AI RESEARCH
CARE: Privacy-Compliant Agentic Reasoning with Evidence Discordance
arXiv CS.CL
•
ArXi:2604.01113v1 Announce Type: new Large language model (LLM) systems are increasingly used to high-stakes decision-making, but they typically perform worse when the available evidence is internally inconsistent. Such a scenario exists in real-world healthcare settings, with patient-reported symptoms contradicting medical signs. To study this problem, we