Enregistré dans:
Détails bibliographiques
Auteurs principaux: Zhou, Zeqi, Wu, Fang, Talaei, Shayan, Zhao, Haokai, Meixin, Cheng, Xu, Tinson, Saberi, Amin, Choi, Yejin
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2506.06020
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866910992117530624
author Zhou, Zeqi
Wu, Fang
Talaei, Shayan
Zhao, Haokai
Meixin, Cheng
Xu, Tinson
Saberi, Amin
Choi, Yejin
author_facet Zhou, Zeqi
Wu, Fang
Talaei, Shayan
Zhao, Haokai
Meixin, Cheng
Xu, Tinson
Saberi, Amin
Choi, Yejin
contents Large language models frequently encounter conflicts between their parametric knowledge and contextual input, often resulting in factual inconsistencies or hallucinations. We propose Self-Reflective Debate for Contextual Reliability (SR-DCR), a lightweight framework that integrates token-level self-confidence with an asymmetric multi-agent debate to adjudicate such conflicts. A critic, deprived of context, challenges a defender who argues from the given passage; a judge model evaluates the debate and determines the context's reliability. The final answer is selected by combining the verdict with model confidence. Experiments on the ClashEval benchmark demonstrate that SR-DCR consistently enhances robustness to misleading context while maintaining accuracy on trustworthy inputs, outperforming both classical debate and confidence-only baselines with minimal computational overhead. The code is available at https://github.com/smiles724/Self-Reflective-Debates.
format Preprint
id arxiv_https___arxiv_org_abs_2506_06020
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle When to Trust Context: Self-Reflective Debates for Context Reliability
Zhou, Zeqi
Wu, Fang
Talaei, Shayan
Zhao, Haokai
Meixin, Cheng
Xu, Tinson
Saberi, Amin
Choi, Yejin
Computation and Language
Artificial Intelligence
Large language models frequently encounter conflicts between their parametric knowledge and contextual input, often resulting in factual inconsistencies or hallucinations. We propose Self-Reflective Debate for Contextual Reliability (SR-DCR), a lightweight framework that integrates token-level self-confidence with an asymmetric multi-agent debate to adjudicate such conflicts. A critic, deprived of context, challenges a defender who argues from the given passage; a judge model evaluates the debate and determines the context's reliability. The final answer is selected by combining the verdict with model confidence. Experiments on the ClashEval benchmark demonstrate that SR-DCR consistently enhances robustness to misleading context while maintaining accuracy on trustworthy inputs, outperforming both classical debate and confidence-only baselines with minimal computational overhead. The code is available at https://github.com/smiles724/Self-Reflective-Debates.
title When to Trust Context: Self-Reflective Debates for Context Reliability
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.06020