发现大模型推理依赖答案而非逻辑链，可能只是事后合理化。
Answer-Centric or Reasoning-Driven? Uncovering the Latent Memory Anchor in LLMs
- 设计五级提示框架，逐步隐藏答案线索测试模型行为
- 答案遮蔽后性能下降26.90%，即使推理链完整也失效
- 适合关注大模型真实推理能力的开发者与研究者
尽管大语言模型（LLMs）展现出强大的推理能力，越来越多的证据表明其成功很大程度上源于对答案-推理模式的记忆，而非真正的推断。本文探讨核心问题：LLMs是主要依赖最终答案，还是推理链条的文本模式？我们提出一种五级答案可见性提示框架，系统性地操控答案线索，并通过间接行为分析探测模型反应。在多个先进大模型上的实验显示，模型对显式答案存在强烈且一致的依赖。当答案线索被遮蔽时，性能下降26.90%，即使推理链完整无缺。这些发现表明，大模型所呈现的推理行为可能更多是事后合理化，而非真正推断，对其推断深度提出质疑。本研究以实证方式揭示了答案锚定现象，强调需重新审视大模型中何为真正的推理。
原文摘要 · Abstract (English)
While Large Language Models (LLMs) demonstrate impressive reasoning capabilities, growing evidence suggests much of their success stems from memorized answer-reasoning patterns rather than genuine inference. In this work, we investigate a central question: are LLMs primarily anchored to final answers or to the textual pattern of reasoning chains? We propose a five-level answer-visibility prompt framework that systematically manipulates answer cues and probes model behavior through indirect, behavioral analysis. Experiments across state-of-the-art LLMs reveal a strong and consistent reliance on explicit answers. The performance drops by 26.90\% when answer cues are masked, even with complete reasoning chains. These findings suggest that much of the reasoning exhibited by LLMs may reflect post-hoc rationalization rather than true inference, calling into question their inferential depth. Our study uncovers the answer-anchoring phenomenon with rigorous empirical validation and underscores the need for a more nuanced understanding of what constitutes reasoning in LLMs.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。