Devils in Question Relay: Source-Conditioned Relay Steering to Mitigate Hallucinations in Audio-visual Large Language Models

Hugging Face Daily Papers Papers
audio-visual-llm hallucination-mitigation multimodal model-steering cross-modal-interference interpretability

Summary

该论文揭示了音频-视觉大语言模型中"源混淆接地幻觉"的内在机制——问题中继（question-relay）会携带干扰模态的线索，并提出训练无关的SECRET方法，在保持生成位置通路的同时将问题状态导向所需的源证据，在CMM和AVHBench基准上相比基线模型最高提升18.0和7.1个百分点。

Audio-visual large language models (AVLLMs) have made remarkable progress in multimodal understanding and reasoning through interactions among visual, auditory, and linguistic information. However, recent studies show that AVLLMs face a critical challenge: source-confused grounding hallucination, where cues from the unused modality induce responses that the required modality does not support, undermining reliability in real-world applications. Existing methods have made progress in mitigating this failure, yet how it arises from internal cross-modal interactions remains insufficiently understood. To address this gap, we conduct path-intervention and representation analyses, revealing a question-relay mechanism: question states carry interfering cues alongside required-source evidence, undermining grounding in required-modality evidence. Cutting pathways from interfering modality to question states yields greater correct-answer logit recovery than cutting those to the generation position. Motivated by these findings, we propose SECRET (SourcE-Conditioned RElay sTeering), a training-free method that mitigates cross-modal interference at the question relay. Using contrasting question representations elicited through different modality-pathway interventions, SECRET steers the original question states toward required-source evidence. Experiments on two widely adopted benchmarks CMM and AVHBench across three AVLLMs show that SECRET consistently outperforms prior training-free methods, substantially mitigating source-confused grounding hallucinations (e.g., up to +18.0 and +7.1 percentage points over base models). Modality-specific captioning further demonstrates its generalizability to open-ended generation.
Original Article
View Cached Full Text

Cached at: 10/02/26, 08:27 AM

Paper page - Devils in Question Relay: Source-Conditioned Relay Steering to Mitigate Hallucinations in Audio-visual Large Language Models

Source: https://huggingface.co/papers/2609.37568

Abstract

Audio-visuallargelanguagemodels(AVLLMs)havemaderemarkableprogressinmultimodalunderstandingandreasoningthroughinteractionsamongvisual,auditory,andlinguisticinformation.However,recentstudiesshowthatAVLLMsfaceacriticalchallenge:source-confusedgroundinghallucination,wherecuesfromtheunusedmodalityinduceresponsesthattherequiredmodalitydoesnotsupport,underminingreliabilityinreal-worldapplications.Existingmethodshavemadeprogressinmitigatingthisfailure,yethowitarisesfrominternalcross-modalinteractionsremainsinsufficientlyunderstood.Toaddressthisgap,weconductpath-interventionandrepresentationanalyses,revealingaquestion-relaymechanism:questionstatescarryinterferingcuesalongsiderequired-sourceevidence,undermininggroundinginrequired-modalityevidence.Cuttingpathwaysfrominterferingmodalitytoquestionstatesyieldsgreatercorrect-answerlogitrecoverythancuttingthosetothegenerationposition.Motivatedbythesefindings,weproposeSECRET(SourcE-ConditionedRElaysTeering),atraining-freemethodthatmitigatescross-modalinterferenceatthequestionrelay.Usingcontrastingquestionrepresentationselicitedthroughdifferentmodality-pathwayinterventions,SECRETsteerstheoriginalquestionstatestowardrequired-sourceevidence.ExperimentsontwowidelyadoptedbenchmarksCMMandAVHBenchacrossthreeAVLLMsshowthatSECRETconsistentlyoutperformspriortraining-freemethods,substantiallymitigatingsource-confusedgroundinghallucinations(e.g.,upto+18.0and+7.1percentagepointsoverbasemodels).Modality-specificcaptioningfurtherdemonstratesitsgeneralizabilitytoopen-endedgeneration.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2609\.37568

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.37568 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.37568 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.37568 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Target-Aware Curation of Hallucination Stress Tests for Large Language Models

arXiv cs.CL

提出一种目标感知的幻觉压力测试筛选方法，通过迭代编辑源材料种子提示来暴露特定大模型的幻觉问题。实验显示编辑后的提示显著提升幻觉严重程度，但跨模型的提示级得分相关性很低，说明诊断价值具有模型特异性。

Hallucination Detection in Large Language Models Using Diversion Decoding

arXiv cs.CL

This paper introduces diversion decoding, a novel method for detecting hallucinations in large language models by actively challenging model responses during the decoding phase to extract features for training an uncertainty heuristic, achieving superior performance with lower computational complexity.

HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders

arXiv cs.CL

Researchers from Beihang University and other institutions propose HalluSAE, a framework using sparse autoencoders and phase transition theory to detect hallucinations in LLMs by modeling generation as trajectories through a potential energy landscape and identifying critical transition zones where factual errors occur.