arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.15938cs.AIcs.CL
收藏

盲点剖析:LLM评判者必须看到什么以及需要多少

The Anatomy of a Blind Spot: What an LLM Judge Must Be Shown, and How Much

Hosho

Nitish Kumar, Prateik Pothuneedi, Greg Bandak

AI总结:

该研究将LLM评判者视为控制论监管者,通过7个公开基准实验发现,向评判者展示的内容比评判者种类更重要,还揭示了不同任务中参考材料的作用,报告了理论预测的不足并推导了相关设计规则。

摘要:

大型语言模型(LLM)生成文本的评估越来越多地委托给充当评判者的LLM,但这类评判者必须看到什么仍未解决。我们将其报告被作为依据的评判者视为控制论意义上的监管者,探究必须向其展示什么、展示多少、以何种形式展示,以及其报告必须准确的内容。在7个公开基准上,我们仅改变向评判者展示的内容。在价格相差10至17倍的3种评判者中,向其展示的内容比评判者的种类更重要。当一个系统根据评判者的反馈修改答案,而该评判者被剥夺了必须遵守的材料时,其在指令遵循和问答任务上的表现最终低于初始水平。在指令遵循任务中,损失随每一轮增加,且当评判者看到用户请求但未看到约束条件,以及使用第二类评判者时,损失仍然存在。在新闻摘要任务中,源文本旁的正确参考摘要抵消了源文本的大部分贡献,而在单答案问题上,参考摘要有所帮助。源文本中约十分之一的内容完成了几乎全部工作,且小型蕴含图与LLM评判者在查找基于事实文本的错误方面表现匹配。我们在运行前提出了83条可测试条款;在运行的81条中,32条达到预期标准,50条朝着预测方向发展,且理论的大多数定量预测均未通过。我们报告了所有不足,并推导了与测量效应相关的设计规则。

英文摘要:

Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown remains unresolved. We treat a judge whose reports are acted on as a regulator, in the sense of cybernetics, and ask what it must be shown, how much, in what form, and what its reports must get right. On seven public benchmarks we varied only what the judge was shown. Across three judges whose prices differ 10- to 17-fold, what the judge was shown mattered more than which judge it was. When a system revised its answers on feedback from a judge denied the material they had to respect, it ended below its starting point on instruction following and question answering. On instruction following the loss grew with each round, and it persisted when the judge saw the user's request but not the constraints and with a second judge family. On news summarisation, a correct reference summary beside the source undid most of the source's contribution, whereas on single-answer questions a reference helped. About a tenth of the source did almost all of its work, and a small entailment graph matched an LLM judge at finding errors in grounded text. We committed 83 testable clauses before running; of the 81 that ran, 32 met their bar and 50 went in the predicted direction, and most quantitative predictions from the theory failed. We report every shortfall and derive design rules tied to measured effects.

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。

↑