From what I can tell so far, it only really knows publicly-available algorithms and never knows how to implement them in the way you want. I’ve tried to get it to write converters (for the purpose I mentioned) along with even basic effects (like comb filters) and it fails every time. Once me and the bot refactor it a million times, it definitely runs without errors (although it’s like trying to get a kid to write code to begin with) but the effect isn’t even close (usually it just ear-rapes the shit out of you and does nothing else).
I think there will probably come a day when that stuff actually works, but in my experience it really only works for those generic questions that everyone has answered a million times that you can basically just copy and paste from Stack overflow if you really need to. It’s probably really good for people who can’t code to begin with (although I think that’s probably a bad thing over all, but who knows), but it’s not even remotely close for DSP. Even if you literally feed it an algorithm from Beads or Minim, it’ll never be able to connect the dots.
(That is the part where I’m failing, though, so your post is still correct. I just feel that if I were a little better at algorithms in general, just going on Wikipedia would be more useful than using a bot)
The only real successes I’ve had were processing images with the Processing language. It wrote me convolution matrices, edge detection and all sorts of crazy shit in the exact way I wanted, but that’s because the information is just chilling on forums and elsewhere in massive quantities and nothing had to be converted from one guy’s data structure to the next. But even after that, I got a crippling feeling of, “I need to learn these algorithms”.