Prompt, Retrieve, or Tune?

Different gaps need different interventions

Prompting changes instructions for one request. Retrieval-augmented generation (RAG) supplies current external facts at inference time. Fine-tuning continues training a pre-trained model on task examples, changing weights so behavior or style becomes more natural. These mechanisms solve different problems.

Use RAG when knowledge changes, citations matter, or access control must select per-user information. Fine-tuning is better suited to stable behavior, specialized formats, terminology, or task patterns demonstrated across many examples. It is a poor way to keep rapidly changing facts current and cannot guarantee exact factual recall.

need current policy facts + citations -> RAG
need consistent domain classification style -> consider fine-tuning
need one clearer instruction -> improve prompt first
need hard permission enforcement -> deterministic code, never tuning

Analogy: Prompting is giving today's directions, RAG is opening a current reference book, and fine-tuning is additional professional training. None replaces a locked door when the requirement is access control.

Warning: This platform does not run fine-tuning. It has no deep-learning framework or accelerator; the module teaches design so learners understand production choices without pretending CPU-only labs can perform them.