1567

6 years later we still need to ask good questions in order to get LLMs to return the results we want, and it can be difficult. Music LLMs hardly ever return a good result and you have to keep regenerating them until they understand the question. 

Understanding a question doesn't necessarily mean it understands the topic it is answering a question about. It can spot patterns, but not always or consistently. It might get the answer half right, but in ways that isn't interesting. Or it can say something interesting but it doesn't work as a whole.

Using AI as a shortcut just to make it easier and more convenient isn't a reason to use it. In music it can generate lots of just-ok content, but it's at the expense of something better you might make if you plugged away at it yourself.

The danger is the shifting baseline of just accepting the mediocre, or stuff that sounds good as a recording but has no real substance when you look under the hood.