Mastery: connect the pieces

1. Idea — Fine-tuning changes behavioral habits in weights; RAG changes evidence available at request time. Prefer RAG for living facts; prefer tuning for stable style/fo…
2. Mechanism — RAG retrieves and inserts passages; fine-tuning runs gradient updates on task examples. Hybrid systems retrieve facts and use a tuned model for tone or structur…
3. Controls — retrieval freshness SLA, tune data mix, LoRA rank, eval for style vs fact, abstain policy
4. Tradeoff — Tuning can lock tone and is slow/expensive to refresh for facts. RAG cites sources and fails when retrieval misses. Doing both raises complexity.…
5. Failures — Facts in weights; RAG without style control
6. Metrics — citation support, freshness lag, style rubric, factual error rate, $/query
7. Ship rule — RAG for authorities with hourly index; small style tune optional; never rely on weights for amendment text.

Neighboring layers (retrieval, serving, policy, human review) still own what fine-tuning vs RAG cannot on the legal research assistant. Do not ask this chapter’s dial to replace permissions or source truth.

Mastery is transfer: take fine-tuning vs RAG into a second scenario that is not the legal research assistant and rebuild the checklist without copying sentences. If you can only recite this chapter’s examples for fine-tuning vs RAG, you have memorized a story, not a model.

Also pin one numeric memory from this fine-tuning vs RAG chapter: If regulations change weekly, a 2-week fine-tune cadence cannot beat an index refreshed hourly on citation freshness. That number is not decoration; it is a template for how claims about fine-tuning vs RAG on the legal research assistant should look in design docs. Scoped specifically to fine-tuning vs RAG / legal research assistant / mastery-connect.

People confuse fine-tuning vs RAG with neighboring buzzwords when debugging the legal research assistant. Before changing prompts, ask whether the broken stage was evidence gathering, the fine-tuning vs RAG judgment itself, validation, or the product action. Fixing the wrong stage creates folklore (“we tried fine-tuning vs RAG and it failed”) that blocks the next team on the legal research assistant. Scoped specifically to fine-tuning vs RAG / legal research assistant / mastery-connect.

Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to fine tuning vs rag rather than generic AI advice.