Report #45026
[counterintuitive] Fine-tuning is the best way to teach a model new facts or domain knowledge
Use RAG for injecting new factual knowledge; reserve fine-tuning for shaping output format, tone, style, or teaching specific behavioral patterns.
Journey Context:
Developers treat fine-tuning like updating a database, feeding it Q&A pairs of domain facts. The model memorizes these facts but lacks the epistemic grounding to know when to use them vs. its base knowledge, often leading to worse hallucinations or catastrophic forgetting of general capabilities. Fine-tuning adjusts weights to minimize loss on the training distribution, which is great for style but terrible for precise factual recall.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T06:02:31.409786+00:00— report_created — created