PulseAugur
中
实时 08:56:49
English(EN) Modular Pretraining Enables Access Control

新的GRAM方法为双重用途能力提供模块化AI访问控制

研究人员开发了梯度路由辅助模块(GRAM),这是一种新颖的预训练方法,旨在解决AI开发中的双重用途困境。GRAM允许在单个AI模型中选择性地禁用特定能力,以极低的成本近似训练经过筛选数据的独立模型的效果。这种方法实现了细粒度的访问控制,允许将敏感知识限制在受信任的部署中,同时保持通用性能。实验表明,GRAM能够有效地隔离病毒学和网络安全等各个领域的特定能力,并且即使在微调后也能保持这种隔离。 AI

影响 能够更精细地控制AI能力,可能减轻与双重用途技术相关的风险。

排序理由 该集群描述了一种在学术论文中发布的新AI模型预训练方法。

在 Alignment Forum 阅读 →

AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →

新的GRAM方法为双重用途能力提供模块化AI访问控制

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
Research
该集群描述了一种在学术论文中发布的新AI模型预训练方法。
Source corroboration
3 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
paper, safety, model release
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
92 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。

报道来源 [3]

  1. arXiv cs.LG TIER_1 English(EN) · Ethan Roland, Murat Cubuktepe, Erick Martinez, Stijn Servaes, Keenan Pepper, Mike Vaiana, Diogo Schwerz de Lucena, Judd Rosenblatt, Addie Foote, Cem Anil, Alex Cloud ·

    模块化预训练实现访问控制

    arXiv:2607.08077v1 Announce Type: new Abstract: AI developers face a dual-use dilemma. An AI capability that helps one user cure a disease can help another synthesize one. This dilemma could be resolved with access control, limiting dual-use AI capabilities to trusted deployments…

  2. Alignment Forum TIER_1 English(EN) · E.Roland ·

    模块化预训练实现访问控制

    <p><i><span>Full author list: Ethan Roland</span></i><span>*</span><i><span>, Murat Cubuktepe</span></i><span>*</span><i><span>, Erick Martinez</span></i><span>*</span><i><span>, Stijn Servaes, Keenan Pepper, Mike Vaiana, Diogo Schwerz de Lucena, Judd Rosenblatt, Addie Foote, Cem…

  3. LessWrong (AI tag) TIER_1 English(EN) · E.Roland ·

    模块化预训练实现访问控制

    <p><i><span>Full author list: Ethan Roland</span></i><span>*</span><i><span>, Murat Cubuktepe</span></i><span>*</span><i><span>, Erick Martinez</span></i><span>*</span><i><span>, Stijn Servaes, Keenan Pepper, Mike Vaiana, Diogo Schwerz de Lucena, Judd Rosenblatt, Addie Foote, Cem…