LLM mental models: ways to think about language models

A mental model is a simple picture of how something works that helps you predict what it will do. Language models are hard to reason about directly, so each guide in this series offers one picture and shows what follows from it for prompting.

Seen as a probability map, a model picks likely continuations, not true ones. Seen as a simulator, it plays whatever role and world the prompt sets up. Seen as a trickster, it is fluent and sometimes wrong, so you generate and then verify. No single picture is complete; together they explain why clear context, examples, verification and limits work. Each essay starts from a question people ask about LLMs, explains it with one picture, and can be read on its own.