I hate to be the contrarian voice, but I really don't like where things are headed with these types of assistive tools.
I like the idea of using it partially to help complement some effort that you've done yourself, but the idea that you're using both an LLM to produce the story and a diffusion model to produce the pictures makes the entire piece of work almost devoid of individual effort.
Using them to produce a private story for your own child is a noble enough effort, but I'm seeing lots of comments on wanting to get these types of things published. With the explosion of these vanity press level dime novel tabloid literary works, it will be that much more difficult to sift out the wheat from the chaff.
Completely agree. People really don't understand where this is going. It's not like society lacks mediocre mid-brow content as it is. LLM's and diffusion models reduce the cost of production to basically zero, and we now have an endless supply mediocre content. Extend it out a few 10x improvements, and we get an endless stream of so-so Netflix shows produced with zero effort. Basically WALL-E made real.
It doesn't look to me like the cost of production was zero and the content really wasn't bad. For anyone who didn't read to the end, the author (operator?) explains his process:
> I typed a series of prompts to ChatGPT to get it to write me a story with the following plot:
> 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.
He mentions elsewhere that generating the images was a lot of work.
So basically he and his friends had an oddball premise for a story, and with some non-trivial amount of labor, he produced a story which was far from the worst thing I've read on HN.
It was actually pretty clever for a kids' story and the illustrations were good. The finished product was better than the prompt. Somehow the process produced a fairy, a plot twist and slabs of butter riding a roller coaster.
Where's the crisis here? A person came up with an idea and used a tool to generate a work product from it. Without that tool he would have needed to hire a professional illustrator and maybe a writer. I bet this took a fair bit of work to do right. But the technology enabled a creative work that wouldn't have occurred otherwise.
I have no doubt there will be tons of crappy spam enabled by ChatGPT too, but frankly this little story wasn't it, it was humorous and I doubt it was a trivial endeavor.
Per authors comments above, it was indeed a trivial endeavor: a few hours to tweak the prompts and a few hours to make the site.
Compare to the most basic children’s book, which would take at least 10 hours to illustrate for a professional (10 picture times one hour per picture).
So then found objects and ready-mades are also a trivial endeavour?
These authors published one of the first stories like that and that alone makes it not trivial.
People are ultimately going to do that. And if we really can't figure out a book written by an AI from that written by a human, then what's the point of all of it? Eventually good authors will still succeed as their writing style will not be squeezed out of the grand summation of all the writings in the world.
1. Entire classes of skilled jobs are eliminated, only the most elite writers and artists can make money
2. Value for story creation gets concentrated in the few companies that own and operate the best LLM's and best stable diffusion models
3. Society becomes more fragmented because we are all watching different content
These are just a few I'm able to think of in 2 minutes.
It's a common mistake to assume all technological progress is an unadulterated good. Just look at all the unanticipated negatives that came out of the computer revolution, like every technological revolution before it.
Agree that the assumption is that progress, no matter what, is good and healthy.
If one considers how inefficient humans are (daily consumption, heating, transport..etc) and how in comparison computers are extremely efficient (once they have been built), you realise where "pure" progress will take us.
Ephemeralization is the term Buckmeister-Fuller used.
I'm not at all assuming all technological progress is an unadulterated good - in fact, nothing is, everything would have a downside if it has an upside. All I'm saying is it's still bound to happen and we can do nothing about it.
It's not the first time that 'entire classes of skilled jobs' have been eliminated. It's the natural side effect of automation and has happened with every generation of new products. Industry automation might have taken millions of jobs away so far to give us better products faster. Tomorrow my job as a usual developer might be gone to these AI tools (though it's a far fetched thought), and I have to be prepared for it, and re-skill/up-skill myself if I ever sense that day is about to arrive.
Another strong agree. These models don't understand anything, they just produce something that is statistically (at many levels) close enough that it will fool most of the people, most of the time.
For those that are increasingly happy to live in, and raise their children in, that fake and simplistic environment, I suppose it's their call. But reading/watching/living in that simpler, dumber environment can't be good for our intellect and higher abilities.
And what the brain produces is not something that is statistically (at many levels) close enough that it will fool most of the people, most of the time?
Have you looked into even basic neuroscience research or perhaps at least personally known someone with, for example, dementia? It would perhaps teach you something about what the mind is and isn't and what we know and don't know and how perhaps AIs might fit in this story.
