arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.07883cs.CL
收藏

超越'我无法完成此请求':通过标签增强缓解LLM中的刚性拒绝

Beyond "I cannot fulfill this request": Alleviating Rigid Rejection in LLMs via Label Enhancement

Southeast University, Nanjing, China(东南大学)

Ying Zhang, Congyu Qiao, Xin Geng, Ning Xu

AI总结:

本文提出LANCE方法,通过标签增强实现安全且灵活的响应,减少LLM中的刚性拒绝问题,提升交互自然性。

摘要:

大型语言模型(LLM)依赖安全对齐来遵守安全请求并拒绝有害请求。然而,传统拒绝机制常导致'刚性拒绝',即通用模板(如'我无法完成此请求')无差别触发拒绝,严重破坏人类与LLM的自然交互。为此,本文提出LANCE,通过变分推断进行标签增强,预测多个拒绝类别间的连续分布。这些细粒度拒绝分布为精炼模型提供多向文本梯度,以中和提示中的危险元素,使LLM生成安全响应,避免刚性拒绝并保持交互自然性。实验表明,LANCE显著缓解刚性拒绝问题,同时维持高安全标准,在帮助性和响应自然性方面显著优于现有基线模型。

英文摘要:

Large Language Models (LLMs) rely on safety alignment to obey safe requests while refusing harmful ones. However, traditional refusal mechanisms often lead to "rigid rejection," where a general template (e.g., "I cannot fulfill this request") indiscriminately triggers refusals and severely undermines the naturalness of interactions between humans and LLMs. To address this issue, LANCE is proposed in this paper to ensure safe yet flexible and natural responses via label enhancement. Specifically, LANCE employs variational inference to perform label enhancement, predicting a continuous distribution across multiple rejection categories. These fine-grained rejection distributions provide multi-way textual gradients for a refinement model to neutralize the hazardous elements in the prompt, so that the LLMs could generate safe responses that avoid rigid rejections while preserving the naturalness of interactions. Experiments demonstrate that LANCE significantly alleviates the rigid rejection problem while maintaining high security standards, significantly outperforming existing baseline models in terms of helpfulness and naturalness of responses.

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。

↑