contradiction-minimal-effect-confidence
IN premise — summaries/2026/08/24/xu-2024-knowledge-conflicts-survey-s3-inter-context-conflict.md
Created 2026-08-25T02:59:02+00:00
Despite producing wrong answers, LLMs do not lower their output confidence when exposed to contradictory context (Chen et al., 2022).
Summary
LLMs keep sounding just as sure about their answers even when the information around them contradicts what they said, so a high-confidence tone is not a reliable signal that the output is actually correct. This means any system relying on self-reported confidence to detect errors or trigger corrections will miss problems, because the model simply won't flag its own uncertainty.