Is It Still a Proof if Nobody Can Read It?
A 167-year-old question in math may have been answered by a man and a machine. The proof is 17 million lines long, and we’re trusting other machines to tell us it’s right.
In 1859, a German mathematician named Bernhard Riemann wrote a short paper with a guess about where certain numbers fall. Those numbers explain how prime numbers are spread out. For 167 years the guess held up, and nobody could prove it. It’s one of the seven problems the Clay Mathematics Institute will pay you $1 million to solve.
On Thursday, a mathematician at Anthropic named Levent Alpöge posted a four-page note on his website. In it he wrote: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.”
There was no paper to go with it, just 17 million lines of code written in Lean, a language computers use to check math step by step. Written out by hand, the proof would run past 250,000 pages. That’s longer than the entire Encyclopedia Britannica, several times over.
A day later, the people who run Lean said they have “full confidence” it’s correct. Several independently built checkers, run by different people, accepted it.
So here’s where we are. For all of history, a proof was something a person could follow. You read it, you nodded, you checked the steps. This one is a proof no person will ever read. We’re taking the word of machines that checked the work of a machine.
Maybe that’s fine. We already fly in planes no single engineer fully understands, and take drugs whose chemistry most doctors couldn’t explain. But math was supposed to be the one place where you didn’t have to take anyone’s word for anything.
The prize committee will want a human-readable paper. Somebody may have to turn 250,000 pages into something people can actually read. That job might take longer than the proof did.