The original point, that LLMs are plagiarising inputs, is a very common and common sense opinion.
There are court cases where this is being addressed currently, and if you think about how LLMs operate, a reasonable person typically sees that it looks an awful lot like plagiarism.
If you want to claim it is not plagiarism, that requires a good argument, because it is unclear that LLMs can produce novelty, since they're literally trying to recreate the input data as faithfully as possible.
