> The mental calisthenics required to justify this stuff must be exhausting.

It's only exhausting if you think copyright ever reasonably settled the matter of ownership of knowledge and want to morally justify an incoherent set of outcomes that they personally favor. In practice it's primarily been a tool for the powerful party in any dispute to hammer others for disrupting their business model. I think that's pretty much the only way attempting to apply ownership semantics to knowledge or information can end up.

Correct.

Knowledge consists of, roughly speaking, thoughts.

(a "justified true belief" - per https://plato.stanford.edu/entries/knowledge-analysis/ - is a kind of thought)

The "thinking" part of a "thinking being" - that also consists of thoughts.

If your knowledges are someone's property, you are someone's property.

A society where all knowledge is proprietary, is a society of ubiquitous slavery.

Maybe multi-layered, maybe fractional, maybe with a smiley-face drawn on top.

Doesn't matter.

Humans have been known to recite entire parts from plays from memory, live in front of audiences even.

And they are legally required to license the play to do that, if it's still in copyright.

Only to perform it, not learn it.

And LLMs perform when you prompt them.

> Yes LLMs can reproduce passages from copyrighted works verbatim but that's only because it "learned" it and it's just telling you what it "knows".

Are you finding people that actually say this?

When it can quote something like that, it's a training error. A popular enough work gets quoted and copied by people online, and then it's not properly deduplicated. It's a very small fraction of works it can do that with, and the cleaner your data the less it happens.

I'll once again quote that stable diffusion launched with fewer weights than training images. It had some accidental memorizations, but there wasn't room for its core functionality to be memorization-based.

This is a perfect example of 'begging the question'. Arriving at a conclusion from a fact assumed as true without evidence. Your reductio does not actually demonstrate that copyright applies to LLMs, because you did not demonstrate how transcoding is comparable to inference, just that LLMs can reproduce some passages from copyrighted works. You could also produce passages from copyrighted works by generating enough random sequences of words, but no one is arguing that is comparable to transcoding. That the people who do not share this conclusion are engaging in motivated reasoning is based only on your assumption and has no logical backing, and is therefore begging the question.