Forum: Story Discussion and Feedback

I was disappointed to find they were both heavily (and unadvertisedly) AI-influenced, and the AI had been poorly utilised. I gave up on both.

There should be serious consideration places by the Web God here to mandate all AI stories be tagged so, or they risk being deleted.

I am one of those that detests AI, and I don't want anybody to think that because I am some kind of "luddite". I actually remember playing around with "Eliza" in the early 1970s, and have actually worked for decades as an IT specialist (both in the Army and civilian).

I recognize when AI can be useful, but it is not in actually creating anything "original" of note. And the more AI I can see in anything, the more I believe the "creator" is just a talentless hack that refuses to take the effort to learn how to create real works and just wants everything done for them.

At least one author very publicly took his ball and went home over a disputed AI generated tag.

That's why I refuse to fat shame authors I spot using AI. Some of them have stories in the 8s and 9s. So, if they took umbrage at being outed, it would damage the site.

AJ

I would also suggest that there should be some humility as to your ability to correctly identify what is AI.

So far I haven't been proved wrong in incorrectly identifying AI-assisted stories but I have been proved wrong in some of my assessments of stories as AI-free.

AJ

You aren't getting agreement from me here.

Doesn't your opinion that I need your agreement show a lack of humility?

AJ

But the way your reply tries to justify your judgements implies that you want it. If you were uninterested in my opinion, why the effort to justify your judgement to me?

Your 'humility' post implies my judgement is suspect. Whatever you think isn't going to make me change my assessments (which are based on evidence within the stories), but I am curious as to why you think my judgement might be suspect.

AJ

Because, actually measuring the accuracy objectively is impossible.

Technically, maybe, but there are clear parallels with other such story-content assessments such as whether an author is male or female.

AJ

You think your judgment is accurate

From aroslav's recent blog post, I'm considerably underestimating the amount of AI content. The recent flood of reincarnation stories are allegedly AI.

AJ

How do you prove that?

In an absolute sense you can't. But, as in statistical analysis, you can be sure beyond reasonable doubt.

AJ

How many death-row inmates in the USA were exonerated after they were sentenced upon a verdict beyond reasonable doubt?

Too effing many. Around 200 in the US alone since 1972 (the year SCOTUS changed how the death penalty is applied in the US in the case Furman v Georgia).

And I'd wager that number is a significant undercount of the ones who are actually wrongfully convicted.

One is sufficient to end the practice. Imagine if that one were you.

The men who wrote the US Constitution drew the opposite conclusion.

Were they secret Christians?

AJ

If we have to execute 1 innocent person in order to execute hundreds/thousands of guilty murderers then it is moral and right.

Are you willing to be that one? Or have your child be that one?

Not to mention the idea that 'the needs of the many outweigh the needs of the few or the one' is the perfect recipe for oppressive authoritarianism and mob rule. The Founders of the US expressly rejected that idea and built a system to protect individual rights from the mob.

mob rule. The Founders of the US expressly rejected that idea and built a system to protect individual rights from the mob.

Isn't that the definition of democracy? Where the mob over-rules the individual? After all, the mob goes and votes and those votes have greater weight than that of one individual...

A constitutional republic IS a form of democracy. they are not mutually exclusive.

Which is why the authors of the US constitution created a constitutional republic, not a democracy.

There was an interesting short story I read by Harry Turtledove about 30 years ago that covered one of the problems of Democracy.

In it, Athens voted to not build up their navy after the Battle of Marathon, and instead of their naval victory at Salamis Greece looses to Xerxes and it becomes a part of the Persian Empire.

Most of them were rather well read, and they knew from many examples in Greek history that unchecked democracy was as much a threat as absolute rule was. In ancient Greece, it was direct rule democracy and almost nothing could be done without a vote.

Something that Rome learned from as well, which is why they formed a Republic and even created the position of "Dictator".

Isn't that the definition of democracy?

Which is, of course, why the Founders rejected the idea of pure, direct democracy. We could fix MANY problems in the US if the Senate actually represented the States, not the people, as was intended.

We could fix MANY problems in the US if the Senate actually represented the States, not the people, as was intended.

The founding fathers were intelligent and made personal sacrifices to build the country. They weren't in it for themselves.

The opposite of the people now in politics. That's the problem.

Umm... Not really.

Umm, yes, really. While it is possible to change things, those changes are difficult. The fact that SCOTUS upended the balance does not change anything about the original intent of the system, any more than the 16th and 17th Amendments do, despite turning it on its head.

If anything we execute far too few people

No offence meant, and this is just my personal opinion, but civilised countries should not be executing people. Period.

Here's the list of countries that executed someone in 2022 (according to Wikipedia)

China (>1000), Iran (>590), Saudi Arabia (146), Egypt(13), Somalia & Somaliland (19), United States (18), Singapore (11), Iraq (4), Kuwait (7), Palestine (5), South Sudan (2), Bangladesh (4), Myanmar (4), Yemen (1), Japan(1), Syria(1), Jordan(1), North Korea (Unknown), Vietnam(Unknown).

It's quite a list and frankly, the United States sticks out like a sore thumb. Or maybe not. Not my country so not place to say. But as one of my countries allies, I do find it horrific that the US is on this list with this kind of compnay.

https://en.wikipedia.org/wiki/Capital_punishment#By_country

We should absolutely be executing people. In the UK it costs about £55K a year to incarcerate a person.

