source: techcrunch ai: why this ceo thinks video games make better training data than the internet
level: business
large language models like chatgpt and claude handle text well but struggle with physical reasoning. they lack an understanding of how objects move through space and time. this shortfall limits progress toward artificial general intelligence. general intuition, a startup backed by jeff bezos and valued at $2.3 billion, believes video game data can bridge that gap. the company recently raised $320 million from investors including coatue, eric schmidt, and researchers from mit and google deepmind.
ceo pim de witte explained on techcrunch's equity podcast that world models trained on gaming data could advance physical ai. the company spun out of gaming platform medal tv. de witte discussed how game environments provide rich, dynamic data about physics and interactions. this data trains models to predict and reason about real-world movement. the approach aims to create ai that generalizes beyond text-based tasks.
the conversation also touched on ethical concerns. de witte acknowledged that models capable of physical reasoning could be used in defense applications. he stressed the need for clear red lines. the startup's focus remains on building foundational world models. these models could eventually power robotics and other embodied ai systems. the funding round signals strong investor confidence in gaming data as a path to more capable ai.
why it matters: using video game data could improve ai's ability to handle real-world physical tasks, from robotics to autonomous systems, by teaching models how objects move and interact.
source: techcrunch ai: why this ceo thinks video games make better training data than the internet