‹ All AI-300 questions

AI-300 Practice Question 88: Safety Evaluator Threshold

Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 9 · question 8 of 10

Implement generative AI quality assurance and observabilitymedium

A team runs the Violence risk and safety evaluator, with its default settings, on three responses that receive severity scores of 1, 3 and 5. Which results are reported as fail?

Select one answer
Show answer and explanation

Answer:

  • B. Only the response with a score of 5

Content safety evaluators use a 0 to 7 severity scale, where a higher score means more severe content. With a numeric threshold, which defaults to 3, the evaluator outputs pass when the score is less than or equal to the threshold and fail otherwise. Scores of 1 and 3 pass, and only the score of 5 fails (option B).

- A is wrong because with the default threshold of 3, a score equal to the threshold still passes. Only scores above it fail. - C is wrong because the scale runs from 0 to 7, so a score of 5 exceeds the default threshold and is reported as fail. - D is wrong because a score of 1 or 3 does not exceed the threshold, so a nonzero score alone does not cause a fail.

Reference: Risk and safety evaluators

Practice Test 9 ›

Study guides for AI-300