Meta ships KernelEvolve, following Google's AlphaEvolve AI kernel-optimization play
RichardsonDx · x · 2026-11-12
Matrix multiplication underpins every AI model, render, and simulation. The naive way to multiply two 4x4 matrices takes 64 multiplications; Volker Strassen cut it to 49 in 1969, and nobody beat that for 56 years — until Google's AlphaEvolve agent found a 48-multiplication procedure. Google used the technique to cut Gemini training time and design next-gen TPUs. Meta has now followed with KernelEvolve, its own system for optimizing AI accelerator kernels, signaling that AI-driven kernel discovery is becoming a competitive frontier among major labs.
More from Infra
- Dell expands AI Data Platform: agents have a data-context problem, not a model problem — DavidLinthicum · 2026-11-12
- Potion UI's local login system turns your PC into a multi-user AI server — cocktailpeanut · 2026-11-12
- Recursive's Eureka Machine ships Flint: a sparse 40B model with 2.5B active for local agentic coding — agihouse_org · 2026-11-12
- AWS Finally Built Its Own AWS Emulator — super_linear · 2026-11-12
- Megakernel is multiprogramming for GPUs: 'GPUs are entering the CPU age' — pranjalssh · 2026-11-12
- Gemini Decision 1 costs $345 per 1M tasks, 3x pricier than leader Drex Max v1.1 — ArtificialAnlys · 2026-11-12