Overview
- AlphaEvolve surpassed Strassen’s 1969 matrix multiplication record, reducing scalar multiplications from 49 to 48 for 4×4 matrices, a major mathematical breakthrough.
- The system has been deployed across Google’s infrastructure, recovering 0.7% of global compute resources and improving AI training efficiency by 1%.
- AlphaEvolve optimized Google’s TPU chip design by eliminating unnecessary components, with changes validated and integrated into future designs.
- In tests on over 50 open mathematical problems, AlphaEvolve rediscovered the best-known solutions 75% of the time and improved them in 20% of cases.
- DeepMind plans to launch a user interface and an early access program for academic researchers, signaling broader potential applications beyond Google.