reipianoA simulated Unitree G1 humanoid learns the opening of "Runaway" with reinforcement learning: 43 of 44 notes on time.github ↗training dashboard ↗policies ↗GPT-2 from scratch124M parameters, pretrained on 9.99B FineWeb-Edu tokens.github ↗model card ↗hugging face ↗PyBoxLocal Python execution, made inspectable: what happens between a command and its result.github ↗site ↗reipaintTeaching Qwen3-8B to paint watercolours with code: 20 RL updates took valid paintings from 10 to 23 of 24.demo ↗models ↗dataset ↗From scratchTwelve ML papers implemented from scratch, from an autograd engine to a diffusion pipeline: the equation, the code, the result.github ↗site ↗SpeakMaxPractice high-stakes conversations with an AI or live with other people, then review a debrief tied to your own words.site ↗OutsNo-limit hold'em from first principles: Socratic lessons, animated explainers, and timed practice drills.site ↗ReisearchSupervisor-worker agents that turn a vague question into a sourced report.github ↗demo ↗Tri-modal agentRoutes between chat, retrieval, and sandboxed code, and asks before it acts.github ↗