What LLMs Are

base model vs chat model

A base model is the raw result of training on mountains of text: it is brilliant at continuing text but has no idea it is supposed to be an assistant. Ask it a question and it might continue with more questions, because that is what often follows a question in its training data. A chat model is that same base model after extra training to follow instructions and answer politely.

The chat version is built in stages on top of the base. First, supervised fine-tuning shows it many examples of instructions paired with good responses. Then preference training rewards the answers humans prefer. The underlying knowledge barely changes; what changes is behavior — the chat model knows a question is a request to be answered, not a pattern to be extended.

If you are using an app like ChatGPT or Claude, you are talking to a chat model; base models are mostly for developers who want a blank, steerable starting point.