Spun out from gaming platform Medal, the new startup is training AI agents using billions of video game clips to navigate both digital and physical environments.
Training AI Through Gameplay: A New Path to Intelligence
General Intuition, a new AI research startup spun out of Medal, has raised $133.7 million in seed funding to build foundation models that understand how objects and agents move through space and time — a field known as spatial-temporal reasoning.
Unlike traditional AI models trained on text or static images, General Intuition is tapping into billions of gameplay clips uploaded annually by Medal’s 10 million monthly users, offering a unique data advantage.
- Medal’s dataset includes edge-case moments, such as dramatic wins or fails.
- These clips offer highly relevant training material for teaching AI how humans react and interact in dynamic environments.
Why Spatial Reasoning Matters for AGI
Current AI systems like LLMs excel at language but struggle with understanding physical context. General Intuition believes true artificial general intelligence (AGI) demands more than just text prediction.
“You lose general intuition around spatial-temporal reasoning when you convert the world into words,” said CEO Pim de Witte.
Their AI agents are trained to perceive the world visually, just as humans do when playing video games. These agents:
- Navigate environments via controller inputs.
- Learn to operate without GPS or external sensors.
- Generalize to new environments they weren’t explicitly trained on.
Backing and Use Cases Beyond Gaming
The impressive seed round was led by Khosla Ventures and General Catalyst, with participation from Raine Group. This early-stage backing signals strong confidence in General Intuition’s long-term vision.
The startup plans to scale up hiring in AI research, engineering, and robotics, targeting use cases such as:
- Search-and-rescue drones that navigate GPS-denied zones.
- Robotic arms and autonomous vehicles that operate like skilled human controllers.
- Next-gen gaming bots that behave more realistically and scale to any player level.
Unlike others developing world models for sale — such as DeepMind’s Genie or World Labs’ Marble — General Intuition is not commercializing their models as game-building tools.
“Our goal is not to compete with game developers,” de Witte emphasized.
Smarter Bots, Better Games
General Intuition’s immediate focus within gaming is to replace traditional deterministic bots — the kind that repeat the same actions every time — with adaptive AI agents.
These new bots can:
- Match a player’s skill level dynamically.
- Keep win rates around 50%, maximizing engagement and retention.
- Enhance the player experience across varied game scenarios.
“It’s not compelling to create a god bot that beats everyone,” said Moritz Baier-Lentz, founding team member and partner at Lightspeed Ventures.
Instead, the goal is to create scalable intelligence that feels natural, human, and responsive — a step closer to real-world capability.
Why This Matters for the Future of AI
General Intuition’s work points to a paradigm shift in how AI is trained and applied. While most major labs focus on language and vision benchmarks, this team is building agents that can:
- Learn from visual inputs alone.
- Predict actions in real-time 3D environments.
- Transfer learnings from gameplay to real-world robotics.
This visual-first, action-based learning could unlock more robust general intelligence than text-based systems can achieve alone.
