Why this CEO thinks video games make better training data than the internet

The race to achieve artificial general intelligence has hit a familiar wall. Large language models like ChatGPT and Claude have mastered the art of text generation. They can write code, draft emails, and summarize articles with impressive fluency. Yet they struggle with the fundamental physics of the real world. They do not inherently understand how objects move through space or interact with gravity. This limitation suggests that text alone is an incomplete training set for true intelligence.

According to recent reporting, a new company called General Intuition is betting that video games hold the key to solving this problem. The core idea is that game engines provide a perfectly structured simulation of physical reality. Unlike the chaotic and often contradictory data found on the open internet, game worlds operate on consistent, hard-coded rules. These rules mimic the laws of physics in a way that is clean and predictable. This consistency makes them ideal for teaching AI systems how the physical world actually works.

Current models are essentially statistical parrots. They predict the next word based on patterns in text data. They lack a grounded understanding of cause and effect in a spatial context. If you ask a standard LLM to describe what happens when a ball is dropped, it can recite a textbook definition. It cannot, however, intuitively grasp the trajectory or the impact without explicit programming. This disconnect between language and physical intuition is a major barrier to generalization.

Video games offer a solution by providing millions of hours of interaction data. In these environments, agents must learn to navigate, manipulate objects, and solve puzzles. The feedback loop is immediate and precise. If an AI agent fails to account for momentum, it crashes. This creates a rich dataset of physical reasoning that text data simply cannot provide. The data is not just about what things are called, but how they behave.

This shift represents a move toward embodied AI. Researchers are increasingly realizing that intelligence is not just about processing information. It is about interacting with an environment. By training models on game data, we are giving them a virtual body to learn from. This approach bridges the gap between abstract language and concrete physical action. It allows models to develop an intuitive sense of space and time.

The implications for the broader tech industry are significant. If General Intuition and similar ventures succeed, we could see a new generation of AI that understands the physical world. This would enable robots to perform complex tasks in unstructured environments. It would also lead to more robust AI systems that are less prone to hallucinations. The focus shifts from pure language processing to physical reasoning and simulation.

What this means for you is that the next wave of AI tools will likely be more grounded in reality. As these models integrate physical intuition, they will become better at planning and execution. You should start exploring how simulation data can enhance your own workflows. Try using an AI assistant to generate code for a simple physics simulation in a game engine. Ask it to create a scenario where an object must navigate an obstacle course. This exercise will help you understand the intersection of code, physics, and AI reasoning. It is a practical step toward mastering the next generation of intelligent systems.