General Intuition Raises $134M to Train AI Agents with Video Game Clips
General Intuition Raises $134M to Train AI Agents with Video Game Clips
General Intuition, a newly launched AI research lab, is making waves with its innovative approach to training artificial intelligence agents using video game clips. The company has secured an impressive $133.7 million in seed funding to accelerate its mission of teaching AI spatial-temporal reasoning—a core skill for understanding how objects and entities move through space and time.
Harnessing Massive Gaming Data
Spun out from Medal, a leading platform for sharing video game clips, General Intuition leverages a unique dataset: over 2 billion gaming videos uploaded annually by 10 million monthly active users. This dataset, representing diverse game environments and gameplay moments, is considered a "data moat" that surpasses alternatives like Twitch or YouTube for training AI agents.
Pim de Witte, CEO of Medal and General Intuition, explained that gamers typically upload their most dramatic wins or losses—providing valuable edge cases for AI training. "You get this selection bias towards precisely the kind of data you actually want to use for training work," he noted.
From Gaming to Real-World Applications
With funding from Khosla Ventures, General Catalyst, and Raine, General Intuition is expanding its team of researchers and engineers. Their aim is to develop AI agents that can interact naturally with the world—starting with gaming, but with applications in areas like search and rescue drones and robotics.
What sets General Intuition apart is its model's ability to understand entirely new environments and accurately predict actions using only visual input. The AI agents move through space by following the same controller inputs a human player would use, making the technology directly transferable to physical systems such as robotic arms, drones, and autonomous vehicles.
Going Beyond World Models
Unlike other labs that sell world models (like DeepMind's Genie or World Labs' Marble), General Intuition does not compete with game developers. Instead, its focus is on creating adaptable AI bots and non-player characters (NPCs) for games—bots that can dynamically adjust difficulty to match player skill, enhancing engagement and retention.
Moritz Baier-Lentz, co-founder and partner at Lightspeed Ventures, emphasized this approach: "It’s not compelling to create a god bot that beats everyone, but if you can scale gradually and fill in liquidity for any player situation so that their win rate is always around 50%, that will maximize their engagement and retention."
Humanitarian Impact and Future Vision
General Intuition's technology also has humanitarian potential, particularly in powering search and rescue drones that must navigate unknown environments without GPS. De Witte’s background in humanitarian work shapes this focus, demonstrating how gaming data can have real-world impact.
Ultimately, the team sees spatial-temporal reasoning as a missing link in the quest for artificial general intelligence (AGI). While large language models dominate headlines, General Intuition argues that true AGI will require a deeper, intuitive understanding of the physical world—something current models lack.
What's Next for General Intuition
- Generating new simulated worlds for training AI agents
- Enabling agents to autonomously navigate unfamiliar physical environments
- Expanding into robotics, drones, and advanced gaming AI
As the company pushes towards these milestones, its pioneering approach could set new standards for spatial intelligence in AI.
