General Intuition Secures $2.3B Valuation to Train AI via Gameplay
General Intuition has raised $320 million at a $2.3 billion valuation, betting that millions of hours of human video game footage can teach AI agents to navigate the physical world. The startup uses action-labeled data from its sister company, Medal, to train models that bridge the gap between virtual simulation and robotics.
The company’s approach centers on the belief that traditional large language models lack a fundamental grasp of spatial-temporal reasoning. By utilizing gameplay clips where every button press is recorded, the firm trains its agents to understand causality and physical constraints, such as the permanence of walls or the movement of shadows. During a recent demonstration, a quadrupedal robot powered by this model navigated an office environment, adjusting to real-world obstacles after being fine-tuned with only eight minutes of external robotics data.
This latest funding round, led by Khosla Ventures with participation from figures including Jeff Bezos and Eric Schmidt, brings the company’s total capital to $454 million. CEO Pim de Witte intends to direct the majority of these funds toward scaling compute capacity through a deal with CoreWeave, while preparing to open the startup’s API to external developers. Unlike competitors seeking to build specific end-user products, de Witte aims to provide a foundational model that simplifies development for others in robotics and simulation.
Despite the technical promise, the transition from simulated environments to large-scale, real-world deployment remains an unproven challenge. De Witte has also drawn clear ethical boundaries, explicitly prohibiting the use of his firm's technology for lethal autonomy or military harm. To address potential workforce displacement caused by AI, the company recently launched Nerve, a platform designed to provide gamers with opportunities in data labeling and robot teleoperation.
Comments (0)
No comments yet. Be the first!