A quantum computing conjecture sat open for 18 years. A research system running an unreleased @AnthropicAI model found a proof by day three, according to Physical Superintelligence. The lab published the paper and the Lean code, so anyone can rerun the checks on the core mathematical result. The Aaronson–Ambainis conjecture gets at a basic question: when can a quantum computer solve a problem exponentially faster than a classical one? The proof says that, in the setting the conjecture studies, you need special structure to get that kind of speedup. PSI hasn’t said which model it used, and I haven’t seen an independent rerun yet. But having a proof other people can check makes this a lot more interesting than another benchmark score.