General Intuition lands $134M seed to teach agents spatial reasoning using video game clips

Summary

Medal, a platform for uploading and sharing video game clips, has spun out a new frontier AI research lab that’s using its trove of gaming videos to train and build foundation models and AI agents that can understand how objects and entities move through space and time – a concept known as spatial-temporal reasoning. Called General Intuition, the startup is betting that Medal’s dataset – which consists of 2 billion videos per year from 10 million monthly active users across tens of thousands of games – surpasses alternatives like Twitch or YouTube for training agents. “When you play video games, you essentially transfer your perception, usually through a first-person view of the camera, to different environments,” Pim de Witte, CEO of Medal and General Intuition, told TechCrunch. He noted that gamers who upload clips tend to post very negative or positive examples, which serve as really useful edge cases for training. “You get this selection bias towards precisely the kind of data you actually want to use for training work.” This data moat is what reportedly attracted the attention of OpenAI, which late last year attempted to acquire Medal for $500 million, per The Information. (Neither OpenAI nor General Intuition would comment on the report.) It’s also what has led to General Intuition’s raising a whopping $133.7 million in seed funding, led by Khosla Ventures and General Catalyst with participation from Raine. General Intuition’s founding team.,Image Credits:General Intuition The startup intends to use the funds to grow its team of researchers and engineers focused on training a general agent that can interact with the world around it, aiming for initial applications in gaming, and search and rescue drones. De Witte says the founding team has already made strides: General Intuition’s model can understand environments it wasn’t trained on and correctly predict actions within them. It’s able to do this purely through visual input; agents only see what a human playe...