RAG versus fine-tune battlecards
Recall concise responses to common stakeholder objections about RAG, fine-tuning, and hybrid designs.
battlecard comparison Objection Why not just fine-tune all the documents? A stakeholder wants the model to 'know' a changing knowledge base. Your line If the content changes or needs citations, it belongs in retrieval first. Fine-tuning can teach answer behavior later, but it should not be our document store. Arguing that fine-tuning is bad in general instead of explaining the source-control problem. It preserves the stakeholder goal while naming the criterion: freshness and provenance are retrieval controls. RAG is enough; why would we ever fine-tune? The system retrieves correct sources but answer style, schema, or rubric behavior keeps failing. RAG…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in