Codex just found an 11x matmul speedup

A kernel-level speedup compounds across every higher-level system that depends on it

I spent a lot of today rewiring the jax-js's WebAssembly kernel codegen and got the 4096x4096 fp32 matmul from 30 GFLOPs/s -> 341 GFLOPs/s (11x faster). was iterating a lot with codex, it made a lot of mistakes (at one point it was convinc

Ranked #9 on backlist 2026-06-20 (20 Jun 2026 UTC) · by @ekzhang1 (Eric Zhang) · view on X ↗

How it ranks: Backlist reads my Twitter/X timeline, scores every tweet for substance with an LLM rubric (not engagement), and publishes the daily top picks with a one-line takeaway. Curated by Surya Dantuluri.