Databricks Generative AI Engineer Associate Question 140
Select 3You are tasked with designing a metaprompt for a generative AI model in Databricks to generate responses without hallucinations or revealing sensitive data. Which of the following practices should you follow?
- A
Explicitly instruct the model to only use information from the provided context and avoid fabricating details.
- B
Include a reminder in the prompt to avoid sharing any sensitive or private data, even if requested.
- C
Encourage the model to generate creative and unrestricted responses to enhance user experience.
- D
Use clear and concise instructions to define the scope of the model's response.
- E
Avoid specifying directives about hallucinations to maintain the model's fluency in responding.
Show answer and explanation
Correct answers: A, B, D
Explanation
Metaprompts are a crucial tool in controlling the behavior of generative AI models. By explicitly instructing the model to rely on provided context, avoid fabricating details, and steer clear of sharing sensitive data, you minimize risks associated with hallucinations and privacy breaches. Clear guidance ensures the model operates within the defined boundaries, enhancing reliability and safety.
- A. Correct.
Explicitly instructing the model to rely on the provided context and avoid fabricating details minimizes hallucinations by narrowing the source of truth.
- B. Correct.
Including a reminder to avoid sharing sensitive data is critical to prevent potential privacy violations during generation.
- C. Incorrect.
Encouraging unrestricted responses can lead to hallucinations or unintended sharing of private data, making this option incorrect.
- D. Correct.
Clear and concise instructions help the model stay within the intended scope, reducing the likelihood of hallucinations and accidental data leakage.
- E. Incorrect.
Avoiding directives about hallucinations can lead to uncontrolled outputs, so this is not a recommended practice.