2026.05.15 | 30B模型刷奥赛金牌；自蒸馏让3B小模型零外挂超能

【目录】本期的 15 篇论文如下：[00:23] 🥇 Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling（通过简单且统一的缩放实现金牌级别的奥赛推理）[01:00] 🤖 Self-Distilled Agentic Reinforcement Learning（自蒸馏智能体强化学习）[01:46] 🧠 MemLens: Benchmarking Multimodal Long-Term Memory in Large Vision-Language Models（MemLens：大型视觉语言模型中多模态长期记忆的基准测试）[02:57] 👁 MemEye: A Visual-Centric Evaluation Framework for Multimodal Agent Memory（MemEye：面向多模态智能体记忆的视觉中心评估框架）[04:00] 🎬 SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer（SANA-WM：高效分钟级世界建模的混合线性扩散Transformer）[04:43] 🎬 Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation（因果强制++：面向实时交互式视频生成的可扩展少步自回归扩散蒸馏）[05:21] 🧬 Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning（达尔文家族：基于MRI信任加权进化合并的无训练语言模型推理扩展）[06:19] 🐾 WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation（WildClawBench：面向真实世界长周期智能体评估的基准）[07:11] 🧠 STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?（STALE：LLM代理能否知晓其记忆何时失效？）[08:03] 🧠 Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems（超越个体智能：基于LLM的多智能体系统中的协作、故障归因与自我进化综述）[08:44] 🎥 Warp-as-History: Generalizable Camera-Controlled Video Generation from One Training Video（扭曲即历史：从单个训练视频实现可泛化的相机控制视频生成）[09:24] 🧠 PREPING: Building Agent Memory without Tasks（PREPING：无需任务构建智能体记忆）[10:04] 🧭 RouteProfile: Elucidating the Design Space of LLM Profiles for Routing（RouteProfile：阐明用于路由的LLM配置文件设计空间）[10:49] 🧠 EvolveMem:Self-Evolving Memory Architecture via AutoResearch for LLM Agents（EvolveMem：面向LLM智能体的自演化记忆架构通过自动研究实现）[11:28] 🧠 ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both（ATLAS：是智能体推理还是潜在视觉推理？一个词就足够了）【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