General Intuition: game clips (1.5T action tokens/month) train one model to control 10+ robots
AI Engineer · youtube · 2026-11-05
AI Engineer interview with Pim de Witte (General Intuition, the team behind Medal). Acting models differ from turn-by-turn LLMs: they take frames and emit actions continuously. Medal's 1B game videos per year and 1.5T+ action tokens per month become training data, and since game controllers already drive most robots, one model can both play games at a high level and control 10+ robot types. Covers world models, contrast with JEPA, accessibility uses, early access and hackathons.
More from Embodied
- Pretraining Makes Robot Policies Reliable but Hurts Posttraining, New Paper Finds — siddkaramcheti · 2026-11-05
- XR Blocks team runs talk, demos and 'World as a Canvas' workshop at UIST 2026 — DuRuofei · 2026-11-05
- YC-backed Ishiki Labs launches Fern, a robotics sim platform that builds high-fidelity sims from 3 teleop episodes — ycombinator · 2026-11-05
- First method to estimate empowerment from offline data for intrinsically motivated exploration — Berkeley/Princeton — ben_eysenbach · 2026-11-05
- GCRL and unsupervised skill learning unified via control-maximization — NeurIPS Oral from Eysenbach and Dayan — ben_eysenbach · 2026-11-05
- Aformic raises $6.95M seed round with $34M in robot purchase orders — MatthewChang · 2026-11-05