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 2402.11436cs.CLcs.AI
收藏

傲慢与偏见:LLM在自我完善中放大自我偏见

Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement

University of California Santa Barbara(加州大学圣巴巴拉分校);Carnegie Mellon University(卡内基梅隆大学)

Wenda Xu, Guanglei Zhu, Xuandong Zhao, Liangming Pan, Lei Li, William Yang Wang

AI总结:

该研究定义并测量LLM自我偏见,发现self-refine会放大该偏见,而更大模型和准确外部反馈可有效缓解并提升下游性能。

摘要:

近期研究表明,大型语言模型(LLM)通过在某些任务上的自我反馈提升性能,但在其他任务上却会出现性能下降。我们发现,这种矛盾源于LLM在评估自身输出时存在偏见。本文中,我们使用两个统计量正式定义了LLM的自我偏见——即偏好其自身生成内容的倾向。我们在翻译、受约束文本生成和数学推理任务上分析了六个LLM(GPT-4、GPT-3.5、Gemini、LLaMA2、Mixtral和DeepSeek)。我们发现,在多种语言和任务中,自我偏见在所有受测LLM中普遍存在。分析表明,尽管self-refine流程提高了模型输出的流畅性和可理解性,但它进一步放大了自我偏见。为缓解此类偏见,我们发现更大的模型规模以及带有准确评估的外部反馈能够显著降低self-refine流程中的偏见,从而带来下游任务的实际性能提升。代码和数据发布于https://github.com/xu1998hz/llm_self_bias。

英文摘要:

Recent studies show that large language models (LLMs) improve their performance through self-feedback on certain tasks while degrade on others. We discovered that such a contrary is due to LLM's bias in evaluating their own output. In this paper, we formally define LLM's self-bias - the tendency to favor its own generation - using two statistics. We analyze six LLMs (GPT-4, GPT-3.5, Gemini, LLaMA2, Mixtral and DeepSeek) on translation, constrained text generation, and mathematical reasoning tasks. We find that self-bias is prevalent in all examined LLMs across multiple languages and tasks. Our analysis reveals that while the self-refine pipeline improves the fluency and understandability of model outputs, it further amplifies self-bias. To mitigate such biases, we discover that larger model size and external feedback with accurate assessment can significantly reduce bias in the self-refine pipeline, leading to actual performance improvement in downstream tasks. The code and data are released at https://github.com/xu1998hz/llm_self_bias.

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。

↑