Can your AI rewrite your code in assembly?
Summary
Daniel Lemire explores whether AI-generated assembly can outperform a classic C++ counting implementation by benchmarking several ARM64/NEON variants created by Grok and Claude. The post presents a range of AI-generated implementations, shows notable reductions in instructions per string and speed, and discusses the limits of AI-optimized code, including production-readiness and the availability of a public GitHub source.