Why this CEO thinks video games make better training data than the internet
TL;DR
General Intuition CEO Pim de Witte argues on TechCrunch's Equity podcast that web and text data alone will not get AI to AGI, because LLMs still struggle with motion, space and time. His bet: video games provide stronger training data for world models, because games expose objects, actions, rules and consequences inside simulated environments. General Intuition spun out of gaming platform Medal TV, is based in New York, and TechCrunch says it is valued at $2.3 billion after closing a $320 million round.
Nauti's Take
The thesis lands because it ties the AGI hype to a concrete technical gap: models need better learning of world states, causality and timing. Still, this is an investor story before it is proof.
A gaming background plus famous backers does not magically create physical intelligence. The useful test is simple: can these models make measurably better decisions in robotics, planning or agents than systems trained mostly on more web data?
Briefingshow
The argument points at a real weakness in current models: chatbots can continue text convincingly, but they only understand the physical world indirectly. If game data creates better world models, the AI race shifts further toward simulation, robotics and autonomous systems. For users, the next leap may come less from larger text corpora and more from data where actions have consequences.
Sources