Name of participant: Shuzhou Yuan
Project’s name: LLM4Edu: Learning Large Language Models for Education
Project description:
The project, Learning Large Language Models for Education (LLM4Edu), aims to advance the role of Large Language Models (LLMs) in educational settings through the creation of specialized datasets and the development of fine-tuned, education-oriented LLMs. LLM4Edu will construct a Chain-of-Thought (CoT) dataset that captures step-by-step reasoning processes, enabling models to guide students toward independent problem-solving rather than directly providing answers. Fine-tuning based on this dataset will focus on improving model patience, scaffolding of student reasoning, and alignment with pedagogical strategies.
An industrial partnership with Macmillan Learning, a leading educational platform where students interact with AI tutors, further grounds the project in real-world applications. Macmillan Learning provides interaction data sourced from authentic student-AI dialogues, offering valuable insights into student learning behaviors and expectations. This data will serve as the foundation for building the CoT dataset. Additionally, LLM4Edu will employ Large Language Models to augment and enrich the collected data, enhancing coverage, diversity, and the quality of reasoning paths represented.
Beyond dataset construction and fine-tuning, the project will also explore model architecture adaptations and evaluation protocols specifically suited for educational tasks. The ultimate goal of LLM4Edu is to create LLMs that are not only technically proficient but also pedagogically effective, capable of serving as patient, supportive, and adaptive AI tutors in diverse learning environments.
Software Campus Partner: TU Dresden and Holtzbrinck Publishing Group
Implementation period: 01.01.2025 – 31.05.2026
