牛哥精选 · 半年

📋 全部 🤖 AI·大模型 ⚡ 效率工具 📝 深度技术 🚀 产品观察 💰 商业科技 🔓 开源项目 🎨 设计创意 📖 阅读推荐 🏷 资源合集 🌱 成长效率

📝 深度技术 arXiv AI 2026-08-10

GRM: Utility-Aware Jailbreak Attacks on Audio LLMs via Gradient-Ratio Masking

音频大模型也难逃越狱攻击？GRM通过梯度比掩蔽实现效用感知攻击，兼顾攻击效果与音频质量，为多模态AI安全防御敲响警钟。

arXiv:2604.09222v2 Announce Type: replace-cross Abstract: Audio Large Language Models (ALLMs) enable spoken interaction but introduce new jailbreak vu…

音频大模型越狱攻击梯度比掩蔽 ai安全多模态模型

📝 深度技术 arXiv NLP 2026-07-30

Prosody-driven Jailbreaks in Audio LLMs: A Controlled Study and Mechanistic Analysis

研究揭示语音大模型可被声调特征越狱攻击，机制分析清晰，ACM MM 2026录用论文。

arXiv:2607.26541v1 Announce Type: cross Abstract: Audio-capable foundation models enable end-to-end spoken interaction, but they also introduce safety…

音频大模型越狱攻击声调分析安全漏洞 acm mm

🤖 AI·大模型 arXiv 机器学习 2026-05-20

Heterogeneity-Aware Dataset Scheduling for Efficient Audio Large Language Model Training

提出异构感知数据集调度方法，提升音频大模型训练效率与效果的新方案。

arXiv:2605.19101v1 Announce Type: cross Abstract: Training general-purpose Audio Large Language Models (ALLMs) across diverse datasets is essential fo…

音频大语言模型数据集调度异构感知训练效率 ai训练优化

📅 日期

2026-05-20 2026-05-19

🐂 牛哥精选

GRM: Utility-Aware Jailbreak Attacks on Audio LLMs via Gradient-Ratio Masking

Prosody-driven Jailbreaks in Audio LLMs: A Controlled Study and Mechanistic Analysis

Heterogeneity-Aware Dataset Scheduling for Efficient Audio Large Language Model Training

📅 日期