A new benchmark shows that 28 LLMs all decline when patient information changes whether a medication-safety rule applies, exposing a warning-retention failure.
Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence , pages=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Evaluating Counterfactual Sensitivity to Patient Information in Medication-Safety Reasoning
A new benchmark shows that 28 LLMs all decline when patient information changes whether a medication-safety rule applies, exposing a warning-retention failure.