Meet AlphaEvolve, the Google AI that writes its own code—and just saved millions in computing costs

Meet AlphaEvolve, the Google AI that writes its own code—and just saved millions in computing costs | VentureBeat

Google DeepMind's AlphaEvolve AI system breaks a 56-year-old mathematical record by discovering a more efficient matrix multiplication algorithm that had eluded human mathematicians since Strassen's 1969 breakthrough.

venturebeat.com

TL;DR: Google DeepMind’s new AI agent, AlphaEvolve, teams up with Gemini LLMs and an evolutionary test-and-refine loop to invent entire computer algorithms—and it’s already live inside Google’s infrastructure. It’s boosted Borg cluster efficiency by 0.7%, trimmed bits off TPU circuits, and even sped up Gemini model training kernels by 23% (cutting overall training time 1%), all while churning out human-readable code.

But AlphaEvolve isn’t just optimizing data centers—it’s smashing math records. Using its gradient-based evolution, it beat Strassen’s 56-year-old 4×4 matrix-multiply record, improved 14 algorithms, and tackled 50+ open problems (nailing or improving 75% of them), including setting a new 11-dimensional “kissing number” high score. With its plug-and-play approach, DeepMind hopes to spin out breakthroughs in material science, drug discovery, and beyond.