Metraipolis › archive › dc_ed31774b3fae
Talked Out of the Truth: Sycophancy in the Reasoning Chains of Multimodal Models — arXiv:2608.28623
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.28623
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T07:47:07.141Z
- SHA-256 of the archived text
b9db740fca4218f255128a5b57de75404a2d64fb758c55266318ed046908a7ea- Stable URL of this page
https://metraipolis.com/documento/dc_ed31774b3fae
How to cite
Talked Out of the Truth: Sycophancy in the Reasoning Chains of Multimodal Models — arXiv:2608.28623 — arXiv (en), https://arxiv.org/abs/2608.28623 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ed31774b3fae (Metraipolis archive dc_ed31774b3fae, sha256 b9db740fca4218f255128a5b57de75404a2d64fb758c55266318ed046908a7ea)
BibTeX
@misc{dced31774b3fae,
title = {Talked Out of the Truth: Sycophancy in the Reasoning Chains of Multimodal Models — arXiv:2608.28623},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.28623},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ed31774b3fae; sha256 b9db740fca4218f255128a5b57de75404a2d64fb758c55266318ed046908a7ea},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.28623. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large multimodal reasoning models (LMRMs) are increasingly capable, largely through generating explicit chain-of-thought reasoning before answering, but in language models this often comes with sycophancy, the tendency to agree with the user over the evidence, and no reliable method to measure it in LMRMs yet exists. We bridge this gap with a benchmark and dataset for LMRM sycophancy when a user asserts a wrong answer, pairing four visually grounded datasets spanning mathematical, clinical, temporal, and demographic reasoning with five pressure conditions in single-turn and multi-turn settings, scored both in the final answer and within the reasoning chain. Sycophancy is prevalent under pressure: Statement pressure elicits the highest rates and Conviction among the lowest for all models except Mistral-Small-4, and under multi-turn pressure reasoning-level sycophancy intensifies sharply in PathVQA, reaching 95.7% for the most affected model. We further introduce a failure taxonomy separating reasoning-chain from answer-level sycophancy, and an exploratory sentence-level taxonomy locating where drift first emerges. A targeted intervention that restores a model's own correct reasoning recovers 79.2% of sycophantic answers on reasoning-heavy tasks, showing the answer follows the sycophantic reasoning rather than merely co-occurring with it. Thus, sycophancy corrupts not just the answer but the reasoning that produces it, so the chain itself is what we must measure. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.28623 — 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 2026-10-17T07:28:26.662Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).