Should we be paying out 55K a year for the likes of Peter Tobin, Harold Shipman, Peter Sutcliffe, Fred and Rose West etc etc? These individuals and many like them, have no redeeming merits, will never 'rehabilitate', will never be of use to society, only a drain. For one prisoner, you can pay for two medical nurses, who will be of use to society. Just those names alone cost the UK taxpayer almost 300K a year. That's money that can be used to repair roads, replace hips, train up apprentices...

Yes I know some of those names are dead now, but you get the point. UK jails have thousands of inmates who will never be safe for society, why are we keeping them in cells till they die? Just save a fortune and execute them straight after trial.

We are achieving nothing by keeping them alive.

Again, just my opinion, but it isn't a question of cost. It's a question of morality. Civilised societies should not be putting people to death. At all. Ever.

It's a question of morality

You are correct. We all have a moral duty to protect the weak from murderers. Also, any male charged with rape for the second time, should be physically (not chemically) castrated. Their right to own sexual organs does not trump the right of a woman to not be raped.

Civilised societies should not be putting people to death. At all. Ever.

Also correct. But a civilised society would not have murderers and rapists running around killing and raping at will. So until we become a civilised society we should execute till we become one.

Putting a pink bow on a rabid dog does not make it any less rabid.

This is clearly not something we will ever agree on.

Some things are neuanced. Shades of grey. I do not believe this is one of those things. No grey. Just black and white.

If murder is wrong (which it is), then State-sanctioned murder is just as wrong. That's why there isn't a single European country on that list I posted.

If murder is wrong (which it is), then State-sanctioned murder is just as wrong. That's why there isn't a single European country on that list I posted.

Maybe not on that list, but the European armies murder people (state-sanctioned murder). They don't call it murder. They call it war. They even murder people who are non-combatants. Their police also kill people. They are taking the lives of other human beings. And if one of those they kill is about to set off a bomb, they aren't called murders. They're called heroes.

Israel isn't on that list either. That's why Sinwar was in prison after being convicted of murder. And look what he did when he was released. How many innocent people were killed because he was not executed?

Isn't abortion legal in the U.K.? Why does the U.K. allow the killing of something just beginning their life? Don't get me wrong, I'm in favor of abortion. But I'm also in favor of capital punishment when the crime justifies it.

And "mercy killing" (active euthanasia) is allowed in multiple European countries, but not in most of the U.S. Why do those European countries allow the taking of a human life? By your definition, it's morally wrong.

And denying abortion to the victim of rape seems to me to be compounding the harm to the victim.

Allowing a murderer to live is compounding the harm to the victim's loved ones.

This is clearly not something we will ever agree on.

Again, true.

But when people argue that the death penalty is wrong, I generally counter with "Why does the killer's right to life supersede the right of the victim to have had a life?"

It's great for causing chaos with those on street corners arguing that the death penalty is inhumane. They never think of the victim, or the victims. Prison is torture pure and simple and you will find that people who espouse prison over the death penalty are vehemently against zoo's. How come it's inhumane to keep an animal in a small cage, yet it's okay to do the same with a human?

Some things are neuanced (sic). Shades of grey.I do not believe this is one of those things. No grey. Just black and white.

Exactly. You kill a person with deliberate intent (as opposed to accidentally by say, a moments inattention whilst driving) outwith of state sanctioned conflict, then you should be automatically executed. No ifs, no buts. If you deem an other persons life to have so little meaning that you think they don't deserve to live, then you don't deserve to have that right to live either.

I don't think the death penalty is immoral, if you have certainty that you are executing the right person.

Which is a whole new proverbial can of worms.

On the whole, I agree, and governments deciding who is right/wrong is a valid concern (and topical given the recent Alexei Navalny revelations).

However, for the majority of potential candidates, there is little doubt as to their guilt. Like the individual jimq2 mentioned below who continued to kill in prison. That individual has no redeeming qualities and will never be rehabilitated, no matter how many well meaning individuals hold up a candle whilst singing.

Also, the UK stopped executions long before the advent of DNA testing (1964 and 1984 respectively). I accept that DNA is not foolproof, but used in accordance with robust policing, it should result in statistically high probabilities of guilt. Given that one of the main reasons for the stoppage of executions in the UK was the very fear that innocent people would be executed (as, ironically, born out by the re investigation of cold cases using modern DNA forensics), the chances of that have become statistically slim.

At the moment, there is no pervasive dissuasion for killing someone. And if you do get caught (which is still a statistically low outcome) then given good legal representation, you could be out in just a couple of years. That isn't a deterrent, it's a minor inconvenience.

I just read this week a really good story of an author who I really appreciate, who had to make a point and "trans-women raping women in prison"

In recent UK news, a trans-man admitted to a men's mental health facility got raped multiple times.

AJ

As I said, not my country, so I would not presume to tell anyone how it should operate. But if I were a US citizen, I would look at that list of countries and be horrified to be in such company.

I am proud that Britain abolished the death penalty over sixty years ago - ten years before I was born.

Do you not find it interesting that there are only two majority "Christian" countries on that list? The USA and South Sudan.

South Sudan has only existed since 2011 and had a Civil war from 2013 - 2018. It's a very young, extremely unstable post-war country that inherited the Death Penalty from the authoritarian Muslin-majority country is split from. What's the US's excuse?

