General Intuition lands $134M seed to teach agents spatial reasoning using video game clips

Medal, a platform dedicated to the sharing of video game clips, has launched a pioneering AI research lab named General Intuition. This startup aims to harness an extensive database of gaming videos to develop AI agents capable of understanding spatial-temporal reasoning, which involves comprehending how objects and entities move through both space and time. General Intuition's dataset is impressive, comprising 2 billion videos annually from 10 million active users across a multitude of games. CEO Pim de Witte stated that the unique nature of gaming clips—often highlighting extreme player experiences—provides valuable training data. Gamers tend to upload clips showcasing both the best and worst moments, creating a dataset rich with useful edge cases. This unique data advantage reportedly caught the eye of OpenAI, which attempted to acquire Medal for $500 million last year, although neither OpenAI nor General Intuition has confirmed this information. The startup's ability to leverage this data has led to an impressive seed funding round of $133.7 million, spearheaded by Khosla Ventures and General Catalyst, with additional support from Raine. The funding will be directed toward expanding General Intuition's team of researchers and engineers, who are focused on developing a general agent capable of interacting intelligently with its environment. Initial applications are projected in sectors such as gaming and search-and-rescue operations, with De Witte noting that their model can engage with environments it hasn't been specifically trained on, making accurate predictions based solely on visual input. The startup's innovative approach is designed to bridge the gap between virtual and physical systems. As De Witte explained, agents are trained to navigate environments as a human player would, utilizing video game controllers. This methodology could significantly benefit robotics, drones, and autonomous vehicles. General Intuition's next objectives include creating simulated worlds for further agent training and enabling agents to navigate completely new physical settings autonomously. Unlike competitors that focus on creating world models for agent training, General Intuition is intentionally avoiding direct competition with game developers and copyright issues. The company's vision extends beyond gaming; its applications could enhance search and rescue drones, which often operate without GPS in unpredictable environments. De Witte and his team believe that mastering spatial-temporal reasoning is essential for advancing towards artificial general intelligence (AGI). While many AI labs concentrate on developing powerful language models, General Intuition asserts that true AGI demands a deeper understanding of the physical world, which conventional models often overlook.