Fine-tuning further trains a model on task-specific examples.

It can improve style, format, or task behavior, but it is not always the best way to add factual knowledge. RAG may be better for changing knowledge.