AI did real maths — here's what that does and doesn't mean

What happened: AI produced original results on Erdős problems — open questions that are easy to state and brutal to solve, with no memorised answer to crib. Cracking one means actual reasoning. So the 'just autocomplete' dismissal doesn't survive contact with the evidence.

The strongest results lean on Lean, a proof checker that verifies every step — which is exactly why they're trustworthy, and also a tell about the shape of the achievement. This is AI generates, formal system filters, in a domain where 'correct' is mechanically checkable. Most of the world isn't like that. And DeepMind itself said the system is 'not AGI'. When the vendor caps the hype, believe the vendor.

So file it correctly: a real research milestone, a narrow one, and a preview of how AI gets useful in serious domains — not by being trusted, but by being checked. The headline 'AI solves famous maths' is true. The implied 'AI is now a mathematician' is not. The interesting bit is the machinery in between.