(I’ve been working on …)

Machine Learning

  • Modded NanoGPT - an LLM speedrunning competition where I have set several records 1. I’ve written a corresponding blogpost on muon. My work here has been extensively tested and used in Andrej Karpathy’s Nanochat (>42k github stars).
  • Prime Environments - creating Reinforcement Learning envs to test and train agents, including SWE Bench
  • Spotify MCP Server - get your LLM to control spotify (>500 github stars!)
  • Semantic Finder - vector search in your browser (>300 github stars!)

Graphics

Footnotes

  1. At the time of writing this, 8 records including: incorporating Flash Attention 3, adding a better matrix sign function, speeding up Muon, improving weight decay, adding batch size scheduling, and inventing an efficient form of multitoken prediction. ^