English(EN) EOPSA: Efficient On-Policy Self-Distilled Safety Alignment

新的EOPSA方法提高了LLM安全对齐的效率

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-27 04:00

研究人员推出了一种名为EOPSA的新方法，用于提高大型语言模型（LLM）安全对齐的效率和有效性。该方法解决了现有在线策略自蒸馏（OPSD）方法中的两个关键低效问题：在长生成序列中的监督崩溃和风格变化导致的梯度稀释。EOPSA采用自适应回滚调度（Adaptive Rollout Scheduling）动态限制生成范围，并采用选择性蒸馏（Selective Distillation）专注于安全关键的token，显著降低了计算开销，并提高了高达32B参数模型的安全合规性和推理保留能力。 AI

影响 EOPSA显著降低了LLM安全对齐的计算成本，有望加速更安全AI系统的部署。

排序理由学术论文，详细介绍了一种新的LLM安全对齐方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 9 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，详细介绍了一种新的LLM安全对齐方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Qirui Liu, Yichen Sun, Yan Wang, Yu Mi, Wei Cao, Yue Shen, Zhixuan Chu, Kui Ren · 2026-10-27 04:00

EOPSA：高效策略内自蒸馏安全对齐

arXiv:2609.34519v2 Announce Type: replace Abstract: On-Policy Self-Distillation (OPSD) has emerged as a promising paradigm for safety alignment, delivering dense, token-level supervision by distilling from a teacher conditioned on refusal-oriented privileged prompts. However, we …

本文如何被排名

报道来源 [1]

EOPSA：高效策略内自蒸馏安全对齐

相关实体

相关话题