It's quite a list and frankly, the United States sticks out like a sore thumb.

And only done to those who commit the most egregious murders. Like Frank Wills, who confessed to the rape of 4 women and the murder of 4 women and 1 man. Or Bryan Jennings, who after raping a 6 year old girl picked her up by the ankles and crushed her skull by bashing her into the ground.

Sorry, I really have no sympathy for people who do the kinds of crimes that put them on death row.

Or Bryan Jennings, who after raping a 6 year old girl picked her up by the ankles and crushed her skull by bashing her into the ground.

And then drowned her. Yet Jennings lived another 46 years while going through retrials and appeals. The victim only lived 6 years total. This murderer lived to be 66 years old after killing her when he was 20. Is that fair?

No, broken up for transplant parts

How very Chinese Communist Party of you.

I say bump that 50% up to 80% or 90%.

AI is nothing more than a computer scanning the existing content on the web and then STEALING it.
Everything already on the web is other people's hard work.
To copy a published (even if only "posted") work, then "revamp" it as new content, is the epitome of PLAGIARISM.

There are already plenty of reports from authors out there in cyberspace, from all genre's of writing/media, that have found their work has been copied, cannibalized by AI, and re-published without due credit. I just hope some of the more powerful authors will bring about legal changes to protect us all.

IF YOU WRITE STORIES OR BOOKS WITH AI INSTEAD OF USING YOUR OWN BRAIN/IMAGINATION TO CREATE... YOU ARE A WORTHLESS HACK AND A FRAUD.
I said what I said and I'll stand by it to the death.

I say bump that 50% up to 80% or 90%.

You'd approve of stories 89% AI-generated NOT being flagged?

AJ

If an author doesn't care enough about their story to figure out how to tell it, why should anybody else give a shit about it?

The problem here is that 'use of an LLM' isn't evidence that the author doesn't care. The same argument was made in the 1980s that word processors ('If the author doesn't care enough about getting it right and needs editing tools, why should anyone else care?'), spell check ('If the author can't even spell without some program helping thm ...'), grammar checkers, brainstorming, editors, beta readers, and on and on.

And I remember endless arguments that anything done in Photoshop couldn't be 'creative' because 'The program does all the work. If they were creative, they'd learn to use the tools the rest of us use!'

Note: I'm absolutely not defending someone prompting, 'Write me a story about a guy and a girl having a wonderful and interesting relationship. Go!' That's as creative as the prompt is (meaning, not much at all).

But someone writing thousands of words of character defintions and story bibles, being extremely careful of continuity, editing for overused phrases, and so forth is someone who cares enough enough their story to figure out how to tell it. They're putting effort into telling the story. They're providing the value of putting a story they're invested in but can't write into a form others can read.

I'm not going to use LLMs to write. Not interested, and I can provide a much better character and story bible in my head than an LLM can.

My point is: 'You used tool X! Your story is bad! No one should care!' is a lousy argument. 'You used tool X poorly! Your story is bad!' is a much better argument, but that's a judgment call.

Stephen King once opined, 'It's not the teller. It's the tale.' To my mind, that's true if the teller is a human guiding an LLM. Is the tale good? Then you should care. Is the tale lousy? Then you shouldn't.

The story is what matters, not what tools you use.

Flipping the oft-stated cliche around, if I had an infinite number of monkeys, and they produced Shakespeare, Shakespeare is still Shakespeare. 'Romeo and Juliet' produced by a mind-bogglingly unlikely series of typewriter-key presses by said monkeys is still 'Romeo and Juliet.' The use of an utterly mindless tool that would produce gibberish in the vast majority of cases does not disqualify the output from being brilliant.

I have no interest in reading the output of the monkeys in nearly all cases. And most LLM prose is bland and iffy at best. But that doesn't mean it's impossible to produce a masterpiece by either method.

There's an entirely separate niche for LLM-produced text that I should mention. If you care deeply about fiction about e.g. green unicorns playing foosball using their horns while giggling virgins cheer them on, you will probably find a serious lack of authors writing in your niche. Prompting the LLM to produce green-unicorn-foosball fiction will produce stories that float your boat. No one else may want to read them, and they may be objectively poor writing, but you get to read stories no one would otherwise write and that will actually surprise you occasionally because you didn't write them, either.

But if you want to publish them? Best to do so on a site where you can tag them 'green-unicorn-foosball' and only people in your niche can find them. You're doing a service to a miniscule community, but the rest of us have no interest in them (green unicorn foosball is not an interest of mine).

I see the point you're making, but there's a qualitative difference between grammar checking, spellchecking, word processing, etc and an LLM writing prose, and it's that all of the other tools you mentioned are polishing prose as opposed to creating it. I'm comfortable drawing a line between those two things.

When it's the LLM writing the prose, it's doing so without the author's voice. How can an author even develop their own voice unless they do their own writing?

Some SOL authors do some of their own writing but use LLM scenes to pad out their stories and increase their productivity. It's noticeably jarring when the author switches from human to AI or AI to human.

AJ

If the author is tweaking and adjusting the prompt as they go, and tweaking the prompted output, it's their voice. If their voice reads like 'generic LLM', they have a lousy voice. But that's a matter of prompting and editing.

