Large language model LLM

A large language model is trained to continue text. Answering questions, writing code and following instructions all come out of that same continuation, steered by training on examples of those tasks.

Most of the odd behaviour you will meet follows from this. The model has no store of facts to check itself against, so a plausible continuation and a true one cost it the same effort. Between calls it remembers only what you put back into the context window. And it reacts to the shape of your input, so one question phrased two ways can get two answers.

Where the pattern is clear, the model is very good at completing it. Show it one worked example of the output you want and it will copy the structure closely. If you take a single technique away from prompting, take that one.