Databricks Generative AI Engineer Associate Question 142
Select 3You are designing a Generative AI application using Databricks to summarize customer feedback. To minimize hallucinations and avoid leaking private customer data, which of the following metaprompt strategies should you implement?
- A
Explicitly instruct the model to avoid making up information or fabricating details.
- B
Include a disclaimer in the prompt to ensure the model repeats any private data it encounters.
- C
Guide the model to only generate responses based on the provided context or input data.
- D
Use clear instructions to prevent the model from guessing answers when uncertain.
- E
Encourage the model to use external APIs to infer missing information not in the input.
Show answer and explanation
Correct answers: A, C, D
Explanation
To minimize hallucinations and protect private data, metaprompts should clearly instruct the model to avoid fabricating information, remain within the context of provided input, and refrain from guessing when uncertain. These strategies help ensure reliable and secure outputs from the model.
- A. Correct.
Explicitly instructing the model to avoid hallucinations helps to set boundaries for its output and reduces the likelihood of fabricated responses.
- B. Incorrect.
Including a disclaimer to repeat private data increases the risk of privacy violations and should be avoided when working with sensitive information.
- C. Correct.
Guiding the model to only generate responses based on the provided context ensures that it does not produce information outside of the input scope, helping to prevent hallucinations.
- D. Correct.
Preventing the model from guessing when uncertain ensures its output remains grounded in the available data and reduces the likelihood of introducing inaccuracies.
- E. Incorrect.
Encouraging the use of external APIs to infer missing information can lead to hallucinations and potential privacy breaches, as the model may generate or access unintended outputs.