Use stakeholder language that redirects a changing-knowledge request from fine-tuning to retrieval.
Pick the reply that redirects the catalog request to the right mechanism. The request sounds urgent. Can we fine-tune on the new catalog before launch? fromTop Your reply decides the architecture. fromBottom Let's retrieve the catalog from the source of truth first; if evals still show a stable response behavior issue, then we can consider a fine-tune. Best. It preserves the launch goal, explains why current facts belong in retrieval, and leaves fine-tuning for repeated behavior after evidence. Yes, training on the catalog will make the model remember the products. This creates stale model memory. Fine-tuning does not keep a…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in