这家$2.3B机器人初创公司用8分钟训练取代100万小时

general intuition raised $320m at a $2.3b valuation led by vinod khosla to advance foundation models for physical ai. the company trains on millions of hours of video game footage to develop spatial-temporal reasoning, requiring only eight minutes of real-world data to control robots. this approach aims to generalize across different embodiments, potentially rendering the current industry practice of collecting hundreds of thousands of hours of specialized data obsolete. the goal is to provide a base model that makes building robotics applications ten times easier, rather than building specific robots.