Gameplay as a Training Ground for Embodied AI

General Intuition is betting that the key to building generalized AI agents lies in video game data. By leveraging hundreds of millions of hours of gameplay from the platform Medal, the company trains models on spatial-temporal reasoning. Unlike competitors who attempt to infer actions from video alone, General Intuition utilizes the specific action labels embedded in game clips—the exact button presses and timing—to teach the model how to navigate environments and understand causality. This approach allows the model to distinguish between the 'self' and the 'environment,' a critical step for physical embodiment.

From Simulation to Physical Reality

The company uses a proprietary world model, referred to as "the gym," to simulate environments frame-by-frame. This simulation serves as the training ground for agents that can eventually control physical hardware, such as quadrupedal robots. In one demonstration, a robot was fine-tuned for navigation using only eight minutes of real-world data, with the model's base intelligence derived entirely from its virtual training. The company claims this model is compatible with any hardware controllable via standard game inputs (keyboard, mouse, or controller), positioning itself as an ecosystem enabler rather than a hardware manufacturer.

A Data-Driven Business Strategy

General Intuition recently raised $320 million at a $2.3 billion valuation, bringing its total funding to $454 million. The capital is primarily earmarked for scaling compute capacity to pre-train future model versions. The company’s strategy involves a "data flywheel": they intend to partner with customers who provide diverse, real-world data, which in turn improves the model's performance across various embodiments. Furthermore, the company is launching "Nerve," a jobs marketplace that allows gamers to participate in data labeling and robot teleoperation, aiming to provide economic opportunities for the demographic most impacted by AI-driven labor displacement. CEO Pim de Witte has also explicitly restricted the use of its technology for lethal autonomous weapons, citing ethical concerns.