AI can create pictures and music from prompts, how soon will it be before they can create novels?

>it’s not the quality of AI writing that’s the problem, it’s the quantity.

Exactly. And while I take the point about publishers, traditional publishers are increasingly irrelevant. Self-publishing is where AI will shine. Human factories with scores of workers will submit prompts and publish under a range of pen names as wide as the fake email names we have now. Will Amazon be able to prove that Arnold Realperson did *not* write a book a week all year long?

And while many readers may be more discerning, a great many will not be. Enough to justify the book spam. Easy money attracts easy morals. Moreover, the next generation will grow up on this stuff. It doesn't have to be good, it just has to be good enough.

I wonder if AI could take an AI story and reverse engineer it to derive the prompts that created the story in the first place.
 
I wonder if AI could take an AI story and reverse engineer it to derive the prompts that created the story in the first place.
I see a more likely scenario where AI would take on the role of promoting stories based on story attributes and review comments. I am not sure that the economics are there to support "factories with scores of workers." Whereas a single individual might be able to derive a secondary income stream from self-publishing a lot of low-cost, low read books, I don't see the profits being sufficient to justify anything at a larger scale.

Eventually, someone is going to take on the role of traditional publishers in the online market and start providing curated lists of authors and stories.
 
Expect a weird proliferation of butlers and social intercourse.
 
An AI that writes bad science fiction... surely there’s a story in there.

How about this idea. A society where humans are banned from complex work, which is performed by AI and robots (including, say, novel writing). Humans are restricted to menial tasks which cannot be easily automated (like, say, working in a fast food outlet). But a small band of free thinkers gathers in secret to read a banned book which, it is said, was written by a human hand. But there is a traitor in their midst and soon the robo-justice system is hunting them down.
 
How about this idea. A society where humans are banned from complex work, which is performed by AI and robots (including, say, novel writing). Humans are restricted to menial tasks which cannot be easily automated (like, say, working in a fast food outlet). But a small band of free thinkers gathers in secret to read a banned book which, it is said, was written by a human hand. But there is a traitor in their midst and soon the robo-justice system is hunting them down.
I love it. Fahrenheit 451 meets RoboCop :LOL:


On a more serious note, I do wonder if all this AI stuff won't spur a societal shift back toward an appreciation for hand-made/human-sourced products for their "novelty."
 
How about this idea. A society where humans are banned from complex work, which is performed by AI and robots (including, say, novel writing). Humans are restricted to menial tasks which cannot be easily automated (like, say, working in a fast food outlet). But a small band of free thinkers gathers in secret to read a banned book which, it is said, was written by a human hand. But there is a traitor in their midst and soon the robo-justice system is hunting them down.
This reminds me of Rush's 2112. It's worth an investment of about 20 minutes of time.
 
How about this idea. A society where humans are banned from complex work, which is performed by AI and robots (including, say, novel writing). Humans are restricted to menial tasks which cannot be easily automated (like, say, working in a fast food outlet). But a small band of free thinkers gathers in secret to read a banned book which, it is said, was written by a human hand. But there is a traitor in their midst and soon the robo-justice system is hunting them down.
Good idea for a novel,
In real life, I would rather have a society in which humans didn't have to work, but were paid to have marginally productive hobbies.
 
As a reader, I wouldn't be interested in books by an AI except for maybe the "novelty"
My bigger concern would be about people trying to pass off AI writing as their own since AI generated works can't be copyrighted.
Film might be different though. As much as a good script matters it can be make or break based on the actors performance. And that's what most people are interested in when it comes to movies. The actors, not the script writing team.
People looks for their favorite actors, but good actors look for good movies with good scripts. When a person says they like an actor, often that is because they like the kind of films that actor signs on to.
 
>it’s not the quality of AI writing that’s the problem, it’s the quantity.

Exactly. And while I take the point about publishers, traditional publishers are increasingly irrelevant. Self-publishing is where AI will shine. Human factories with scores of workers will submit prompts and publish under a range of pen names as wide as the fake email names we have now. Will Amazon be able to prove that Arnold Realperson did *not* write a book a week all year long?

And while many readers may be more discerning, a great many will not be. Enough to justify the book spam. Easy money attracts easy morals. Moreover, the next generation will grow up on this stuff. It doesn't have to be good, it just has to be good enough.

I wonder if AI could take an AI story and reverse engineer it to derive the prompts that created the story in the first place.
Well , yes.
Since every prompt is logged you can do train a model on the inverse operation: create a prompt from the following text.
Problem being the current 2,000 tokens limit. Which will probably vanish in a couple of years.
 

If language models were the only models then, effectively, AI would be a very limited realm, but we have all kinds of models : convolutional nets, generative models, transformers ( aka language models), and a miriad of training techniques : reinforcement learning, supervised learning, unspervised learning , genetic algorithms... the list goes on and on and it is evolving so fast it is hard to keep the pace just to be up to date with the list of techniques and models.
The number of people dedicated to ML and AI is so large it has really gotten into a steady pace and the only thing that will probably slow it down is the lack of ( even more ) data. Language models are trained with almost all the curated text available worldwide.
 
