Rarely do we see judges actually doing their jobs, but this is one of those cases.
Indeed, lyrics cases are among the clearest ones as an infringing output is a direct way to prove both that the language model illegally reproduced someone’s lyrics during training, as well as copied same lyrics unchanged in the output. This means that the AI corps are on the hook more than the users when using unlicensed material. So now, they have to pay compensation.
I never trusted GPT from the outset (for many technical and non-technical reasons) and I never used it, but I now trust it even less and will absolutely avoid to work with companies that use it a lot. Intuitively, from pure observation of how it affects users, ChatGPT seems programmed to sabotage the user. It is like a test of identifying the most masochistic users out there, compiling a database of their vulnerabilities and most secret fears, for the purpose of exploiting said vulnerabilities against the poor users.
At the end of the day, when you steal, at least try to change the fucking lyrics. Everyone does it. I don’t understand how ChatGPT can be this incompetent. It’s gotta be intentional.