2026.07.02 | 感知准则揭示AI可靠性鸿沟；TurboServe实现流式视频生成降本增效

【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：[00:33] 🔍 PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception（感知准则：将多模态评估校准至人类感知）[01:34] 🎬 TurboServe: Serving Streaming Video Generation Efficiently and Economically（TurboServe：高效经济地提供流式视频生成服务）[02:30] 🧠 MemSyco-Bench: Benchmarking Sycophancy in Agent Memory（MemSyco-Bench：评估智能体记忆中的谄媚行为基准）[03:26] 🚀 ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving（ELDR：面向PD分离式MoE服务的高效专家局部感知解码路由）[04:25] 🧮 Domain Arithmetic: One-Shot VLA Adaptation under Environmental Shifts（领域算术：环境变化下的单样本视觉-语言-动作模型适配）[05:15] 🧠 Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning（基于非对称互变分学习的多模态连续推理）[06:05] 🔬 CausalMix: Data Mixture as Causal Inference for Language Model Training（因果混合：将数据混合视为语言模型训练的因果推断问题）[07:04] 🔍 Perceive-to-Reason: Decoupling Perception and Reasoning for Fine-Grained Visual Reasoning（感知到推理：解耦感知与推理以进行细粒度视觉推理）[07:58] 🤖 ABot-M0.5: Unified Mobility-and-Manipulation World Action Model（ABot-M0.5：统一的移动与操作世界动作模型）[08:49] 🌱 Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity（Seed2.0 模型卡：迈向应对真实世界复杂性的智能前沿）[09:41] 🤖 ASPIRE: Agentic /Skills Discovery for Robotics（ASPIRE：面向机器人的智能体技能探索与自主编程）[10:28] 🧬 BioInsight: Multi-Agent Orchestration for Interactive Biomedical Knowledge Discovery（生物洞察：面向交互式生物医学知识发现的多智能体编排系统）[11:31] 🔀 The State-Prediction Separation Hypothesis（状态预测分离假说）[12:24] 🤖 AutoTrainess: Teaching Language Models to Improve Language Models Autonomously（AutoTrainess：教语言模型自主改进语言模型）[13:20] 🌍 Valdi: Value Diffusion World Models（Valdi：价值扩散世界模型）【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