Databricks Certified Generative AI Engineer - Associate
Get started today
Ultimate access to all questions.
Question: 16
You are designing a metaprompt for a generative AI application in Databricks that will handle sensitive customer information, such as phone numbers and addresses. Your primary objective is to minimize hallucinations and prevent the leakage of private data. Which two approaches should you include in your metaprompt to achieve this? (Select two)
Real Exam
Community
LLeetQuiz
Explanation:
Explanation
B. Clearly stating in the metaprompt that sensitive information like phone numbers or addresses should not be generated ensures that the model avoids outputting private data. This explicit instruction helps prevent data leakage by directly telling the model what types of information to exclude from responses.
D. Instructing the model to avoid generating or referencing data not explicitly provided helps minimize hallucinations and ensures the model only uses the provided input data. This approach reduces the risk of the model inventing or fabricating information that wasn't part of the original input.
Why other options are incorrect:
- A: Avoiding limitations would actually increase the risk of data leakage and hallucinations by giving the model too much freedom.
- C: Repeating all input data could actually increase the risk of exposing sensitive information rather than protecting it.
- E: A temperature setting of 1.5 would encourage more creative outputs, which could lead to more hallucinations and unpredictable behavior, contrary to the goal of minimizing hallucinations.
Comments
Loading comments...