ScholarPulse 日报 2026-07-26

2026-07-26 学术简报：2 篇。本地部署的开源大语言模型在纵向研究数据准备任务中表现优异，平均任务完成率达87.9%，为治理受限研究提供可行路径。

今日速览

序号	标题	来源	日期	主题	推荐等级
1	Agentic coding without the cloud: evaluating open-weight large language models on longitudinal data preparation tasks	arXiv	2026-07-23	RAG	高
2	PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning	arXiv	2026-07-23	AI-Agent	高

重点论文与技术动态

1. Agentic coding without the cloud: evaluating open-weight large language models on longitudinal data preparation tasks

来源：arXiv
日期：2026-07-23
作者/机构：Mack Nixon, Liam Wright, Yevgeniya Kovalchuk, Alison Fang-Wei Wu, Martin Danka, Andy Boyd
主题标签：RAG, arXiv
推荐等级：高
分类：cs.AI, cs.CL

一句话结论

本地部署的开源大语言模型在纵向研究数据准备任务中表现优异，平均任务完成率达87.9%，为治理受限研究提供可行路径。

核心内容

研究聚焦纵向人口研究数据准备瓶颈，构建开源框架评估AI代理效能，涵盖类别调和与多波合并等任务定义。
框架基于英国队列研究六波数据清洗脚本，包含自动化评估工具验证LLM生成的R代码及输出数据。
31-35B参数模型在20个任务（创建102变量）中平均完成率达87.9%，接近基准上限，证明消费级硬件可行性。

方法与数据

数据集源自英国队列研究清洗脚本，任务定义明确涵盖数据准备关键操作。
摘要未明确其他数据细节。

价值判断

值得关注：避免敏感数据外传，符合研究治理要求，解决云端传输约束。
可复用点：框架开源（GitHub链接），可直接复用于其他治理受限研究的数据准备任务。
局限/待核查：摘要未明确提及具体局限。

摘要

Large language models (LLMs) and agents are now widely used tools in code development, with data typically sent to third-party cloud-based models. Their adoption in research using personal data is constrained by governance requirements that typically prohibit data transmission to external services. Locally deployable open-weight models offer an alternative since sensitive data never leave the local environment. We introduce an open-source framework for evaluating the efficacy of AI agents powered by open-weight LLMs on one of the most persistent bottlenecks in research on longitudinal population studies: data preparation. The framework comprises: a curated ground-truth dataset (cleaning scripts preparing six sweeps of data from a British cohort study), task definitions encompassing tasks such as category harmonization and multi-wave merging, and automated routines for evaluating the LLM-produced R code and outputted data. We benchmark LLMs across the (consumer grade) deployment spectrum to assess their efficacy in 20 data preparation tasks (creation of 102 variables). Current state-of-the-art, 31-35B parameter models almost saturated our benchmark ("average task completion" up to 87.9%). The performance of open-weight LLMs running on consumer-grade hardware shows promise of a viable path toward AI-assisted data preparation in governance-restricted research settings. Our framework is publicly available at: https://github.com/UCL-ARC/RRBench.

2. PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning

来源：arXiv
日期：2026-07-23
作者/机构：Yipeng Shi, Zhipeng Ma, Yue Wang, Qitai Tan, Yang Li, Peng Chen
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.AI

一句话结论

PATS通过政策感知动态训练脚手架显著提升了长时域LLM代理强化学习的性能。

核心内容

弱策略在长时域RL中重复失败，产生无信息轨迹，限制有效优化。
PATS将技能重构为动态训练脚手架，将策略轨迹转换为证据卡，通过任务特定评估动态调整后续上下文。
随策略改进，移除冗余上下文以减少依赖，优化策略使用环境奖励，部署时丢弃脚手架。

方法与数据

方法：动态调整上下文以适应策略进化，使用任务特定评估优化轨迹。
数据：摘要未明确具体数据集细节。

价值判断

值得关注：在ALFWorld和WebShop上性能提升最高18.6%，在七个搜索增强QA基准上减少32.1%提示令牌。
可复用点：动态训练脚手架框架可推广至其他代理强化学习任务。
局限/待核查：摘要未明确局限。

摘要

In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization. Existing skill-centric methods improve exploration by optimizing, filtering, or internalizing reusable skills. However, they remain centered on the skills themselves rather than being designed as adaptive training-time support for the evolving policy. To address this, we propose a policy-centric training paradigm that reframes skills as a dynamic training scaffold. Our framework, Pats, converts rollout groups from the latest policy into evidence cards and uses task-specific evaluation to adjust the context used in subsequent rollouts. Concrete guidance helps weak policies to complete challenging tasks. As policy improves, redundant context is revised or removed to reduce reliance on explicit guidance while preserving useful rollout variation. The policy is optimized with environmental rewards using standard RLVR, and the training scaffold is discarded at deployment. On ALFWorld and WebShop, Pats improves over strong baselines by up to 18.6%. Across seven search-augmented QA benchmarks, it remains competitive while using 32.1% fewer prompt tokens than the baseline.