Firstly, we are very impressed with the musical productions produced via AI (including synth vocals). The legal and other ramifications have been amply discussed here so that is not my subject.
What has totally failed to impress is the accuracy of music related technical inquiries to AI sources. As a production geek and an old guy I often get a bit overwhelmed learning new products. For example moving older product presets to the latest version can be tedious at times. EVERY inquiry I made using AI for help on this subject has been wrong even though I clearly stated the products under question and what I was attempting to accomplish. Every one. From using outdated sources to simply stating incorrect methodology.
That piqued my interest and I started making test inquiries ... all related to music production. A conservative estimate would be that 75% of them were simply wrong.
Having touched on the above I will also mention that medical related inquiries have also been wrong. While I would never trust AI for such I do feel that it is dangerous for those less inquisitive. From our love of outdoors/hiking to music production to mountain biking to our love of historical architecture we often get erroneous answers that are sometimes hilarious.
On a positive note

I have a friend who owns several tech companies that specialize in healthcare. Compliance with privacy regulations is a big issue in the healthcare field. I'm very aware of it having spent my career in healthcare. He used AI to write a book about how to deal with the issue. BUT for that operation he provided all the vetted sources on the subject he could find and AI "wrote" the book using only that multitude of sources and their references. The result was incredible. He and others read every word and made practically no changes ... even to footnotes. So AI can "write" when it knows what it's "talking" about! I would assume this would apply to any application of it where the user is supplying the sources, e.g., one or multiple websites, etc.
Bud