2026.06.24 | Qwen-AgentWorld超越GPT-5.4；NatureBench揭示AI创新瓶颈

【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：[00:31] 🌍 Qwen-AgentWorld: Language World Models for General Agents（Qwen-AgentWorld：面向通用智能体的语言世界模型）[01:28] 🧪 NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?（NatureBench：编码智能体能否复现《自然》系列论文的已发表SOTA？）[02:24] 🤖 AOHP: An Open-Source OS-Level Agent Harness for Personalized, Efficient and Secure Interaction（AOHP：面向个性化、高效与安全交互的开源操作系统级智能体框架）[03:16] 📱 MobileForge: Annotation-Free Adaptation for Mobile GUI Agents with Hierarchical Feedback-Guided Policy Optimization（移动锻造：基于分层反馈引导策略优化的无标注移动GUI智能体自适应方法）[04:04] 🤖 MemGUI-Agent: An End-to-End Long-Horizon Mobile GUI Agent with Proactive Context Management（MemGUI-Agent：一种具有主动上下文管理的端到端长时移动GUI智能体）[04:57] 🧠 LingxiDiagBench: A Multi-Agent Framework for Benchmarking LLMs in Chinese Psychiatric Consultation and Diagnosis（灵犀诊断基准：用于评估大语言模型中文精神科咨询与诊断能力的多智能体框架）[06:01] 🔒 FedOT: Ownership Verification and Leakage Tracing via Watermarks for Federated LDMs（联邦扩散模型的所有权验证与泄漏追踪水印方法）[06:57] 🧠 OpenThoughts-Agent: Data Recipes for Agentic Models（开放思维智能体：用于智能体模型的数据配方）[08:03] 🤖 Escaping the Self-Confirmation Trap: An Execute-Distill-Verify Paradigm for Agentic Experience Learning（逃离自我确认陷阱：一种用于智能体经验学习的执行-提炼-验证范式）[09:02] 🔺 FLAT: Feedforward Latent Triangle Splatting for Geometrically Accurate Scene Generation（FLAT：用于几何精确场景生成的前馈潜在三角面片喷溅）[09:58] 🦃 Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning（文本到图像模型是归纳主义的火鸡吗？一个用于因果推理的反事实基准）[11:03] 🧪 DiffusionBench: On Holistic Evaluation of Diffusion Transformers（DiffusionBench：扩散变换器的全面评估基准）[11:49] 🚗 FlowR2A: Learning Reward-to-Action Distribution for Multimodal Driving Planning（FlowR2A：学习多模态驾驶规划中奖励到动作的分布）[12:53] 🔍 DREAM: Dense Retrieval Embeddings via Autoregressive Modeling（DREAM：通过自回归建模实现密集检索嵌入）[13:49] 🔍 ReMMD: Realistic Multilingual Multi-Image Agentic Verification for Multimodal Misinformation Detection（ReMMD：面向多模态 misinformation 检测的真实多语言多图像智能验证框架）【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