arXiv 2410.016772024-10-24cs.AI

思维混乱：通过 Typoglycemia 揭示大型语言模型心理学

Mind Scramble: Unveiling Large Language Model Psychology Via Typoglycemia

University of Science and Technology of China (USTC)（中国科学技术大学）；Squirrel AI（松鼠AI）；National University of Singapore (NUS)（新加坡国立大学）；NanYang Technological University (NTU)（南洋理工大学）；The Hong Kong University of Science and Technology (HKUST)（香港科技大学）

Miao Yu, Junyuan Mao, Guibin Zhang, Jingheng Ye, Junfeng Fang, Aoxiao Zhong, Yang Liu, Yuxuan Liang, Kun Wang, Qingsong Wen

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出 LLM Psychology 方法，借助 Typoglycemia 乱序文本实验从字符、词和句子层面揭示 LLM 的类人认知行为、鲁棒性差异及稳定认知模式。

摘要：

对大型语言模型（LLM）外部行为和内部机制的研究，已在处理物理世界复杂任务方面展现出前景。研究表明，诸如 GPT-4 等强大的 LLM 正开始表现出类似人类的认知能力，包括规划、推理和反思。本文提出一条名为 LLM Psychology（大语言模型心理学）的研究路线与方法论，利用人类心理学实验考察 LLM 的认知行为与机制。我们迁移心理学中的 Typoglycemia（词序混乱）现象，以探究 LLM 的“心智”。人类大脑依赖语境和词汇模式理解乱序文本，而 LLM 与之不同，采用独特的编码与解码过程。通过在字符、词和句子层面开展 Typoglycemia 实验，我们观察到：（I）LLM 在宏观尺度上表现出类人行为，例如任务准确率降低、token/时间消耗增加；（II）LLM 对乱序输入呈现不同的鲁棒性，因此 Typoglycemia 可作为无需新数据集的模型评估基准；（III）不同任务类型具有不同影响，复杂逻辑任务（如数学）在乱序形式下更具挑战性；（IV）每个 LLM 在各任务中都具有独特且一致的“认知模式”，揭示其心理过程中的一般机制。我们对隐藏层进行了深入分析以解释这些现象，为未来 LLM Psychology 研究和更深层次的可解释性铺平道路。

英文摘要：

Research into the external behaviors and internal mechanisms of large language models (LLMs) has shown promise in addressing complex tasks in the physical world. Studies suggest that powerful LLMs, like GPT-4, are beginning to exhibit human-like cognitive abilities, including planning, reasoning, and reflection. In this paper, we introduce a research line and methodology called LLM Psychology, leveraging human psychology experiments to investigate the cognitive behaviors and mechanisms of LLMs. We migrate the Typoglycemia phenomenon from psychology to explore the "mind" of LLMs. Unlike human brains, which rely on context and word patterns to comprehend scrambled text, LLMs use distinct encoding and decoding processes. Through Typoglycemia experiments at the character, word, and sentence levels, we observe: (I) LLMs demonstrate human-like behaviors on a macro scale, such as lower task accuracy and higher token/time consumption; (II) LLMs exhibit varying robustness to scrambled input, making Typoglycemia a benchmark for model evaluation without new datasets; (III) Different task types have varying impacts, with complex logical tasks (e.g., math) being more challenging in scrambled form; (IV) Each LLM has a unique and consistent "cognitive pattern" across tasks, revealing general mechanisms in its psychology process. We provide an in-depth analysis of hidden layers to explain these phenomena, paving the way for future research in LLM Psychology and deeper interpretability.

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。