You're right that going from mixing the two can be jarring. I'm not going to agree that it's always jarring. I've read work by people who mix the two that isn't, but that's because they know how to use the tool.

If the author is tweaking and adjusting the prompt as they go, and tweaking the prompted output, it's their voice.

We're definitely never going to agree on that. I can't imagine why anyone would even think that, unless they don't understand what is meant by an author's voice.

James Patterson's proteges write their novels according to strict guidelines provided by James Patterson. James Patterson provides the plot and even helps to edit the finished story. But the stories don't read like James Patterson's voice.

AJ

I agree that we're probably not going to agree.

To me, an author who works only with 'LLM as a tool' and some editing isn't going to have a voice other than that. Thus, that's their voice. They have no other one.

I haven't read the Patterson stories, but that actually doesn't surprise me. No one is supposed to think Patterson wrote them entirely himself, and the various aspiring authors working with him are also trying to get going themselves. They're likely allowed - even encouraged - to have a bit of their voice bleeding into the story.

The interesting question is whether an AI that has read author X's stories can write in author X's voice. Maybe? Maybe not? But an author trying to make their own stories would be wise to not specify someone else's voice. That would result in something 'not their voice.'

The interesting question is whether an AI that has read author X's stories can write in author X's voice.

I'm going with a resounding 'no' on this one unless that work was part of the training, and even then, I'd be highly skeptical.

I've fed all 15,000,000 words of my work into AI, and tried pretty much everything possible, and I can't even get it to write a character consistently, let alone write in my voice.

This is all experimentation, and so far, it's failed miserably, which is what I had expected. A next-word prediction engine with a PRNG isn't going to be able to write in anyone's voice in a way consistent enough to pass for their work.

On the other hand, as a fact-checker, it's done an amazing job of catching minor historical errors (and a few doozies, too).

By 'read' I meant 'part of the training.' Do any LLMs know J.K. Rowling's voice? Stephen King's? Shakespeare's? Can they reproduce it to a reasonable facsimile?

I'm not sure and haven't tried the experiment.

For untrained work, I agree, I don't see that being possible. Maybe with a LoRA (low-rank adapter, a way of customizing models by training a small addition, not the whole model), but LoRAs seem uncommon for LLMs (while highly common for image models).

I have no interest in doing it, but I'm curious about whether it's possible. I am interested in fact-checking, style checking, and so forth, though.

I am interested in fact-checking, style checking, and so forth, though.

I've built some sophisticated skills in Grok to do this, including a custom dictionary, multi-language support, my own style guide, etc. So far, this only works in Grok Build (CLI). I have not tried to get it working in the GUI/App.

The skills I have are:
typocheck
proofread
factcheck
outline
StoryBible
ManageWiki
ManageArcs/Calendar

Just today I had Grok help me lay out a new story arc, along with dates, locations, and so on, via natural language, and it produced a working calendar with notes for me to write from.

I'm happy to share the skills. Send me a PM and we'll sort it out.

I had Grok help me lay out a new story arc, along with dates, locations, and so on, via natural language, and it produced a working calendar with notes for me to write from.

That wouldn't work for me. I once tried writing from an outline. I didn't get far. The writing was stilted and it was boring for me to write with everything planned out.

When I begin, I know the basics to a story, like the conflict and resolution, but not much detail. I go where the story takes me, writing toward the ending I already know. Which is why I often have to go back and revise something I had written earlier. That's how I get enjoyment from writing.

And then the second step of that enjoyment is trying to make every word and sentence the best it can be. That's the editing or rewriting or whatever you want to call it. That's where most of my time is spent.

The first part is the creative part; the second the technical part.

That wouldn't work for me. I once tried writing from an outline.

The outline is more of a list of events/items I planned. This isn't something the AI came up with independently. IN effect, it's my brain dump, but better organized.

It's akin to my writing notes in the past, but now I have a skill to ensure I keep it all on track — it'll compare what I wrote with the calendar/outline and ask me if I intentionally left it out, etc.

I'm not really doing anything I wasn't doing before; I just have an assistant, as it were.

Effort isn't the metric I'm measuring here; if it's what you want to measure then go for it, but that's a completely different conversation.

I'm responding to your own use of the word 'effort':

If you as a storyteller aren't going to put effort into telling the story, what value are you providing?

If you didn't mean 'effort', or wish to measure it, why did you make a point of it?

If I went and dug out the remote from wherever it's gone off to, and then changed the batteries that are surely dead, and then turned on my tv and found a channel showing the princess bride, I've put effort into making the tv emit a story I want to tell in a well-written format.

And you have had the less creative input on 'The Princess Bride' as someone prompting 'write me a story.' We agree - that's not creative. It's such a low level of effort as to fail your own 'effort' test (which you are apparently disavowing?) or my 'creativity comes from the human's influence over the work' view.

Photoshop is unquestioningly not creative; it is a tool through which creativity is expressed.

We agree.

If an LLM is a tool on par with photoshop, why is it creative?

I didn't say the LLM was creative. Somehow you seem to have misread the passage you quoted ('LLMs are exactly as capable of 'creativity' as the humans using and prompting them are').

Trying to put it differently: Photoshop is exactly as capable of 'creativity' as the human using it and prompting it to do things. LLMs are exactly as capable of 'creativity' as the human using it and prompting it to do things. The creativity of both is the same, and the creativity of resulting works spans the same range for the same reasons.

