Rohan Paul (@rohanpaul_ai)T2·Models
88 pts
A new Columbia and MIT paper reports that Claude autonomously discovered a faster method for computing selected entries of a product of two long, thin matrices, which classic reductions turn into speedups for 3SUM and all-pairs shortest paths. 3SUM now runs in O(n^1.9992) and APSP in O(n^2.9995, tiny gains that nonetheless refute the long-standing 3SUM and APSP hypotheses. The method was found in a single AI session using 16M output tokens with no human input and was formally checked in Lean 4.