Right, but if an AI company buys some vanishingly uncommon book, digitizes it, shreds it, and adds the information it contains to their permanent digital library and digests its contents into an AI that is then publicly accessible...are they making that book less accessible, or more?

▲

scarmig 18 hours ago | parent | next [-]

You've got to compare it to the alternative. Books have a half-life, and the vast majority of these books being purchased are grody, moldering ex-lib copies of books that no one has read in decades. Their other likely outcome is mulching.

▲

margalabargala 18 hours ago | parent [-]

Right, that's my point.

These generally are not books people care about. The information contained therein was doomed.

Now the information has been digitally preserved and a digestion of the information will be made publicly available.

	▲	michaelmrose 17 hours ago \| parent [-]
		[dead]

▲

halsafar 18 hours ago | parent | prev | next [-]

Can you get the exact text back out with a prompt or not? Having or not having a book isn't fuzzy.

▲

skeledrew 17 hours ago | parent | next [-]

Funnily the argument made just a few months ago by many rights holders who wanted their pound of flesh was that, if prompted a certain way, exact text could be retrieved.

▲

margalabargala 18 hours ago | parent | prev | next [-]

Having or not having a book is absolutely fuzzy. If you have a translation, do you have the book? Even if, like the Odyssey, there are hundreds of wildly varying translations? What about an abridged copy? What about the Sparknotes version? If you have a copy of Pride And Prejudice And Zombies, do you have a copy of Pride And Prejudice? Certainly more so than if you have neither.

▲

fluoridation 16 hours ago | parent [-]

>If you have a translation, do you have the book?

No, you have a translation.

>Even if, like the Odyssey, there are hundreds of wildly varying translations?

Precisely why translations are not considered equivalent to the original text.

>What about an abridged copy? What about the Sparknotes version?

An abridged copy is not a copy of the unabridged version.

>If you have a copy of Pride And Prejudice And Zombies, do you have a copy of Pride And Prejudice?

No.

I'm honestly surprised these were the questions you chose to ask, when you could have asked what if you have 90% of the pages, or what if most of the pages are missing pieces because the book was shot with a shotgun, or what if the book was scanned and OCRed and all the "rn"s were replaced with "m"s and all the lower case Ls with ones. Hell, is a scan of the book close enough to having the book, or is it far enough that one can no longer be said to have the book anymore?

▲

margalabargala 9 hours ago | parent [-]

My opinion is different from yours.

If I have a translation of a book, I think I have more of that book than if I had nothing at all. It's fuzzy.

▲

fluoridation 9 hours ago | parent [-]

You don't have the book, you have someone else's interpretation of the book's contents, re-expressed into a language you can read. Both steps can involve a loss or distortion of information, either because the translator doesn't fully grasp the original language or context, or because in the re-expression they chose to leave out details that were relevant to you. The more distant the original language to yours, the more translation involves interpretation, too. A translation is really not too different from a commentary; it's just a different text.

▲

margalabargala 8 hours ago | parent [-]

That sounds to me like fuzzily having a version of the book. That is, it's more like having the book, than having nothing would be.

You seem to be arguing that a translation, etc is not literally having the book, which is something that has always been my stance as well.

▲

fluoridation 8 hours ago | parent [-]

You're trying to use the Socratic method to show that the havingness of the book is a spectrum, and I'm taking the position of a hardliner who considers that "having the book" means having the original string of symbols from beginning to end, and anything besides that is not having the book. I'm trying to show you that your line of argumentation is uncompelling to such a person.

▲

margalabargala 8 hours ago | parent [-]

I was never using the Socratic method. I was asking rhetorical questions, and then gave the (my) answer at the end.

We just have different opinions about what it means to have a book. I think that having a copy of "Pride and Prejudice and Zombies" is more like having a copy of "Pride and Prejudice", than having no book at all is like having a copy of that book. You disagree and that's fine.

▲

fluoridation 7 hours ago | parent [-]

Asking rhetorical questions as a form of argumentation is the Socratic method.

	▲	margalabargala 7 hours ago \| parent [-]
		If one is using the Socratic method, rhetorical questions are one tool they might employ. The reverse is not true. Just because someone asks a rhetorical question, does not mean they are using the Socratic method.

▲

close04 14 hours ago | parent | prev [-]

The scan and destroy method is what a judge allowed to do in order to have a copy of the book in the training dataset. With the physical copy destroyed there's still only 1 copy in circulation. Once "inside" an LLM I don't know if anyone decided unequivocally that it's copyright infringement or not, and if that counts as a second copy.

There's no technical reason why an LLM couldn't reproduce verbatim some of the training material. It's sort of a lossy statistical compression engine. Enough of the info will survive to the output in the original form. With the amount of data and the commercial nature it's hard to argue fair-use. But nobody tested this in court. I'm not even sure the US wants to ever test this. Why even attempt something that has a non-0 chance to sabotage your most promising industry/bubble in ages?

▲

GPerson 18 hours ago | parent | prev | next [-]

They’re not supposed to be storing a copy. What they’re doing is destroying their copy after training a model on it.

	▲	derektank 18 hours ago \| parent \| next [-]
		No, US copyright law allows them to keep a single digital copy. The hypothetical issue is with them maintaining two copies, one digital and one physical, when they only purchased one.
	▲	monocasa 18 hours ago \| parent \| prev [-]
		They're absolutely keeping the digitized copies. They're not going to just train a single model.

▲

Natsu 18 hours ago | parent | prev [-]

AIs are weirdly bad at quoting stuff in my experience.

But you'd think that the Library of Congress and such would actually prevent stuff from vanishing just by collecting it themselves.

▲

margalabargala 18 hours ago | parent [-]

Bad at quoting, good at digesting and regurgitating. The concepts are preserved even if quotes aren't.

I'd rather a digital copy exist in someone's hands than a rotting physical copy.

	▲	subscribed 16 hours ago \| parent [-]
		But you don't have access to this copy. The digital copy is removed and all you get is paraphrased content. Some frontier models have been explicitly forbidden from recalling exact quotes in system prompt. It almost seems like you're suggesting that having Claude generate a paraphrased book is as good as having the original book but i don't think that could be your intention?