1Z0-184-25 Question 131
Single answerYou are designing a Retrieval-Augmented Generation (RAG) solution in Oracle Cloud Infrastructure for an internal Q&A chatbot. Your source documents are kept in Oracle Object Storage and frequently updated with new or revised policies. End users report that the chatbot is sometimes responding with outdated policy information. Which approach is most critical to ensure the chatbot always references the latest policy documents during retrieval?
- A
Regularly rebuild or update the document index when policies change, ensuring the retrieval component accesses the most recent documents.
- B
Reduce the model� generation temperature to zero so it strictly adheres to the data from its initial training.
- C
Increase the LLM� token size so that older document versions are kept for historical reference along with new versions.
- D
Use a higher-capacity compute shape on OCI for the LLM to handle more concurrency and avoid stale responses.
Show answer and explanation
Correct answer: A
Explanation
When using a RAG pattern in OCI, a critical maintenance step is ensuring that your knowledge base or index is kept in sync with the latest documents. Oracle� best practices recommend monitoring for changes in source data�such as in Oracle Object Storage�and re-building or updating search or vector indexes on a schedule or trigger event. This ensures the retrieval process selects the most recent content before handing it off to the LLM for generation, mitigating the risk of outdated responses. See Oracle� documentation on implementing RAG solutions for additional indexing and re-ingestion strategies.
- A. Correct.
Correct. RAG solutions rely on an updated index or embedding store to retrieve the latest content. Whenever the underlying documents are revised or replaced, the index used by the retrieval pipeline must be refreshed so the chatbot always has access to the newest policies.
- B. Incorrect.
Incorrect. Lowering the temperature only reduces the creative variability of the model's responses; it does not inherently refresh the knowledge base or guarantee references to newly updated documents.
- C. Incorrect.
Incorrect. Increasing the token size expands how much text the model can process at once, but without re-indexing, the chatbot may still pull outdated or stale references. Token size alone does not ensure accuracy against changing documents.
- D. Incorrect.
Incorrect. Allocating more compute is useful for scaling throughput and reducing latency, but it does not fix stale data. The retrieval step must be pointed at an updated and valid data source to remain accurate.