The Bet
I've seen a lot of AI demos. Most of them feel like magic tricks—flashy, but brittle. You show the robot walking, and it's charming. Then you ask it to open a door, and it spins in circles like a confused puppy.
General Intuition's demo wasn't charming.
It was unnerving.
The quadruped didn't just walk up to me. It circled. It paused. It looked me in the eye—well, the camera lens that passed for one—and then kept going. No programmed path. No pre-scripted motion. Just… presence.
That's the point.
Pim de Witte didn't build this company to make robots. He built it to build intuition.
And he's betting $320 million that the key to artificial intuition isn't more compute, or better transformers, or even more data.
It's the button presses.
The ones you didn't even know you were making.
The Data Moat
Most AI companies scrape video. They feed it into models, hoping the system will guess: "Was that a jump? A crouch? A dodge?"
It's like trying to learn to play piano by watching someone's fingers move and guessing which keys they pressed.
General Intuition didn't do that.
They started with Medal.
A platform where gamers upload clips of their best plays. Not just the highlights. The messy, repetitive, 47-minute Fortnite run where you respawned six times and still didn't win. The hours spent grinding in Elden Ring just to get the timing right on a parry.
And every single one of those clips? Embedded with the exact timestamp of every button pressed. Every time you tapped X to jump, held RT to sprint, flicked the right stick to aim.
That's not video.
It's intention.
That's the difference between seeing someone walk and knowing why they chose to walk.
Most AI models see pixels.
General Intuition's model sees decisions.
From Pixels to Physics
I asked de Witte how long it took to get the robot to walk.
"Eight minutes."
I laughed.
He didn't.
The robot wasn't trained on robotics data. It wasn't fine-tuned on LIDAR scans or force sensors.
It was fine-tuned on eight minutes of footage from a camera mounted on its head—real-world data, collected on the street, in natural lighting, with actual people walking by.
The model didn't need to learn how to move.
It already knew.
It just needed to map its virtual body onto a physical one.
That's the magic.
The same brain that played Fortnite for 100 hours straight is now navigating a cluttered office.
It doesn't know it's a robot.
It doesn't know it's in the real world.
It just knows: "This is a space. I am here. I must explore."
And that's what intuition looks like.
Not a prediction.
A posture.
The Investors
You don't get Jeff Bezos, Eric Schmidt, and Khosla Ventures to write checks for $320 million because you have a cool demo.
You get them because you have a moat.
And General Intuition's moat isn't software.
It's data.
The kind of data you can't buy.
You can't scrape it.
You can't synthesize it.
It's the messy, human, unfiltered record of what people do when they think no one's watching.
Khosla put it bluntly: "In LLMs, reasoning was the quantum leap. In world models, intuition is. And the only way to get that is from human action data."
They're not investing in a robotics company.
They're investing in the first true embodiment dataset.
And they know: once you have this, you don't need to train on real-world robots anymore.
You just need to simulate.
The Platform Play
General Intuition isn't building a self-driving car.
They're building the engine.
They're not selling robots.
They're selling the ability to make robots that don't act like robots.
"We're not gonna build a self-driving car company," de Witte told me. "We'll make it 10x easier for the next person to build one."
That's why the API is coming this summer.
They're not trying to be OpenAI.
They're trying to be the new CUDA.
The foundational layer.
The thing that makes everything else possible.
And the customers? They're not just robotics startups.
They're gaming studios who want NPCs that feel alive.
Factory owners who need digital twins that predict failures before they happen.
Search and rescue teams who want drones that can navigate collapsed buildings without GPS.
This isn't about robots.
It's about making the world feel alive to machines.
Ethics & Nerve
I asked de Witte about military use.
"We drew a line," he said. "No lethal autonomy. Ever."
He didn't say it like a PR statement.
He said it like someone who's seen what happens when technology escapes its ethics.
He spent three years with Doctors Without Borders.
His chief of staff quit Palantir over ICE.
This isn't a Silicon Valley startup pretending to be ethical.
It's a Dutch engineer who grew up on RuneScape servers and decided the future shouldn't be built by people who don't care who gets left behind.
That's why they launched Nerve.
A jobs platform for gamers.
Not to replace them.
To give them a stake.
You used to earn XP.
Now you can earn a paycheck.
Label data. Teleoperate a robot. Train the next model.
It's not charity.
It's justice.
The Open Question
Is this real?
Can a model trained on Fortnite really generalize to a warehouse floor?
Can a quadruped that learned to walk from 200 million hours of gameplay handle snow?
Rain?
A broken stair?
I don't know.
But I know this: the only thing more dangerous than a machine that thinks like a human… is a machine that acts like one.
And if General Intuition is right?
The next wave of AI won't be smarter.
It'll be more alive.
And we'll have gamers to thank for it.