Hallucinations and Trust

1. What is the first layer of defense against hallucinations as described in the lesson?

Human-in-the-loop validation Grounding and citations Thresholds and refusals Structured-output validation

2. Why is setting evidence thresholds important in the context of AI suggestions?

It guarantees all suggestions are validated by AI It prevents high-confidence suggestions from being accepted automatically It ensures only human suggestions are accepted It minimizes the need for human review

3. What role does the human-in-the-loop play in the decision-making process according to the lesson?

They solely validate AI suggestions They introduce subjectivity in the suggestions They make the final determination based on AI suggestions They are optional and not necessary for decision-making

4. Which of the following is a red flag indicating poor management of hallucinations?

Using a golden evaluation set Logging override rates Allowing high-confidence suggestions to auto-finalize Implementing structured-output validation

Continue: The Design Method → Review this lesson Retake quiz