2026.06.10 | 快手可灵长视频理解新突破；ABot-Earth三维生成仅需十分钟

【目录】本期的 15 篇论文如下：[00:32] 🎥 Kwai Keye-VL-2.0 Technical Report（快手可灵-VL-2.0技术报告）[01:24] 🌍 ABot-Earth 0.5: Generative 3D Earth Model（ABot-Earth 0.5：生成式三维地球模型）[02:13] 🤖 Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution（角色代理：通过双角色进化引导LLM代理）[03:08] 🔧 Retrospective Harness Optimization: Improving LLM Agents via Self-Preference over Trajectory Rollouts（回顾性装备优化：通过轨迹展开上的自我偏好改进LLM智能体）[04:06] 🐝 SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research（搜索蜂群：面向长周期深度研究的代理型大语言模型委派智能）[05:02] 🎥 MemDreamer: Decoupling Perception and Reasoning for Long Video Understanding via Hierarchical Graph Memory and Agentic Retrieval Mechanism（MemDreamer：通过分层图记忆与智能检索机制解耦感知与推理实现长视频理解）[06:05] 📊 Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories（数据记者智能体：将数据转化为可验证的多模态故事）[07:01] 🎭 SCAIL-2: Unifying Controlled Character Animation with End-to-end In-Context Conditioning（SCAIL-2：通过端到端上下文条件控制统一受控角色动画）[07:58] 🔀 Flow-DPPO: Divergence Proximal Policy Optimization for Flow Matching Models（Flow-DPPO：面向流匹配模型的散度近端策略优化）[09:05] 🏅 WorldOlympiad: Can Your World Model Survive a Triathlon?（世界奥林匹克：你的世界模型能经受三项赛考验吗？）[10:00] 🎯 Rethinking the Divergence Regularization in LLM RL（重新思考大语言模型强化学习中的散度正则化）[10:56] 💋 Lip Forcing: Few-Step Autoregressive Diffusion for Real-time Lip Synchronization（唇部强制：用于实时唇部同步的少步自回归扩散）[11:57] 🤖 EEVEE: Towards Test-time Prompt Learning in the Real World for Self-Improving Agents（EEVEE：面向真实世界测试时提示学习的自改进智能体）[12:52] 🧠 One Token per Multimodal Evidence: Latent Memory for Resource-Constrained QA（每多模态证据一个令牌：面向资源受限问答的潜在记忆）[13:49] 🤖 Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic tasks in Real-World Professional Fields（工作流健身房：面向真实世界专业领域的长周期计算机使用代理任务评估）【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【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在小宇宙查看该单集文稿