Imagine artificial intelligence that can not only talk, but also understand space and time much like a human. That is exactly what the new startup General Intuition, which was spun off from the game video-sharing platform Medal, is working on. The project recently announced that it had raised $133.7 million (approximately CZK 3.1 billion) in a seed funding round. The money came primarily from Vinod Khosla, founder of Khosla Ventures, who was one of OpenAI's earliest investors. Other investors include General Catalyst and Raine Group. Moritz Baier-Lentz, who oversees gaming investments at Lightspeed, joined the team part-time as a founding member.
General Intuition focuses on developing foundation models for artificial intelligence agents capable of spatiotemporal reasoning. This means they can understand how objects move through space and time. According to Pim de Witte, CEO of both Medal and General Intuition, this is crucial to achieving artificial general intelligence (AGI). De Witte explains that text models lose a great deal of information because people describe the world in words while omitting details about space and movement.
The advantage of data from the gaming world
The entire idea stems from the vast amount of data Medal collects. Every year, approximately 2 billion videos are uploaded to the platform by 10 million monthly active users across tens of thousands of games. Players typically share extreme moments—major victories or crushing defeats. This creates a dataset full of edge cases that are ideal for training artificial intelligence. De Witte says that such data has a "selection bias toward exactly the kind of information you want to use for training."
This data attracted the attention of major industry players. Last year, offers emerged to acquire Medal, including one reportedly made by OpenAI for $500 million (approximately CZK 11.6 billion). De Witte admits that they initially considered the offers, but then realized the value of what they had. Instead of selling, they decided to use the data themselves to build unique models.
How world models work
General Intuition is building on the concept of world models, which are neural networks designed to generate virtual environments. These models are intended to teach artificial intelligence to predict actions in 3D space. For example, a robot could predict when a glass of water is about to fall off a table and catch it before it hits the ground. De Witte sees potential in controlling devices whose inputs can be mapped to a keyboard, mouse, or game controller.
The first applications are expected to target drones for search and rescue operations that navigate unfamiliar environments without GPS. This is connected to De Witte's experience in humanitarian work. Another area is the gaming industry, where they want to create intelligent bots and non-player characters that dynamically adjust the difficulty. Instead of unbeatable "god bots," they should maintain a balanced game so that players have around a 50% chance of winning and remain engaged.
What the future holds
The General Intuition team has already demonstrated that its models can understand environments on which they were not trained and predict actions based solely on visual input. The agents see the same things as human players and navigate using control inputs, an approach that can be transferred to physical systems such as robotic arms or autonomous vehicles. Nevertheless, it is a risky bet—the path to perfect world models remains a subject of debate within the industry.
De Witte predicts that gaming companies will become attractive acquisition targets for major artificial intelligence labs as interest in world models grows. He chose to pursue his own path thanks to Medal's data, but warns others that they are at an informational disadvantage. The better the models become, the less data they will need.
General Intuition operates out of New York and Geneva, where it is building a European research and engineering hub. This helps with recruiting talent and collaborating with the European artificial intelligence community.
Sources: theverge.com
