General Intuition CEO: Video game data superior for AI training
Startup General Intuition proposes that video game data is more effective for training artificial intelligence than internet data. Their goal is to develop artificial general intelligence (AGI) capable of understanding the physical world.
While large language models like ChatGPT excel at text processing, they struggle with understanding how objects move through space and time. Startup General Intuition believes video game data can fill this gap to achieve artificial general intelligence (AGI).
The company's CEO argues that the dynamic and interactive nature of gaming data, which includes information on object interactions and environmental physics, is crucial for developing AI that can generalize knowledge. True AGI requires an understanding of causality and the physical world, something current text-based models lack.
General Intuition plans to collect data from various gaming platforms and develop algorithms capable of learning from these simulated environments. This approach diverges from traditional AI training methods that rely heavily on vast internet datasets, often at the expense of spatial and temporal reasoning.
The startup aims to create AI systems that can better navigate and interact with physical environments, a capability vital for applications in robotics and autonomous systems. Utilizing gaming data could significantly accelerate progress towards these more general AI capabilities.