CHECK combines database fact-checking with an ensemble-consistency classifier to flag medical LLM hallucinations, with reported AUCs of 0.95-0.96, but the headline '31% to 0.3%' reduction is a comparison of question context, not of the detection system.
Hallucination rates and reference accuracy of chatgpt and bard for systematic reviews: compar ative analysis
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
REJECT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Trustworthy AI for Medicine: Continuous Hallucination Detection and Elimination with CHECK
CHECK combines database fact-checking with an ensemble-consistency classifier to flag medical LLM hallucinations, with reported AUCs of 0.95-0.96, but the headline '31% to 0.3%' reduction is a comparison of question context, not of the detection system.