规范顺序问题：当大型语言模型作为多值关系的不可靠知识库时
The Canonical Order Problem: When Large Language Models Are Unreliable Knowledge Bases for Multi-Valued Relations
浏览论文内容
中文总结 AI 辅助
本文发现LLMs生成多值关系时存在规范顺序问题，内部排序导致偏离该顺序时可靠性显著降低，并通过机制分析揭示了三阶段生成过程。
中文摘要 AI 辅助
大型语言模型（LLMs）因其在预训练期间获取的大量知识，越来越多地被用作知识库（KBs）。虽然许多工作专注于提取单一关系三元组，但大多数现实世界的关系是多值的，需要生成实体集合。在本文中，我们研究了LLMs如何表示和生成多值关系。我们发现了规范顺序问题：LLMs内部的概率分布根据规范顺序（例如，字母顺序或时间顺序）组织许多多值关系。通过机制分析，我们表明LLMs中的集合生成可以分为三个阶段：（1）候选实体的检索，（2）内部排序，以及（3）下一个元素的选择。因此，当提示旨在构建偏离这种内部规范顺序的知识库时，LLMs在生成多值关系的完整集合时的可靠性会显著降低。
英文摘要
Large language models (LLMs) are increasingly used as knowledge bases (KBs) due to the vast amount of knowledge they acquire during pre-training. While many works focus on extracting single relational triples, most real-world relations are multi-valued and require generating sets of entities. In this paper, we investigate how LLMs represent and generate multi-valued relations. We identify the canonical order problem: The probabilistic distributions inside LLMs organize many multi-valued relations according to a canonical ordering (e.g., alphabetical or chronological). Through mechanistic analysis, we show that set generation in LLMs can be thought of in terms of three phases: (1) retrieval of candidate entities, (2) internal sorting, and (3) selection of the next element. As a result, prompts aiming to construct KBs that deviate from this internal canonical ordering lead to a markedly reduced reliability of LLMs when aiming to generate complete sets for multi-valued relations.
发表机构
- University of Augsburg（奥格斯堡大学）
- ScaDS.AI & TU Dresden（ScaDS.AI与德累斯顿工业大学）
- Bosch Center for Artificial Intelligence（博世人工智能中心）
机构由 AI 辅助整理，请以论文原文为准。
