LLM Cognition Is Sequential, Not Parallel

An LLM describes its own cognitive architecture — token by token, one path at a time, no true parallelism or lateral leaps, just sequential prediction with attention context.