Claude Fable 5 solved a 12-year quantum riddle, but a human beat it by a day
We are officially entering the era where neural networks do actual, rigorous math instead of just hallucinating plausible-sounding nonsense. An AI has just generated a mathematically flawless proof, and it didn't even need a human peer review to confirm it.
Researchers from Harvard and MIT decided to test Anthropic's newest brain, Claude Fable 5, on a headache that quantum physicists have been nursing since 2014. They targeted the famous Farhi-Goldstone-Gutmann hypothesis, which deals with how well a quantum algorithm called QAOA can organize a ring of microscopic, rebellious spins that absolutely hate pointing in the same direction. For over a decade, scientists could only run numerical simulations to guess that the formula worked, but nobody could write down a bulletproof proof.
Instead of letting the AI just yap out some text and calling it a day, the team hooked it up to Lean 4, a notoriously pedantic computer program that compiles math and rejects any step that has even a microscopic whiff of logical error. The researchers did the heavy lifting of translating quantum physics definitions into code, left a giant gaping hole where the actual proof should be, and told the AI to fill in the blanks.
The AI solved the mystery by finding a hidden symmetry and borrowing tools from an entirely different field called quantum signal processing. It repeatedly wrote code, got yelled at by the Lean compiler, fixed its errors, and eventually produced a flawless, machine-verified proof.
But the universe has a hilarious sense of timing. Just twenty-four hours before this research was published, a human mathematician named Kunal Marwaha published his own independent proof of the exact same hypothesis. He used the exact same quantum signal processing trick, though he confessed he didn't do it alone—he was heavily back-seat driving with ChatGPT 5.5 Pro and Claude Opus 4.8.
This isn't a story about silicon overcoming biology, but rather a wild glimpse into how science happens now. Instead of lone geniuses scribbling on blackboards, breakthroughs are turning into a race between different human-AI centaurs running the same algorithms at the same time. The bottleneck has officially shifted from finding the answers to simply knowing which questions are actually worth asking.
Source: arXiv
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.