Large language models like ChatGPT excel in processing text but fall short in grasping how objects move through space and time — a vital component for true artificial general intelligence. New York-based startup General Intuition, valued at $2.3 billion, believes video game-derived data offers a richer training ground for smarter AI.
- General Intuition raised $320 million to develop AI using video game data.
- Gaming data better captures spatial and temporal dynamics than internet text.
- Investors include Coatue, Eric Schmidt, MIT researchers, and Google DeepMind.
What happened
General Intuition, a startup based in New York and backed by Amazon founder Jeff Bezos, announced a $320 million funding round that values the company at $2.3 billion. Key investors include Coatue, Eric Schmidt, and research teams from MIT and Google DeepMind. The company originated from the gaming platform Medal TV and now focuses on developing advanced AI models.
Unlike conventional AI models trained largely on vast internet text corpora, General Intuition is prioritizing data derived from video games. These environments provide detailed simulations of objects moving through time and space, offering more dynamic and physical context that large language models typically lack.
Why it matters
Current large language models, like ChatGPT or Claude, effectively generate and understand text but struggle with the complexities of physical reality and temporal movement. This limits their ability to produce intelligence that generalizes across diverse tasks and environments, a core feature necessary for achieving artificial general intelligence (AGI).
By training AI on video game data, which inherently involves modeling objects in dynamic, spatially coherent worlds, General Intuition aims to develop 'world models' that better reflect real-world physics and temporal progression. This could accelerate progress toward AI systems capable of more general and practical reasoning beyond language.
What to watch next
General Intuition's progress in refining AI through gaming data could reshape strategies around AGI development, offering an alternative to purely language-based models. How their technology performs in applications requiring physical reasoning or real-world interaction will be a critical measure of success.
Ethical considerations are also on the horizon, as AI trained in these environments may have defense or surveillance applications. Monitoring how General Intuition navigates potential ethical boundaries and real-world deployment will be an important aspect for stakeholders and regulators.