General Intuition raises $320M to use video game data to train robots

A company has raised $320 million to develop AI models trained on billions of video game clips rather than real-world robot data. The approach uses gameplay videos that capture humans perceiving environments and deciding how to move, along with embedded labels showing exactly what button presses players made and when. The company argues that this video-based training approach is more effective than text descriptions or simulated data for creating AI systems that can perceive, predict, and act in physical environments. This funding reflects growing investment in physical AI, a dominant topic in robotics development.