AI Meets Mathematics Education: A Case Study on Supporting an Instructor in a Large Mathematics Class with Context-Aware AI
A fine-tuned AI model achieved 75.3% accuracy answering student questions, with over a third of its responses rated equal to or better than the instructor's.
A team of researchers from EPFL and the University of Lausanne has published a case study demonstrating how a fine-tuned AI can effectively support instruction in a large university mathematics course. The study, focused on a Calculus I class, involved developing a system to answer students' questions on a discussion forum. The core of the system was a lightweight language model, fine-tuned on a dataset of 2,588 historical student-instructor Q&A pairs to ensure its responses were pedagogically aligned with the course's specific context and materials.
When benchmarked against 150 representative questions annotated by five instructors, the AI model achieved an accuracy of 75.3%. Notably, in 36% of cases, the AI's responses were rated by experts as equal to or better than the original instructor answers. A post-deployment survey of 105 students revealed they highly valued the system's immediate availability and its consistency with course content, though they still relied on instructor verification for final trust. The research underscores the viability of a hybrid human-AI workflow, where AI handles scalable, immediate support, freeing the instructor to focus on complex interventions and verification, thereby enhancing the learning ecosystem in large-enrollment settings.
- The AI model was fine-tuned on 2,588 historical Q&A pairs from the specific Calculus I course for context-aware responses.
- It achieved 75.3% accuracy on an expert-annotated benchmark, with 36% of its answers rated equal to or better than the instructor's.
- A survey of 105 students confirmed the tool's value for immediate, aligned support, while highlighting the need for a trusted human-in-the-loop.
Why It Matters
This model provides a scalable blueprint for using reliable, course-specific AI to augment teaching support in large classes, improving student access without replacing instructor expertise.