South KoreaEducation2024 – ongoing

KAIST

LLM virtual teaching assistant for graduate AI programming

KAIST evaluated a virtual teaching assistant through 3,869 classroom interactions

WorkingIndependent study

Problem

A graduate AI programming course enrolled students with different academic backgrounds. Students needed help with course content and coding outside live teaching sessions.

Approach

KAIST researchers deployed a virtual teaching assistant during the fall semester of 2024. It retrieved course materials and used GPT-4o-mini to answer questions through a web interface. Human instructors retained responsibility for critical and grade-related questions.

Results

3,869student-assistant question-response interactions analyzed in the classroom studyIndependent study[1]

477students enrolled in the course, rather than confirmed assistant usersIndependent study[1]

Takeaway

The deployment measured use and student perceptions. Interaction volume did not establish an improvement in learning outcomes.

Sources

A Large-Scale Real-World Evaluation of an LLM-Based Virtual Teaching Assistant
Association for Computational LinguisticsJul 2025Research paper or report

Source summaries

[1]Association for Computational LinguisticsJul 2025

KAIST researchers reported a classroom deployment and analyzed 3,869 interactions. The course enrolled 477 students, and the evaluation included surveys of student perceptions.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026