← Browse

World's Fair 2025

The Coherence Trap: Why LLMs Feel Smart (But Aren’t Thinking) - Travis Frisinger

Travis Frisinger

Overview

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.

Who should watch

Key takeaways

Notable quotes

*LLMs are not intelligent. They're coherent. And that's their superpower.*
*The magic happens in the collaborative dance. So stop chasing intelligence and start designing for structured residents.*

Watch on YouTube →

Unofficial community note. Prefer the recording for nuance.