Hi,
I’ve been using ChatGPT for 10 months for a range of things and have generally come to use it for a range of admin tasks. I have the paid version which is great for creating resources for teaching in school as that is my main job. However, I would like to use it for music production admin.
There’s a couple of things I’d like it to do.
Organise my sample collections. I’ve got thousands of samples and it would be great to reorganise them in a more logical and accessible way.
Working in Ableton, I create a set with bounced audio that mirrors the structure of a master set. There are the same number of tracks in the sets and the names of the tracks are the same. I then copy the bounced clips over into the master so I end up with the master set containing different songs that I can play live.
It would be great to get ChatGPT to do some of the dull bouncing and copying stuff. How feasible is that?
Anyone using AI for music production admin - I’d love to know what you’ve got it doing for you.
I’ve done this kind of thing. Not total re-orgs, but things like ‘find all the percussion samples that are short’. I was using copilot (with claude sonnet). Now on opencode/sonnet. But any AI assistant that runs on your local machine and can actually see your files should work.
EDIT: e.g. claude cowork. But that’s not a recommendation, I’ve not used it.
EDIT2:
claude cowork anecdote from a friend
not coding but used Claude Cowork.
I downloaded my last 20 Ocado (online grocery delivery here in the UK), took photos of my larder, spices and fridge and put them all in a folder. Then I added a cooking philosophy.txt file with our dietary requirements, general cuisines we like to cook, dishes we prefer. It then constructed me a menu and went to Ocado and added relevant items to my basket and has created individual recipe cards for each day/dish. All this took about 3 hours for it to do (but it was running in the background so no problem for me!). Bloody brilliant.
In my experience, ChatGPT is not a particularly good AI, and it is created by one of the most evil organizations in the field.
I personally can’t recommend using that. There are better alternatives, may I suggest you take a look at Claude, it is a better choice both in terms of quality and company ethics.
Don’t bother. Use a dedicated tool like Sononym.
Ableton has a Python API, if it exposes the methods you need you can easily achieve what you want with Claude Code.
I use NotebookLLM so I don’t have to read manuals when researching devices.
My first prompt typically is:
briefly explain the basic paradigms of the device, and list 5-10 small features I am probably not aware of
And then I use it for checking more esoteric stuff in manuals. It’s still wrong about 20% of the time, but overall it saves me a lot of time.
Other than that, I used AI to downsize my studio (today I only have about a third of the gear I had 4 months ago), that was a surprisingly successful way of dealing with all the emotional baggage of letting go of things. The result is some cash, lots of free space, and streamlined production experience that incidentally also reduced the amount of admin work necessary for music production.
I don’t think it is wise to recommend openclaw to anyone, because most people do not understand the risks of this thing completely fucking up your life.
Notebook LLM on a android phone - is a super tipp, thank you ! Not sure if its running locally, but the answers are actually very good. The feature to create learning cards - is a nice idea. Not sure how well that runs, but the overall Q/A was very good.
I tested the learning cards aswell, and - its actually also very nice - you get a quiz and learn stuff about your equipment. Exciting.
They do. Documenting the API is a community effort (mostly from reverse-engineering the controller scripts), because Ableton won’t provide documentation.
That’s how Faderfox and other small manufacturers create the scripts for their controllers.
I tried this with Claude to organise a PDF library. The results were mixed, but part of it sort of depends on how you prompt. If you don’t give clear guidance it will make its own assumptions.
My suggestion would be to make a copy of your library and experiment with that, then if its all good you can replace the old one if you’re happy with the results (or have another shot with another copy).
Also - use the best model possible - it will likely nail it whereas the old ones might fumble around and you have to correct things. I found Claude would get VERY excited and run off and start actioning things before I’d given a ‘confirmed, please proceed’ - so be sure to set in your instructions that you do not want Claude to run off and start actioning things without your explicit permission.
