs were offered, then retracted after errors were found. It was a mathematical Everest, perpetually unconquered, yet everyone assumed the view from the top would confirm its truth.

What makes Alpöge's discovery particularly wild is the sheer simplicity of the solution. Mathematicians knew that finding a counterexample should be straightforward if one existed. It was just a matter of sifting through the infinite possibilities to find that one elusive function that had the constant Jacobian determinant and wasn't reversible. Turns out, the AI found one that was so simple, a "smart undergraduate" could've spotted it, as one user noted back in 2017.

Alpöge's AI-generated function is a three-dimensional marvel with a Jacobian determinant of -2. It takes multiple input points and shoves them into the same output point, making it impossible to reverse. Boom. Conjecture debunked for all dimensions greater than two. (The original two-dimensional version is still standing, for now.) The best part? Its brevity meant other mathematicians could verify it almost instantly.

This isn't the first time AI has crashed the math party, with models like OpenAI disproving other long-standing problems. But this one feels different. It wasn't about generating a mind-bendingly complex proof; it was about intelligently searching for a needle in a haystack of polynomial mappings. It suggests AI isn't just a proof-builder, but a powerful tool for unearthing unexpected mathematical objects.

Which, if you think about it, is both impressive and slightly terrifying. What other mathematical "truths" are just waiting for a smart AI to tweet their demise?