Medal, a platform for uploading and sharing video game clips, has established General Intuition, a frontier AI research lab that raised $133.7 million in seed funding to build foundation models teaching agents spatial-temporal reasoning.
The startup will use Medal’s dataset of two billion videos annually from 10 million monthly active users across tens of thousands of games to train AI agents capable of understanding how objects and entities move through space and time, reports TechCrunch. Khosla Ventures and General Catalyst led the funding round, with participation from Raine.
Pim de Witte, chief executive of Medal and General Intuition, noted that creating text to describe the world results in information loss, particularly around spatial-temporal reasoning capabilities.
The founding team claims General Intuition’s model can understand environments it was not trained on and correctly predict actions within them using purely visual input. Agents see only what human players would see and move through space by following controller inputs, an approach the company states can transfer to physical systems, including robotic arms, drones and autonomous vehicles.
De Witte said gamers who upload clips tend to post highly negative or positive examples, serving as useful edge cases for training. Medal’s dataset reportedly attracted acquisition interest from OpenAI, which attempted to purchase Medal for $500 million late last year, according to The Information. Neither OpenAI nor General Intuition commented on the report.
General Intuition plans to focus on creating bots and non-player characters that surpass traditional preprogrammed characters, while also powering search and rescue drones to navigate unfamiliar environments without GPS. The startup operates as a public benefit corporation and stated it will not develop technology to replace game developers, designers, or artists.
“Our goal is not to produce models that replaces game developers,” says de Witte.
Moritz Baier-Lentz, a founding member of General Intuition and partner at Lightspeed Ventures, noted the bots can scale to any difficulty level, maximising player engagement by maintaining approximately 50 per cent win rates through gradual difficulty adjustments.
The company’s next milestones include generating new simulated worlds for training agents and autonomously navigating entirely unfamiliar physical environments. Unlike competitors, including DeepMind and World Labs, selling world models for training agents and content creation, General Intuition positions its world models as training infrastructure rather than products to avoid copyright issues.
