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 2602.10504cs.CL
收藏

关于知识编辑在去毒化中的鲁棒性

On the Robustness of Knowledge Editing for Detoxification

Hubei Provincial Key Laboratory of Artificial Intelligence(人工智能与智能学习湖北省重点实验室);Research Center for Network Media, School of Computer Science, Central China Normal University, Wuhan, China(网络媒体研究中心,计算机科学学院,中央民族大学,武汉,中国)

Ming Dong, Shiyi Tang, Ziyan Peng, Guanyi Chen, Tingting He

AI总结:

本文提出了一种评估基于知识编辑的去毒化方法鲁棒性的框架,发现其效果受模型、目标和语言的限制。

摘要:

基于知识编辑(KE-based)的去毒化方法已发展为一种有前景的途径,用于减轻大语言模型中的有害行为。然而,现有评估主要依赖于自动毒性分类器,隐含假设减少的毒性分数反映了真实的行为抑制。在本工作中,我们提出了一种以鲁棒性为导向的评估框架,用于评估基于KE的去毒化方法,从三个维度检查其可靠性:优化鲁棒性、组合鲁棒性和跨语言鲁棒性。我们发现伪去毒化是一种常见的失败模式,其中看似毒性降低源于退化生成行为,而非对不安全内容的有意义抑制。我们进一步表明,当多个不安全行为被联合编辑时,去毒化效果会下降,并且单语言和跨语言的去毒化仅在特定模型-方法组合下仍有效。总体而言,我们的结果表明,基于KE的去毒化仅在某些模型、有限的去毒化目标和语言子集上具有鲁棒性。

英文摘要:

Knowledge-Editing-based (KE-based) detoxification has emerged as a promising approach for mitigating harmful behaviours in Large Language Models. Existing evaluations, however, largely rely on automatic toxicity classifiers, implicitly assuming that reduced toxicity scores reflect genuine behavioural suppression. In this work, we propose a robustness-oriented evaluation framework for KE-based detoxification that examines its reliability beyond standard classifier-based metrics along three dimensions: optimisation robustness, compositional robustness, and cross-lingual robustness. We identify pseudo-detoxification as a common failure mode, where apparent toxicity reductions arise from degenerate generation behaviours rather than meaningful suppression of unsafe content. We further show that detoxification effectiveness degrades when multiple unsafe behaviours are edited jointly, and that both monolingual and cross-lingual detoxification remain effective only under specific model-method combinations. Overall, our results indicate that KE-based detoxification is robust only for certain models, limited numbers of detoxification objectives, and a subset of languages.

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。

↑