If language models were the only models then, effectively, AI would be a very limited realm, but we have all kinds of models : convolutional nets, generative models, transformers ( aka language models), and a miriad of training techniques : reinforcement learning, supervised learning, unspervised learning , genetic algorithms... the list goes on and on and it is evolving so fast it is hard to keep the pace just to be up to date with the list of techniques and models.
The number of people dedicated to ML and AI is so large it has really gotten into a steady pace and the only thing that will probably slow it down is the lack of ( even more ) data. Language models are trained with almost all the curated text available worldwide.
Would you be willing to expand upon some of the other, more significant AI approaches being used. From the article, I gather ChatGPT is using an LLM, Large Language Model, but those buzzwords are really the total of my knowledge in this area. If they can be effectively summarized, perhaps a simple paragraph on some of the leading approaches would be interesting, at least to me.
 
Would you be willing to expand upon some of the other, more significant AI approaches being used. From the article, I gather ChatGPT is using an LLM, Large Language Model, but those buzzwords are really the total of my knowledge in this area. If they can be effectively summarized, perhaps a simple paragraph on some of the leading approaches would be interesting, at least to me.
Some of the models:
Ensembles - Normally a set of models trained with different sets of data emit a vote the output of the model is the result of the mayority voting process.
GAN - Generative adversarial models : they create data similar to the training data. It has two components : the generator and the discriminator; the generator is trained to fool the discriminator producing models that it can't distinguish from authentic models.
Reinforcement Learning : a training method based on rewards. The model is rewarded when it gives correct answers and "punished" when it yields incorrect results. The agent usually stores knowledge as belief states which are updated as it explores its surroundings.
Genetic algorithms : This is a method for solving problems with agents which are subjected to survival rules. The agent population's codes are subjected to modifications and mutations.... It works , but it is a slow process.
But back to my point we are not remotely near the apex of what AI can achieve and we are already looking at some areas where they match human capacity.
 
I wonder if AI-generated copy will start finding its way into AI data sets, causing it to start parroting itself.
The data is usually curated, but it could happen. Also GANs( Generative Adversarial Networks) can be trained to differentiate between man made and generated content ( up to a certain point), which can help avoid the process you describe.
 
It all depends on what you're worried about. I don't think that AI-generated art nor AI-generated writing are going to replace complex, high-quality art or writing because of the inherent limitations of the ML models these things are built on. No amount of prompt engineering is going to create a coherent 100,000-word novel, unless they completely change how machines "think." Nor will it really replace good art.

What it absolutely will do is come for the low end of both of these things with a vengeance. For instance, it's pretty difficult to get a truly good book cover with generated AI. There are artifacts, things you'd like to clean up, etc, and you don't have the power to do that with the AI. You are still much better off finding a talented person to do something on commission. But, if you want to make 50 of them and don't care too much about the quality, 1) you were never going to hire someone for that anyhow, and 2) it's actually easier to just accept the randomness and gunk that gets added in with generation.

Similarly with writing, you can't effectively tell a whole story with it. But you absolutely can summarize a chapter of your work with it or create a quick quote. I've found AI to be super helpful in creating summaries. You can craft quick blurbs with ease. Again -- it's the low end. Sure, it does put some people out of work; if you are a writer charging a premium for 200-word advertisements or copy for magazines, AI is coming for you. But I wasn't going to hire you to do a tiktok ad that isn't worth that much to me or a short paragraph from an editor demanding the cliffs notes. I might click a button to generate something acceptable and use that.

I'd read a book that was written with AI, but I'd apply the same standards I'd apply to any human author, and the large-scale quality of those for now is far beneath that.
 
It all depends on what you're worried about. I don't think that AI-generated art nor AI-generated writing are going to replace complex, high-quality art or writing because of the inherent limitations of the ML models these things are built on. No amount of prompt engineering is going to create a coherent 100,000-word novel, unless they completely change how machines "think." Nor will it really replace good art.

What it absolutely will do is come for the low end of both of these things with a vengeance. For instance, it's pretty difficult to get a truly good book cover with generated AI. There are artifacts, things you'd like to clean up, etc, and you don't have the power to do that with the AI. You are still much better off finding a talented person to do something on commission. But, if you want to make 50 of them and don't care too much about the quality, 1) you were never going to hire someone for that anyhow, and 2) it's actually easier to just accept the randomness and gunk that gets added in with generation.

Similarly with writing, you can't effectively tell a whole story with it. But you absolutely can summarize a chapter of your work with it or create a quick quote. I've found AI to be super helpful in creating summaries. You can craft quick blurbs with ease. Again -- it's the low end. Sure, it does put some people out of work; if you are a writer charging a premium for 200-word advertisements or copy for magazines, AI is coming for you. But I wasn't going to hire you to do a tiktok ad that isn't worth that much to me or a short paragraph from an editor demanding the cliffs notes. I might click a button to generate something acceptable and use that.

I'd read a book that was written with AI, but I'd apply the same standards I'd apply to any human author, and the large-scale quality of those for now is far beneath that.
Most of what you say applies only to the current models.

With curated data, and a lot of work , an early AI generated one of the first samples of quality art seven years ago.

Models have improved enormously over the past seven years I am sure current models will look like toys in seven years.
Regarding stories, the current window of 32 tokens doesn't allow to generate anything longer than a short story, but again, this is the current state of technology.
I truly wonder what will happen in the next ten years and what role will art play in our lives ... will it remain only as a hobby?
 

Similar threads


Back
Top