2025.07.21 | dLLM新型安全漏洞,现有防御不足;俄语语音合成,数据与标注是核心。

HuggingFace 每日AI论文速递

本期的 10 篇论文如下:

[00:20] 😈 The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMs(隐藏在面具后的恶魔:扩散大语言模型的一种新兴安全漏洞)

[01:12] 🎤 A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models(解决俄语语音生成模型中语音与韵律挑战的数据中心框架)

[02:07] 🧩 Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning(Franca:用于可扩展视觉表示学习的嵌套套娃聚类)

[02:49] 🚀 Mono-InternVL-1.5: Towards Cheaper and Faster Monolithic Multimodal Large Language Models(Mono-InternVL-1.5:迈向更经济、更快速的单体多模态大语言模型)

[03:24] 🎨 CSD-VAR: Content-Style Decomposition in Visual Autoregressive Models(CSD-VAR:视觉自回归模型中的内容-风格分解)

[04:27] 🚀 RedOne: Revealing Domain-specific LLM Post-Training in Social Networking Services(RedOne:揭示社交网络服务中领域专用LLM的后训练)

[05:08] 🤝 Inverse Reinforcement Learning Meets Large Language Model Post-Training: Basics, Advances, and Opportunities(逆向强化学习与大语言模型后训练的结合:基础、进展与机遇)

[05:41] 🚫 Mitigating Object Hallucinations via Sentence-Level Early Intervention(通过句子级早期干预缓解物体幻觉)

[06:20] ⚡ The Generative Energy Arena (GEA): Incorporating Energy Awareness in Large Language Model (LLM) Human Evaluations(生成式能源竞技场 (GEA):在大型语言模型 (LLM) 人工评估中融入能源意识)

[07:41] 📈 Quantitative Risk Management in Volatile Markets with an Expectile-Based Framework for the FTSE Index(波动市场中基于期望分位数框架的定量风险管理:以富时指数为例)

【关注我们】

您还可以在以下平台找到我们,获得播客内容以外更多信息

小红书: AI速递

To listen to explicit episodes, sign in.

Stay up to date with this show

Sign in or sign up to follow shows, save episodes and get the latest updates.

Select a country or region

Africa, Middle East, and India

Asia Pacific

Europe

Latin America and the Caribbean

The United States and Canada