AIP-C01 · Question #92
A company uses Amazon Bedrock to generate technical content for customers. The company has recently experienced a surge in hallucinated outputs when the company's model generates summaries of long…
The correct answer is B. Use Retrieval Augmented Generation (RAG) with an Amazon Bedrock knowledge base. Apply C. Configure Amazon Bedrock guardrails to block any generated output that matches patterns that. Two solutions are needed. Option B (RAG with Amazon Bedrock knowledge base) directly addresses hallucinations by grounding the model's responses in the actual source documents-retrieved chunks are injected into the prompt as factual context, preventing the model from…
Question
A company uses Amazon Bedrock to generate technical content for customers. The company has recently experienced a surge in hallucinated outputs when the company's model generates summaries of long technical documents. The model outputs include inaccurate or fabricated details. The company's current solution uses a large foundation model (FM) with a basic one-shot prompt that includes the full document in a single input. The company needs a solution that will reduce hallucinations and meet factual accuracy goals. The solution must process more than 1,000 documents each hour and deliver summaries within 3 seconds for each document. Which combination of solutions will meet these requirements? (Select TWO.)
Options
- AImplement zero-shot chain-of-thought (CoT) instructions that require step-by-step reasoning with
- BUse Retrieval Augmented Generation (RAG) with an Amazon Bedrock knowledge base. Apply
- CConfigure Amazon Bedrock guardrails to block any generated output that matches patterns that
- DIncrease the temperature parameter in Amazon Bedrock.
- EPrompt the Amazon Bedrock model to summarize each full document in one pass.
How the community answered
(32 responses)- A16% (5)
- B75% (24)
- D6% (2)
- E3% (1)
Explanation
Two solutions are needed. Option B (RAG with Amazon Bedrock knowledge base) directly addresses hallucinations by grounding the model's responses in the actual source documents-retrieved chunks are injected into the prompt as factual context, preventing the model from fabricating details. Option C (Bedrock guardrails to block outputs matching fabricated patterns) adds a safety net by filtering responses that contain hallucinated or inaccurate content before delivery. Option A (zero-shot CoT) can improve reasoning quality but adds latency that may violate the 3-second requirement and does not guarantee factual grounding. Option D (increasing temperature) is counterproductive-higher temperature increases randomness and makes hallucinations worse, not better. Option E reproduces the current failing approach of summarizing the full document in one pass.
Topics
Community Discussion
No community discussion yet for this question.