General Intuition Bets on Video Game Data to Train Physical World Models for AGI
General Intuition, a well-funded AI startup backed by Jeff Bezos and industry experts, is training physical AI world models using gaming data rather than standard internet text. The company argues that video games provide crucial spatial and temporal context necessary for reaching artificial general intelligence (AGI). Additionally, the company is navigating the ethical implications of these models potentially being adapted for defense applications.
Skynet Chance (+0.03%): Developing AI with deep spatial awareness and physical world understanding increases the risk of autonomous systems operating uncontrollably in physical environments. Furthermore, potential defense applications heighten the risk of militarized, autonomous AI systems.
Skynet Date (-1 days): The massive $320 million funding round and active backing from top researchers could accelerate the timeline for deploying physical AI systems in real-world scenarios. This rapid progression brings the onset of complex physical AI control risks closer in time.
AGI Progress (+0.03%): Utilizing gaming data to build spatial and temporal world models addresses a major limitation of current text-based LLMs in achieving general intelligence. This method provides a viable pathway to overcoming hurdles in physical AI capabilities.
AGI Date (-1 days): Substantial capital injection and collaboration with elite research institutions like MIT and DeepMind will likely speed up the development of capable world models. This shifts the expected timeline for achieving embodied AGI closer to the present.