UK Startup Turns Gaming Inputs Into Training Data for World Models, With LeCun Advising
nordicinst · x · 2026-09-28
WIRED reports that Worldmodeldata, a UK startup advised by Yann LeCun, is converting players' thumbstick and button sequences in 3D games into training data for world models that can navigate the physical world.
- A growing camp including Fei-Fei Li and Yann LeCun argues LLMs trained only on text are ill-suited for physical tasks like autonomous driving or robotic arms, spurring a shift to world models.
- World models need paired visual and action data (video plus grip force, torque, etc.), which barely exists on the internet.
- Unskilled players' gaming moves offer a trove of cause-and-effect data at low cost.
More from Embodied
- University of Tokyo grows self-healing living human skin on robotic finger — TinfoilTricorn · 2026-09-28
- Delta Intelli launches Δ₀, a humanoid foundation model with 69-DoF whole-body control — jeasinema · 2026-09-28
- Stanford's HomeBody gives frontier VLMs spatial memory and composable humanoid skills — CyberRobooo · 2026-09-28
- Meta's Muse AI assistant watches everything, critics call it a parasite on human life — CarissaVeliz · 2026-09-28
- British Startup Turns Gamer Inputs Into Training Data for Physical-World AI — Wired AI · 2026-09-28
- AUAR ships robotic microfactories in containers to build homes up to six stories — lukas_m_ziegler · 2026-09-28