Reproduce “Towards Just-in-Time Adaptive Feedback: Enhancing Student Learning via Knowledge-Grounded LLM” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Younghun Lee, Amir Bralin, Nobel Sanjay Rebello, Dan Goldwasser
Researchers at Purdue University developed a knowledge-grounded Large Language Model framework that delivers Just-in-Time adaptive feedback on student strategy essays in large-scale STEM courses. This intervention reduced the error rate on a targeted quiz question from over 50% to less than 10%, demonstrating how AI can overcome the feedback bottleneck and promote deeper conceptual understanding.
Key insights
- Students, regardless of their proficiency, predominantly preferred LLM feedback targeted at "novice-level" learners, emphasizing clarity and concept understanding over linguistic complexity.
- Domain-knowledge grounding through expert annotations was crucial for effective feedback and learning outcomes, even when the LLM's direct error type classification accuracy was suboptimal.
- Students voluntarily engaged in iterative conversations with the LLM, and the degree of revision in their strategy essays showed a positive correlation with achieving correct understanding.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.