The gamer who said no to OpenAI

In October 2025, the reason became clear. De Witte launched General Intuition, a new AI lab that raised $133.7 million in seed funding from Khosla Ventures and General Catalyst. [2] It is one of the largest seed rounds ever in gaming and AI. The thesis: anyone who wants to understand the future of artificial intelligence should stop focusing on text and start with video.

The difference between describing and understanding

Large language models can describe the world, but they do not understand it. That, at least, is De Witte's conviction. "When people create text to describe the world, a lot of information is lost," he told TechCrunch. "You lose the intuition for how things move and interact in space and time." [3]

The solution lies in so-called world models: AI systems that do not merely predict words, but can simulate physical reality. In 2024, Google DeepMind demonstrated with Genie 2 how such a model can generate a playable 3D world from a single image. [4] The bottleneck? Training data. High-quality video of people performing complex spatial tasks is scarce and expensive to collect.

Medal.tv offers exactly that. Ten million monthly active users upload two billion clips per year from tens of thousands of different games. [5] But the interesting part is not the quantity. Gamers do not share random moments, they upload their best performances and their most spectacular failures. That selection bias, normally a problem in data science, becomes an advantage here. Edge cases are precisely what you need to make AI robust.

The quiet strategy

What General Intuition does not do is equally notable. While DeepMind and World Labs sell their world models as products for AI training and content creation, De Witte keeps his models for internal use. [6] The commercial focus is on NPCs and bots that dynamically adapt to the player's skill level.

That choice is strategically sound. By not competing with game developers, the company avoids the legal minefields around copyright that plague other AI companies. The lawsuit against Stability AI is still ongoing, with a court hearing scheduled for September 2026. [7]

"Our goal is not to build models that compete with game developers," De Witte said. [3] The first products are expected in the first half of 2026.

Beyond gaming

The ambitions extend beyond entertainment. De Witte previously worked at Médecins Sans Frontières, where he co-developed MapSwipe, an app that allows volunteers to analyse satellite images for humanitarian purposes. [8] That background now inspires a second application area: search-and-rescue drones capable of navigating without GPS in unfamiliar terrain.

The underlying idea is that spatial reasoning is a crucial ingredient for artificial general intelligence that language models fundamentally lack. Most AI labs are scaling their language models ever larger. General Intuition is betting on a different modality.

Whether that bet pays off remains uncertain. World models are young enough that the race is still wide open. According to Hartmann Capital, total funding in the sector amounted to $315 million in Q4 2024 alone. [9] Decart recently reached a valuation of $3.1 billion, while World Labs raised $230 million at a valuation of more than one billion. [10]

The stakes

The question hanging over this story is more fundamental than which company wins. It is about whether the next breakthrough in AI will come from language or from imagery, from description or from simulation. De Witte has chosen his position. With two billion video clips per year and investors who were willing to forgo half a billion dollars to get involved, he at least has the means to test his thesis.

First products in 2026. Then we will know more.