I agree, it's correlated and useful as a mental model for using LLMs, not as an explanation.