This is an excerpt from Sources by Alex Heath, a newsletter about AI and the tech industry, delivered only to The Verge subscribers once a week.
Around the middle of last year, Pim de Witte began contacting a handful of leading AI labs to see if they would be interested in using data from Medal, its popular video game capture platform, to train their agents.
Within weeks, it became clear that Medal’s data was more valuable to labs than expected. “We received several acquisition offers very quickly,” he told me. (He declined to name names, but it was reported that OpenAI was offering $500 million.) “Initially, we were very interested in these offers,” he said of the offers, but “that was mainly due to the fact that we didn’t understand what we were sitting on.”
He had read Google DeepMind’s research paper showing that gaming data can be used to teach AI to navigate a 3D environment. However, interest from AI labs made him realize that his data from Medal, which receives about 2 billion video uploads per year from tens of thousands of video games, could be used to develop a unique fundamental model for extending AI to the real world.
“It’s a pretty big gamble.”
Today, Pim de Witte announced that Medal is creating a new AI lab called General Intuition, which has raised a $133.7 million seed round. The money in the round came primarily from Vinod Khosla, founder of Khosla Ventures and an early investor in OpenAI. Other investors include General Catalyst and the Raine Group. Moritz Baier-Lentz, who oversees Lightspeed’s gaming investments, is also joining the startup part-time as a founding team member.
Khosla thinks general intuition could be as impactful in the field of AI agents as OpenAI was on how people use large language models. It’s his company’s biggest startup check since it backed OpenAI in 2018. “It’s a pretty big gamble,” he told me. “They have a unique data set and a unique team.”
Unless you’re steeped in the world of AI, you probably haven’t heard much about global models yet. It is a branch of research that trains AI to have spatial understanding like a human. The idea is that a robot could, for example, predict when a glass of water is going to spill when knocked off a table and catch it before it falls. More practically, AI researchers are increasingly turning to world models as a way to train agents that can reliably generate and interact with 3D space.
Among major AI leaders, Demis Hassabis, CEO of Google DeepMind, has been the strongest advocate of global models and their importance in achieving AGI. Google recently introduced Genie 3, a model that generates a video game-like environment from scratch when you navigate through it. There are also a handful of startups working on similar models, including Fei-Fei Li’s World Labs, which this week released its own demo of a model that generates interactive videos in real time.
For General Intuition, the goal is to control any type of device that can be mapped to a keyboard and mouse or have an input pattern similar to that of a game controller, according to de Witte. He expects the startup’s first model to be used by search and rescue drones, but sees potential for applications in other areas, including humanoid robots and self-driving cars.
Just as LLMs were initially trained on Internet text data, de Witte believes that gaming environments will allow AI to reliably predict appropriate actions to take in the physical world. “Games are basically the only testable domain for space-time reasoning,” he explained. “You can separate a good deed from a bad deed, that’s why it’s so valuable.”
It’s still a risky bet. The correct technical path to developing global models is hotly debated in the AI industry, and as even Khosla pointed out to me, it’s unclear which data will ultimately prove most valuable. Members of De Witte’s first research team have published notable research in the field, but the startup still competes with better-funded giants like Google. “Someone will win big in this market,” said Khosla, who told me he thought it was an area in which “companies worth hundreds of billions of dollars, or potentially billions of dollars, would be built.”
De Witte predicts that gaming companies will become prime targets for AI labs as interest in global models intensifies. Its decision to launch General Intuition was driven by the realization that, with Medal’s data, it is in the unique position of being more than just a data provider. However, he cautioned me that others might find it difficult to resist licensing checks and acquisition offers from big AI labs.
“You’re at an information disadvantage,” he replied when I asked if he had any advice for the gaming industry. “The better these models get, the less data they’re likely to need.”
