发表机构
KTH Royal Institute of Technology（皇家理工学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对GenAI在高等教育中削弱深度学习的悖论，提出NuPaD框架，通过结构化问题学习协议与专用教科书，将GenAI转化为个性化导师，促进亚原子物理的深度学习。
AI 中文摘要
生成式人工智能（GenAI）在高等教育中的迅速普及引入了一个关键的教学法悖论：尽管这些系统拥有非凡的信息检索与综合能力，但其默认的即时提供未经请求的答案的操作模式，却积极削弱了构建真正科学理解所依赖的认知过程。本文提出了NuPaD（核物理与粒子物理——深度学习导师），这是一个专为研究生阶段亚原子物理课程设计的新型教学框架。对于此类高级课程，教学自然转向探究驱动、基于问题的学习，这使其成为利用GenAI探索复杂、开放性问题而非仅仅查询既定事实的理想环境。该框架由三个紧密耦合且易于使用的组件构成：一个强制执行结构化基于问题学习协议的主要智能体指令文件；一本为隐私保护的本地GenAI精确解析而优化的专用教科书；以及一份弥合教科书与GenAI之间知识鸿沟的简洁配套文件，同时还包括一套用于专门任务的综合动态技能集。我们详细阐述了模块化NuPaD框架的架构原则、Markdown原生教科书格式的设计理念以及底层的GenAI调控原则。通过重新定义学生与模型之间的交互循环，该框架将GenAI从被动的答案引擎转变为主动的个性化导师，确保其加速而非绕过深度学习与科学推理的发展。
英文摘要
The rapid adoption of generative artificial intelligence (GenAI) in higher education has introduced a critical pedagogical paradox: while these systems possess extraordinary capacity for information retrieval and synthesis, their default operational mode of supplying immediate, unprompted answers actively undermines the cognitive processes upon which genuine scientific understanding is built. This paper presents NuPaD (Nuclear \& Particle Physics -- Deep Learning Tutor), a novel pedagogical framework designed for the graduate-level subatomic physics curriculum. For such advanced courses, instruction naturally shifts toward inquiry-driven, problem-based learning, making it an ideal environment to use GenAI to explore complex, open-ended physical questions rather than merely querying established facts. The framework consists of three tightly coupled and easy-to-use components: a primary agent instruction file that enforces a structured problem-based learning protocol, a purpose-built textbook optimized for precise parsing by privacy-preserving local GenAI, and a concise companion file that bridges the knowledge gap between the textbook and GenAI, alongside a comprehensive dynamic skill set for specialized tasks. We explain in detail the architectural principles of the modular NuPaD framework, the design philosophy of the Markdown-native textbook format, and the underlying GenAI regulation principles. By redefining the interaction loop between student and model, this framework transforms GenAI from a passive answer engine into an active, personalized tutor, ensuring that it accelerates rather than bypasses the development of deep learning and scientific reasoning.
