The problem with nightshade or similar tools is, that if you leave the changes it makes at too weak a setting, then it can be pretty easily removed. For example GAN upscalers that pre date modern “AI”, were pretty much built to remove noise or foreign patterns. And if you make the changes strong enough that they can’t be removed by these models (because so much information was lost), then the image looks like shit. Its really difficult to strike a balance here.

Does anyone know if non destructive editing will be possible with 3.0 or if its a feature for the next ten years?

Tbh, no. At least regarding your technical critisisms. Try out a LLM with a large context window. Claude3 is currently the best. Feed it a chapter or 10 of a book and then ask it to write another chapter, about an unrelated topic and it will just do it. No problem. Creative writing/condensing/summarising is one of the strengths of these things, if given enough info in the active context. Just asking for a style doesn’t really work, but its also not necessary here. Reproducing akkurate facts however is a big problem, which makes using them dangerous. So the retaining of information would actually have to be manually checked by a human, as LLMs are prone to hallucinations and will make errors or changes to fit the writing style. It can write nice sounding paragraphs about spatial or logical impossibilities any day though. Finally, quality greatly depends on the people feeding the LLM, because low effort will also produce low effort outputs.

And I don’t really follow what you mean with your comment of “hideous mockery”. The people who can read and understand the original version, will do so and never read the simplified version instead of the original. Only individuals “incapable” of dealing with the original would even interact with the simplified version. Isnt it better if at least part of the authors message can be conveyed, rather than nothing? That person never even touching on the ideas of that book? I am currently writing a SciFi book myself and I would definitely want to personally look over and check the simple version of it. But I would see it more like something in the way of an audio book. Reading doesn’t have to be difficult and I would want as many people as possible to have the ability to read my book, regardless of their reading level.

All that said, I am talking about the concept here, the specific implementation of the image above is likely going to be some half assed garbage.

On first thought this seems like its such a weird usecase for AI. However, I don’t actually think its completely useless, turning more complex books into children’s books while maintaining their lessons and ideas is pretty interesting. And that is something that LLMs can realistically also achieve, not just hype bullshit. Getting grade schoolers to read Nietzsche and them actually understanding something, is a very fun thought to me. I don’t think this will have any impact on the reading comprehension of teenagers or above. Those that can’t handle the original text, aren’t going to read the simplified one. But getting young children acquainted with “grown up” books and their topics and ideas could be a good thing. When its not just about the rabbit in the mushroom house etc. It might even encourage the parent to (re)read the book with the child together, one the original and one the simplified version. Also useful for illiterate persons learning to read, as reading children’s books can be uncomfortable for an adult.