I think it depends on how you use it. If you write a one-line prompt, you'll get a "generic" result. With more guidance, you can get something much more personal. For example, you could imagine a story and let ChatGPT pick the actual words.
At the end, you can put an arbitrary amount of work to get a generated story. You can treat ChatGPT as a co-author who will help you do the parts you want (e.g. find ideas, find text structure, write the actual sentences). The end result depends on how much work you want to do yourself.
This IMO is a handwavy criticism of AI-generated content by people judging content who have the foreknowledge that the content was AI-generated.
To put it another way, pretend it was a year (or more) ago and ChatGPT/Stable Diffusion were ~unknown. You stumble across this kids book. Be completely honest here: would your reaction be "there's no way a human created this book"?
-> Hedgehog making bread -> Loves making bread -> One day bread stolen -> Butter everywhere -> Hedgehog sets out to find culprit -> Finds culprit -> Opens butter themed amusement park with culprit -> The End
There is no overlapping theme or logic to the story. Children's book will have some form of gimmick and stick to it (young humans aren't all that bright) so you can have a baker hedgehog that encounters adversity (bread is gone) and maybe tie it in with a human culprit (bread thief, didn't say please or something), but the butter amusement park is out of place and uncanny.
I think that's pretty different than GP's point about art needing to have "voice." It seems entirely clear to me that AI will get to the point where it can generate something coherent in terms of plot/theme/logic - that's just a matter of time.
> Once upon a time there was porcupine. Every day, she baked a loaf of bread. One day the loaf disappeared. Because of that, she had too much butter. Because of that, things got slippery. Until finally she opened the first ever butter theme park.
It seems like the LLM did exactly what the author wanted
I read the prompt and I can see the human’s intention: write a story about how you turn something bad into something good. But the AI transitions didn’t land and instead of getting a heartwarming story of a clever hedgehog, it just feels like a badly executed Deus Ex Machina.
-> Caterpillar eats food -> Caterpillar eats more food -> Caterpillar eats even more food -> Caterpillar gets sick -> Caterpillar eats less food -> Butterfly.
The succession of events have a thread that join them together. I'm pretty sure ChatGPT could write it as well if you give it the above summary, but the hedgehog story doesn't have that and feel uncanny because of it.
Take a look at Penguin in Peril for example.
Cats are hungry -> they go to see a film -> they are still hungry so they decide to steal a penguin -> the penguin escapes -> the penguin swims through a fountain back to the zoo -> the cats get put in prison.
I dunno maybe it's still fairly coherent, but we've gone from "completely impossible" to "bad children's books" in like 2 years. How many more years do you think it will take to get to "good children's book"? I bet less than 10.
Where the Wild Things Are - a fever dream about rebellion and unrealized sexuality, hero's journey into the latent world and back.
Green Eggs and Ham - picky eating works on its own, or can stand in for just about any other unpleasant experience. The repetition mirrors the real-life experience of getting used to something you don't like.
Can't remember the name, about fuzzy little monkey people living in a hut? - really cozy, about the warmth and safety of being a cuddled-up tiny person without worry.
There's no throughline in this one. The prickly porcupine stands in for a stressed mom. The fairy stands in for a child who messes up the kitchen. The solution is play all day and leave the kitchen dirty? Doesn't hold together.
I'd take out the filler "journey" bit, add in some conflict between the mom and kid, and make the ending more realistic (fairy uses magic to make icing for the cake?). That might bring it up to library-filler tier book.
To me, the art looks really impressive though. Again, no throughline - I'd expect a progression that goes from lonely porcupine -> in the wilderness -> sharing the stage -> back at home but happy
THIS. This is exactly the power of these models. The AI generated an entire story in some extremely quick amount of time and an experience Human can edit and curate the results. It generates first drafts trivially.
The reason I might not have guessed it was a machine creation is that the prose is adequate on a technical level, and previous markers for machine creation included wonky prose. But going forward we'll all learn new markers for machine-created dreck.
Never is a really, really long time.
> Art, in this case illustration and narrative, requires a coherent viewpoint.
Why can't an AI have a coherent viewpoint, especially in the context of a children's book, where that viewpoint need only be incredibly simple? "John has lost his <insert item>, feels sad about that, and wants to get it back" is a coherent viewpoint, and certainly something AI can (if not now, at some point in the future) write a coherent story about.
> Voice is special because of the uniqueness of the human life behind it, which I realize sounds a bit airy-fairy, but is very true once you start digging into what works and what doesn't.
So what works and what doesn't? You're saying all of this at a high level and giving a timeline of forever on your prediction, but then you say it's true if you dig in without doing that.
> It's also why some pieces really don't work for some people. The "a kid could draw that" critiques of various modernism styles come from this place.
The thing about modernism is that it requires outside context to understand. It is a movement in response to art movements that preceded it, and the fact that its style uses simplistic shapes and lines is because of what it's a response to. You need to understand that history to appreciate why it is what it is.
In that sense, you're absolutely right - a child can't produce modern art, because even if it looks visually similar to existing art, it lacks that context.
The problem with that argument is that it's totally irrelevant to the topic at hand - children's books. They do not need history or context - those things serve no purpose because, for obvious reasons, they would be lost on the audience. All you have to hit is the superficial level - if you book looks and reads like a great children's book, it's a great children's book!
