Isn't the whole basis of the legal success of the AI companies so far that training an LLM is "transformative", i.e., something that would not apply in any way to archival?