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 2602.00459cs.CL
收藏

什么是对LLM重要的?来自摘要的行为主观和计算证据

What Matters to an LLM? Behavioral and Computational Evidences from Summarization

Univ. Grenoble Alpes(格勒诺布尔大学);CNRS(国家科学研究中心);Inria(法国国家信息与自动化技术研究所);Grenoble INP(格勒诺布尔研究所);LIG(实验室)

Yongxin Zhou, Changshun Wu, Philippe Mulhem, Didier Schwab, Maxime Peyrard

AI总结:

本文通过行为和计算分析研究LLM在摘要任务中的重要性选择机制,揭示了LLM在重要性模式上的一致性及注意力头与层对重要性的预测作用。

摘要:

大型语言模型(LLMs)现在在摘要任务上处于最先进的水平,但驱动其信息选择的内部重要性概念仍然隐藏。我们通过结合行为和计算分析来研究这一点。从行为上,我们为每篇文档生成一系列长度受控的摘要,并根据每个信息单元被选择的频率推导出经验重要性分布。这些结果表明,LLMs在重要性模式上趋于一致,与预LLM基线显著不同,并且LLMs按家族而非大小聚类。从计算上,我们发现某些注意力头与经验重要性分布匹配良好,而中后期层对重要性具有强预测性。这些结果为理解LLMs在摘要中优先考虑什么以及这种优先级如何在内部表示提供了初步见解,为解释和最终控制这些模型中的信息选择开辟了道路。

英文摘要:

Large Language Models (LLMs) are now state-of-the-art at summarization, yet the internal notion of importance that drives their information selections remains hidden. We propose to investigate this by combining behavioral and computational analyses. Behaviorally, we generate a series of length-controlled summaries for each document and derive empirical importance distributions based on how often each information unit is selected. These reveal that LLMs converge on consistent importance patterns, sharply different from pre-LLM baselines, and that LLMs cluster more by family than by size. Computationally, we identify that certain attention heads align well with empirical importance distributions, and that middle-to-late layers are strongly predictive of importance. Together, these results provide initial insights into what LLMs prioritize in summarization and how this priority is internally represented, opening a path toward interpreting and ultimately controlling information selection in these models.

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。

↑