World Models: Computing the Uncomputable

Co-written essay with Pim DeWitte, CEO of General Intuition ($133.7M seed). A comprehensive primer on world models -- AI systems that learn to predict and simulate physical world dynamics from observation, distinct from LLMs that process text.

Core Argument

World models represent a fundamentally different approach to AI than LLMs: they learn physics and causality from video/sensor data rather than language patterns. This makes them critical for robotics, autonomous systems, and any domain where understanding physical reality matters. McCormick argues world models may drive superhuman capabilities in domains where LLMs plateau, complementing rather than replacing language models. The essay covers history (from early simulation to modern neural approaches), competing architectures, and the massive compute implications.

RDCO Relevance

• Compute infrastructure: world model training demands reshape data center and pipeline architecture
• Agent design: world models as a component in autonomous systems that need to predict consequences of actions
• General Intuition's approach: learning from action-labeled gaming clips as training data pipeline pattern

Note: Co-authored with portfolio-adjacent CEO (General Intuition). Likely sponsored/promotional content.