大语言模型在研究级数学问题上的失败模式:分类与实证刻画
Failure Modes of Large Language Models on Research-Level Mathematics: A Taxonomy and an Empirical Characterisation
AI总结:
本文通过分析“First Proof”基准测试中LLM的错误,识别出四种失败模式(F1-F4),并审计Gemini 2.5 Flash生成的证明,发现前提走私(F2)普遍存在且无法被引文验证检测,提出应构建推理时管道预防而非事后检测。
摘要:
“First Proof”基准测试[1]向最强大的公开可用LLM提出了十个研究级数学问题,发现它们始终错误——不是沉默,而是自信流畅地错误。本文探究原因。基于First Proof附录A中每个问题的事后分析,我识别出四种失败模式:引文捏造(F1)、前提走私(F2)、静默问题重构(F3)和局部到全局兼容性缺口(F4)。随后,我审计了Gemini 2.5 Flash在基准测试问题1、2和5上生成的八个一次性证明,使用了专门为揭示F1和F2而构建的两个工具。核心发现对任何将检索增强生成(RAG)视为明显解决方案的人来说都令人不安:八个证明中没有一个是确认的捏造引文,但每一个都至少包含一个作为“基本结果”或“标准论证”断言且未附任何证明的承重主张。这种失败模式——F2,前提走私——在设计上对引文验证不可见。我引入的前提审计工具以100%精度(5/5个法官确认的标记为真阳性)和该语料库中50%的证明级召回率标记它。分类和审计共同表明,正确的长期目标是构建推理时管道以防止这些失败模式发生,而不仅仅是事后检测它们。索引术语——大语言模型,数学推理,幻觉,前提走私,失败模式分类。
英文摘要:
The "First Proof" benchmark [1] posed ten research-level mathematics questions to the strongest publicly available LLMs and found them consistently wrong-not silent, but confidently, fluently wrong. This paper asks why. Working from the per-question post-mortems in First Proof's Appendix A, I identify four failure modes: citation fabrication (F1), premise smuggling (F2), silent problem reformulation (F3), and local-to-global compatibility gaps (F4). I then audit eight one-shot proofs generated by Gemini 2.5 Flash on Questions 1, 2, and 5 of the benchmark, using two instruments built specifically to surface F1 and F2. The central finding is uncomfortable for anyone who sees retrieval-augmented generation (RAG) as the obvious fix: not one of the eight proofs contained a confirmed fabricated citation, yet every single one contained at least one load-bearing claim asserted as a "fundamental result" or "standard argument" with no justification attached. That failure mode-F2, premise smuggling-is invisible to citation verification by design. A premise-audit instrument I introduce flags it at 100% precision (5/5 judge-confirmed flags are true positives) and 50% proof-level recall in this corpus. The taxonomy and the audit together suggest that the right long-term objective is building inference-time pipelines that prevent these failure modes from occurring, not just detecting them after the fact. Index Terms--Large language models, mathematical reasoning, hallucination, premise smuggling, failure-mode taxonomy.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。