arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.14673cs.CL
收藏

洞察大语言模型长上下文失效现象:当Transformer知道却不说时

Insights into LLM Long-Context Failures: When Transformers Know but Don't Tell

Johns Hopkins University(约翰斯·霍普金斯大学)

Taiming Lu, Muhan Gao, Kuai Yu, Adam Byerly, Daniel Khashabi

AI总结:

本研究针对大语言模型长上下文位置偏差问题,通过探测隐藏表征发现其存在“知而不言”的信息检索与利用脱节现象,并分析提取时间与准确率的关系,为理解Transformer底层机制提供了新视角。

摘要:

大语言模型(LLM)存在位置偏差,难以利用长上下文中间或末尾的信息。本研究通过探测LLM的隐藏表征探究其长上下文推理能力。我们发现,尽管LLM编码了目标信息的位置,却往往无法在生成准确回复时利用该信息。这揭示了信息检索与利用之间的脱节,即“知而不言”现象。我们进一步分析了提取时间与最终准确率的关系,为理解Transformer模型的底层机制提供了洞见。

英文摘要:

Large Language Models (LLMs) exhibit positional bias, struggling to utilize information from the middle or end of long contexts. Our study explores LLMs' long-context reasoning by probing their hidden representations. We find that while LLMs encode the position of target information, they often fail to leverage this in generating accurate responses. This reveals a disconnect between information retrieval and utilization, a "know but don't tell" phenomenon. We further analyze the relationship between extraction time and final accuracy, offering insights into the underlying mechanics of transformer 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。

↑