Deepthika S, Pawni Yadav, B. Sankar · 2026 · arXiv
DOI: 10.48550/arXiv.2610.31579 · open-access PDF
La investigación sobre la sycophancy pregunta si los grandes modelos de lenguaje (LLM) se inclinan hacia lo que sus usuarios quieren oír. Nos preguntamos la pregunta inversa: cuando un LLM es confiado, consistente y equivocado, ¿los usuarios se inclinan hacia él? En un experimento dentro de sujetos, 32 participantes aprendieron hasta cinco temas de ciencia chateando con cinco asistentes construidos sobre el mismo modelo (GPT-5-mini). Desconocido para los participantes, el briefing de cada asistente tenía 0, 25, 50, 75 o 100% de sus hechos reemplazados por falsedades plausibles, y cada asistente hablaba con plena confianza.
Machine translation (Meta M2M100); the English original follows.
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, t
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.31579, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange