Scholarlink

The integration of Large Language Models (LLMs) holds significant promise for enhancing programming education by providing personalized, immediate feedback and fostering student engagement through adaptive learning. Existing research demonstrates that LLMs can offer meaningful learning support when designed with pedagogical considerations. However, challenges such as hallucinations in feedback and learners’ negative attitudes towards automated instructional solutions hinder broader adoption in educational contexts. Many existing solutions fail to promote critical thinking or address diverse learner needs. This paper proposes design of an interactive story-based environment for teaching programming languages, utilizing LLMs and automated assignment grading through a plug-in for Visual Studio Code (VSC). The goal is to provide contextual, pedagogically relevant tasks and feedback to students, employing Socratic questioning to encourage active participation and critical thinking. We adapt an existing VSC plug-in framework to support PHP and other common languages, designing middleware that enriches student prompts and redirects them to a custom-tuned curriculum-driven Swedish LLM. This architecture integrates meta-prompts based on pedagogical strategies into LLM interactions, employing a Socratic dialogue approach rather than providing direct answers. Anticipated outcomes include increased student engagement through storyline-based tasks and personalized feedback within the VSC environment, alongside better alignment of LLM interactions with pedagogical objectives. By presenting the underlying architecture of the prototype, we contribute to the use of generative AI in software engineering education. Our work highlights the potential of AI-powered tools in education to improve learning while addressing ethical considerations and ensuring need for thoughtful implementation to avoid amplifying biases or diminishing the role of teachers. Further studies are recommended to evaluate the impact of LLM interactions on student learning outcomes and to explore adaptability in real-life educational simulations.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Programming Education with LLMs and NPCs: A Dialogical Learning Framework for VS Code

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Andrey Shcherbakov,
  • Dennis Biström,
  • Truong An Pham,
  • Leonardo Espinosa-Leal

摘要

The integration of Large Language Models (LLMs) holds significant promise for enhancing programming education by providing personalized, immediate feedback and fostering student engagement through adaptive learning. Existing research demonstrates that LLMs can offer meaningful learning support when designed with pedagogical considerations. However, challenges such as hallucinations in feedback and learners’ negative attitudes towards automated instructional solutions hinder broader adoption in educational contexts. Many existing solutions fail to promote critical thinking or address diverse learner needs. This paper proposes design of an interactive story-based environment for teaching programming languages, utilizing LLMs and automated assignment grading through a plug-in for Visual Studio Code (VSC). The goal is to provide contextual, pedagogically relevant tasks and feedback to students, employing Socratic questioning to encourage active participation and critical thinking. We adapt an existing VSC plug-in framework to support PHP and other common languages, designing middleware that enriches student prompts and redirects them to a custom-tuned curriculum-driven Swedish LLM. This architecture integrates meta-prompts based on pedagogical strategies into LLM interactions, employing a Socratic dialogue approach rather than providing direct answers. Anticipated outcomes include increased student engagement through storyline-based tasks and personalized feedback within the VSC environment, alongside better alignment of LLM interactions with pedagogical objectives. By presenting the underlying architecture of the prototype, we contribute to the use of generative AI in software engineering education. Our work highlights the potential of AI-powered tools in education to improve learning while addressing ethical considerations and ensuring need for thoughtful implementation to avoid amplifying biases or diminishing the role of teachers. Further studies are recommended to evaluate the impact of LLM interactions on student learning outcomes and to explore adaptability in real-life educational simulations.