2026.07.01 | Orca世界模型验证状态预测范式；Dockerless实现无环境代码验证

【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] 🌍 Orca: The World is in Your Mind（虎鲸：世界在你心中）[01:30] 🧪 Dockerless: Environment-Free Program Verifier for Coding Agents（无Docker：面向编码智能体的无环境程序验证器）[02:22] 🎭 DOPD: Dual On-policy Distillation（双在线策略蒸馏）[03:20] 🚀 BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding（BlockPilot：基于扩散的推测解码的实例自适应策略学习）[04:09] 🧩 Scenes as Objects, Not Primitives: Instance-Structured 3D Tokenization from Unposed Views（场景即对象，而非基元：基于未标定视图的实例结构化3D分词化）[05:00] 🎨 GEAR: Guided End-to-End AutoRegression for Image Synthesis（GEAR：引导式端到端自回归图像合成）[05:59] 🧩 Multi-Block Diffusion Language Models（多块扩散语言模型）[06:42] 🧬 Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks（进化微调：学习在371个优化任务中发现）[07:33] 🔧 SkillHone: A Harness for Continual Agent Skill Evolution Through Persistent Decision History（SkillHone：基于持久决策历史的持续智能体技能演进框架）[08:22] 🎥 MemLearner: Learning to Query Context memory for Video World Models（MemLearner：学习为视频世界模型查询上下文记忆）[09:13] 🧠 Managing Procedural Memory in LLM Agents: Control, Adaptation, and Evaluation（LLM智能体中的程序性记忆管理：控制、适应与评估）[10:08] 🧠 DataEvolver: Self-Evolving Multi-Agent Data Construction for Text-Rich Image Generation（DataEvolver：面向文本丰富图像生成的自我进化多智能体数据构建框架）[11:10] 🔊 RedVox: Safety and Fairness Gaps in Speech Models Across Languages（RedVox：跨语言语音模型中的安全性与公平性差距）[12:00] 🧠 Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs（基于元认知反馈的强化学习激发大语言模型可靠的置信度表达）[12:56] 🧠 Little Brains, Big Feats: Exploring Compact Language Models（小大脑，大成就：探索紧凑型语言模型）【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