An Anthropic researcher let Claude run unsupervised on physics problems for six weeks. It finished a black hole calculation left incomplete since 2003 and found a hard new ceiling on black hole mass in a model of quantum gravity.
An Anthropic researcher let Claude run unsupervised on physics problems for six weeks. It finished a black hole calculation left incomplete since 2003 and found a hard new ceiling on black hole mass in a model of quantum gravity.
0 vue
Publié le 20/11/2026 17:57
0:29
An Anthropic researcher let Claude run unsupervised on physics problems for six weeks. It finished a black hole calculation left incomplete since 2003 and found a hard new ceiling on black hole mass in a model of quantum gravity.
1 commentaire
?
The unfinished problem: the Hawking-Page transition, where a particle gas collapses into a black hole as spacetime heats up. Claude showed the transition stays sharp even with quantum effects, matching general relativity's prediction.
The second problem used modular bootstrap, a method for constraining 3D quantum gravity theories. Claude found the lightest possible black hole mass can't exceed c divided by 9.1266, and proved that's the lowest value this method can produce.