Both are tools. An uncreative user just doing the minimum results in an uncreative work. A creative user using the tool properly results in a creative work.

Your analogy about Lazeez fails on two points:

1) Lazeez is giving a fundamentally non-creative prompt, on the level of 'Write me a story about.' He isn't trying to be creative.

2) Lazeez is prompting human authors to use their creativity within a bounded topic. The creativity is on the part of the humans.

Mind you (and I've said it before), 'creative' is a terribly ill-defined term. There is no fundamental reason known to humans why a computer cannot be as creative as a human. Humans are Turing machines, computationally, and no more (as far as we know). So are computers. And, if those who believe in determinism are correct, no human is, ever has been, or ever can be creative by the common definitions.

All of that said: I tend to agree that current LLMs are, at best, very weakly 'creative' by the common colloquial definition (yes, I said 'exactly' before; this is a very slight equivocation). Thus the analogy to Photoshop.

They are surely not less creative than Photoshop. If a graphic designer using Photoshop can produce a creative work, so can an author using an LLM.

They could potentially be more creative than Photoshop. If so, an author using an LLM would have a slight advantage in producing creative works over a graphic designer using Photoshop.

Putting a brush to canvas or pencil to paper isn't computation, so the limit of being a Turing machine doesn't apply.

I disagree. Either the brain is directing the brush or pencil, in which case computation is computation, or the brush / pencil are doing random things, in which case random is random. Now, philosophically, I absolutely agree that humans might do something outside of the scope of a Turing machine. However, we have no proof of that. We don't even have a model of computation that's more powerful than a Turing machine, much less any proof that such a thing is possible.

If human intelligence is no more powerful than a Turing machine, a computer can do anything human intelligence can, in theory, given enough computational resources. If human intelligence can exceed what a Turing machine can do, our models of computation are insufficient, and we have no idea whether a machine can be built that reaches those levels of computation.

But Photoshop does what the user instructs it to do, nothing more and nothing less. An LLM does more than that; it fills in gaps that the user failed to specify.

Yes and no, in my opinion. Photoshop is great at 'gap-filling'. A human is hardly painting every pixel in Photoshop. 'Replace this texture with a blur' fills a mountain of gaps. Some blurs look good. Some blurs look awful.

Note that I'm also speaking of LLM+human as a team. The requirement isn't that the LLM produce every word, the requirement is that the LLM is involved as part of the process of creating the work. The human can edit word choice and so forth.

But part of this is going to become an argument over whether the 'gaps' matter, and that's going to vary from reader to reader. Going back to the James Patterson example, some people only want to read something Patterson actually wrote. The gaps filled in by a co-author grate on them. Some people are reading for 'story' and only vaguely care about word choice or authorial voice. Some people are in the middle. There is a wide range of options.

But we're dancing around the original point here. The person in question had a story they wanted to tell. They're lousy at stringing words together in an aestheticly pleasing fashion. The LLM has no story to tell, but it's very good at stringing words together. The two fit together much in the same way as someone who has an eye to visual style and design but is lousy at using traditional tools to edit visual compositions fits together with Photoshop.

A person using the LLM to translate their idea to a finished story is doing work, is putting in effort and producing a document they believe tells the story they want to tell. You may not like that document; that's fine. But faulting them for 'not putting in the effort' seems to be, at best, a strawman, in much the same way as a pre-Photoshop graphic designer faulting one using Photoshop as 'not putting in the effort.'

And part of this is necessarily philosophical. If one's view of 'an author' is a person who chooses each word themself with no help, no use of tools will help.

Philosophically, is a 'James Patterson + coauthor' book more, or less, of a 'James Patterson' book than a 'James Patterson + LLM' book would be? There are arguments either way. If one asserts that LLMs are fundamentally uncreative, then the use of a coauthor dilutes Patterson's creativity more than the use of an LLM does, but the resulting work might arguably be more creative, since there are two sources of creativity.

Both answers are valid. I suspect that how one answers that may influence how one thinks about LLM use in fiction.

And we have fundamental agreement: lousy prompting ('Tell me a story about X, Y, and Z') is fundamentally not very creative and low-effort.

The question is whether 'Tell me a story about this person [4000 words of text digging into their personality, motives, behaviors, speaking style, etc] who does [4000 words of text talking about their adventures] in the company of [8000 words of text about their friends, companions, etc] and are opposed by [4000 words of text about the bad guy, in detail]' is 'effort' and 'creative' or not. If the person doing the prompting has great ideas but expresses them in lousy prose, perhaps the combination keeps the idea but improves the prose? They're certainly putting in a lot of work trying to present their ideas to readers in a format readers want.

That was my original point - that the claim that anyone who uses an LLM to tell a story 'doesn't care enough about their story to figure out how to tell it' or '[isn't] going to put effort into telling the story' is a strawman. The first part is trivially obvious: they've clearly figured out how to tell it. You may not like what they've figured out, but that's beside the point: they've figured out how to tell their story. And the second point is only true to the extent that they've failed to use effort in prompting the LLM to achieve worthy results. No one says you need to like those results, but that's properly a judgment on the results, not a judgment on the level of effort.

Unless, of course, one believes that it's inherently impossible to use effort to prompt an LLM, or that no level of effort in prompting an LLM can affect the quality of its output. But those seem like trivially absurd statements, don't they?

