General Intuition thinks large language models hit a wall when it comes to understanding physical movement through space and time. Their solution? Train AI on gaming data instead. The New York startup just snagged $320 million from Coatue, Eric Schmidt, and researchers at MIT and Google DeepMind, pushing its valuation to $2.3 billion. CEO Pim de Witte explains how world models built from game physics could be the key to truly generalizable AI. They even got a robot navigating an office with just eight minutes of real-world data. The company recently turned down an acquisition offer from OpenAI to stay independent and is launching Nerve, a marketplace where gamers can label data or do teleoperation work to preempt job displacement. Why it matters: If gaming data can crack physical reasoning, it might bridge the gap between chatty LLMs and AI that actually understands how the world works.