Dongxi 东锡 NLP· @dongxi_nlp · X·· 2026-08-03AI 评分20
AI 导读
第31周论文 探索式建模——模型探索多条有效路径，每种可能性都保持清晰。 StateAct——智能体先读取应用状态，像素沦为后备方案，动作更加有据可依。 论文： 探索式建模：解锁第三条预训练轴与端到端生成 StateAct：程序状态优先于像素，用于长时程计算机操作智能体
正文 · 原文
Papers of Week 31
Explorative Modeling
The model explores many valid paths.
Each possibility stays sharp.
StateAct
Agents read app state first. Pixels become a fallback. Actions become more grounded.
Papers:
Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
来源：Dongxi 东锡 NLP · x.com