Building Educational Chatbots: Project Overview

This lecture presents the project description for building educational chatbots tailored for EPFL courses. The instructor discusses the context of the project, highlighting the relevance of ChatGPT and its underlying technologies, such as transformers and pretrained language models. Students are encouraged to design systems that incorporate skills like fine-tuning, prompting, and evaluation. The project is structured into three main stages: collecting preference data, training a generator model, and improving the model. In the first stage, students will gather ranked pairs of responses from ChatGPT to create a dataset. The second stage involves training a generator model using the collected preferences, while the final stage focuses on refining the model to specialize in multiple-choice questions. The instructor also outlines the advising policy, available computational resources, and grading criteria, emphasizing collaboration and equal contribution among team members. This comprehensive overview equips students with the necessary framework to successfully complete their projects.