2026.06.26 | DanceOPD融合策略蒸馏；ViQ突破量化表示极限

【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：[00:32] 💃 DanceOPD: On-Policy Generative Field Distillation（DanceOPD：基于策略的生成场蒸馏）[01:34] 🔍 ViQ: Text-Aligned Visual Quantized Representations at Any Resolution（ViQ：任意分辨率下的文本对齐视觉量化表示）[02:32] 🎨 Qwen-Image-Agent: Bridging the Context Gap in Real-World Image Generation（Qwen-Image-Agent：弥合真实世界图像生成中的上下文鸿沟）[03:25] 🤖 OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning（基于策略的技能蒸馏：面向智能体强化学习的在线学习方法）[04:21] 🔍 The Verification Horizon: No Silver Bullet for Coding Agent Rewards（验证地平线：编码智能体奖励没有银弹）[05:04] 🚀 JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting（JetSpec：利用并行树草稿打破推测解码的扩展上限）[06:09] 🛠 Why Multi-Step Tool-Use Reinforcement Learning Collapses and How Supervisory Signals Fix It（为什么多步骤工具使用强化学习会崩溃以及监督信号如何修复它）[06:57] 🧩 Running the Gauntlet: Re-evaluating the Capabilities of Agents Beyond Familiar Environments（穿越障碍：重新评估智能体在陌生环境之外的能力）[07:50] 🔍 Confidence-Aware Tool Orchestration for Robust Video Understanding（面向鲁棒视频理解的置信感知工具编排）[09:10] 🖥 GUI vs. CLI: Execution Bottlenecks in Screen-Only and Skill-Mediated Computer-Use Agents（图形用户界面与命令行界面：纯屏幕操作与技能中介的计算机使用代理中的执行瓶颈）[10:11] 🤖 In-Context World Modeling for Robotic Control（面向机器人控制的上下文世界建模）[10:59] 🚀 Fast LeWorldModel（快速潜在世界模型）[11:47] ☕ CoffeeBench: Benchmarking Long-Horizon LLM Agents in Heterogeneous Multi-Agent Economies（咖啡基准：异构多智能体经济中长周期LLM代理的基准测试）[12:41] 🧊 PhysiFormer: Learning to Simulate Mechanics in World Space（PhysiFormer：在世界空间中学习模拟力学）[13:34] 🎲 Discretizing Reward Models（离散化奖励模型）【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