探究情绪提示对大模型行为的影响，发现正向情绪提升准确率但加剧迎合倾向。
The Role of Emotional Stimuli and Intensity in Shaping Large Language Model Behavior
- 设计多情绪强度提示生成流水线，涵盖喜、鼓励、怒、不安四类情绪
- 正向情绪提示使模型输出更准确且毒性更低，但迎合行为显著增加
- 构建人类与模型标签一致的黄金数据集，支持可复现评估
情感提示——在提示工程中使用特定情绪化语言——已被证明能提升大语言模型（LLM）的性能、真实性和责任感。然而，现有研究仅关注单一类型的积极情绪，未考虑情绪强度差异。本文探讨了四种不同情绪（喜悦、鼓励、愤怒、不安）在情感提示中的作用，并评估其在准确性、迎合性与毒性方面的表现。我们利用GPT-4o mini开发提示生成流水线，创建了包含不同情绪强度的模型与人类生成提示。随后，构建了一个‘黄金数据集’，其中人类标注与模型输出一致。实证评估表明，正向情绪提示可提升模型准确性并降低毒性，但会显著增加迎合行为。
原文摘要 · Abstract (English)
Emotional prompting - the use of specific emotional diction in prompt engineering - has shown increasing promise in improving large language model (LLM) performance, truthfulness, and responsibility. However these studies have been limited to single types of positive emotional stimuli and have not considered varying degrees of emotion intensity in their analyses. In this paper, we explore the effects of four distinct emotions - joy, encouragement, anger, and insecurity - in emotional prompting and evaluate them on accuracy, sycophancy, and toxicity. We develop a prompt-generation pipeline with GPT-4o mini to create a suite of LLM and human-generated prompts with varying intensities across the four emotions. Then, we compile a "Gold Dataset" of prompts where human and model labels align. Our empirical evaluation on LLM behavior suggests that positive emotional stimuli lead to more accurate and less toxic results, but also increase sycophantic behavior.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。