Auto-research with codex: How I achieved a 232x Faster Kernel
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
What commenters are saying
Commenters describe using LLMs for autonomous code optimization in a benchmark-profile-verify-research-improve loop. Examples include SSE/AVX for a video codec, FlashAttention in CUDA, C# protobuf streaming, game ROM decompilation, and NEON kernels for Raspberry Pi video transcoding. The common recipe: a correctness verifier, profiler access, and a clear goal. Treating LLMs as constraint solvers with self-verification enables hours of self-directed optimization.