But faulting them for 'not putting in the effort' seems to be, at best, a strawman,

It seems to me that you're equating the effort of providing 16,000 words of LLM prompts (a figure I can imagine Michael Loucks achieving but probably in the 99th percentile) to get, say, a pretty much oven ready 80,000 word novel, with the effort put in by someone writing and editing and refining an 80,000 word novel themselves. SOL shows that having the LLM write some or all of the prose greatly enhances productivity, but the end product won't have the author's voice throughout, at best only in fragmented parts.

Do any LLMs know J.K. Rowling's voice? Stephen King's? Shakespeare's? Can they reproduce it to a reasonable facsimile?

A N other using an LLM trained on J K Rowling's works and prompting 'in the style of' has a better chance of approximating JKR's voice than JKR using an LLM with only a prompt, whether the LLM is trained on her works or not. Allegedly 'in the style of' is the LLM facility most feared by best-selling authors.

AJ

It seems to me that you're equating the effort of providing 16,000 words of LLM prompts (a figure I can imagine Michael Loucks achieving but probably in the 99th percentile) to get, say, a pretty much oven ready 80,000 word novel, with the effort put in by someone writing and editing and refining an 80,000 word novel themselves.

I'm not equating. I'm just saying it's not a negligible or trivial amount of effort.

If one writer is better at writing, that 80,000 word novel will take much less effort than an 80,000 word novel from another author. Does that make the second one better, because that author put in more effort?

SOL shows that having the LLM write some or all of the prose greatly enhances productivity, but the end product won't have the author's voice throughout, at best only in fragmented parts.

A subset of a subset of works on SoL prove that. That's like saying most writers are awful because there are a lot of awful works on SoL. It's not much proof of anything.

It seems to me that you're equating the effort of providing 16,000 words of LLM prompts (a figure I can imagine Michael Loucks achieving but probably in the 99th percentile) [...]

Heck, 16,000 may be too few for the example I'm giving. Say someone has a detailed story. It may take them 30k or more words to explain it to the LLM. The LLM isn't 'writing' the story. It's turning blah, utilitarian story plot points into more pleasing writing.

I'm absolutely not saying most people do that. Nor that LLM-generated works where someone isn't doing that are good (mostly, they're not - maybe entirely). But I am saying it's a legitimate case where the work was both 'written by LLM' and 'human-level creative.'

The LLM isn't 'writing' the story. It's turning blah, utilitarian story plot points into more pleasing writing.

I haven't found that to be the case. All the prose I've generated via an LLM, testing with different models, is mediocre at best and sh-te most of the time.

All the prose I've generated via an LLM, testing with different models, is mediocre at best and sh-te most of the time.

As a writer you might think that, but what do other readers think?

Some of the highest rated recent stories on SOL are partly or wholly AI-generated. As a writer I think the quality is poor - everything is so superficial, the characterisation, the plot, and the third option needed to make this sentence look AI-generated. Yet readers lap it up.

Crash-landed on an alien planet with no way of repairing your ship and getting home to see your family again? Not a single word expressing how the protagonist feels about the situation.

AJ

I'm comparing this to an author whose writing is:

There was this guy. He looked like . He went to school and met his friends. After school, they went off to dinner. While at dinner he met a really interesting girl. She looked like . [And so on, and so on.]

For a writer with writing skills, yes, AI prose is mediocre. For someone who has what might be interesting characters and plot, but struggles to make them anything but purely utilitarian, it's an upgrade.

I actually know people like that. As far as I know, none of them have tried AI writing. One of them spent years on writing skills and produces decent (not amazing, but decent) prose now. The other is likely always going to be hopeless. Great ideas; terrible expression of them.

For a writer with writing skills, yes, AI prose is mediocre. For someone who has what might be interesting characters and plot, but struggles to make them anything but purely utilitarian, it's an upgrade.

I know writers who craft much better prose than I do, so I see your point.

The challenge in doing this is the predictive nature of the output, slightly mitigated by the PRNG. I've done side-by-side comparisons of various models, and while the output isn't identical, the patterns are, and the creativity with prose is limited.

So, yes, it might well be an upgrade, but there's something seriously missing. But that's my 2¢.

If one writer is better at writing, that 80,000 word novel will take much less effort than an 80,000 word novel from another author.

I can't believe you're equating writing speed with quality. That's me put in my place - my best is probably around 2,000 words in a day.

A subset of a subset of works on SoL prove that. That's like saying most writers are awful because there are a lot of awful works on SoL. It's not much proof of anything.

That's SOL put in its place too.

Taking the half-dozen or so highly rated SOL authors who I'm 100% confident have taken to AI to write significant chunks of their recent stories, the contrast between their pre-AI and post-AI work is stark. From the writing style and voice, the former are consistently attributable to their authors, whereas the latter could have been written by any Tom,Dick or Harry because large parts of them actually were.

Heck, 16,000 may be too few for the example I'm giving. Say someone has a detailed story. It may take them 30k or more words to explain it to the LLM.

30k is obviously a made-up figure, it's nearly half a traditional novel. If someone is capable of lucidly writing 30k words, they might as well write the other half too and get something written in their own voice that's copyrightable, instead of a chunk of vanilla same-old. It also helps them to learn their craft properly, which they don't get with a 'ghostwriter'.

