This is a fragment Sources: Alex Heatha newsletter about artificial intelligence and the tech industry, distributed once a week only to The Verge subscribers.
Around the middle of last year, Pim de Witte began contacting several leading artificial intelligence labs to see if they would be interested in using data from Medalits popular video game cutting platform to train its agents.
Within a few weeks, it became clear that Medal’s data was more valuable to the labs than he had expected. “We received multiple takeover offers very quickly,” he told me. (He declined to give names, but yes has been reported that OpenAI offered $500 million). “We were very interested in them initially,” he said of the offers, but “that was mostly a result of us not understanding where we were sitting.”
He read Google DeepMind research article showing that game data can be used to teach artificial intelligence to navigate in a 3D environment. But interest from artificial intelligence labs made him realize that his data from Medal, which receives approximately 2 billion videos from tens of thousands of video games each year, could be used to develop a unique, basic model for extending artificial intelligence to the real world.
“It’s a pretty big bet.”
Today, Pim de Witte announced that Medal is creating a recent artificial intelligence lab called General Intuition, which has raised a $133.7 million seed round. Funding for the round comes primarily from Vinod Khosla, founder of Khosla Ventures and one of the early investors in OpenAI. Other investors include General Catalyst and Raine Group. Moritz Baier-Lentz, who oversees Lightspeed’s gaming investments, is also joining the startup part-time as a founding team member.
Khosla believes General Intuition can have as much impact on AI agents as OpenAI has on the way humans employ gigantic language models. It’s the biggest audit of his company since backing OpenAI in 2018. “It’s a pretty big assumption,” he told me. “They have a unique data set and a unique team.”
Unless you’re saturated in the world of artificial intelligence, you probably haven’t heard much about world models yet. This is a field of research that trains artificial intelligence to understand space like a human. The idea is that the robot could, for example, predict when a glass of water would be spilled after being shot off the table and grab it before it falls. From a practical standpoint, AI researchers are increasingly using world models as a way to train agents that can reliably generate and interact with 3D space.
Among the top AI leaders, the most ardent supporter of global models and their importance in achieving AGI is Google DeepMind CEO Demis Hassabis. Google recently unveiled Genie 3, a model that generates a video game-like environment from scratch as you move around it. Several startups are also working on similar models, including Fei-Fei Li’s World Labs released its own demo this week model generating interactive video in real time.
According to de Witte, General Intuition’s goal is to control any type of device that can be assigned to a keyboard and mouse or that has an input pattern similar to a game controller. 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 autonomous cars.
Just as college-educated people were initially trained in online text data, de Witte believes that gaming environments will unlock the ability of artificial intelligence to reliably predict the right actions to take in the physical world. “Games are basically the only verifiable domain of spatial-temporal reasoning,” he explained. “You can separate good action from bad action, which is why it is so valuable.”
Still, it’s a risky bet. The correct technical path for developing world models is a hotly debated topic in the AI industry, and as even Khosla noted, it’s unclear which data will ultimately prove to be the most valuable. Members of de Witte’s early research team have published noteworthy research in this field, but the startup still competes with better-funded giants like Google. “Someone is going to be very successful in this market,” said Khosla, who told me he thinks this is an area where “multi-hundred billion-dollar, potentially even trillion-dollar companies will be built.”
De Witte predicts gaming companies will become prime acquisition targets for AI labs as interest in world models grows. His decision to found General Intuition was based on the realization that Medal’s data put it in a unique position to be more than just a data provider. But he warned me that others might find it challenging to resist license checks and takeover offers from gigantic AI labs.
“You are at an information disadvantage,” he said when I asked if he had advice for the gaming industry. “The better these models are, the less data they are likely to need.”