On a somewhat related note, I think AI art will eventually be seen as a movement somewhat akin to modernism. In response to the fact that art has become increasingly elite and insular (the whole fact that a modern painting can't be fully appreciated just by looking at the painting - you must know art history, which most people don't have the time or inclination to learn), people will be inclined to appreciate the flood of beautiful but democratized art that can be made by anyone. Just as the generation of artists before modernism thought it was awful, so will the current generation of artists about AI art. I'm a believe that art should be about aesthetics, not history, and AI art is certainly great for that.
True but irrelevant. If we want to get to decent art, we'll need something well beyond an LLM.
Humans generate stories using introspection. Every writer is their own first reader. They rapidly and iteratively try out story elements to create ideas and feelings. They then find the right words, again seeking out particular feelings. After iterating on the words for a while, they then have others read the words: partners, friends, eventually an editor. Those people all experience feelings, which get observed and sometimes articulated to the writer, driving further revisions.
LLMs can create story-shaped things. They can write glib, imitative dreck. But to write a new, compelling book, they need to be able to iterate against high-quality simulations of reader emotions. That may be something that we can create in the lab, but if we do, it won't be called an LLM.
Like, AI art generators don't just have issues with human hands or eyes, they also create buildings which are architecturally incoherent. I don't know how you solve that with mere data or refinements.
Tell me you don't understand LLM's without telling me you don't understand LLM's
From HN FAQ:
> Avoid generic tangents. Omit internet tropes. Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something.
You can't teach someone who is unwilling to learn.
I'm curious how many modern children's books you've read. There are a ton of very shitty... Remixes with poor cheap art. Especially books for cities and holidays. I could see this eating into that
Perhaps, if the cover was engaging enough. However this seems like the kind of story that my child (4 1/2) might read once and then never pick up again. She'd be like "ew, who wants butter all over their kitchen?"
Much of children's best-selling fiction is extremely derivative. The way the market works, it's nearly impossible to get things published that aren't extremely derivative.
To the extent that, I think this might actually be a good place to start - with auto-generated derivative garbage - and then fix the "problems" with the plot through editing.
My first thoughts on this story were pretty negative. The plot is almost incoherent. There's not much conflict. But it does have a little bit of a mystery, and it does get resolved.
Although much of it came from out-of-nowhere (The Theme Park), Alice's Adventures in Wonderland and many other extremely popular children's books are literary nonsense.
The illustration definitely does not fit together well, and is not what you'd expect to get from a book you paid $20 for.
But, overall, it's really not bad.
I could see a huge market for parents wanting custom books like this generated for their kids about all their zany interests. If your kids loves baking & cute animals & mystery - they'd probably love this story.
Maybe it costs $1 or $2 - it's definitely not a big business - it's probably not going to "disrupt" the market (because it's so entrenched I don't even think disruption is possible).
If this is too much literary nonsense, I'd love to see what it could do if you asked for the same story "In the style of AA Milne" or Robert Lewis Stevenson or James Mathew Barrie or Beatrix Potter.
I can definitely see a near future where almost every parent does this for their kid once or twice a year.
But of course, it's the adults buying/distributing these books so you still need to "sell" it to them first.
(seriously, there's some so-cheap-I'm-not-sure-why-they-bothered kids' cartoons on Netflix, if you look for them—you're on the right track if it looks like it was rendered with mid-grade 2005 desktop 3D rendering tech using stock models and stock scenes with way-too-little decoration or clutter, despite in-fact dating no earlier than 2015. A lot of them have whole scenes that are basically just the characters talking in circles. The plots are always extremely straightforward and about 1/5 as much plot as they should need to fill the time, even by very generous reckoning. Lots of interpersonal conflict that amounts to nothing, existing just so they can have two characters have some conflict in a scene to absolutely no thematic-, message-, characterization-, or plot-related end. Most turn out to have been originally produced in India or Romania or something, if you start digging. It's crazy. But young kids will watch these.)
Trash in, trash out.
Just because there is a bunch of trash on Netflix and elsewhere doesn't mean we should pollute the kid's world more with it.
Especially at the age of reading. 1-5 is the most important age by far...
I don't know how we get this myth to stop perpetuating but it does not ring true to me at all. LLM's has proven to have "novelty" many times now.
Even so, it's a fun and interesting tool to use alongside traditional media, so I'm excited to see it improve.
Kid's books aren't intended to be novel. It's not like a four year old is going to say, "Man, this is a retread of Hamlet!" They are often re-tellings packaged up in a new way to appeal to kids in the current generation. Sometimes as simple as removing sexist, classist, and/or racist elements.
There's only so many words in the english language. If I ask ChatGPT to write me a story about how a I turn into a super hero tomato that saves the world against the villainy of pasta, that's a novel story I'm pretty sure.
Art, in this case illustration and narrative, requires a coherent viewpoint. In literature this is often referred to as the authorial voice. Voice is special because of the uniqueness of the human life behind it, which I realize sounds a bit airy-fairy, but is very true once you start digging into what works and what doesn't. When you read something and wonder why it's so engaging when it's not _that_ much different than other very similar pieces, that's often what's at work.
It's also why some pieces really don't work for some people. The "a kid could draw that" critiques of various modernism styles come from this place.
I don't begrudge people experimenting with art through LLM generation, and perhaps it can be a great tool for scaffolding ideas, I grant that. However, I do worry about the deluge of poorly conceived LLM content that will make the good stuff harder to find. We had a glorious age where the gatekeepers had lost most of their power. With the flood of content about to be released, I'm afraid we'll find ourselves beholden to them once again.