arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.01378cs.CLcs.AIstat.ML
收藏

自回归语言模型中的上下文依赖性与可靠性

Context Dependence and Reliability in Autoregressive Language Models

Department of Informatics, University of Oslo(信息学院,奥斯陆大学);Department of Electronic Systems, Aalborg University(电子系统系,奥胡斯大学)

Poushali Sengupta, Shashi Raj Pandey, Sabita Maharjan, Frank Eliassen

AI总结:

本文提出RISE方法,用于自回归语言模型中区分关键上下文元素与冗余信息,提升解释的可靠性和稳定性。

摘要:

大型语言模型(LLMs)通过利用大量上下文来生成输出,这些上下文通常包括提示中的冗余信息、检索到的段落和交互历史。在关键应用中,识别哪些上下文元素实际上影响输出至关重要,因为标准解释方法难以处理冗余和重叠的上下文。输入的微小变化可能导致属性分数的不可预测变化,从而损害可解释性并引发诸如提示注入等风险。本文解决区分关键上下文元素与相关元素的挑战。我们引入了RISE(冗余不敏感的解释评分)方法,该方法量化了每个输入相对于其他输入的独立影响,最小化冗余的影响,提供更清晰、稳定的属性。实验表明,RISE比传统方法提供更稳健的解释,强调了条件信息对可信LLM解释和监控的重要性。

英文摘要:

Large language models (LLMs) generate outputs by utilizing extensive context, which often includes redundant information from prompts, retrieved passages, and interaction history. In critical applications, it is vital to identify which context elements actually influence the output, as standard explanation methods struggle with redundancy and overlapping context. Minor changes in input can lead to unpredictable shifts in attribution scores, undermining interpretability and raising concerns about risks like prompt injection. This work addresses the challenge of distinguishing essential context elements from correlated ones. We introduce RISE (Redundancy-Insensitive Scoring of Explanation), a method that quantifies the unique influence of each input relative to others, minimizing the impact of redundancies and providing clearer, stable attributions. Experiments demonstrate that RISE offers more robust explanations than traditional methods, emphasizing the importance of conditional information for trustworthy LLM explanations and monitoring.

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。

↑