Retrieval augmented generation: Which option best fits this requirement?

Grounding answers in frequently changing documents without retraining is the purpose of retrieval augmented generation.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Fine-tuning the model on the document corpus so the updated knowledge becomes embedded directly in the model weights.

   Almost right as a listed technique, but fine-tuning bakes knowledge into weights and requires re-tuning as documents change.
2. Moving inference to edge devices so responses are generated locally and closer to where employees ask their day-to-day questions.

   Plausible but wrong: edge deployment addresses location and latency, not grounding answers in changing documents.
3. Adopting an agentic architecture so the system can chain tools and take multi-step actions on the employee's behalf.

   Plausible but wrong: agentic design enables action-taking, not the retrieval-based grounding the requirement describes.
4. Retrieval augmented generation, which grounds responses in current documents retrieved at query time without retraining. ✓

   Correct because IV.A.3 lists retrieval augmented generation, which injects up-to-date documents at inference without changing model weights.

How to remember it

Grounding answers in frequently changing documents without retraining is the purpose of retrieval augmented generation.

More Understanding How to Govern AI Deployment and Use questions

• An open-source model: Which model-type characteristic best satisfies this governance constraint? →
• Perform or review an impact assessment on the selected: Which activity directly meets this deployment-stage →
• The data-use clause governing whether the vendor may: Which term in the vendor agreement most directly →
• All 424 Understanding How to Govern AI Deployment and Use questions →