LLM answer checking needs smarter review to reduce mistakes effectively
Risk Is Not Review Value: Wrong-Answer Exposure Under Bounded Review Budgets
Artificial IntelligenceComputation and Language
Summary
Big language models (LLMs) often give more answers than people can check before relying on them. The authors found that just looking at which answers are risky isn’t enough to decide which ones to check first. They propose a smarter way to pick answers for review by considering how fixable the mistakes are, how important they are, and how much it costs to fix them. This approach helps reduce the number of wrong answers users see, even with limited time to review. Their tests show that smarter prioritization keeps the same number of detected problems but lowers how many mistakes still slip through after fixing what can be repaired.
large language modelsanswer reviewrisk assessmenterror detectionreview budgetrepairabilitywrong-answer exposureevaluation metricsbenchmarking
Authors
SangJin Park, Myungsub Choi, Jineok Kim, Minseung Kang
Abstract
LLM assistants often produce more answers than humans can review before users see them. Most evaluations ask whether an answer is wrong, unsupported, or low-confidence. Bounded review budgets instead ask which answers should be checked first under a fixed review budget. Risk alone is not enough: a high-risk answer may be hard to repair, while a moderately risky answer may be directly correctable from available evidence. For generated-answer evaluation, we model review prioritization as exposure reduction, where review value combines estimated wrongness, intervention affordance, impact, and cost. We evaluate review queues with Wrong-Answer Exposure Ratio (WAER), the fraction of wrong answers left unreviewed, and post-repair residual exposure (PRRE), the fraction still exposed after deterministic benchmark-supported repairs. PRRE uses repairability rules that do not numerically reuse the affordance scores used for ranking. On a 720-item TAT-QA/SciFact stress benchmark, review-value ranking keeps answer-level WAER nearly unchanged at 20% budget (0.605 vs. 0.600) but lowers PRRE from 0.881 to 0.716. These results show that trustworthy LLM evaluation should measure not only error detection, but also how limited review capacity reduces exposed wrong answers.