General Intuition, a New York-based AI startup, has secured $320 million in funding, elevating its valuation to $2.3 billion. This investment underscores the company’s innovative approach to training AI agents using video game environments to develop real-world applications.
Founded by Pim de Witte, General Intuition emerged from Medal, a platform where gamers share video clips. The vast repository of gameplay data from Medal serves as the foundation for training AI models in spatial-temporal reasoning—essentially, understanding movement through space and time.
Unlike traditional methods that rely solely on video footage, General Intuition leverages action labels embedded in these clips. These labels detail the exact button presses and timings of players, providing a richer dataset for training AI agents. This approach enables the development of models capable of interpreting on-screen information and responding appropriately, a feat that large language models (LLMs) struggle to achieve.
During a demonstration at their R&D facility, the company showcased an AI agent playing a game continuously for 100 hours. Remarkably, the same AI model powered a quadrupedal robot navigating the office environment. With just eight minutes of real-world data collected from street environments, the robot adapted to its surroundings, highlighting the model’s versatility.
General Intuition’s strategy centers on creating a unified model that can generalize from virtual gameplay to physical embodiment. This vision has attracted significant investment from prominent backers, including Khosla Ventures, General Catalyst, Jeff Bezos, Eric Schmidt, and researchers from Google DeepMind and MIT.
The recent funding will primarily enhance the company’s computational capacity, facilitating the pre-training of the next model iteration. Additionally, plans are underway to make their API more accessible by the end of the summer.
Vinod Khosla, whose firm led the funding round, emphasized the significance of human action and reaction data in games. He believes this data is crucial for developing AI with human-like intuition, marking a potential quantum leap in AI capabilities.
General Intuition’s innovative use of video game data to train AI agents represents a significant shift in AI development. By focusing on spatial-temporal reasoning through interactive gameplay, the company aims to bridge the gap between virtual simulations and real-world applications. This approach not only accelerates AI training but also opens new avenues for deploying AI in diverse environments, from robotics to autonomous systems.
