Why this CEO thinks video games make better training data than the internet
What changed
General Intuition CEO argues that training AI with video game data can produce stronger artificial general intelligence than the internet’s text alone. Large language models like ChatGPT excel at processing language but struggle to understand physical movement through space and time, a key skill for flexible intelligence. Video games, with their rich, dynamic environments and interactive physics, offer structured training data that better reflect how the real world works in three dimensions over time.
Why builders should care
For AI developers and founders pushing toward AGI, this challenges the current strategy of relying almost exclusively on large text corpora sourced from the internet. Language-only training limits model capabilities in reasoning about events, causality, and spatial relationships. Using video game environments as training datasets could accelerate advancing models’ world understanding and decision-making skills, improving practical applications in robotics, simulations, and autonomous systems requiring real-time physical interaction.
The practical takeaway
Incorporating video game data sets changes the training focus from static text to active embodied experience. Builders working in domains requiring spatial reasoning or physical control can expect more robust AI agents that generalize better outside narrow language tasks. This approach also potentially lowers reliance on costly real-world data gathering since games provide rich scenarios at scale with built-in physics and rule sets, speeding up experimentation and iteration.
What to watch next
Observe if more AI startups and labs start building pipelines around gaming engines and simulated environments instead of expanding just text datasets. Watch how models trained on such data perform when integrated with real-world robots or virtual agents. Track investments and partnerships between gaming companies and AI developers aiming to monetize high-fidelity game data for model training. This shift could redraw priorities in AI research and infrastructure over the next few years.
AI Quick Briefs Editorial Desk
