General Intuition Raises $134M Seed for AI Spatial Reasoning

Medal Launches General Intuition: An AI Research Lab Focused on Spatial-Temporal Reasoning

Medal, a well-known platform for sharing and uploading video game footage, has established a new AI research laboratory. This new venture, named General Intuition, leverages the extensive collection of gaming videos available on the platform to develop and train foundation models and AI agents.

Understanding Movement in Space and Time

The core focus of General Intuition is on spatial-temporal reasoning – enabling AI to comprehend how objects and entities move within both space and time. This capability is considered crucial for advanced artificial intelligence development.

A Unique Dataset for AI Training

The startup believes its dataset, comprising 2 billion videos annually from 10 million monthly active users across numerous games, offers a significant advantage over datasets sourced from platforms like Twitch or YouTube for agent training.

Pim de Witte, CEO of both Medal and General Intuition, explained to TechCrunch that gamers frequently share clips showcasing particularly positive or negative experiences. This creates a valuable selection bias, providing precisely the type of data most beneficial for training AI models.

Reported Acquisition Attempt and Seed Funding

According to reports, OpenAI previously attempted to acquire Medal for $500 million. While neither OpenAI nor General Intuition commented on this, the interest highlights the value of Medal’s data assets.

General Intuition has secured $133.7 million in seed funding, led by Khosla Ventures and General Catalyst, with additional participation from Raine. This substantial investment will fuel the company’s growth and research efforts.

Applications in Gaming and Beyond

The funds will be allocated to expanding the team of researchers and engineers dedicated to training a versatile agent capable of interacting with the surrounding world. Initial applications are anticipated in the gaming industry, as well as in the development of search-and-rescue drones.

The founding team has already demonstrated promising results. Their model can successfully understand and predict actions within environments it wasn’t specifically trained on, relying solely on visual input mirroring a human player’s perspective.

Transferring Skills to Physical Systems

This approach, the company asserts, can be readily adapted to physical systems such as robotic arms, drones, and autonomous vehicles, which are often controlled by humans using video game controllers.

Future Milestones and Commercialization Strategy

General Intuition’s immediate objectives include generating new simulated worlds for agent training and enabling autonomous navigation in completely unfamiliar physical environments.

The company’s commercialization strategy distinguishes it from competitors focused on building world models. While General Intuition is also developing world models, these are not intended as the primary product.

Avoiding Copyright Concerns

Unlike companies like DeepMind and World Labs, which are selling their world models (Genie and Marble, respectively), General Intuition is prioritizing alternative use cases to mitigate potential copyright issues.

“Our goal is not to produce models that compete with game developers,” de Witte stated.

Enhancing Gaming Experiences with Intelligent Bots

Instead, the startup’s gaming applications will concentrate on creating bots and non-player characters (NPCs) that surpass the capabilities of traditional, preprogrammed bots.

Moritz Baier-Lentz, a founding member of General Intuition and partner at Lightspeed Ventures, explained that the goal is to create bots that can dynamically adjust difficulty levels. Maintaining a player win rate around 50% will maximize engagement and retention.

Humanitarian Applications and the Pursuit of AGI

De Witte’s background in humanitarian work also influences the startup’s focus on powering search-and-rescue drones, which often operate in unfamiliar environments without GPS assistance.

Ultimately, de Witte and Baier-Lentz believe that General Intuition’s core competency – spatial-temporal reasoning – is a vital component in the pursuit of artificial general intelligence (AGI).

The Importance of Intuition Beyond Language

While many AI labs are concentrating on developing increasingly powerful large language models (LLMs), General Intuition contends that true AGI necessitates capabilities that LLMs inherently lack.

“As humans, we create text to describe what’s going on in our world, but in doing so, you lose a lot of information,” de Witte said. “You lose general intuition around spatial-temporal reasoning.”