2026.06.15 | AI视频精准控镜；智能体推理细粒度优化

【赞助商】OpenClaw快报每天五分钟，听听 OpenClaw 快报，带你了解最新动态和业内讨论传送门 https://www.xiaoyuzhoufm.com/podcast/6a1732a2dffa135d0ab5ef43【目录】本期的 15 篇论文如下：[00:30] 🎥 OmniDirector: General Multi-Shot Camera Cloning without Cross-Paired Data（OmniDirector：无需配对数据的通用多镜头相机克隆）[01:25] 🤖 APPO: Agentic Procedural Policy Optimization（智能体程序策略优化）[02:23] 🧠 Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents（记忆是重构的，而非检索：面向LLM智能体的图记忆机制）[03:16] 🤖 From Chatbot to Digital Colleague: The Paradigm Shift Toward Persistent Autonomous AI（从聊天机器人到数字同事：向持久自主人工智能的范式转变）[04:12] 🎼 Orchestra-o1: Omnimodal Agent Orchestration（管弦乐队-o1：全模态智能体编排框架）[05:12] 🔧 HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry（HarnessX：一个可组合、自适应且可演化的代理框架铸造厂）[06:16] 🎥 Rethinking RAG in Long Videos: What to Retrieve and How to Use It?（重新思考长视频中的检索增强生成：检索什么以及如何使用？）[07:19] 🎬 OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence Chains（OmniVideo-100K：一个通过结构化脚本和证据链进行音视频推理的数据集）[08:36] 🤖 From AGI to ASI（从通用人工智能到超级人工智能）[09:34] 🧠 Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO（小型模型是GRPO中策略级多样性的天然探索者）[10:32] 🛡 RedAct: Redacting Agent Capability Traces for Procedural Skill Protection（RedAct：为保护程序技能而屏蔽智能体能力痕迹）[11:21] 👁 LLM Agents Can See Code Repositories（LLM智能体能够“看见”代码仓库）[12:14] 🩺 Measuring Epistemic Resilience of LLMs Under Misleading Medical Context（测量大语言模型在误导性医疗语境下的认知韧性）[13:15] 🔄 Skip a Layer or Loop It? Learning Program-of-Layers in LLMs（跳过一层还是循环它？学习大语言模型中的层程序化执行）[14:12] 🎨 RepFusion: Leveraging Multimodal Priors for Denoising in Representation Space（RepFusion：利用多模态先验在表示空间中进行去噪）【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速递在小宇宙查看该单集文稿