A pilot assistant occasionally produces confident-sounding but false answers. Which evaluation criterion targets this specific symptom?
Select an answer to reveal the explanation.
Short Explanation
Confident and false is the signature of an ungrounded answer. Groundedness checks are built to catch that by tracing each answer back to a reliable source.
Full Explanation
Groundedness evaluation is specifically designed to catch the case where a model states something with confidence that is not actually supported by any reliable source. Checking whether the answer can be traced back to a reliable source rather than being fabricated directly targets the confident-but-false symptom.
The relevance criterion is incorrect because it addresses whether the topic is on point, not whether the content is factually accurate.
The harmful content criterion is incorrect because a false answer is not necessarily offensive or dangerous; the symptom described is inaccuracy, not harm.
The policy compliance criterion is incorrect because it addresses adherence to organisational content rules, not factual basis.
For the exam, match the confident-but-false symptom to groundedness specifically, not to the other evaluation criteria.