Testing ChatGPT for Violin Practice Planning: A Practice Lab Experiment

Over the winter holidays, I started looking into using ChatGPT for my violin practice planning. I have been using analog or digital methods of practice planning for years and I was interested in experimenting with using AI to help me plan my practice. While the AI generated practice plans weren’t perfect, I found many of them helpful and have continued to experiment with modifications and enhancements to my prompts over time.

This article is intended to document this ongoing experiment, and share what worked, what didn’t, and how musicians can start using AI as a practice aid.

The prompt

I asked ChatGPT to create a 60 minute practice plan for me. I provided the AI with the overall practice structure I wanted to follow, the techniques, and the repertoire that I wanted to work on. I received a practice plan that was feasible, fit within my allocated amount of practice time, and ensured I didn’t skip foundational work like scales.

While the resulting plan had some quirks, such as practice suggestions that didn’t align well with the exercises I told the model I wanted to practice, I found the plan to be a useful tool to for guiding my practice, especially on days when I am tired and likely to skip some of the more difficult technical work.

I started with the following very simple natural language prompt:

I’d like a 60 minute practice plan for violin. For my technical work, I’d like to start with a three octave scale. Then I’d like to play some position etudes from Whistler’s Introducing the Positions followed by vibrato exercises. Then, I'd like to work on a spiccato exercise from Ševčík, Op. 3/40 Variations. For my repertoire, I am currently studigy the second and third movements of Ysaÿe Sonata No. 4. I am just starting to work on the second movement. It has three main sections. I’d like to spend a little time on each section. I’ve been working on the third movement for longer. Mostly I’m working to play the movement more fluidly with fewer stops. At the end of practice I’d like to do a short sight reading session. 

Since my initial experiment, I have tried more advanced prompting techniques, including plans that rotate through scales based on day of week, and allocate time between technique and repertoire in a structured way. While the result still needs some refinement, you can find the full prompt on GitHub.

What worked

Throughout my experiment, I tried different prompts to help me plan practice sessions. They were all able to allocate time to technical exercises and repertoire that I wanted to practice. In many cases, the simpler prompts generated a more straightforward, usable output.

• Divided practice into distinct time-based sections focusing on specific technical work or repertoire
• Included the technical work and repertoire I specified
• Added reflection prompts for practice journaling when I asked
• Created a comprehensive practice plan covering scales, technique, and repertoire
• Produced a repeatable structure for daily practice
• Included specified priorities in the practice session

I found that ChatGPT was good at organizing practice around specified technical and repertoire work, and produced reasonable time allocations across practice areas.

Why I the prompt worked

• I specified the duration of the practice session
• I specified the materials and repertoire that I wanted to use during my practice
• I described practice goals for certain passages and movements

In general, my instructions were simple enough that the model didn’t produce an overly complex plan, but also detailed enough that the model could design my practice in alignment with my personal goals.

What didn’t work

While my AI generated practice plans were all useable with some slight modification, I found that ChatGPT wanted to provide helpful practice suggestions even when it did not have enough pedagogical insight to accurately provide any instruction.

ChatGPT often fell short when it moved from organizing practice to prescribing practice strategies. When I gave the AI exercises and pieces that I intended to practice, it often suggested how to practice certain difficult passages, frequently defaulting to the instruction “practice in rhythms,” where it didn’t always make sense to do this. In one case, it recommended reading one whole line ahead while practicing sight reading where reading one bar ahead may have been a more reasonable suggestion.

I did find that ChatGPT can help prioritize what to practice as long as you have a comprehensive list of your practice assignments. But, in some cases, the suggested time allotments in the AI practice plan didn’t always align with my own ideas, but this can be mitigated with more careful prompting. In one practice plan, it suggested that I work on viola setup and positioning for 5 minutes, something that had come up in my previous violin lesson, and I didn’t feel like it was necessary to spend as much time on this activity.

Improvements

Here are the next refinements I’d like to test in future versions of my prompts:

• Specify how much practice time to dedicate to technique versus repertoire.
• Limit practice instructions to those drawn from lesson notes or transcripts.
• Build in flexibility for different session lengths and focus areas.
• Include a method for cycling through scales by day of week or practice session
• Build week-long practice plans.
• Build a dedicated ChatGPT practice planning project to centralize prompts and context files.

Tips for prompting ChatGPT for music practice plans

• Be specific in your prompt. Rather than telling the AI that you want to work on spiccato in general, name an exercise or an exercise book that you are using. “I’d like to work on spiccato using exercise 13 from Sevcik 40 variations.”
• Be realistic about how much time you want to practice and what you want to focus on in a particular session. If you have many areas to focus on and a shorter available practice time, consider writing a prompt to develop a 7 day practice plan that rotates between a set of technical focal points rather than trying to get through each focal point in one practice session.
• Ask for extras if you want them. If you are trying to start practice journaling, ask the AI to write some reflective prompts to use in your journal. This is a great way to get inspiration on ways to reflect on practice.
• Provide additional context: Upload files showing previous practice plans or practice plan formats you’d like to emulate. Or alternately, outline your desired format in your prompt.
• Signal to the AI “Write this as a step-by-step plan I can follow when I open my music and start practicing.”

Conclusion

ChatGPT is good at organizing music practice, especially when given specific materials and goals, but weak at pedagogical instructions.

AI is just another tool for musicians to use, and using AI for practice planning is just one of many options. Many experienced musicians already have practice plans that work well for them, however AI generated practice plans can be useful for students, including adult learners, who might otherwise have difficulty prioritizing elements of practice or want a written plan to help keep them on track.

The value of using AI to generate practice plans was that it could turn my priorities into plans quickly. I found that having basic daily plans tied to my weekly practice goals can help keep me on track.

If you want to try this yourself, I’ve included a simple natural language prompt in the prompt section of this article to get started and a more robust practice planning prompt on GitHub.