I think they're talking about the misconception that LLMs can only ever regurgitate their training data verbatim enough to constitute mass copyright violation.

I think they're talking about the misconception that LLMs can only ever regurgitate their training data verbatim enough to constitute mass copyright violation. And that that's therefore "stealing"