Travis Frisinger

This talk argues that Large Language Models (LLMs) excel due to coherence, not intelligence. While LLMs can produce outputs that feel remarkably insightful and intelligent, they lack genuine understanding, intent, or desire. The speaker proposes that coherence is a system property, not a cognitive one, and explains how LLMs construct meaning on demand through pattern alignment within a high-dimensional latent space, rather than retrieving stored knowledge.