Fine-tuning
Continuing training on a smaller, curated set of examples to specialise a model or teach it a format.
Also written: finetuning, fine-tune, SFT, supervised fine-tuning.
After pretraining (The first and most expensive stage: learning to predict the next token across a very large body of text.), a model is trained further on a much smaller collection of example conversations showing the behaviour you want. This is where a text completer becomes something that answers questions.
As a thing you might do yourself, fine-tuning is usually the wrong first answer. It is good at teaching a format or a voice, poor at teaching facts, and it is almost always worth trying a better prompt and retrieval first — both are faster, cheaper and reversible.