General Intuition lands $134M seed to teach agents spatial reasoning using video game clips (techcrunch.com)

Medal has spun out General Intuition, a new AI lab that just raised $133.7M in seed funding (led by Khosla Ventures and General Catalyst) to train agents that understand space and time from video-game clips. General Intuition is leveraging Medal’s vast dataset — ~2 billion clips per year from 10 million monthly users across thousands of games — arguing that first-person, high-signal gameplay footage (often extreme success/failure edge cases) is uniquely useful for learning how objects and agents move and interact. The dataset reportedly even drew acquisition interest from OpenAI last year, underscoring its strategic value. Technically, General Intuition trains models that learn spatial-temporal reasoning purely from visual inputs and controller-action traces: agents only “see” what a human player would see and learn to act via game controls. The team says their models generalize to environments they weren’t trained on and can predict correct actions, enabling transfer to physical systems (robotic arms, drones, autonomous vehicles) and scalable non-player characters for games. Rather than selling world models as products, the startup focuses on agent behaviors and simulated-world generation for training, with target applications including gaming, search-and-rescue drones, and broader embodied AI — a capability its founders argue is a missing ingredient on the path to AGI.