about
projects
blog

Music Tutor Project

I have been teaching fully online courses since 2019. Over the past five years, I've found that the model that I started with, no longer seems to apply in the modern classroom. When I first received by certification to teach fully online courses, we were presented with a series of requirements and best practices with regards to accessibility, contact hours, etc. For the most part, my online courses followed a certain template.


Each week, students would have a module to complete. They would start by watching some lecture videos and reading assignments. Along the way, students would have to complete quizzes, listening assignments (more quizzes), and some form of discussion or written response assignment. I tried my best to make these assignments fun and educational (such as, write your own blues!). For some of the discussion assignments, I would ask a somewhat open ended question about a particular moment in history, and ask for the students thoughts on the matter. Of course, I started to run into trouble with these assignments specifically when Chat GPT 3.5 came out.


Essentially, ChatGPT teachers of online courses in a strange place. As a teacher, we can almost always tell if a student has plagiarized, and it was usually easy enough to even find proof of the plagiarism. However, when ChatGPT started being used by students to answer my discussion prompts, it presented an even more complicated problem. I could tell that the student didn't write the response, and since I was also using ChatGPT quite a bit, I could tell by the style of the writing, that it was coming from ChatGPT. However, there wasn't really a way to prove definitively that the response was coming from an LLM. Apps like GPT-Zero started popping up, as a way to shed light on the issue, but I decided that I wanted to avoid a punitive approach.


Rather than include spooky language in more course syllabus about the consequences of using LLMs to get out of doing the written assignments, I wanted to encourage students to use these tools in order accelerate their learning. I had spend much of the past year working on what was essentially a RAG application for You Tube videos, and I wanted to leverage some of that experience to make a sort of educational helper for my lecture courses. What if students could have free access to an LLM that could help explain any difficult concepts from the course materials. I could create embeddings the textbook, and maybe even expand to my recorded lecture transcriptions. My ideas around how to incorporate the listening components are still a little fuzzy currently, but I am excited for where those challenges can take the app as well.

Essentially, my music tutor app allows students, who are enrolled in my courses, to login, and chat with an assistant about the course materials. After the tutoring session, the students can take a quiz, to check their understanding of the material, and receive feedback from the LLM, based on their written responses. I have disable the paste feature in the UI, which of course is not fool proof, but will make it more of a pain for students to just copy and past another ChatGPT response to the quiz question. I am still brainstorming ideas for how the app will deal with the musical components, but again, I am excited to see what solutions to try out!