General Intuition CEO Bets on Video Games for AGI Training Data

General Intuition CEO believes that video game data offers a more effective training ground for achieving artificial general intelligence (AGI) than traditional internet data. The CEO argues that current large language models, while proficient with text, lack the crucial ability to understand how objects and entities move through space and time. This spatial-temporal reasoning is considered essential for developing generalized intelligence.

The company, General Intuition, is positioning itself to leverage this insight by utilizing gaming data. The rationale is that the dynamic and interactive nature of video games provides a rich source of information about cause and effect, physics, and object manipulation. This contrasts with the static and often unstructured nature of internet text, which is less adept at teaching these fundamental aspects of real-world interaction.

By focusing on the complex environments and simulated physics within video games, General Intuition aims to train AI models that can better grasp the physical world. This approach could lead to AI systems that are more capable of performing tasks requiring a deep understanding of motion, interaction, and consequence, moving beyond mere linguistic comprehension. The company's strategy suggests a shift in how AI training data is perceived, prioritizing simulated environments that mimic real-world dynamics over vast repositories of text.

This perspective challenges the prevailing reliance on internet-scale text datasets for AI development. The CEO's assertion highlights a potential bottleneck in current AI progress and proposes a novel solution rooted in the immersive and interactive qualities of video games. If successful, this method could significantly accelerate the development of more robust and generally intelligent AI systems.