Should AI Write books?

[Crezth]

It can, but it's inadequate. A good DM has to be flexible and creative, which an AI DM cannot do.

So you think, but you can actually simulate flexibility and creativity. It might not satisfy you or anyone with a heart, but what is sufficient for the mass market? There’s a lot of people on roll20 mewling for your scarce “good DMs” who will pay and maybe handsomely for a good-enough simulation. Such is the power of separation among modern humans.

 

[Joij21]

It's a tool, and people who have some background in literary theory can come up with ideas to refine the output, also with trial and error ^^
To paraphrase the epicurians, even a donkey can see that ai itself cannot produce good fiction.

It will only get better, there's already some top literature (like Moby Dick or the Illiad) that's already in the public domain which you can use as the training data for it's ability to improve.

And that's the stuff without legal issues, we all know that in more darker circles outright theft for training sources will and already is taking place. These more illicit forks will likely be the best versions of cgpt and corruption of various companies will simply take advantage of the lack of regulations and the geriatric nature of our politicians & regulators and in the dead of the night purchase such versions on the dark web in crypto for themselves then proclaim that the models were their own and trained by ethical sources when they weren't. Or rather they'll just do a superficial training session with legal content on top of what they buy to proclaim it's their own in house version.

 

[Kyriakos]

It will only get better, there's already some top literature (like Moby Dick or the Illiad) that's already in the public domain which you can use as the training data for it's ability to improve.

And that's the stuff without legal issues, we all know that in more darker circles outright theft for training sources will and already is taking place. These more illicit forks will likely be the best versions of cgpt and corruption of various companies will simply take advantage of the lack of regulations and the geriatric nature of our politicians & regulators and in the dead of the night purchase such versions on the dark web in crypto for themselves then proclaim that the models were their own and trained by ethical sources when they weren't. Or rather they'll just do a superficial training session with legal content on top of what they buy to proclaim it's their own in house version.

Cgpt may already be "trained", even if less directly, by passages from fictional books. It is why unintentional breech of copyright can be an issue. You can prompt it even outright to mimic the style of an author, with mixed results.
I think it can be interesting to try to come up with a general plot of how it strings sentences together, although it will be only a subjective idol to form - after all, if one was serious about rigorousness there, they can just read the accompanying data about how cgpt works.
While a cgpt (or similar) fiction piece can be helped by the very ambiguous nature of literature in regards to what the author calculates in the text and how the text is actually picked up by the reader (even the same reader will see different things in time), my sense is that for any "ai" fictional story to be of some interest you will need copious amounts of work by the human who directs it with prompts and other refining processes, which is why I do feel that currently you are better off writing everything yourself from the start.
All that doesn't mean one can't find fun in using ai for writing; anything can be interesting, if you have the intelligence and will to extract from it. After all, you are reacting to a new text, which is bound to allow for some useful reflections ^^

 

[Joij21]

Cgpt may already be "trained", even if less directly, by passages from fictional books. It is why unintentional breech of copyright can be an issue. You can prompt it even outright to mimic the style of an author, with mixed results.
I think it can be interesting to try to come up with a general plot of how it strings sentences together, although it will be only a subjective idol to form - after all, if one was serious about rigorousness there, they can just read the accompanying data about how cgpt works.
While a cgpt (or similar) fiction piece can be helped by the very ambiguous nature of literature in regards to what the author calculates in the text and how the text is actually picked up by the reader (even the same reader will see different things in time), my sense is that for any "ai" fictional story to be of some interest you will need copious amounts of work by the human who directs it with prompts and other refining processes, which is why I do feel that currently you are better off writing everything yourself from the start.
All that doesn't mean one can't find fun in using ai for writing; anything can be interesting, if you have the intelligence and will to extract from it. After all, you are reacting to a new text, which is bound to allow for some useful reflections ^^

Yes but it will get better precisely because there are human programers whose task it is is to prune the training sessions for these AI until they become more perfect for the programmer's intentions.

