English(EN) $C$-$\Delta\Theta$: Circuit-Restricted Weight Arithmetic for Selective Refusal

新的 C-ΔΘ 方法将 LLM 安全拒绝嵌入模型权重

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-09 04:00

研究人员开发了一种名为 C-ΔΘ（电路受限权重算术）的新方法，以提高大型语言模型的安全性。该技术旨在离线训练阶段直接将拒绝能力嵌入模型权重，而不是依赖于昂贵的推理时干预。通过识别和更新模型参数的一小部分，C-ΔΘ 创建了一个即插即用的编辑检查点，可以有选择地拒绝有害请求，同时保留其通用能力。 AI

影响该方法可以通过将干预从运行时转移到离线训练来降低 LLM 安全的运营成本。

排序理由该项目是一篇研究论文，详细介绍了一种新的 LLM 安全方法。[lever_c_demoted from research: ic=1 ai=1.0]

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: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该项目是一篇研究论文，详细介绍了一种新的 LLM 安全方法。[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: 86 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

arXiv cs.CL TIER_1 English(EN) · Aditya Kasliwal, Pratinav Seth, Vinay Kumar Sankarapu · 2026-07-09 04:00

$C$-$\Delta\Theta$: 限制电路的权重算术用于选择性拒绝

arXiv:2602.04521v2 Announce Type: replace Abstract: Modern deployments require LLMs to enforce safety policies at scale, yet many controls rely on inference-time interventions that add recurring compute cost and serving complexity. Activation steering is widely used, but it requi…