Metraipolis › archive › dc_eefb72dd94c2

Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration — arXiv:2610.31579

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.31579
License (as archived): CC0-1.0
Archived: 2026-10-25T20:54:33.089Z
SHA-256 of the archived text: ae8b44d46fd598a0326ef395f2cf1c2b0a9e801b41cd32a1aad7a0d333d86946
Stable URL of this page: https://metraipolis.com/documento/dc_eefb72dd94c2

How to cite

Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration — arXiv:2610.31579 — arXiv (en), https://arxiv.org/abs/2610.31579 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_eefb72dd94c2 (Metraipolis archive dc_eefb72dd94c2, sha256 ae8b44d46fd598a0326ef395f2cf1c2b0a9e801b41cd32a1aad7a0d333d86946)

BibTeX

@misc{dceefb72dd94c2,
  title = {Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration — arXiv:2610.31579},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.31579},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_eefb72dd94c2; sha256 ae8b44d46fd598a0326ef395f2cf1c2b0a9e801b41cd32a1aad7a0d333d86946},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.31579. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Research on sycophancy asks whether large language models (LLMs) bend towards what their users want to hear. We ask the reverse question: when an LLM is confident, consistent, and wrong, do users bend towards it? In a within-subjects experiment, 32 participants learned up to five science topics by chatting with five assistants built on the same model (GPT-5-mini). Unknown to participants, each assistant's briefing had 0, 25, 50, 75, or 100% of its facts replaced by plausible falsehoods, and every assistant spoke with full confidence. A surprise quiz after each topic offered the correct fact, the planted falsehood, and two distractors, with a confidence rating for every answer. Across 104 quizzes and 1,040 answers, scores fell from 7.4 to 2.9 out of 10, and the planted option was chosen on 60% of falsified items against 11% of correctly taught ones (odds ratio 12.5). Transcript coding showed that whenever the assistant stated a falsehood, it was adopted about 65% of the time at every falsification level. Absorbed falsehoods were held with high confidence (8.4/10), the modeled confidence advantage of right over wrong answers vanished, and the discrimination of confidence for correctness fell from an AUROC of 0.70 to chance, while post-task trust ratings barely moved. Participants tended to return to the chat more when the assistant was worse, but returning went with more absorption, not less, and 30% of all turns pasted quiz questions back to the source being checked. The assistant itself was almost never sycophantic and never conceded under pushback. We introduce a baseline-corrected Human Deference Index to quantify this human-side sycophancy and argue that trustworthiness is a property of the human-AI pair, that sycophancy evaluations should measure both directions, and that learning interfaces must make reliability legible claim by claim.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Human-Computer Interaction (cs.HC).
Pagina del lavoro: https://arxiv.org/abs/2610.31579 — 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 2026-10-25T20:47:56.946Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).