AJ

I can't believe you're equating writing speed with quality. That's me put in my place - my best is probably around 2,000 words in a day.

I was, rather explicitly, doing the opposite, saying that a slow writer, who puts in 'more effort', doesn't create a better book simply because of that 'effort.'

Faster isn't necessarily better. Slower isn't necessarily better. And writer X, who does 10x as much 'work' as writer Y, might produce a better, or worse, book than writer Y does with all of that 'effort.'

Taking the half-dozen or so highly rated SOL authors who I'm 100% confident have taken to AI to write significant chunks of their recent stories, the contrast between their pre-AI and post-AI work is stark.

This is an interesting point. I wonder who those authors are (not kidding, nor being sarcastic). Mind you, I haven't been reading a lot of SoL work recently, so I've missed the trend. I pretty much agree, though - if someone is a good writer, actually using LLMs to write prose will, almost certainly, drag the quality down.

I'm not 100% sure of that, though. If writer X has 20 hours to write, is the story better if 10 of that is spent on 'filler' and 10 on 'the good stuff', or 5 on filler via AI and 15 on 'the good stuff?' That's purely a hypothetical, and I don't know the answer. The 'filler' will surely be worse. Will the high points be higher? I don't know.

30k is obviously a made-up figure, it's nearly half a traditional novel. If someone is capable of lucidly writing 30k words

An LLM will take a prompt of the form:
X is male, tall, with brown hair, blue eyes, fit, athletic, smiles a lot, ...
Y is female, short, with red hair, green eyes, fit, athletic, tends to giggle, is a warm and caring person, ...
X and Y meet at a basketball game by accident when Y trips and knocks X over, making him spill his soda on himself
She fusses over him ineffectively, but it's very charming.
They agree to go out to dinner
X finds himself really liking Y and puts out a real effort on the date.
[...]
Story rules: X is a genuinely nice guy, but he has a troubled past because [...].
Y is sensitive to the issues X has because [...]
This creates conflict at point Z.
[...]
Chapter outlines:
Chapter 1: [...]
Chapter 2: [...]
Chapter 3: [...]
[...]

And so forth. Someone can write quite a lot of words filling that out.

Before I derail (which I was about to): there might be a fully developed story there. But our hypothetical author doesn't have a voice, unless that voice is 'utilitarian.' To them, the key is the plot and characters, not how the plot turns into prose.

Pretending either 1) it's easy for anyone to develop the ability to write good prose or 2) everyone even can develop that ability is silly. It's easy for some people, hard for others, and impossible for still others.

But our author still has a good story to tell. Is it better to tell them 'Nope, you're not capable of telling that story in a form others will enjoy' or 'If you use this tool, you can tell this story in a form others will enjoy. It won't be as good as if you could really write, but it'll be a lot better than you can do on your own'?

You're assuming they want to 'learn their craft.' They may just want to tell stories in a way that lets them tell more stories sooner. Or they may be incapable of 'learning their craft.'

And 'copyrightable' is tricky. If there is 'substantial human input', either on the prompting end via very specific prompting or on the output end by substantial editing, the resulting work is copyrightable, per current legal thinking. And that's the hypothetical in this case.

What's the motivation of your hypothetical author? Do they have a good story to tell? How do they know, if all they have is some ideas for characters and a plot outline?

Prose-writing ability is most likely to be a bell curve. Everyone who can hold a conversation and write a letter will be clear of the lower tail and capable of writing a story. If they want to write better prose, getting an LLM to write it for them won't help them learn, any more than riding in the back of a taxi will help them to learn the mechanics of driving.

On the other hand, if your hypothetical author has a good story to tell, the quality of prose matters less than actual story-telling ability. That's an area where not only will an LLM not help, but probably be detrimental as it straitjackets it.

Some of the best-selling storytellers are actually not that good at prose (not mentioning J K Rowling, E L James). SOL has examples of authors whose prose-writing skills are poor, but whose stories earned above average ratings because they're such good storytellers.

Using an LLM seems to be a lose-lose situation unless the primary criterion is speed.

AJ

What's the motivation of your hypothetical author? Do they have a good story to tell? How do they know, if all they have is some ideas for characters and a plot outline?

They believe they have a good story to tell. A lot of 'real authors' start with less than 'ideas for characters' and 'a plot outline,' so that's entirely plausible.

Everyone who can hold a conversation and write a letter will be clear of the lower tail and capable of writing a story.

Maybe, but I know people who are very good conversationalists, lousy letter-writers, and their prose is utterly stark and utilitarian. And that's after serious attempts to improve.

So, in this example, they have no 'story-telling ability', but they have a story to tell.

I'm not going to 'name and shame,' but there are stories on SoL that have moderately positive reviews, many of which praise the ideas and 'the story', but are mechanically so bad I couldn't make myself read them. A somewhat bland LLM translation of those stories would be a serious upgrade for me, because I wouldn't be rolling my eyes four times a sentence at the author's failing around tense, first-person vs third-person, head-hopping, spelling, and on and on and on.

I might roll my eyes at the 'AI slop', but I personally can tolerate more of that than endless mechanical problems. I'm very tolerant of a tense glitch here, a spelling problem there, and so forth. Several in nearly every sentence? I can't do that and just read for story.

