> This way, it will live forever instead, scanned and trained on, conjoined with the rest of our knowledge in a magic machine.

But that's just it, it won't live on forever because, from the perspective of preservation, training is a lossy, noninvertible transformation. The LLM cannot legally produce the book verbatim, it will only spit out a regurgitation of the information, chopped and mingled into a broad information space.

Furthermore, these "magic machines" are not the property of the public. They are owned by a handful of corporations who want to charge you continuously for every token output by the machine. So, not only is the original text locked away forever behind company walls, you now need to pay for access to an approximation of the original contents which you can no longer even verify as being correct because the source is no longer accessible.

If you are cool with this, from a cost perspective you are cool with a deal whereby I trade you access to a definite resource for a one time fee of $N for, instead, a perpetual cost of $M to you every month/day/hour for access to an amalgam in which you cannot even determine what proportion of the resource you are actually getting. You're basically saying you're cool with me selling you some unknown portion of wine for a monthly subscription price instead of selling you a definitive amount of wine for a one time fee. lol.

Are you under the impression they scan the book, train on it, then destroy the digital copy? Because that's not what's happening. They scan it, and hold it forever to train future models on. The scan still exists, not available to the general public but that's no different than if they had bought the books and kept them in a private library closed to the public.