Just handwaving because it's "just statistics" is not useful.
As for the environment - well, I have my opinion on social media and online content and even most print content and most books. Will AI make that worse? Yes. But it's already bad enough that this is not a meaningful change. The good stuff has always been accessible and for a long time.
In Roald Dahls short story "The Great Automatic Grammatizator", it becomes impossible to separate the generated from the human. The machine initially is published under fake names, but what most don't realize is that for even established authors the profession no longer exists and they are left with the choice of destitution or licensing their names under which the the machines books can continue to be published.
"And worse is yet to come. Today, as the secret spreads, many more are hurrying to tie up with Mr Knipe. And all the time the screw turns tighter for those who hesitate to sign their names.
This very moment, as I sit here listening to the howling of my nine starving children in the other room, I can feel my own hand creeping closer and closer to that golden contract that lies over on the other side of the desk.
Give us strength, Oh Lord, to let our children starve"
Thank you for saying that. The tech is really cool but it gives utterly untalented storytellers the tools to inflict really bad content on the world. The insipid and pointless plot of this one is a case in point. (I do admire the OP’s get-something-done adventurous spirit of trying out the tech and making something, kudos, but not their artistic sensibilities on this one unfortunately.) At the same time it will allow others to create fantastic stuff. We are going to need some better filters.
Seems like in another era you might get upset at the invention of the record player, because now poor people would have the ability to listen to music without having to pay to see it live.
Yes, better tools for making art means more art will be generated, and lowering the bar means less talented people might start making up the bulk of the new art that is generated, but new mediums inevitably spark creativity in a handful of emerging artists, and that enables them to create masterpieces that could not have been created before.
counterpoint; separating the wheat from the chaff is already terribly difficult, and perhaps this will provide the necessary motivation to produce an effective method of doing so.
Google, once upon a time, solved a similar problem. With all the information on the internet, how do you find the good stuff? The relevant stuff? I don't think it does this job at the moment - and perhaps flooding the world with content will create the conditions for the next major disruptor to create something that actually works rather than this ad-bloated nonsense we have now.
It already takes a ton of my time seeking out books and other media that align with what I want. Humans have done a spectacular job of creating stuff that's minimally acceptable to make a sale. AI isn't creating the problem you're talking about, even if it is making it worse.
Perhaps it will become bad enough that it becomes worthwhile to fix.
The fix for the flood of crappy Chinese products so far has been recommendation engines. And those are gamed (Amazon reviews, goodreads comments, etc)… so no fix yet.
Yep. Part of the problem is the set of incentives surrounding these huge centralized databases of consumer preference - the folks holding the reins have no reason to fix the situation because they make money off the circumstances as well.
This was effectively the situation with the internet when the Goog came in and wrecked everyone's cash cow - because it was valuable enough to take the hit and prospect for future value to undermine the status quo.
I see two ways to break the current stalemate:
- Some 'new google' comes by and does to google what google did to Yahoo et al.
- Somebody creates a system that does the same job, but distributed. You control your recommendation algorithm, so you can prune it as you want.
I really really want the second one, but I haven't yet figured out how one would get the necessary data. There's a bootstrapping problem here - in order for recommendation algorithms to work, you need a ton of folks creating data for them. In order to get those folks, you would need to have a good enough recommendation algorithm to attract them.
I'm wondering now, though, if data on user preferences is for sale somewhere. Like, if I wanted to bootstrap such a system, is there some adtech business somewhere that has information on what books / music / etc people like, and how much would it cost to get that in there?
But now, of course, because you're looking at spending a bunch of money to bootstrap - you're hobbled by the need to be profitable to make that back. Stuck in Trapitalism.
It can easily be polished and that _does_ take work. So it's not zero effort. Yes, there will be a lot of generated content and that will probably lead to a recalibration of what's important and what not (in life, work, art, etc).
I like the idea of using it partially to help complement some effort that you've done yourself, but the idea that you're using both an LLM to produce the story and a diffusion model to produce the pictures makes the entire piece of work almost devoid of individual effort.
Using them to produce a private story for your own child is a noble enough effort, but I'm seeing lots of comments on wanting to get these types of things published. With the explosion of these vanity press level dime novel tabloid literary works, it will be that much more difficult to sift out the wheat from the chaff.