General Intuition is betting that video game data, not internet text, is the key to achieving artificial general intelligence. The startup's CEO argues that large language models like ChatGPT and Claude excel at text but struggle with understanding how things move through space and time. This spatial-temporal understanding, they contend, is essential for intelligence that truly generalizes.
The company's thesis challenges the prevailing approach in AI research, which has focused on ever-larger text datasets. By contrast, General Intuition believes gaming environments provide richer, more dynamic training data that captures physical interactions. This could address a blind spot in current models, which often fail at tasks requiring real-world reasoning.
No specific funding amounts, revenue figures, or user metrics were disclosed in the coverage. The startup's approach remains in the early stages, with no public benchmarks or demos yet available. The company has not released details on game partnerships or data acquisition methods.
The implications are significant: if successful, this method could accelerate progress toward AGI by providing training data that captures cause-and-effect in physical space. Critics may question whether game physics adequately represents real-world complexity, or whether this approach scales efficiently compared to text-based training.
AI researchers have long debated whether simulated environments can bridge the gap to human-level reasoning. General Intuition's bet is that play, not prose, teaches machines to think.