Fine-tuning

Fine-tuning - additional training of a ready-made model on your data set. The output is a model that by default writes in the desired style or solves a narrow problem - without long instructions every time.

Honestly: a marketer rarely needs fine-tuning. In 90% of cases, the same thing can be solved with a good system prompt, a couple of examples (few-shot) and RAG - faster and without training costs. Fine-tuning is justified when the task is massive, narrow and stable: thousands of generations of the same type in one rigid format.

My priority rule: first prompt, then few-shot, then RAG, and only if all this does not meet the bar - fine-tuning. Most tasks do not reach the last step.