Full story
Gaming data to robotics
General Intuition, the AI agent and world model startup spun out of Medal, raised $320 million at a $2.3 billion valuation to train agents that can act across space and time using action-labeled gameplay clips.
“Insider Brief - General Intuition raised $320 million in Series A funding at a $2”
TechCrunch described a demonstration at General Intuition’s New York office where its agent “has been playing for 100 hours straight,” before a quadrupedal robot approached as “the same brain powering the agent playing the game is powering the robot.”
The company’s thesis centers on the action labels embedded in Medal’s footage—“records of exactly what buttons a player pressed and when”—as the key ingredient for teaching spatial-temporal reasoning.
TechCrunch also said it took “just eight minutes of real-world robotics data” to fine-tune the AI model for the quadruped after the street-collected data was gathered outside the office.
Backers and training plan
General Intuition’s Series A round was led by Khosla Ventures and included participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers at Google DeepMind and MIT, with the round bringing total disclosed funding to $454 million after a $134 million launch raise last October.
TechCrunch reported that “The vast majority of the round will go toward scaling compute capacity,” with a deal with CoreWeave planned for pre-training the next version of the model.
The company also earmarked “a slice” of the funding for making its API more broadly available by the end of summer, as it aims to sell the agentic model itself rather than the world model directly.
In an earlier round-up, AI Insider said General Intuition trains large action foundation models on action-labeled gameplay clips from Medal’s “17 million monthly active users,” and it is developing world models to generate training environments for AI agents.
Ethics and what’s at risk
General Intuition’s approach is framed around building systems that can generalize from gameplay to simulation to embodiment, but TechCrunch cautioned that “getting such a model to hold up in the physical world, at scale, hasn’t yet been done.”
“General Intuition raises $320M at $2B valuation to scale AI training with gameplay data The Medal spinoff is betting that millions of hours of gaming footage can teach AI something textbooks never could: intuition”
In its own description of the company’s direction, TechCrunch quoted de Witte saying, “We view this as just the next stage of future pre-training,” positioning the world model as “the gym” internally rather than the end product.
The stakes for investors and customers hinge on whether the action-data thesis can translate into useful real-world behavior, with TechCrunch describing how the world model learned that “walls are walls, ladders are for scaling, and shadows lengthen as the sun moves.”
Beyond demos, the company’s broader plan ties the funding to expanding compute and broadening access to its API, while TechCrunch said it ultimately wants to “sell the agentic model itself” and make the training environment available for agent development.
