AlphaEvolve sets new record for matrix multiplication exponent: omega to 2.371177
新智元 · wechat · 2026-09-01
DeepMind's Pushmeet Kohli announced that AlphaEvolve, Gemini-powered coding agent, combined with theorists Josh Alman and Vassilevska Williams, improved the matrix multiplication exponent omega from 2.371339 to 2.371177 — a gain of 1.62e-4 comparable to most progress of the past 40 years, and the first substantive AI contribution to this classic open problem. The paper is on arXiv.
The approach had three parts: reformulating combination loss analysis to scale from 25K to 7M parameters on GPUs; an ML-designed optimizer (Adam, Softmax parameterization, Sinkhorn-Knopp, implicit differentiation) contributing 0.97e-4; and AlphaEvolve evolving the optimizer itself (5 GPU-hours) adding 0.63e-4. Humans still supplied the framework and verification paths while AI pushed the search space to its limits — reaching omega=2 will need genuinely new ideas, but squeezing existing methods to their ceiling is now irreversibly AI's home turf.
More from Research
- Opus-assisted open design: a $31k low-field head MRI sketched in hours — DreamFly_13 · 2026-10-18
- semiconductor.diy: open-source guides to make real chips in a garden shed — jwt0625 · 2026-10-18
- Work in progress: geometric structure of GPG projections as Markov chain coarsening — michaelchchoi · 2026-10-18
- Backprop-free pretraining gets 2.3x speedup as Noise method crushes Dust record on 8 GPUs — inductionheads · 2026-10-18
- q3dviewer: open-source Qt+OpenGL tool for painter-like point cloud editing — 4310sy · 2026-10-18
- Smarter Every Day's Destin to defend PhD dissertation on impact fusion — CatAstro_Piyush · 2026-10-18