General Intuition raises $320M to use video game data to train robots - The Robot Report

General Intuition US Inc. has secured $320 million in Series A funding, catapulting its valuation to $2.3 billion. The round was led by General Catalyst and included investments from Amazon founder Jeff Bezos and former Google CEO Eric Schmidt. This latest injection brings the company’s total funding to $454 million, following a $134 million raise in October. General Intuition plans to channel the capital into building AI models capable of perceiving, predicting, and acting in both virtual and physical environments.

A Radical Data Strategy

While many robotics firms struggle to collect extensive real-world demonstrations or rely on synthetic simulations, General Intuition takes a strikingly different approach. The company taps into billions of gameplay clips uploaded to Medal, a platform co-founded by General Intuition’s CEO Pim de Witte that lets users share gaming moments. These clips capture humans perceiving a digital environment and deciding how to move through it—precisely the kind of decision-making context the company argues is essential for training physical AI.

Beyond Text: Learning from Action

According to the company, text-based models only provide descriptions of reality, which are insufficient for grounding AI in physical action. The game clips offer a richer training signal because they include embedded action labels that record exactly which button a player pressed and when. This timing data gives General Intuition deeper insight into how humans make decisions under varying conditions—a level of detail that far surpasses static text or raw video.

From Words to Worlds

General Intuition contends that human intelligence emerged over millennia of interaction and exploration, through an endless cycle of intent, action, and consequences. The company believes that truly intelligent machines must move beyond language to operate in real and simulated worlds, acquiring a “general intuition of reality.” Its models are designed to learn from unique, action-labeled video datasets across countless gaming environments, yielding agentic systems that can perceive, anticipate, and improvise.

Roadmap and Next Steps

Founded in 2015, General Intuition has been developing two core types of models: action models that decide what to do next, and world models that predict the outcomes of actions. With the new funding, the company intends to scale its compute capacity and focus on pretraining the next version of its model. It also aims to make its API more broadly available this summer, according to TechCrunch, signaling a push toward wider industry adoption.

The source for this article is https://www.therobotreport.com/general-intuition-raises-320m-uses-video-game-data-train-robots/.