Summary: World models represent a promising frontier in AI research, enabling machines to understand and interact with the physical world in a human-like way. A new startup, General Intuition, is leveraging vast gaming data to develop AI agents capable of spatial reasoning and real-world interaction. Backed by major investors, this approach could revolutionize applications from drones to robotics, though challenges and competition remain fierce.
The Rise of General Intuition and the Power of Gaming Data
Last year, Pim de Witte, founder of the popular video game clipping platform Medal, began reaching out to leading AI labs to explore how Medal’s extensive gaming data might train AI agents. Medal processes roughly 2 billion video uploads annually from tens of thousands of games, offering a rich dataset for AI research.
The response from AI labs was overwhelming, with multiple acquisition offers arriving quickly—including a reported $500 million offer from OpenAI, which de Witte declined. This interest revealed the untapped potential of Medal’s data to build foundational AI models that extend into the real world.
Building on this momentum, de Witte announced the spinout of a new AI lab called General Intuition, which recently secured a $133.7 million seed round. The funding primarily comes from Vinod Khosla, founder of Khosla Ventures and an early investor in OpenAI, alongside General Catalyst and the Raine Group. Moritz Baier-Lentz, Lightspeed’s gaming investments lead, also joined as a founding team member.
What Are World Models?
World models are an emerging branch of AI research focused on teaching machines spatial understanding similar to humans. Instead of just processing data, these models enable AI agents to predict and interact with their environment—for example, anticipating when a glass might fall from a table and acting to prevent it.
Google DeepMind CEO Demis Hassabis has been a vocal advocate for world models, highlighting their importance in advancing artificial general intelligence (AGI). Google’s recent demo of Genie 3, which generates dynamic video game-like environments, exemplifies this technology. Other startups, like Fei-Fei Li’s World Labs, are also developing real-time interactive video generation models.
Why Gaming Data Matters for AI
General Intuition aims to control devices that use keyboard, mouse, or game controller inputs. Their first applications may include search and rescue drones, with potential expansion into humanoid robots and self-driving cars.
De Witte explains that gaming environments provide a uniquely verifiable domain for spatial-temporal reasoning. Unlike many real-world scenarios, games allow clear differentiation between good and bad actions, making them ideal for training AI to predict appropriate responses.
The Future Potential and Challenges of World Models
Despite the excitement, the path to developing effective world models is still debated within the AI community. The most valuable data sources remain uncertain, and startups like General Intuition face competition from well-funded giants like Google.
Khosla believes this field could spawn multiple companies worth hundreds of billions or even trillions of dollars. However, success will require navigating technical challenges and market competition.
Looking Ahead: The Impact on Gaming and AI Industries
De Witte predicts that gaming companies will become prime acquisition targets as AI labs seek valuable data to advance world models. His decision to launch General Intuition reflects a desire to be more than just a data provider, leveraging Medal’s unique dataset to lead in this space.
He also cautions the gaming industry about the risks of licensing data too early, noting that as models improve, they may require less data to achieve breakthroughs.
As world models continue to develop, their ability to bridge AI with real-world understanding promises to reshape technology and industry in profound ways.