Question 24
Domain 1 — AI Governance and Risk ManagementAn IS auditor identifies that an AI model occasionally invents nonexistent medical test results. Which of the following recommendations would BEST mitigate this risk?
Correct answer: D
Explanation
Top-p sampling limits generation to the smallest set of tokens whose cumulative probability meets a threshold, so lowering it makes the model choose from a narrower, more probable set. That reduces random or low-probability outputs, which helps mitigate hallucinations such as inventing nonexistent medical test results.
Why each option is right or wrong
A. Increasing the temperature
B. Enabling frequency penalties on rare words
C. Increasing the model context
D. Decreasing the top-p sampling
Top-p sampling (nucleus sampling) is a generation-control parameter, not a legal or compliance control, and lowering it constrains token selection to a smaller cumulative-probability set at each step. In practice, reducing the threshold from a higher value to a lower one narrows the candidate pool and suppresses low-probability continuations, which directly addresses fabricated outputs such as nonexistent lab values. By contrast, increasing randomness would widen the output space and make hallucinations more likely.