2026.06.30 | 实时编辑视频流；巧用视野胜参数。

【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] 🎬 LiveEdit: Towards Real-Time Diffusion-Based Streaming Video Editing（LiveEdit：迈向基于实时扩散的流式视频编辑）[01:20] 🧠 Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent（扩展智能体视野而非参数规模：以35B智能体达到万亿参数级性能）[02:15] 🤔 Agentic Abstention: Do Agents Know When to Stop Instead of Act?（智能体式弃权：智能体知道何时该停止而非行动吗？）[03:09] 💻 TUA-Bench: A Benchmark for General-Purpose Terminal-Use Agents（TUA-Bench：面向通用终端操作代理的基准测试）[04:12] 🗿 Trimming the Long-Tail of Visual World Modeling Evaluation（修剪视觉世界模型评估中的长尾分布）[05:06] 🧠 Video-MME-Logical: A Controlled Diagnostic Benchmark for Video Temporal-Logical Reasoning（视频MME-逻辑：一个用于视频时序逻辑推理的受控诊断基准）[05:57] 📊 Beyond IID: How General Are Tabular Foundation Models, Really?（超越独立同分布：表格基础模型的泛化能力究竟如何？）[06:40] 🏭 AsyncOPD: How Stale Can On-Policy Distillation Be?（异步OPD：策略蒸馏可以容忍多旧的数据？）[07:38] 🧠 ReFreeKV: Towards Threshold-Free KV Cache Compression（ReFreeKV：迈向无阈值KV缓存压缩）[08:38] 📱 Monte Carlo Energy Aggregation for Mobile 3D Gaussian Splatting（面向移动端三维高斯泼溅的蒙特卡洛能量聚合方法）[09:39] 🔧 TACO: Tool-Augmented Credit Optimization for Agentic Tool Use（工具增强信用优化：面向智能体工具使用的GRPO变体）[10:33] 🎥 Bridging VideoQA and Video-Guided Agentic Tasks via Generalized Keyframe Extraction（通过广义关键帧提取桥接视频问答与视频引导的智能体任务）[11:36] 🔍 Interleaved Speech Language Models Latently Work In Text（交错式语音语言模型在文本中潜在地工作）[12:27] 🤖 OSWorld2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks（OSWorld2.0：面向长时间跨度的真实世界任务的计算机使用智能体基准测试）[13:15] 🌍 DreamForge-World 0.1 Preview: A Low-Compute Real-Time Controllable World Model（DreamForge-World 0.1 预览版：一种低计算量、实时可控的世界模型）【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