Medal, a platform for uploading and sharing video game clips, has established a new frontier AI research lab. It leverages rich game videos to train and build foundational models and AI agents that can understand how objects and entities move through space and time, a concept known as spatiotemporal reasoning.
The startup, called General Intuition, is betting that Medal’s dataset, which consists of 2 billion annual videos from 10 million monthly active users across tens of thousands of games, will outperform alternatives like Twitch and YouTube in training agents.
“When you play a video game, you’re essentially transferring your perception to different environments, usually through the first-person perspective of a camera,” Pim de Witte, CEO of Medal and General Intuition, told TechCrunch. He pointed out that gamers who upload clips tend to post very negative and positive examples, which can serve as edge cases that are very useful for training. “This selection bias is exactly the kind of data you actually want to use for your training efforts.”
The data moat reportedly caught the attention of OpenAI, which sought to acquire Medal for $500 million late last year, according to The Information. (Neither OpenAI nor General Intuition will comment on this report.)
This also led to General Intuition raising a massive $133.7 million in seed funding led by Khosla Ventures and General Catalyst with participation from Raine.
The startup hopes to use the funding to grow its team of researchers and engineers focused on training integrated agents that can interact with the world around them, with initial applications in gaming and search-and-rescue drones.
De Witte said the founding team has already made progress, with General Intuition’s models able to understand the environment in which it was not trained and accurately predict behavior within that environment. This can be done purely through visual input. The agent only sees what the human player sees and moves through space according to controller input. The company says this approach can naturally transfer to physical systems such as robotic arms, drones, and self-driving cars that are often controlled by humans using video game controllers.
tech crunch event
san francisco
|
October 27-29, 2025
General Intuition’s next milestones are twofold. generating new simulated worlds to train other agents, and autonomously navigating completely unfamiliar physical environments.
This technological approach shapes the company’s technology commercialization plans and differentiates it from competitors that are building global models.
General Intuition also builds world models to train agents, but those models are not products. Unlike other world model makers like DeepMind and World Labs, which sell their world models Genie and Marble for agent training and content creation, respectively, General Intuition focuses on other use cases to avoid copyright issues.
“Our goal is not to create a model that competes with game developers,” De Witte said.
Instead, the startup’s gaming applications center on creating bots and non-player characters that can outperform traditional “deterministic bots,” or pre-programmed characters that produce the same output every time.
“[Bots]can handle any level of difficulty,” Moritz Baier-Lentz, founding member of General Intuition and partner at Lightspeed Ventures, told TechCrunch. “It’s not mandatory to create a Godbot that wins for everyone, but if you scale incrementally and meet fluidity for every player situation so that the win rate is always around 50%, you will maximize engagement and retention.”
De Witte also has a background in humanitarian work, which explains the company’s focus on motorizing search and rescue drones. Drones sometimes have to navigate unfamiliar environments and extract information without GPS.
Ultimately, de Witte and Baier-Lentz believe that General Intuition’s core capability, spatiotemporal reasoning, is a key part of the race toward artificial general intelligence (AGI). While leading AI labs are focused on building ever-more powerful large-scale language models, General Intuition believes true AGI requires what LLMs fundamentally lack.
“As humans, we create texts that explain what’s going on in the world, but a lot of information gets lost in the process,” de Witte says. “General intuition about spatiotemporal reasoning is lost.”
