exactly how the new work is created is important when it comes to derivative works.
does it use a copy of the original work to create it, or a vague idea/memory of the original work’s composition?
when i make music it’s usually vague memories. i’d argue that LLMs have an encoded representation of the original work in their weights (along with all the other stuff).
but that’s the legal grey area bit. is the “mush” of model weights an encoded representation of works, or vague memories?
