General Intuition, a New York startup backed by Jeff Bezos valued at $2.3 billion, has closed a $320 million funding round to pursue artificial general intelligence through video game data. CEO Pim de Witte argues that large language models struggle to understand physical movement and spatial relationships, skills essential for true generalisation. The company claims gaming environments offer superior training material compared to the open internet because they provide consistent physics and clear cause-and-effect logic. Investors including Eric Schmidt, researchers from MIT, and Google DeepMind have joined the company’s list of backers. The firm spun out from Medal TV, a gaming platform, and intends to build world models capable of simulating physical reality rather than just processing text.
This approach matters because current AI systems lack reliable reasoning about how objects interact in three-dimensional space. Training on chaotic internet data often leads to hallucinations regarding physics, whereas games offer controlled scenarios where gravity and collision rules remain constant. If successful, these models could improve robotics and autonomous navigation without relying on expensive real-world sensors. The technology also raises questions about ethical boundaries, particularly regarding potential military applications.
- Physics rules in games are consistent unlike real-world data
- Current text models fail at spatial reasoning tasks
- Defense applications present significant ethical concerns
