RAG vs fine tuning: how to choose for a production LLM feature

When you are building a production LLM feature that needs to be factually accurate, you will eventually face this decision: retrieve the facts at query time, or bake the knowledge into the model throu