Databricks Generative AI Engineer Associate Question 109
Select 3You are training a generative AI model for a customer support chatbot. After testing the model, you notice the following issues in its responses: 1) It provides overly generic answers, 2) It occasionally gives inaccurate information, and 3) It suggests actions that could lead to unsafe outcomes for the user. What steps should you prioritize to qualitatively assess and improve the model's responses?
- A
Manually review generated responses for accuracy and relevance.
- B
Implement a keyword matching algorithm to filter unsafe responses.
- C
Test the model against a diverse set of user scenarios to identify edge cases.
- D
Rely on the model's confidence scores to determine response quality.
- E
Establish evaluation criteria focusing on safety, factual correctness, and clarity.
Show answer and explanation
Correct answers: A, C, E
Explanation
Qualitatively assessing generative AI responses requires a combination of manual review, scenario testing, and well-defined evaluation criteria. These steps help identify common issues such as generic, inaccurate, or unsafe outputs and provide a roadmap for improving the model's performance. Keyword matching or reliance on confidence scores alone is insufficient for ensuring high-quality and safe responses.
- A. Correct.
Manually reviewing responses for accuracy and relevance is a critical qualitative assessment step to identify issues such as incorrect or generic answers.
- B. Incorrect.
While a keyword matching algorithm may help identify unsafe responses, it is not sufficient for comprehensive qualitative assessment and improvement of generative AI responses.
- C. Correct.
Testing against diverse scenarios helps uncover edge cases and identify gaps in the model's performance, such as unsafe or low-quality outputs.
- D. Incorrect.
Confidence scores alone are not a reliable indicator of response quality, as a model can be confident in generating incorrect or unsafe responses.
- E. Correct.
Establishing evaluation criteria ensures there is a systematic approach to assessing safety, factual correctness, and clarity, which are key quality metrics for generative AI systems.