GPT5.5 pro and DeepMind's secret Aletheia agent independently cracked a Shafarevich conjecture proof

lmthang · x · 2026-09-28

Mathematicians Carlo and Mark Shusterman posted a resolution of the probabilistic version of Shafarevich's conjecture (posed by Liu–Wood and Sawin–Wood), revealing the main proof idea was discovered independently by GPT5.5 pro and Aletheia, a stealth internal agent at Google DeepMind post-trained in early July by Thang Luong's team. Follow-up applications include settling a Nikolov conjecture on universal perfect groups via an experiment where the agent pattern-matched problems to methods in reverse.

Original post →

More from AGI Musings

AGI Musings channel →