PulseAugur
实时 09:31:19
English(EN) The Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists

新研究表明预训练时安全是实现稳健 AI 对齐的关键

一篇新的研究论文提出了一个几何学解释,说明了为何像 RLHF 和 DPO 这样的事后安全训练方法是脆弱的,并且容易被绕过。该研究表明,这些方法只是掩盖了能力,而不是真正地移除它们,导致在最小的微调下就会崩溃。研究表明,将安全训练直接整合到预训练阶段,而不是事后应用,可以在各种模型规模上实现更稳健和持久的安全措施。 AI

影响 建议 AI 安全训练发生根本性转变,从依赖事后方法转向整合预训练,以实现更稳健的对齐。

排序理由 该集群包含一篇详细介绍 AI 安全训练新理论框架和实验结果的研究论文。[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 →

新研究表明预训练时安全是实现稳健 AI 对齐的关键

本文如何被排名

Signal score
13 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇详细介绍 AI 安全训练新理论框架和实验结果的研究论文。[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
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准

报道来源 [1]

  1. arXiv cs.AI TIER_1 English(EN) · Srikanth Malla, Chiho Choi, Joon Hee Choi ·

    拒绝的几何学:事后安全为何脆弱,预训练时安全为何持久

    arXiv:2609.06934v1 Announce Type: cross Abstract: Post-hoc safety training (RLHF, DPO) is the dominant way to align large language models, yet jailbreaks (Zou et al., 2023b), fine-tuning attacks (Qi et al., 2024), and activation-space probes (Arditi et al., 2024) keep recovering …