DeepMind uses AlphaEvolve to improve matrix multiplication bound
rohanpaul_ai · x · 2026-08-21
Google DeepMind used the AI tool AlphaEvolve to improve the best-known theoretical bound for matrix multiplication.
Key Result:
- Improved the matrix multiplication exponent ω from < 2.371339 to < 2.371177.
- The gain is comparable in magnitude to most improvements over the past 40 years.
Methodology:
- Transformed the proof into a non-convex optimization problem, rebuilt using JAX and parallel tensor operations.
- Increased recursion level from 3 to 4, growing optimizable parameters from 25k to 7 million.
- Gradient descent contributed 0.97×10^-4; AlphaEvolve modified the optimization code to raise the total improvement to 1.62×10^-4.
- The result was certified with exact rational arithmetic.
Related event: DeepMind Improves Matrix Multiplication Bound with AlphaEvolve(2 posts)→
More from Research
- Anthropic releases protein binder design dataset with wet lab validation — anshulkundaje · 2026-08-21
- Paper 'SCALE' Accepted to EMNLP 2026 — RatnRajiv · 2026-08-21
- Lingjie Liu Receives ACM SIGGRAPH Significant New Researcher Award — thoma_gu · 2026-08-21
- Pre-training sets low-rank geometry, RL fine-tunes shape — yunta_tsai · 2026-08-21
- Mars is the experiment robot learning can't run on Earth — Dr_Singularity · 2026-08-21
- Building 1800 Genes in One Tube: Implications for AI Screening — NikoMcCarty · 2026-08-21