2026.10.08 | 递归状态按误差分配精度;世界动作模型扩展上下文。

2026.10.08 | 递归状态按误差分配精度;世界动作模型扩展上下文。

12分钟 ·
播放数50
·
评论数0

【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 www.xiaoyuzhoufm.com

【目录】
本期的 15 篇论文如下:

00:29 🗜 STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization(STEPQuant:Delta-Rule 递归状态量化中的误差何时与何处重要)
01:12 🤖 Long-WAM: Scaling the Context of World-Action Models(Long-WAM:扩展世界-动作模型的上下文)
01:53 🤖 nanoMuse: An Open-Source Personal Agent for Every Device You Own(nanoMuse:面向你拥有的每台设备的开源个人智能体)
02:33 🎮 Recursive Game Creator: An Agentic Product-Level Experience-Oriented Game Harness(递归游戏创作者:面向体验的智能体产品级游戏开发框架)
03:17 🧠 Questioning the Questions: Sustaining Self-Evolution in Reasoning Models(质疑问题本身:维持推理模型的自我进化)
04:09 ⚡ GRACE: Generation-aware latent compression for efficient video generation(GRACE:面向高效视频生成的生成感知潜在压缩)
04:48 🎭 DecepEval: A Benchmark for Evaluating Deception in LLM Agents(DecepEval:用于评估LLM智能体欺骗行为的基准)
05:33 🔮 VepAgent: Bridging Causal-Transition via Tool-Augmented Reinforcement Learning for Video Event Prediction(VepAgent:通过工具增强强化学习桥接因果转移以实现视频事件预测)
06:14 🎬 SGF+: Decoupling Gradient Flows for Autoregressive Video Generation(SGF+:解耦自回归视频生成的梯度流)
06:58 🤖 UniWAM: Unified World-Action Model(UniWAM:统一世界-动作模型)
07:36 🧠 Semifactual Credit-Augmented Policy Optimization(半事实信用增强策略优化)
08:15 🗂 RunningTab: Direct Workspace Interaction with Environment-Side Tabs(RunningTab:基于环境侧标签的直接工作区交互)
08:54 🔊 WorldSonus: Bringing Sound to Worlds(WorldSonus:为世界带来声音)
09:42 🧩 Tetris3D: 3D Scene Generation With Objects That Fit Together(Tetris3D:生成物体相互契合的3D场景)
10:25 🤖 ReSAIL: Mitigating Collapse in Iterative Agent Self-Distillation(ReSAIL:缓解迭代式智能体自蒸馏中的崩溃)

【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