arXiv 2501.097752025-01-24cs.CLcs.AI

多项选择题：推理使大型语言模型（LLMs）更加自信，即使它们是错误的

Multiple Choice Questions: Reasoning Makes Large Language Models (LLMs) More Self-Confident Even When They Are Wrong

Politecnico di Milano（米兰理工大学）；Universidad Politécnica de Madrid（马德里理工大学）

Tairan Fu, Javier Conde, Gonzalo Martínez, María Grandury, Pedro Reviriego

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文研究了推理对LLM在多项选择题中自信心的影响，发现提供推理（思维链）会使LLM对答案更自信，即使答案错误，揭示了LLM估计概率的内在局限性。

摘要：

评估LLM最广泛使用的方法之一是多项选择题（MCQ）测试。MCQ基准使得大规模测试LLM在几乎任何主题上的知识成为可能，因为结果可以自动处理。为了帮助LLM回答，可以在提示中包含一些被称为few shots的示例。此外，可以要求LLM直接用所选选项回答问题，或者先提供推理然后再给出所选答案，这被称为思维链。除了检查所选答案是否正确外，评估还可以查看LLM估计的响应概率，作为LLM对该响应信心的指示。在本文中，我们研究LLM对其答案的信心如何取决于模型是被要求直接回答还是在回答前提供推理。对七个不同模型中广泛主题的问题进行评估的结果表明，当LLM在回答前提供推理时，它们对自己的答案更有信心。无论所选答案是否正确，这种情况都会发生。我们的假设是，这种行为是由于推理改变了所选答案的概率，因为LLM基于输入问题和支持所选推理来预测答案。因此，LLM估计的概率似乎具有内在局限性，应予以理解以便在评估程序中使用它们。有趣的是，在人类中也观察到了相同的行为，对于人类来说，解释答案会增加对其正确性的信心。

英文摘要：

One of the most widely used methods to evaluate LLMs are Multiple Choice Question (MCQ) tests. MCQ benchmarks enable the testing of LLM knowledge on almost any topic at scale as the results can be processed automatically. To help the LLM answer, a few examples called few shots can be included in the prompt. Moreover, the LLM can be asked to answer the question directly with the selected option or to first provide the reasoning and then the selected answer, which is known as chain of thought. In addition to checking whether the selected answer is correct, the evaluation can look at the LLM-estimated probability of its response as an indication of the confidence of the LLM in the response. In this paper, we study how the LLM confidence in its answer depends on whether the model has been asked to answer directly or to provide the reasoning before answering. The results of the evaluation of questions on a wide range of topics in seven different models show that LLMs are more confident in their answers when they provide reasoning before the answer. This occurs regardless of whether the selected answer is correct. Our hypothesis is that this behavior is due to the reasoning that modifies the probability of the selected answer, as the LLM predicts the answer based on the input question and the reasoning that supports the selection made. Therefore, LLM estimated probabilities seem to have intrinsic limitations that should be understood in order to use them in evaluation procedures. Interestingly, the same behavior has been observed in humans, for whom explaining an answer increases confidence in its correctness.

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。

AI总结：

摘要：

英文摘要：

相关论文