DeepMind’s New AI Found A Strange New Way To Think
Two Minute Papers
Jun 5, 2026
Dr. Károly Zsolnai Fehér explains how DeepMind’s AlphaProof Nexus solves complex, long-unsolved mathematical problems by wrapping unreliable AI models in a competitive 'tournament' loop. By using a formal verification language and a judge that ranks iterative attempts, the system generates reliable proofs from unreliable parts, proving that system architecture is now as critical as model intelligence.
Key insight: AlphaProof Nexus solved nine decades-old mathematical problems that humans couldn't crack for 56 years, at a cost of only a few hundred dollars per problem.
