General Intuition's $2.3B bet: video games train better robots

general intuition secured $320m in funding to advance ai models that transfer learning from video games to physical robots. by using action-labeled gameplay from its sister company medal, the startup creates agents that understand spatial-temporal reasoning without traditional engines. demos showed a quadrupedal robot navigating offices after just eight minutes of real-world data collection. the approach tackles the high cost and scarcity of robotics training data, signaling a shift toward 'pre-training' agents in simulated game worlds before deployment.