General Intuition CEO Says Video Games Outperform Internet for AI Training

The CEO of General Intuition said video-game gameplay data is a superior source for training artificial general intelligence systems compared to internet text.

This shift in training philosophy suggests that the current reliance on large language models may be insufficient for AI to understand the physical world. By prioritizing interactive environments over static text, developers hope to bridge the gap between digital reasoning and real-world application.

Based in New York, General Intuition is an AI startup backed by Jeff Bezos ^[1]. The company has reached a valuation of $2.3 billion ^[1] after raising $320 million in funding ^[1]. The firm is now targeting millions of hours of gameplay to refine its models ^[1].

The CEO said that video games provide rich spatial-temporal interaction data. This type of data helps AI develop an intuition about how objects move and interact, a capability that text data alone cannot convey ^[1], ^[2].

While the internet provides a massive repository of human knowledge, it is largely composed of descriptions of actions rather than the actions themselves. Video games allow AI agents to experience cause and effect in a simulated environment. This process creates a feedback loop where the AI can test hypotheses about physics and spatial logic in real time ^[2], ^[4].

This approach focuses on the development of agents that can operate in the real world. The goal is to move beyond chatbots and toward systems that possess a functional understanding of three-dimensional space ^[1], ^[3].