{"id":"trust-me-im-wrong","title":"Trust Me, I’m Wrong: LLMs Hallucinate with Certainty Despite Knowing the Answer","authors":[{"given":"Adi","family":"Simhi"},{"given":"Itay","family":"Itzhak"},{"given":"Fazl","family":"Barez"},{"given":"Gabriel","family":"Stanovsky"},{"given":"Yonatan","family":"Belinkov"}],"first_public":"2025-02-18","date_note":"First arXiv version.","date_source":"https://arxiv.org/abs/2502.12964","year":2025,"venue":"EMNLP 2025 Findings","paper_url":"https://aclanthology.org/2025.findings-emnlp.792/","doi":"10.18653/v1/2025.findings-emnlp.792","arxiv":"2502.12964","relationship":"uses","bibtex_key":"trustmeimwrong2025","record_url":"https://yuntiandeng.com/wildchat/papers/trust-me-im-wrong/","citation_title":"Trust Me, I’m Wrong: LLMs Hallucinate with Certainty Despite Knowing the Answer","citation_type":"paper-conference","bibtex_type":"inproceedings","container_title":"Findings of the Association for Computational Linguistics: EMNLP 2025","citation_url":"https://aclanthology.org/2025.findings-emnlp.792/","citation_year":2025,"citation_note":"EMNLP 2025 Findings proceedings paper.","pages":"14665–14688","publisher":"Association for Computational Linguistics","code_url":"https://github.com/technion-cs-nlp/Trust_me_Im_wrong","category":"safety","relationship_label":"Uses WildVis","dataset_version":"Queries WildChat and LMSYS-Chat-1M through WildVis to validate help-seeking prompt styles; the main experiments use Natural Questions and TriviaQA.","summary":"Uses WildVis to check that prompt variations in a hallucination study resemble real human requests. The paper studies cases where models give confident incorrect answers after small prompt changes despite consistently answering the original question correctly.","finding":"The authors use WildVis searches and WildChat examples to validate the help-seeking language used in their evaluation prompts.","evidence_url":"https://aclanthology.org/2025.findings-emnlp.792.pdf","evidence_note":"Section 3.1: prompt construction and validation. Appendix F.2: WildVis searches of WildChat and LMSYS-Chat-1M."}