[论文解读] Measuring reasoning capabilities of ChatGPT

Adrian Groza|arXiv (Cornell University)|Oct 8, 2023

Artificial Intelligence in EducationComputer Science被引用 3

一句话总结

本论文通过使用Prover9和Mace4进行形式化验证，评估了ChatGPT在100个源自MALOGA库的精选逻辑谜题上的推理能力。仅有7%的回复在内容和理由上均正确；平均每道题存在7个逻辑错误，生成文本中26.03%被归类为有缺陷，揭示了诸如无依据断言和逻辑不一致等系统性推理错误。

ABSTRACT

I shall quantify the logical faults generated by ChatGPT when applied to reasoning tasks. For experiments, I use the 144 puzzles from the library \url{https://users.utcluj.ro/~agroza/puzzles/maloga}~\cite{groza:fol}. The library contains puzzles of various types, including arithmetic puzzles, logical equations, Sudoku-like puzzles, zebra-like puzzles, truth-telling puzzles, grid puzzles, strange numbers, or self-reference puzzles. The correct solutions for these puzzles were checked using the theorem prover Prover9~\cite{mccune2005release} and the finite models finder Mace4~\cite{mccune2003mace4} based on human-modelling in Equational First Order Logic. A first output of this study is the benchmark of 100 logical puzzles. For this dataset ChatGPT provided both correct answer and justification for 7\% only. %, while BARD for 5\%. Since the dataset seems challenging, the researchers are invited to test the dataset on more advanced or tuned models than ChatGPT3.5 with more crafted prompts. A second output is the classification of reasoning faults conveyed by ChatGPT. This classification forms a basis for a taxonomy of reasoning faults generated by large language models. I have identified 67 such logical faults, among which: inconsistencies, implication does not hold, unsupported claim, lack of commonsense, wrong justification. The 100 solutions generated by ChatGPT contain 698 logical faults. That is on average, 7 fallacies for each reasoning task. A third ouput is the annotated answers of the ChatGPT with the corresponding logical faults. Each wrong statement within the ChatGPT answer was manually annotated, aiming to quantify the amount of faulty text generated by the language model. On average, 26.03\% from the generated text was a logical fault.

研究动机与目标

• 评估ChatGPT在多样化形式逻辑谜题上的推理可靠性。
• 创建一个包含100个非图形化逻辑谜题的基准数据集，用于评估大语言模型的推理能力。
• 对大语言模型生成回复中的逻辑推理错误进行分类与量化。
• 为未来模型评估与改进提供有标注的错误推理示例。

提出的方法

• 从MALOGA库中选取144个谜题，排除含有图形输入的题目，最终形成100道题的基准测试集。
• 使用Prover9和Mace4在等式一阶逻辑中对正确解答进行形式化验证。
• 将每道谜题以标准提示方式提交给ChatGPT-3.5，并收集完整回复。
• 人工标注每份回复，识别并分类逻辑错误，如无依据断言和不一致的推理。
• 量化有缺陷文本的比例，并计算每道解答的平均逻辑谬误数量。
• 基于重复出现的错误模式，开发出包含67种不同类型的推理错误分类体系。

实验结果

研究问题

• RQ1 ChatGPT在多大程度上能正确解答多样化形式逻辑谜题？
• RQ2 大语言模型生成解答时，常见的逻辑推理错误有哪些类型？
• RQ3 大语言模型推理输出中，逻辑不一致和无依据断言的普遍程度如何？
• RQ4 是否可以使用标准化的逻辑谜题基准来评估和比较大语言模型的推理能力？

主要发现

• 在100道题的基准测试中，仅有7%的ChatGPT回复在内容和理由上均正确。
• 平均每道解答包含7个逻辑错误，100份回复共累计698个错误。
• 约26.03%的ChatGPT生成文本被归类为含有逻辑错误。
• 本研究识别出67种不同的推理错误类型，包括逻辑不一致、无依据断言和错误的论证。
• 100道谜题的基准数据集已公开，旨在挑战大语言模型的推理能力，超越简单的算术或模式匹配。
• 推理错误分类体系为未来大语言模型的系统性评估与改进奠定了基础。