By Emanuela Furfaro (University of Washington) and Simone Mosciatti
Information
LLteacher is an open-source AI-based tool that we designed to help students work through their homework assignments using a guided large language model (LLM). It has an instructor-facing side, where instructors can write exercises along with solutions and access students’ interactions with the LLM, and a student-facing side, where students interact with the LLM to solve assigned problems. LLteacher is instructed to never provide answers to students, but rather to prompt them with questions and reflections that help them arrive at the correct solution. Throughout the assignment, students engage in a conversation with the LLM, enabling assignments that go beyond traditional homework while ensuring that the LLM’s responses are grounded in the instructor’s solutions. The instructor interface allows instructors to easily customize the tool by adjusting the level of help the LLM provides to support distinct pedagogical goals, such as recalling prior knowledge, discovering new concepts, and critically evaluating LLM outputs. In addition, by providing feedback on students’ ideas and attempts, LLteacher helps students who make errors, putting them back on the right track rather than penalizing them. LLteacher is an open-source project, with code written in Python using the Django framework. In this demonstration, we show how to use LLteacher from both the instructor and student perspectives, using a medium-sized undergraduate statistical computing class in R as context. We provide examples of different types of assignments that we created in this course and illustrate how we reviewed students’ interactions.