(May 14, 2023 at 5:09 am)Belacqua Wrote:(May 14, 2023 at 3:24 am)emjay Wrote: I actually know what you're talking about there (which is rare) having at least watched the film, if not read the book.
Each to their own of course, but personally I don't care if this thing is highly curated/censored, and not 'edgy', since I'm primarily using it as something like a talking encyclopaedia, in which case I am generally looking for the conventional wisdom. But of course I understand your use case is completely different, so I understand your frustrations as well. I do agree it's a bit of a shame if it means it's missing the essence of some of these 'edgy' books. But since from its perspective it doesn't figuratively 'know' if it's reading fiction or fact, art or opinion - since the training data is ultimately just a massive list of words/word-parts - I still think the censorship is probably a wise trade-off to prevent it becoming a 'shitbot'.
I will say though, it is opening me up to your kind of way of thinking as well, though probably in ways you wouldn't necessarily approve of in that I've asked it to write several poems, many of which I've found pretty meaningful and beautiful, but I'm guessing that from your perspective you'd still see that as a negative? Ie thinking that on-demand AI-generated art lacks some essential human element that the AI can't or won't capture, the desire/passion that's gone into it etc (if I've understood what you were trying to say before)? While I understand that, I'm still personally appreciating this bottom-up approach to art; and am happy enough if it's able to capture the exact feeling I want it to portray, and happier still to see its many interpretations.
At the very least, I think it will be a stepping-stone into the art/literature world, not just in seeing how you literate-types
If you're getting results from the AI that work for you, I would never speak against that. I agree that watching how it expresses and organizes things is fascinating and could be very helpful. A clear articulation of even our most common thoughts is a pleasure to read. "What oft was thought but ne'er so well expressed."
My italics. Indeed. My sentiments exactly.
Quote:I think where it differs from human literature is in the range of its possibilities for expression. A whole lot of art is based in irony of various kinds. William Blake's poems, for example, are almost all written in the voice of a character who isn't Blake. We are to see through the character's expression to a further meaning which fits in which Blake's own system.
Dorian Gray is meant to be a very appealing character, in a lot of ways, and when I was young I thought of him as a kind of aspiration. At the same time, he's horrible. So there is that double or triple layer in books -- they are not just a simple statement of events but a challenge, a puzzle, a lasting unresolvable image.
Currently I'm reading Nabokov's Lolita with a group of older Japanese ladies. Nabokov is in some ways an heir of Oscar Wilde since he believes that the quality of a book is not in its morality but in its beauty. And as we read it there is a constant tension between the extreme beauty of the book's writing vs. the extreme evil of its main character (the man who rapes Lolita). The whole book is told in the man's voice, and he attempts to justify himself or somehow persuade us that his love should make him forgivable. But even as he tells it, the reader can see through what's going on and understand a lot more than he tells us. Just in this it is an incredible literary performance -- the book tells us things that its narrator would never say. And on top of this there are all kinds of puns, word games, and references which I don't think an AI could ever manage.
So this is what I think makes literature different from what an AI can do, at least for the foreseeable future. An AI might generate a sequel to Harry Potter or some other kids' book that works on a simple level, but good literature does way more than that.
It is pretty good for what it's doing... I have for instance been exploring things with it such as the notion of analogies, of which it's capable of processing both explicitly and implicitly... ie if you put in 'food is to hunger as drink is to x', 'what is x?', it gives the correct, albeit verbose, answer of 'thirst' as well as, when further pressed, giving the/a correct common denominator of 'satisfies'. And also, in terms of layers, I wanted to know if it's possible/if it could write a poem that changes its meaning on the second pass based on stuff you learn reading the first pass (something that happens a lot in films/books in my experience; sometimes they just need two readings/watchings to get the most out of them). It didn't really know what I meant, but we're getting there
, and it did nonetheless have some interesting things on similar lines to say. I have to say, the reductionist in me is having a whale of a time exploring all this stuff from the bottom up But yes, in terms of complex, carefully crafted literature... in terms of irony, subtext, different perspectives etc... it could never compete on that score and isn't claiming to, since it's 'only' claiming to be a natural language model, not a model of human cognition, the latter of which is required to carefully craft/orchestrate complex works of art/literature. I quote 'only' because what it does do is awesome, but it's just not claiming to also model human cognition. But who knows what the future holds, of which as you can probably see from this thread, I'm pretty ambivalent about... about whether we should go any further in its R&D.
