Fine-tuning updates a model using additional examples after its initial training. It can improve consistency on a specialized task, teach a response format or adapt behavior to a domain when the training examples represent the work the model must do. It is not automatically the best way to add changing facts.
Fine-tuning fits cases such as always answering in a fixed format, matching a brand's writing style or sorting tickets into a company-specific set of categories, where you have many good examples. If the goal is to answer questions from a changing document set, retrieval-augmented generation is usually simpler. Try a well-written prompt first, since prompt engineering often gets most of the way there.
For frequently updated knowledge, retrieval may be easier to maintain. Fine-tuning needs careful data preparation, evaluation and versioning to avoid teaching errors or degrading other capabilities. Related: Low-Rank Adaptation (LoRA), Knowledge Distillation.