Metraipolis › archive › dc_cb3c536bb391

When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA — arXiv:2609.08934

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.08934
License (as archived)
CC0-1.0
Archived
2026-10-16T12:49:21.195Z
SHA-256 of the archived text
c224404fdd584b1e66afc4ed0f4bd9b954e5d273b3f42040d5f5b45e4be57552
Stable URL of this page
https://metraipolis.com/documento/dc_cb3c536bb391

How to cite

When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA — arXiv:2609.08934 — arXiv (en), https://arxiv.org/abs/2609.08934 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cb3c536bb391 (Metraipolis archive dc_cb3c536bb391, sha256 c224404fdd584b1e66afc4ed0f4bd9b954e5d273b3f42040d5f5b45e4be57552)
BibTeX
@misc{dccb3c536bb391,
  title = {When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA — arXiv:2609.08934},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.08934},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cb3c536bb391; sha256 c224404fdd584b1e66afc4ed0f4bd9b954e5d273b3f42040d5f5b45e4be57552},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.08934. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.08934 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-16T12:30:25.206Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).