General Intuition, a New York startup backed by Jeff Bezos, has raised $320 million to build artificial general intelligence using data from video games. The company argues that large language models lack the ability to understand physical movement through space and time, a gap they aim to fill by training world models on gaming environments. This funding round includes investment from Coatue, Eric Schmidt, and researchers at MIT and Google DeepMind. The firm spun out of Medal TV and recently declined a reported acquisition offer from OpenAI to maintain independence. CEO Pim de Witte discussed the approach on TechCrunch’s Equity podcast, highlighting how eight minutes of real-world data helped a robot navigate an office quickly. The team also noted ethical concerns regarding potential use in defense applications.
The strategy matters because current AI struggles with generalising physical tasks beyond text processing. By training on simulated physics and spatial reasoning, the models may eventually control robots or navigate complex environments more effectively than text-only systems. The company is also launching Nerve, a marketplace connecting gamers to data labeling and teleoperations work to address job displacement fears. This approach could provide a practical path toward physical AI without relying solely on massive datasets of human text.
* Eight minutes of real-world data enabled a robot to navigate an office efficiently.
* The company rejected an OpenAI acquisition offer to preserve its mission focus.
* Nerve connects gamers to data labeling and teleoperations tasks.
