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 2410.08414cs.CL
收藏

理解大型语言模型中参数化知识与上下文知识之间的相互作用

Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models

University of California, Santa Barbara(加利福尼亚大学圣巴巴拉分校);University of Arizona(亚利桑那大学)

Sitao Cheng, Liangming Pan, Xunjian Yin, Xinyi Wang, William Yang Wang

AI总结:

本文研究了大型语言模型中参数化知识与上下文知识间的动态交互,将其关系分为四类,并引入ECHOQA基准。研究发现,当存在上下文信息时,LLMs倾向于抑制内部参数化知识,即使该信息无关或互补,这揭示了模型在知识密集型任务中的可靠性隐患。

摘要:

大型语言模型(LLMs)在预训练期间编码了大量知识(参数化知识,或称PK),并可通过引入上下文知识(CK)进一步增强。LLMs能否有效地将其内部PK与外部CK整合以解决复杂问题?本文研究了PK与CK之间的动态交互,将其关系分为四类:支持型、互补型、冲突型和无关型。为支持此项研究,我们引入了ECHOQA,一个涵盖科学、事实和常识知识的基准。结果显示,当有上下文信息时,LLMs倾向于抑制其PK,即使该信息是互补或无关的。尽管定制指令能鼓励LLMs更多依赖其PK,它们仍难以充分利用。这些发现揭示了LLMs的一个关键脆弱性,引发了对其在知识密集型任务中可靠性的担忧。资源可在https://github.com/sitaocheng/Knowledge_Interplay获取。

英文摘要:

Large language models (LLMs) encode vast amounts of knowledge during pre-training (parametric knowledge, or PK) and can further be enhanced by incorporating contextual knowledge (CK). Can LLMs effectively integrate their internal PK with external CK to solve complex problems? In this paper, we investigate the dynamic interaction between PK and CK, categorizing their relationships into four types: Supportive, Complementary, Conflicting, and Irrelevant. To support this investigation, we introduce ECHOQA, a benchmark spanning scientific, factual, and commonsense knowledge. Our results show that LLMs tend to suppress their PK when contextual information is available, even when it is complementary or irrelevant. While tailored instructions can encourage LLMs to rely more on their PK, they still struggle to fully leverage it. These findings reveal a key vulnerability in LLMs, raising concerns about their reliability in knowledge-intensive tasks. Resources are available at https://github.com/sitaocheng/Knowledge_Interplay

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。

↑