Medal, a service for sharing and uploading video game clips, has launched a new AI research lab that leverages its vast collection of gaming videos to develop foundational models and AI agents capable of understanding how objects and entities move through space and time—a concept known as spatial-temporal reasoning.
The new venture, named General Intuition, is banking on Medal’s extensive dataset—which includes 2 billion videos annually from 10 million monthly users playing thousands of different games—offering a superior resource for training AI agents compared to platforms like Twitch or YouTube.
“When you’re gaming, you essentially project your perception—often from a first-person camera—into various settings,” explained Pim de Witte, CEO of both Medal and General Intuition, in an interview with TechCrunch. He pointed out that users tend to share clips that highlight either extreme successes or failures, which provide valuable edge cases for training AI. “This creates a selection bias toward exactly the kind of data that’s most useful for training purposes.”
This unique data advantage reportedly caught the eye of OpenAI, which, according to The Information, made a $500 million acquisition offer for Medal late last year. (Both OpenAI and General Intuition declined to comment on the matter.)
This same data trove has also helped General Intuition secure an impressive $133.7 million in seed funding, with Khosla Ventures and General Catalyst leading the round and Raine also participating.
The founding team of General Intuition.
Image Credits:General Intuition
The company plans to use the new capital to expand its team of engineers and researchers, with the goal of developing a general-purpose agent capable of interacting with its environment—starting with applications in gaming and search-and-rescue drones.
According to de Witte, the founding team has already made significant progress: General Intuition’s model can interpret environments it hasn’t previously encountered and accurately anticipate actions within those spaces. The model relies solely on visual data; agents only perceive what a human player would see and navigate using controller inputs. This method, the company claims, can be seamlessly applied to real-world systems such as robotic arms, drones, and autonomous vehicles, which are often operated by humans with game controllers.
General Intuition’s upcoming goals are twofold: to create new simulated environments for training agents and to enable autonomous navigation in completely new physical settings.
This technical strategy is influencing how the company approaches commercialization and distinguishes it from other firms developing world models.
Although General Intuition is also developing world models for agent training, these models are not their main product. In contrast to companies like DeepMind and World Labs, which market their world models Genie and Marble for agent training and content creation, General Intuition is focusing on alternative applications to steer clear of copyright complications.
“We’re not aiming to develop models that compete with game creators,” de Witte stated.
Instead, the startup’s focus in gaming is on building bots and non-player characters that outperform traditional “deterministic bots”—preprogrammed characters that always behave the same way.
“[These bots] can adapt to any skill level,” said Moritz Baier-Lentz, a founding member of General Intuition and partner at Lightspeed Ventures, in a conversation with TechCrunch. “The goal isn’t to make an unbeatable bot, but to dynamically adjust difficulty and ensure that players’ win rates hover around 50%. This keeps players engaged and coming back.”
De Witte’s background in humanitarian work also shapes the company’s mission to develop search-and-rescue drones that can operate in unknown environments and gather information even without GPS.
Ultimately, de Witte and Baier-Lentz believe that General Intuition’s core capability—spatial-temporal reasoning—is essential in the pursuit of artificial general intelligence (AGI). While most leading AI labs are focused on scaling up large language models, General Intuition argues that true AGI requires something fundamentally different from what LLMs offer.
“As humans, we use language to describe our world, but in doing so, a lot of information is lost,” de Witte said. “You lose the innate intuition about how things move and interact in space and time.”
