General Intuition was spun out of Medal, co-founder and CEO Pim de Witte’s gaming company that lets users upload and share video game clips. Those clips gave the startup access to hundreds of millions of hours of gameplay, but de Witte says the more important asset is the action data attached to the footage: the record of which buttons players pressed and when they pressed them.
That data is central to the company’s pitch. General Intuition is using it to train models that understand movement through space and time, with the aim of building agents that can operate in software environments and eventually in the physical world. In one demo, the same underlying model powered an AI agent playing a Fortnite-like game and a quadrupedal robot navigating the company’s office.
“Our agent has been playing for 100 hours straight,” chief product officer Kent Rollins said.
The robot, using a single camera, moved around the office in an exploration mode after being fine-tuned with eight minutes of real-world robotics data, according to the company. That training data was collected outside rather than in the office where the robot was shown operating.
“The same brain powering the agent playing the game is powering the robot,” de Witte said.
General Intuition is not selling the simulated world model itself. The company treats that environment as a training system, internally referred to as “the gym,” while the commercial focus is the agentic model. De Witte argues that gameplay data gives the model a clearer way to distinguish between its own actions and the environment around it.
“We view this as just the next stage of future pre-training,” de Witte said. “We have a single model that can respond to Fortnite information on the screen and take action, but also to real-world dynamics in a way that an LLM could never.”
The company’s approach remains unproven at scale. AI systems that transfer from simulation into real-world robotics typically require large amounts of physical-world data, which can be expensive and slow to collect. General Intuition’s bet is that human gameplay can provide a faster path because it includes both visual information and action labels.
Most of the new funding will be used to increase compute capacity. General Intuition has a deal with CoreWeave and plans to focus on pre-training the next version of its model. Part of the round will also support broader API availability by the end of summer.
Vinod Khosla said the company’s proprietary data was a major reason his firm invested. “If you look at LLMs, when reasoning emerged, it was a quantum leap,” Khosla said. “In world models, I think the quantum leap is the emergence of intuition in the AI, a human intuition-like capability. The human action data and reaction data you have in games is the key part to the emergence of intuition.”
General Intuition has already attracted acquisition interest, according to the company, including an offer involving Medal’s data. De Witte and co-founders Eloi Alonso, Adam Jelley and Vincent Micheli are not pursuing a sale, and Khosla described the company as more than a potential data acquisition. “At this point, it would be a data acquisition, which is sort of uninteresting,” Khosla said.
The company is also drawing boundaries around how its technology can be used. De Witte said General Intuition will not allow its agents to be used to harm people, though he said applications such as search and rescue would be acceptable. “We don’t want to be an escalatory part of the system,” de Witte said. “Let’s say I were to come out and say, ‘We’re doing lethal autonomy.’ What do you think would happen in other countries?”
General Intuition has also launched Nerve, a jobs marketplace aimed at gamers. The platform begins with data-labeling work and could later expand into robot teleoperation and other tasks, according to the company.
De Witte says General Intuition wants to operate as a model provider rather than build end-user companies in every category. The startup currently has customers in gaming, simulation and robotics, and plans to choose partners that can help expand the range of environments and physical systems its model learns from. “We’re not gonna build a self-driving car company,” de Witte said. “We’re gonna make it 10 times easier for the next person to build a self-driving car company.”
The company has tested its model with a quadruped, drones, other devices and driving games. De Witte said the system can work across controllable environments. “It works on anything that you can control using a game controller or a keyboard mouse,” de Witte said.
For General Intuition, the next stage will be proving that its gameplay-driven training approach can translate into reliable performance across real-world systems. The company’s proprietary data helped secure major investor backing, but its long-term position will depend on whether that data advantage can keep improving models as they move beyond demos and into customer use cases.
This analysis is based on reporting from TechCrunch.
Image courtesy of General Intuition.
This article was generated with AI assistance and reviewed for accuracy and quality.
