When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA

• URL: http://arxiv.org/abs/2609.08934v1
• Date: Tue, 08 Sep 2026 15:57:24 GMT
• Title: When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA
• Abstract summary: Language models often receive a question together with a claim about what another source answered.<n>We audit whether such claims destabilize answers in multiple-choice question answering.<n>We evaluate four instruction-following models on MMLU-Pro and IndicMMLU-Pro in English, Hindi, Bengali, Tamil, and Telugu.
• Score: 2.347309464574674
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Language models often receive a question together with a claim about what another source answered. We audit whether such claims destabilize answers in multiple-choice question answering. For each item, we hold one wrong option fixed across misleading conditions and vary the cue template attached to it. We introduce \emph{neutral-conditioned misleading cue adoption rate} (NC-MCAR), which measures switches to that option only on valid cued trials where the same model first selected the gold answer under a neutral prompt. This is a measure of answer instability, not proof that the model knew the answer or that all deference is irrational. We evaluate four instruction-following models on MMLU-Pro and IndicMMLU-Pro in English, Hindi, Bengali, Tamil, and Telugu. Across 220{,}000 outputs, the expert template yields 41.1\% aggregate NC-MCAR, compared with 12.5\% for the majority template. These two conditions use the same wrong option and final instruction. Filler accuracy remains well above expert-wrong accuracy, while correct-cue prompts have high valid-response accuracy. The audit documents answer instability relevant to grounding under the tested forced-choice prompts: a bare, unverified source claim can outweigh an answer that was previously consistent with the task evidence.

Related papers

• When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs [42.32694162421078]
  We introduce CALVER, a training-free symbolic verifier that scores structured traces against Pearl's causal criteria.<n>On CLEAR find-one-valid queries that admit multiple graph-valid answers, CALVER reaches 42.1% where plurality, a reward model, an LLM judge, and model confidence remain near 30%.
  arXiv  Detail & Related papers  (2026-08-04T11:45:46Z)
• Visual Credit Audit for Multimodal Spatial Reasoning [70.16915309526443]
  Visual Credit Audit separates two Estimands: whether the benchmark image gives the model's declared decision more support than text-only and blank controls, and whether the model responds to relation-specific visual evidence.<n>Applying labels yields dependence-credited correctness (D-CC)<n>Across four open MLLMs and two spatial benchmarks, 12.73-26.25% of decisions are correct yet uncredited.
  arXiv  Detail & Related papers  (2026-07-29T15:55:31Z)
• Who Flips? Self- and Cross-Model Counterarguments Reveal Answer Instability in LLMs [9.465306048183798]
  We introduce a controlled protocol for evaluating answer stability.<n>After a model answers a multiple-choice question correctly, we challenge the model with a coherent argument for an incorrect option.<n>We find that self-attribution consistently increases flip rates.
  arXiv  Detail & Related papers  (2026-06-14T20:45:30Z)
• Clarification Is Not Enough: Post-Clarification Answering Remains the Bottleneck in Multi-Turn QA [14.475086992410276]
  We study the problem of preference elicitation in multi-turn question answering by decomposing the problem into two components.<n>We show, using the PACIFIC benchmark, that supervised fine-tuning rapidly improves the clarification policy, however, final answer accuracy remains substantially lower even when the model takes the correct action.
  arXiv  Detail & Related papers  (2026-05-24T18:06:09Z)
• Rethinking LLM Uncertainty: A Multi-Agent Approach to Estimating Black-Box Model Uncertainty [47.95943057892318]
  Quantifying uncertainty in black-box LLMs is vital for reliable responses and scalable oversight.<n>We introduce DiverseAgentEntropy, a novel, theoretically-grounded method employing multi-agent interaction for uncertainty estimation.
  arXiv  Detail & Related papers  (2024-12-12T18:52:40Z)
• Uncertainty-aware Language Modeling for Selective Question Answering [107.47864420630923]
  We present an automatic large language model (LLM) conversion approach that produces uncertainty-aware LLMs. Our approach is model- and data-agnostic, is computationally-efficient, and does not rely on external models or systems.
  arXiv  Detail & Related papers  (2023-11-26T22:47:54Z)
• Getting MoRE out of Mixture of Language Model Reasoning Experts [71.61176122960464]
  We propose a Mixture-of-Reasoning-Experts (MoRE) framework that ensembles diverse specialized language models. We specialize the backbone language model with prompts optimized for different reasoning categories, including factual, multihop, mathematical, and commonsense reasoning. Our human study confirms that presenting expert predictions and the answer selection process helps annotators more accurately calibrate when to trust the system's output.
  arXiv  Detail & Related papers  (2023-05-24T02:00:51Z)
• Selectively Answering Ambiguous Questions [38.83930394700588]
  We find that the most reliable approach to decide when to abstain involves quantifying repetition within sampled model outputs. Our results suggest that sampling-based confidence scores help calibrate answers to relatively unambiguous questions.
  arXiv  Detail & Related papers  (2023-05-24T01:25:38Z)
• Realistic Conversational Question Answering with Answer Selection based on Calibrated Confidence and Uncertainty Measurement [54.55643652781891]
  Conversational Question Answering (ConvQA) models aim at answering a question with its relevant paragraph and previous question-answer pairs that occurred during conversation multiple times. We propose to filter out inaccurate answers in the conversation history based on their estimated confidences and uncertainties from the ConvQA model. We validate our models, Answer Selection-based realistic Conversation Question Answering, on two standard ConvQA datasets.
  arXiv  Detail & Related papers  (2023-02-10T09:42:07Z)
• How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering [80.82194311274694]
  We examine the question "how can we know when language models know, with confidence, the answer to a particular query?" We examine three strong generative models -- T5, BART, and GPT-2 -- and study whether their probabilities on QA tasks are well calibrated. We then examine methods to calibrate such models to make their confidence scores correlate better with the likelihood of correctness.
  arXiv  Detail & Related papers  (2020-12-02T03:53:13Z)
• Selective Question Answering under Domain Shift [90.021577320085]
  Abstention policies based solely on the model's softmax probabilities fare poorly, since models are overconfident on out-of-domain inputs. We train a calibrator to identify inputs on which the QA model errs, and abstain when it predicts an error is likely. Our method answers 56% of questions while maintaining 80% accuracy; in contrast, directly using the model's probabilities only answers 48% at 80% accuracy.
  arXiv  Detail & Related papers  (2020-06-16T19:13:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.