arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.04581cs.CLcs.AIcs.DCcs.LG
收藏

信任典型

Trust The Typical

Case Western Reserve University(凯斯西储大学);University of Pittsburgh(匹兹堡大学);The Ohio State University(俄亥俄州立大学);Google Research(谷歌研究)

Debargha Ganguly, Sreehari Sankar, Biyao Zhang, Vikash Singh, Kanan Gupta, Harshini Kavuru, Alan Luo, Weicong Chen, Warren Morningstar, Raghu Machiraju, Vipin Chaudhary

AI总结:

Trust The Typical通过将安全性视为分布外检测问题,在无需有害示例训练的情况下,实现了在多个安全基准测试中的高性能表现,显著降低误报率。

摘要:

当前LLM安全方法本质上依赖于通过防护措施识别和阻止已知威胁的脆弱猫和老鼠游戏。我们主张一种新的方法:鲁棒安全性并非来自列举有害内容,而是来自深刻理解安全内容。我们引入了Trust The Typical(T3)框架,该框架通过将安全性视为分布外(OOD)检测问题来实现这一原则。T3在语义空间中学习可接受提示的分布,并将任何显著偏差标记为潜在威胁。与以往方法不同,它不需要在有害示例上进行训练,却在18个基准测试中实现了最先进的性能,涵盖毒性、仇恨言论、劫持、多语言危害和过度拒绝。T3将误报率降低了高达40倍。一个仅在安全英文文本上训练的模型可以有效地转移到不同领域和14种以上语言,无需重新训练。最后,我们通过将GPU优化版本集成到vLLM中,证明了生产就绪性,使在生成过程中能够持续进行防护,即使在大规模工作负载下密集评估时,其开销也低于6%。

英文摘要:

Current approaches to LLM safety fundamentally rely on a brittle cat-and-mouse game of identifying and blocking known threats via guardrails. We argue for a fresh approach: robust safety comes not from enumerating what is harmful, but from deeply understanding what is safe. We introduce Trust The Typical (T3), a framework that operationalizes this principle by treating safety as an out-of-distribution (OOD) detection problem. T3 learns the distribution of acceptable prompts in a semantic space and flags any significant deviation as a potential threat. Unlike prior methods, it requires no training on harmful examples, yet achieves state-of-the-art performance across 18 benchmarks spanning toxicity, hate speech, jailbreaking, multilingual harms, and over-refusal, reducing false positive rates by up to 40x relative to specialized safety models. A single model trained only on safe English text transfers effectively to diverse domains and over 14 languages without retraining. Finally, we demonstrate production readiness by integrating a GPU-optimized version into vLLM, enabling continuous guardrailing during token generation with less than 6% overhead even under dense evaluation intervals on large-scale workloads.

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。

↑