But you still need to buy the sheet music first, all the AI Labs used pirated materials to learn from.
There's two angles to the lawsuits that are getting confused - the largest one from the book publishers (Sarah Silverman et al) attacked from the angle that the models could reproduce copyrighted information. This was pretty easily quelled / RHLF'd out (used to be that if ChatGPT started producing lyrics a supervisor/censor would just cut off it's response early - tried it now and ChatGPT.com is now more eloquent, "Sorry, I can't provide the full lyrics to "Strawberry Fields Forever" as they are copyrighted. However, I can summarize the song or discuss its themes, meaning, and history if you're interested!")
But there's also the angle of "why does OpenAI have Sarah Silverman's book on their hard drive if they never paid her for it? This is the lawsuit against Meta regarding books3 and torrenting, seems like they're getting away with the "we never redistributed/seeded!" but it's unclear to me why this is a defense against copyright infringement.
Not only would the musician have to buy the sheet music first, but if they were going to perform that piece for profit at an event or on an album they'd need a license of some sort.
This whole mess seems to be another case of "if I can dance around the law fast enough, big enough, and with enough grey areas then I can get away with it".
As a student in a school band that debated whether to choose Pirates of the Caribbean vs Phantom of the Opera for our half time show, I remember the cost of the rights to the music was a factor in our decision.
The school and library purchased the materials outright, again, OpenAI Meta et al never paid to read them, nor borrowed them from an institution that had any right to share.
I'm a bit of an anti intellectual property anarchist myself but it grinds my gears that, given that we do live under the law, it is applied unequally.
There's two angles to the lawsuits that are getting confused - the largest one from the book publishers (Sarah Silverman et al) attacked from the angle that the models could reproduce copyrighted information. This was pretty easily quelled / RHLF'd out (used to be that if ChatGPT started producing lyrics a supervisor/censor would just cut off it's response early - tried it now and ChatGPT.com is now more eloquent, "Sorry, I can't provide the full lyrics to "Strawberry Fields Forever" as they are copyrighted. However, I can summarize the song or discuss its themes, meaning, and history if you're interested!")
But there's also the angle of "why does OpenAI have Sarah Silverman's book on their hard drive if they never paid her for it? This is the lawsuit against Meta regarding books3 and torrenting, seems like they're getting away with the "we never redistributed/seeded!" but it's unclear to me why this is a defense against copyright infringement.