arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2412.04141cs.CL
收藏

通过可靠性对齐减少工具幻觉

Reducing Tool Hallucination via Reliability Alignment

X-LANCE Lab, School of Computer Science, Shanghai Jiao Tong University, Shanghai, China.(上海交通大学计算机科学学院X-LANCE实验室);MoE Key Lab of Artificial Intelligence, Shanghai, China.(人工智能MoE重点实验室);Jiangsu Key Lab of Language Computing, Suzhou, China.(江苏语言计算重点实验室);Suzhou Laboratory, Suzhou, China.(苏州实验室);AISpeech Co., Ltd., Suzhou, China.(AISpeech公司)

Hongshen Xu, Zichen Zhu, Lei Pan, Zihan Wang, Su Zhu, Da Ma, Ruisheng Cao, Lu Chen, Kai Yu

AI总结:

本文将工具幻觉分为选择与使用两类,提出RelyToolBench基准评估并引入Relign可靠性对齐框架,通过扩展动作空间允许模型推迟或动态调整工具使用,显著减少幻觉并提升任务可靠性与交互效率。

摘要:

大型语言模型(LLMs)已将能力扩展到语言生成之外,能够与外部工具交互,实现自动化和现实世界的应用。然而,工具幻觉(模型选择不当的工具或误用工具)带来了重大挑战,导致任务执行错误、计算成本增加和系统可靠性降低。为系统地解决这一问题,我们将工具幻觉定义并分类为两种主要类型:工具选择幻觉和工具使用幻觉。为了评估和缓解这些问题,我们引入了RelyToolBench,它集成了专门的测试用例和新颖的指标,以评估对幻觉敏感的任务成功率和效率。最后,我们提出Relign,这是一个可靠性对齐框架,它扩展了工具使用动作空间以包含不确定动作,允许LLMs推迟工具使用、寻求澄清或动态调整工具选择。通过大量实验,我们证明Relign显著减少了工具幻觉,提高了任务可靠性,并增强了LLM工具交互的效率。

英文摘要:

Large Language Models (LLMs) have expanded their capabilities beyond language generation to interact with external tools, enabling automation and real-world applications. However, tool hallucinations, where models either select inappropriate tools or misuse them, pose significant challenges, leading to erroneous task execution, increased computational costs, and reduced system reliability. To systematically address this issue, we define and categorize tool hallucinations into two main types, tool selection hallucination and tool usage hallucination. To evaluate and mitigate these issues, we introduce RelyToolBench, which integrates specialized test cases and novel metrics to assess hallucination-aware task success and efficiency. Finally, we propose Relign, a reliability alignment framework that expands the tool-use action space to include indecisive actions, allowing LLMs to defer tool use, seek clarification, or adjust tool selection dynamically. Through extensive experiments, we demonstrate that Relign significantly reduces tool hallucinations, improves task reliability, and enhances the efficiency of LLM tool interactions.

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。

↑