As soon as I entered the research and development floor of General Intuition’s New York office, Pim de Witte, 31, the company’s co-founder and CEO, directed my attention to a monitor on a standing desk. It looked like someone was playing something like Fortnite. It wasn’t a person.
“Our agents have been playing for 100 hours straight,” Kent Rollins, the company’s chief product officer, said with a smile.
Before I could get lost in the sight of an AI navigating the game’s virtual environment, I heard the electronic footsteps of a large quadruped robot approaching.
“The same brain that powers the agents that play the game also powers the robots,” de Witte told me.
Josh Duplantis, a data analyst with a laptop and streaming a live feed from the robot’s single camera, chimed in to explain that the robot’s default mode is “exploration.”
Using its camera and its unique eyes, the giant bug-like bot approached me, circled me, and continued into the office. Occasionally, I would pinch a chair leg or run into the wrong trash can, like a toddler who hasn’t yet learned how his body relates to the world around him. Duplantis said it took just eight minutes to fine-tune the AI model for quadrupeds from real-world robotics data. Additionally, that data was collected on the street, not in offices, where the bots currently travel.
General Intuition’s raison d’être is an agent model that can generalize everything from gameplay to simulation to embodiment. And the model’s ability to figure out its place in the world has secured the support of some heavy hitters.
On Thursday, General Intuition announced it had raised $320 million at a valuation of $2.3 billion, confirming TechCrunch’s previous reporting. This round brings General Intuition’s total public funding to $454 million, following a $134 million round raised during its launch last October.
The startup is a spinout from de Witte’s other company, Medal, which allows gamers to upload and share video game clips. The hundreds of millions of hours of uploaded gameplay provided the initial data set for training General Intuition’s models for spatiotemporal reasoning, or understanding how to move through space and time.
But the key element wasn’t the gameplay footage. It was an action label embedded in those clips that recorded exactly which buttons the player pressed and when. De Witte said most competitors try to infer actions from video alone, but he argues that’s not enough.
“We see this as just the next step in future pre-training,” De Witte said. “We have a single model that can perform actions in response to on-screen Fortnite information, but it can also respond to real-world dynamics in a way that LLM never could.”
At one point, de Witte set me up with a laptop running General Intuition’s world model, a simulated environment that is generated frame by frame rather than rendered by a traditional game engine. As I often do when testing world models, I ran straight into a series of walls. In other demos I’ve tried, the agent you control would sometimes just pass through, but not in this demo. From millions of hours of gameplay, we’ve somehow learned that walls are walls, ladders are for climbing, and shadows lengthen as the sun moves.
To common intuition, this world model is not a product. It’s a training environment (internally known as a “gym”). The company eventually hopes to sell the agent model itself, and de Witte argues that action data embedded in gameplay can help the model distinguish between “self” and “environment” in a way that allows for a deeper understanding of causal relationships.
While it’s impressive to see General Intuition’s technology in the demo, it’s not the only company trying to solve this problem. Moreover, large-scale application of such models to the physical world has not yet been done. Most of these types of approaches require vast amounts of real-world data that are collected over time and money. General Intuition believes that gameplay is a scalable shortcut.
Investors agree with the bet, too. General Intuition’s latest round was led by Khosla Ventures with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers from Google DeepMind and MIT.
The bulk of the round will go toward expanding computing power. General Intuition has a contract with CoreWeave and will focus on pre-training the next version of the model. Slices are being allocated to make that API more widely available by the end of summer.
The company’s Vinod Khosla, who led the round, said he was drawn to De Witte’s vision and the company’s proprietary data position.
“If you look at LLM, when inference came along, it was a quantum leap,” Khosla told me in a phone interview. “I think the quantum leap in world models is the emergence of intuition in AI, an ability similar to human intuition. The human behavioral and reaction data available in games is an important part of the emergence of intuition.”
Vision is a company that transcends generations.
General Intuition is not alone in realizing that Medal’s human behavior data is a critical piece of the puzzle for building dynamic world models and common agents. Briana Martin, the company’s chief of staff, said the company was born in part after Mr. Medal turned down an acquisition offer from a major research institute. Since then, there have been other offers as well.
De Witte and his co-founders Eloy Alonso, Adam Gerry and Vincent Micheli are not interested in an acquisition, and the startup’s investors are not yet looking for an exit. The quantity and quality of proprietary data that General Intuition has through Medal is one of the reasons Khosla believes the startup is a generational bet rather than an M&A target. It has the potential to become the backbone of generalized agents and world models in simulations and the real world.
“Right now, it’s going to be data acquisition, which is kind of uninteresting,” Khosla said.
That bet also includes trusting Mr. de Witte’s values.
The entrepreneur spent three years working in the humanitarian sector, including Médecins Sans Frontières. As such, he has drawn clear lines on how General Intuition’s technology will be used, and no agents will be used to harm humans.
“We don’t want to be an escalating part of the system,” De Witte said. “Suppose I come forward and say, ‘We have lethal self-government,’ what do you think will happen in other countries?”
The military restrictions come as Silicon Valley becomes increasingly bullish on war, but de Witte says he is happy his model will be used for search-and-rescue missions.
De Witte is Dutch and many of his team are European, which shapes the company’s identity. He said Martin was brought in in part because of her decision to publicly leave Palantir because of her work with U.S. Immigration and Customs Enforcement.
“I don’t understand why Silicon Valley would do something like this,” he says. “There’s a reason I’m not there.”
De Witte’s ethics do not simply restrict what the model does not do. De Witte, a gamer who made $1.5 million building and hosting a private RuneScape server as a teenager, also thinks about what happens to those who are left out of the way AI models work.
General Intuition recently launched a platform called Nerve, a job marketplace where gamers can use their existing setup to earn money. Those who sign up can start with labeling data and eventually move on to remote control of robots and other tasks. De Witte said Medal’s user base is exactly the generation most exposed to displacement by AI, and he wants them to be interested in what happens next.
data flywheel
De Witte wants General Intuition to become an ecosystem enabler like Anthropic and OpenAI, a model provider that allows other companies to build on top of its technology. Currently, the startup has a small number of customers in the gaming, simulation, and robotics fields.
“We’re not building a self-driving car company,” De Witte said. “We’re making it 10 times easier for the next person to start a self-driving car company.”
The company says that once its API is in the hands of more customers, it will be able to test its capabilities in a variety of use cases, including testing robots on digital twins on factory floors, powering human-like bots inside game studios, and sending quadrupeds to navigate dangerous environments.
Quadrupedalism is the first physical manifestation that General Intuition has tried out in the real world, but it has also experimented with drones and other devices, including testing models in driving games.
“Anything that can be controlled using a game controller or keyboard and mouse will work,” De Witte says.
One of our goals is to be able to build a data flywheel.
“We will select customers who can diversify the implementations for which this generalized foundation model will serve as the backbone,” De Witte said. “So we will prioritize choosing customers based on whether they can provide us with real-world data that is interesting and useful to advance our research, and whether they have agile in-house teams that allow us to be truly embedded partners and learn from each other.”
Khosla said General Intuition’s proprietary data has helped it get this far, and the company’s ability to continue collecting data that no one else has is essential. Especially since, despite impressive demonstrations, whether the transition from simulation to the real world can be sustained at scale is still an open question that no one has fully answered.
Correction: The headline previously incorrectly stated how much General Intuition raised in this round. The error has been fixed.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
