Retrieval-augmented generation and fine-tuning are two of the most common ways to adapt a large language model to a specialized domain. The source article breaks down the mechanical difference: RAG leaves the model's weights untouched and instead feeds it relevant documents or database records at inference time, while fine-tuning changes the weights themselves through additional training on domain-specific data.
The article frames the choice in terms of what kind of adaptation is needed. RAG is the right tool when the domain knowledge is specific, current, or subject to change, because updating a retrieval index is cheaper and faster than retraining a model. Fine-tuning is better suited when the goal is to alter how the model behaves — its tone, output format, or reasoning style — rather than to inject new facts.
The two techniques are not mutually exclusive. The source suggests that the decision comes down to practical trade-offs: how often the underlying knowledge changes, whether the team can afford retraining, and whether the application needs the traceability that retrieval provides. In many real-world systems, fine-tuning and RAG are used together, with each handling the part it is best suited for.