Project Goals/Deliverables
Creation of a teachable large language model (LLM) companion that can be used by students in any course.
Creation of a new "learning by teaching" framework.
Creation of a new evaluation method: exam to test the concepts learned by an LLM companion, indicative of the concepts mastered by the student who taught the LLM.
Implementation of the teachable LLM companion and the "learning by teaching framework" in two course trials: one in computer science and another in psychology.
Publication of two research articles: one article in computer science describing the new teachable LLM companion and one article in psychology reporting the results of the evaluation of the teachable LLM companion and the "learning by teaching" framework.
Project Team
Pascal Poupart, Professor (Computer Science)
Igor Grossmann, Professor (Psychology)
Prashanth Arun, Graduate Student (Computer Science)
Brent McCready-Branch, Undergraduate Research Coordinator (Teaching Innovation Incubator)
Erya Xu, Undergraduate Research Coordinator (Teaching Innovation Incubator)
Relevant Resources
Eliciting self-explanations improves understanding (Chi et al., 1994).
The relative benefits of learning by teaching and teaching expectancy (Fiorella & Mayer, 2013).
Wisdom in context (Grossmann, 2017).
Explaining contentious political issues promotes open-minded thinking (Elnakouri, 2024).
Mind in Society: Development of Higher Psychological Processes (Vygotsky, 1978).
Project Resources
There are currently no project resources. Once they become available, they will be posted here.
Connect with the Learning by Teaching a Large Language Model team!
Inquiries about the Learning by Teaching a Large Language Model can be directed to tii@uwaterloo.ca
Project Summary
This project will develop and evaluate a large language model (LLM) companion that students will use to enhance their learning experience. Unlike most AI systems that engage in some form of tutoring or question answering, we will flip the roles. The LLM agent will act as an ignorant virtual student, and the human student will be responsible for teaching concepts to the LLM companion. The human student will effectively learn by explaining concepts to the LLM - a key element of effective learning and probing one’s knowledge gaps. Plus, it could be a lot more interesting and engaging than traditional forms of learning.
We will design a special exam to test the concepts learned by the LLM companion, indirectly assessing whether the human student also knows those concepts. In a randomized control trial in the classroom, we will further assess student self-efficacy, conceptual knowledge, and general engagement in the class.
Proposed Project Impact
The LLM companion will enhance the learning experience of students in two courses (one in computer science and one in psychology). Assuming the trials are successful, at the end of the project, the system will be made available to instructors of courses throughout the University of Waterloo and beyond. A startup may be launched.
Project Updates
Project updates to be shared here!