 

[Gorbles]

Genuinely meant question. Statistical algorithms do sample libraries of existing art and “collage” it together. But as long as people work out who owns what and property owners enforce the rights to content libraries against each other I don’t see the issue with plagiarism. YouTube is already built to tell if you uploaded a picosecond of Pink Floyd and will appropriately nuke your video from orbit, sparing the record labels the indignation of a picosecond of lost revenue.

This plagiarism thing will naturally require some regulation, some arbitration, as all industry does in order to function properly. But at that point AI art is just a sophisticated tool that happens to produce exactly the slop that corporations needs to keep people staring, slack-jawed, at colorful moving images, possibly on a nonstop basis, forever. There’s something beautiful in that, don’t you think?

You seem to be phrasing things positively, when they don't seem like positive things. We have a cryptic quota around here already, aye?

Feels like I'm in the middle of someone doing a bit about "we live in a society".

 

[Crezth]

Oh, I'm just maintaining a stiff upper lip. I love modern civilization, and phones, and tire spikes, and Slurpee™ machines, and I'm very excited about the new things they're building. You've got to be, if you don't want to be Left Behind.

 

[Hygro]

Hell yeah hail to the big gulp

 

[Valka D'Ur]

Last night I got into another Dune discussion at TrekBBS. Someone brought up the idea of a musical parody of Dune, based on The Sound of Music.

I've worked on that show. Even 40 years later some of it's still running around in my head.

Now the question is... should I actually try this the hard way, or should I cheat and get an AI to do it?

I wonder if the AI would know that it isn't safe to do anything rhythmic out on the desert, since rhythm attracts sandworms?

It's gonna be hard to write a waltz without rhythm...

 

[Hygro]

Sometimes when I code, chatgpt's greatest help is to annoy me with how wrong it is and then I feel motivated to correct it like I would a forum post that gets my goat.

 

[Crezth]

The "hard way" being with a stylus and clay tablet, of course.

 

[Gorbles]

Sometimes when I code, chatgpt's greatest help is to annoy me with how wrong it is and then I feel motivated to correct it like I would a forum post that gets my goat.

* ChatGPT

 

[Hygro]

sometimes i don't care about that gorbly

 

[Gorbles]

Sometimes!

Now you tell me.

 

[Hygro]

Well, sometimes I do, Gorbles!

 

[caketastydelish]

AI is not quite there to publish novels but eventually it will.

 

[Thorgalaeg]

For now it only regurgitates existing information in a lame and boring way. The day it regurgitates existing information in a charmy and interesting way will be like most human writers and will be there.

 

[Kyriakos]

For now it only regurgitates existing information in a lame and boring way. The day it regurgitates existing information in a charmy and interesting way will be like most human writers and will be there.

Just see it as monkeys with typewriters which use entire passages as keys.

 

[Hygro]

For now it only regurgitates existing information in a lame and boring way. The day it regurgitates existing information in a charmy and interesting way will be like most human writers and will be there.

I can get it to "regurgitate" novel and interesting things. Just like how some people can find information on google immediately and others cannot at all, it is its own skill.

You'd be amazed how well it responds to being shamed, and how much better it codes when you give it metaphors.

 

[Crezth]

Maybe the solution is a different mindset. If everyone here just tells themselves "It's impossible for a machine to be as smart, creative, interesting, or funny as even an average human writer," then you can live the rest of your lives knowing with conviction that everything you read that you like was actually written by a human after all. Boom, AI art no longer an issue - ever!

 

[Thorgalaeg]

I can get it to "regurgitate" novel and interesting things. Just like how some people can find information on google immediately and others cannot at all, it is its own skill.

You'd be amazed how well it responds to being shamed, and how much better it codes when you give it metaphors.

It reacts amazingly well, the problem is when you ask it to write something open theme, then it will come with the most dull story ever. The more freedom you give it the worse the result. It is the same as image models, in fact both things are the same ultimatelly.