Databricks Generative AI Engineer Associate Question 111
Select 3You are evaluating the responses generated by a fine-tuned language model integrated into a customer service chatbot. During testing, you identify that some responses are vague, overly verbose, or potentially harmful (e.g., providing incorrect advice). What are the common issues you should address to ensure high-quality and safe responses?
- A
Ensure the responses are factually correct and grounded in reliable data.
- B
Verify that the responses are concise and aligned with user intent.
- C
Check that responses never include any form of bias or offensive language.
- D
Allow the model to generate creative responses without any review to preserve its originality.
- E
Ensure that the responses are consistent across repeated queries with slight variations in phrasing.
Show answer and explanation
Correct answers: A, B, C
Explanation
To qualitatively assess responses, it is crucial to focus on factual correctness, alignment with user intent, and ethical considerations such as avoiding bias or offensive language. These factors ensure that the AI system generates high-quality and safe outputs, which are essential when deploying the model in real-world applications.
- A. Correct.
Ensuring factual correctness and grounding in reliable data is critical to avoid misinformation and ensure safe and useful responses.
- B. Correct.
Concise and user-aligned responses improve quality and usability, ensuring that the generated content meets the user’s needs without unnecessary verbosity.
- C. Correct.
Checking for bias or offensive language is essential for safety and ethical considerations, as the model must not propagate harmful or problematic content.
- D. Incorrect.
Allowing the model to generate creative responses without review risks introducing errors, unsafe content, or responses misaligned with the intended use case.
- E. Incorrect.
Consistency can be important, but it is not as critical as safety and quality. Some variability in phrasing is acceptable as long as the response remains accurate and aligned with the query.