Game Data, Not LLMs: How Startup General Intuition Sees the Path to AGI

General Intuition, a startup backed by investment from Jeff Bezos, says it has a fundamentally new approach to achieving artificial general intelligence. In the company's view, today's large language models, from ChatGPT to Claude, are exhausting what text data can give them, and cannot fully develop general intelligence without an understanding of the physical reality of space and time.

Why LLMs fall short of AGI

Text models show impressive ability in language processing: creative writing, analysis, reasoning based on knowledge. But they remain fundamentally blind to one critical area: the physics of space and time. These models have no built-in understanding of how objects move, how they interact under forces, how they react to proximity to one another, or how causes produce effects in the physical world.

An AI system needs this understanding to generalize its knowledge beyond the realm of text and apply logic in a real, physical context. Text data, even the largest collections of descriptive text, cannot replace the experience of observing objects move, fall and collide, and how they respond to forces.

Game data as a key to physical reasoning

This is where video games come in. In game worlds, physical laws are all modeled and captured in data: how a character jumps, how an object falls, how objects interact in a collision, how sound propagates, how light reflects off surfaces. These high-frequency, multidimensional datasets contain billions of examples of spatiotemporal behavior, exactly the information that LLMs cannot extract from text, or even from unannotated video.

General Intuition sees game data as an underused resource for training AI systems to integrate knowledge about physics, causality and spatial relationships. If a model trains on text and game data together, with explicit semantics of physical interactions, it may learn not only to describe the world in words but also to predict and model it, reason about its physical structure and draw causal inferences. The startup believes this is a critical step toward more general and adaptive intelligence.

Bezos's investment and what it means for the industry

Jeff Bezos, known for backing ambitious, fundamental research and risky bets, has invested in General Intuition. This signals that the company's hypothesis is drawing attention from experienced investors and influential figures in the AI community. However, the details of the funding round itself, including the amount invested, the company's stage and the other investors, are not known from public sources.

What this means for the future of AGI

If General Intuition's concept is right, building AGI requires more than scaling up text models: it requires a fundamentally different architectural approach, one that synthesizes and integrates knowledge from many data sources. That shifts the industry's focus from the question "how do we make LLMs bigger?" to questions such as: which other types of data are critical for general intelligence? How can different modalities be integrated into a single system more effectively? What role does active learning through simulation and game environments play? The debate over the fundamental ingredients of AGI is becoming more pressing and more conceptually diverse.