arXiv：cs.CL· Richard Zhe Wang·· 6 小时前
从 LLM 内部状态检测金融问答中的"自信幻觉"
Confidently Wrong: Detecting Hallucinations in Financial Question Answering from LLM Internal States
AI 导读
在真实财报问答基准 FinQA 和 TAT-QA 上，八次重采样答案全部一致的"自信回答"中仍有 15-23% 是错误的。研究者在 Qwen3-8B、Llama-3.1-8B 和 Gemma-2-9B 的残差流上训练线性探针，检测幻觉的 AUROC 达 0.68-0.77，优于 token 对数概率和模型自评等输出层基线（0.55-0.63）。
正文
Abstract:Large language models (LLMs) in financial applications fail most consequentially when they are confidently wrong. Hedged, uncertain answers invite scrutiny, whereas confident errors silently degrade downstream decisions without warning. We ask how reliably such confidently wrong answers, or confident hallucinations, can be detected from a model's internal activations, and whether those activations carry information beyond its output-level confidence signals. We train linear probes on the residual stream and evaluate them on two established numerical question-answering (QA) benchmarks built from real filings, FinQA and TAT-QA. Behavioral confidence is measured as the agreement among eight resampled answers to the same question, and probe effectiveness is compared against baselines, such as token log-probabilities and the model's own True/False self-assessment of its answer. Our findings show that among confident answers, those for which all eight resamples agree, 15-23% are wrong on FinQA. There the probes have a significant advantage over output-level baselines in detecting hallucinations, holding 0.68-0.77 AUROC while the best baselines fall to 0.55-0.63, across Qwen3-8B, Llama-3.1-8B, and Gemma-2-9B. Our results suggest that probing can be a cost-effective triage mechanism for routing LLM answers to human review and quality control procedures in high-stakes financial applications. Code and data are available at this https URL.
| Comments: | Accepted at the 7th ACM International Conference on AI in Finance (ICAIF '26), Milan, Italy, November 14-17, 2026. 8 pages, 2 figures, 5 tables |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.11414 [cs.CL] |
| (or arXiv:2607.11414v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.11414 arXiv-issued DOI via DataCite |
|
| Related DOI: | https://doi.org/10.1145/3858063.3858685
DOI(s) linking to related resources |
Submission history
From: Richard Wang [view email]
[v1]
Mon, 13 Jul 2026 11:22:25 UTC (114 KB)
[v2]
Tue, 13 Oct 2026 21:03:09 UTC (114 KB)
来源：arXiv：cs.CL · arxiv.org