DeepMind’s New AI Found A Strange New Way To Think
DeepMind's AlphaProof Nexus solved 9 previously-unsolved Erdős problems using a novel ELO-ranked judge AI.
“Do not look at where we are. Look at where we will be two more papers down the line.”
DeepMind's AlphaProof Nexus attempted roughly 350 of Paul Erdős's open problems and solved 9 of them for a few hundred dollars each, cracking decades-old questions no human had resolved. Its key innovation is a cheap judge AI that ranks two candidate Lean-formalized proofs by an ELO-style score and picks the better one even when both are flawed. It matters because it extends AI's rapid climb from basic arithmetic to genuine open-research mathematics.