Databricks Generative AI Engineer Associate Question 141
Select 3You are designing a metaprompt for a generative AI model that summarizes user-uploaded documents without hallucinating or exposing any private data from the documents. Which of the following practices should you follow while writing the metaprompt?
- A
Include explicit instructions to avoid generating content that is not present in the document.
- B
Encourage the AI to use external knowledge freely for context, even if the information is not in the document.
- C
Specify that the AI must only summarize based on the given document and not infer or assume missing details.
- D
Avoid mentioning the importance of privacy or data security to keep the prompt concise.
- E
Add a reminder in the metaprompt to flag any ambiguous or incomplete user inputs for clarification.
Show answer and explanation
Correct answers: A, C, E
Explanation
Effective metaprompts to minimize hallucinations and prevent leaking private data must provide clear and explicit instructions to the AI. This includes grounding the output in the provided input, avoiding unwarranted assumptions, emphasizing the importance of privacy, and addressing ambiguous inputs. These practices ensure the AI generates accurate and secure outputs aligned with user expectations.
- A. Correct.
Explicitly instructing the AI to avoid generating content that is not present in the document helps prevent hallucinations by ensuring the output is grounded in the provided input.
- B. Incorrect.
Encouraging the AI to use external knowledge freely increases the risk of hallucinations and introducing irrelevant or inaccurate information, which is not a best practice.
- C. Correct.
Specifying that the AI must summarize strictly based on the provided document ensures that the model does not make unwarranted inferences or assumptions, reducing the likelihood of hallucinations.
- D. Incorrect.
Neglecting to mention privacy or data security can lead to unintentional leaks of private data, which is a critical concern when handling sensitive documents.
- E. Correct.
Including a reminder to flag ambiguous or incomplete inputs ensures that the AI seeks clarification where necessary, preventing misinterpretation of the user's intent and improving the quality of the output.