I find it super interesting that were at a point in time where we would find it faster to ask a computer to randomly generate a short sequence of notes rather than just, like, writing them out.
This is a good example of a class of problem where it would be so much easier for me to describe any of several workflows:
punch it into Dorico directly, write it on some staff paper, put it in logic and have it output notes,
or just stick with Arban/Rabath/Rubank/Dotzaur/et al because, functionally, most useful possible permutations of exercises have been written out somewhere by some one (with the added benefit that they were probably written out by people thinking about speicfic issues like string crossings or the break on a clarinet)
Any of those seems infinitely faster it would be to figure out how to get a LLM to randomly stumble on the exercise based on its training data.
Even if I am a just a musician and not a tech bro, I'm happy that we live in a world where that seems like an easier route to randomly generate that material than to just write it out.
I had an engineering professor in college who talked about the amount of math he was able to do in his head early in his career before taking a job that needed a calculator. It wasn’t long after he took that job that he noticed he lost most of his advanced mental math skills.
On the music stuff specifically, writing it out (or figuring it out on a piano) is useful because you’re going to need to commit intervals to memory anyway if your goal is to be a serious musician/composer/whatever.
Just like you might want to practice reading a foreign language with paragraphs you haven’t seen before, not the paragraphs you just wrote yourself. Certainly a useful skill to be able to write yourself, but not necessarily the skill you want to practice every time you want to practice reading.
I don’t see the problem.
I'd love to have something that generates the same right hand picking pattern with a different left hand.
Also, I think there are licks that you have to be very precise with fingering or picking pattern. It would be cool to generate those.
The same reason you might want to use an LLM to generate language or a meal plan or anything really: it's interactive and you can customize it arbitrarily on the fly instead of browsing pages of content until something kinda lines up with what you want over and over again.
(Everybody else in my life uses go's on default "your direction is up" and it's worthless. You follow directions and have no idea which way you're going or how far :-()
We always wondered if we were missing out on some insight or intuition by not having to churn through the mathematical analysis they did because simulation was so easy for us. On one hand, the computer can tackle situations we could never come up with a good approximation for to studying analytically. On the other hand, maybe we’d be asking our computers smarter questions if we had to grind through that stuff by hand in our own research for a few years before getting high-quality simulations.
I’m an amateur and partially self-taught so I don’t know how common that is.
"Please give me a python script that writes the chromatic 7th intervals from C to B in eighth notes in a convenient output format like lilypond or musicxml."
The script it gave me worked perfectly, outputting lilypond to stdout and a musicxml file that imported into musescore with no issues.
This blog post has some .ABC files you can see as an example:
Anyway, anecdote time: My wife used to play random chords in the guitar. She explained me that if you follow some rules it would sound good. I don't remember the details, something like
X -> Y or Z
Y -> X or W
Z -> X
W -> X or Y
where X, Y, Z, W are actually A, B, C, D, E, F or G, but I can't remember which ones. (Did I mention I don't play the guitar?)
So I made a program that played in the speaker 4 notes per chord, and followed those rules to select the next chord. It was surprisingly good for the low effort I made. It was not fantastic or groundbreaking, but not horrible at least.