General Intuition just raised $134 million in seed funding to build AI agents that can navigate the physical world using spatial-temporal reasoning trained on billions of gaming videos. The Medal spinout is betting that its 2 billion annual video game clips offer a superior dataset for training agents compared to platforms like YouTube or Twitch, attracting massive investor interest after OpenAI reportedly tried to acquire the parent company for $500 million last year.
General Intuition just landed one of the largest seed rounds in AI history, raising $134 million to teach artificial agents how to move through space and time using an unlikely teacher: video game footage. The startup, spun out from gaming clip platform Medal, is betting that billions of gaming videos hold the key to building AI that can navigate the physical world with human-like spatial reasoning. The massive funding round, led by Khosla Ventures and General Catalyst with participation from Raine, signals investor confidence in what CEO Pim de Witte calls a fundamentally different approach to artificial general intelligence. While tech giants pour resources into ever-larger language models, General Intuition argues that true AGI requires something text-based systems fundamentally lack: spatial-temporal reasoning. The company's secret weapon is Medal's unprecedented dataset of 2 billion gaming videos per year from 10 million monthly active users across tens of thousands of games. This treasure trove reportedly caught OpenAI's attention last year, when the AI giant attempted to acquire Medal for $500 million, according to The Information. Neither company commented on the reported acquisition attempt, but the interest makes perfect sense given OpenAI's push into AI agents and robotics. "When you play video games, you essentially transfer your perception, usually through a first-person view of the camera, to different environments," de Witte told TechCrunch. He noted that gamers who upload clips tend to post extreme moments - spectacular wins or crushing defeats - creating "selection bias towards precisely the kind of data you actually want to use for training work." This edge case advantage sets Medal's dataset apart from alternatives like Twitch streams or YouTube gameplay videos, which often focus on entertainment rather than learning edge cases. The company has already demonstrated impressive early results: their models can understand environments they weren't trained on and correctly predict actions within them using only visual input. Agents see exactly what human players see and navigate by following controller inputs - an approach that translates naturally to physical systems like robotic arms, drones, and autonomous vehicles that humans often control with gaming controllers. General Intuition's initial commercial applications span two distinct areas: gaming and humanitarian robotics. In gaming, the startup is developing intelligent bots and non-player characters that can dynamically adjust difficulty levels, moving beyond traditional "deterministic bots" that produce identical outputs every time. "It's not compelling to create a god bot that beats everyone," explained Moritz Baier-Lentz, a founding member and partner at . "But if you can scale gradually and fill in liquidity for any player situation so that their win rate is always around 50%, that will maximize their engagement and retention." The humanitarian angle reflects de Witte's background in aid work and addresses a critical gap in current robotic capabilities. Search and rescue drones often must navigate unfamiliar environments without GPS, making spatial reasoning crucial for mission success. Unlike competitors building world models like 's Genie or World Labs' Marble for content generation, General Intuition is positioning its technology as an intelligence layer rather than a content creation tool. "Our goal is not to produce models that compete with game developers," de Witte emphasized, sidestepping potential copyright issues while focusing on applications where spatial reasoning creates clear value. The company's next milestones include generating simulated worlds for training other agents and enabling autonomous navigation of entirely unfamiliar physical environments. These capabilities could unlock applications far beyond gaming, from warehouse robotics to space exploration, where pre-programmed responses prove inadequate. The $134 million war chest positions General Intuition to compete with well-funded AI labs while maintaining focus on a specific but crucial capability. As the AI industry grapples with the limitations of language models in understanding physical reality, General Intuition's gaming-trained approach offers a compelling alternative path toward more capable artificial agents.
