On this week's AI board

MTPLX: faster local LLMs on a Mac, same output

Mac app and CLI that uses the built-in multi-token prediction heads of models like Qwen to speed up local decoding.