Maybe, but I know people who are very good conversationalists, lousy letter-writers, and their prose is utterly stark and utilitarian. And that's after serious attempts to improve.

There's nothing to stop them writing the story in a conversational style. Isn't that how 'The Princess Bride' is written? And if you want an example of utilitarian prose on SOL, it has been observed that Ernest Bywater's prose is like a government report, and that doesn't seem to have done him much harm.

I think you're just making up excuses for your hypothetical author.

So, in this example, they have no 'story-telling ability', but they have a story to tell.

I don't equate story-telling ability with prose-writing ability. Mystery Writer and anotherp08 are examples of good storytellers with poor prose writing. I dread to think what LLMs would have done to their stories if given prose-writing roles rather than just spelling and grammar-checking.

AJ

By 'conversational style', do you mean dialogue? Because some people are absolutely miserable at dialogue.

I can partly agree with you, and partly disagree. As I said, there are stories on SoL that get decent reviews that I simply can't read. If our hypothetical author is only capable of writing stories they themselves wouldn't read, using an LLM to make something that would read would be a fairly compelling option.

but I know people who are very good conversationalists,

That was what you said about your hypothetical wannabe author.

AJ

Yes. They're good conversationalists. They're miserable at written communication, including dialogue. The two are not at all incompatible.

Internal processing of qualia, which LLMs don't have?

I didn't say LLMs are Turing-equivalent. They're not. 'LLMs' weren't even a factor in my discussion of whether (using your words) 'the act of painting or composing or whatever is a computational one.' Introducing LLMs changes the entire point of the comment.

The question is whether a Turing machine can have qualia and process them. I can't find any reason why they can't, unless qualia are supernatural and not processed by the physical brain, or unless the physical brain corresponds to a model more powerful than Turing machines. Since there is currently no such model, nor (obviously) is there any proof that qualia are non-physical in nature, I'm willing to assume other Turing machines could have qualia and process them internally.

Putting it a different way: we know (*) human beings are capable of having qualia. We know they process them. We do not know the details of human cognition, but our current understanding is that human cognition is not qualitatively more powerful than a Turing machine. Thus, either our understanding is wrong and there is another, more powerful, model of processing that humans have, or our understanding is right and a sufficiently powerful 'computer' with sufficiently powerful 'software' can have qualia and process them, since we already have ~8 billion such 'computers' processing them already.

If a 'computer' cannot, where is the line? Can e.g. a dolphin have qualia? A whale? An octopus? Maybe, then, we build future 'computers' around biological components, thus neatly sidestepping the problem.

Or is this something only humans have? A 'soul?' If so, then certainly you're right, and the act of creation is not purely 'computational.' But, for me, that presupposes facts not in evidence.

An LLM is not making choices; it's outputting a sequence of statistically probable tokens.

Granted, but irrelevant, in the same way that Photoshop isn't making choices, it's outputting pixels according to its algorithm as prompted by the user. This is only true if the author actually doesn't care on both the prompting side and the output side. That's an after-the-fact decision. You're treating it as if it's a before-the-fact decision - the mere use of the tool means the author doesn't care.

If Lem pitched his books over a wall and said, 'Here, translate this as-is with no help from me, either in interpreting my writing or in judging your translation. Whatever you do is fine', that's what you're saying all LLM-users, without exception, do.

If the LLM user reads the output and says, 'Nope, this is a bad wording. And this. And this. And this, this, this, and this. Let's fix all of these', either by prompt-editing or output-editing (or, most likely, both), then the LLM user is not uncaring, and your decision to treat them as if they're uncaring is prejudicial.

Again: you're essentially saying there is only one possible way to use this tool, and thus by using it, all possible uses of it are 'bad.'

Maybe they put in a lot of work alongside those words, but fundamentally if some portion of the output was generated by an LLM then that portion is a statistically probable sequence of tokens and not a sentence that was deliberately crafted to convey an idea or a scene or a feeling or a moment.

'Maybe they put in a lot of work outside of the parts Photoshop did programmatically, but fundamentally if some portion of the output was generated by an algorithm than that portion is computational and not deliberately crafted to convey a visual message.'

Leaving out whether that is true or false (because it omits both the direction of the user in invoking those algorithms and the direction of the user in editing the output), even if we assume it's false, the argument is that, because some portion of the output is 'tainted,' the entire work is 'tainted.' That can be true for you, but it's - again - stating that the tool is always 'bad.'

Are most uses of the tool 'bad?', in the case of LLMs? Almost certainly. There's a lot of slop out there.

But is every use of the tool 'bad?', as your thesis requires? That seems unlikely, especially since it's really not difficult to envision someone using LLMs collaboratively, treating both the input (prompting) and output (prose) as iterative, tweaking both repeatedly to obtain well-written final works that meet the author's intent.

That's really where I entered this. The notion that it's fundamentally impossible to use LLMs in a responsible manner to produce fiction, and that therefore every author who uses one must be judged as lazy (not putting in the effort), uncaring, etc, seems quite unreasonable to me.

Perhaps all work touched by an LLM (as we know them today) will be 'inferior' to the human prose a skilled author can produce. That seems entirely reasonable. But is all work touched by an LLM prompted by an unskilled author inferior to the work that author could produce by spending the time 'just writing?' That's an entirely different question.

Whether intentional or not, this has the effect of gatekeeping. It says, in effect, 'If you don't have the skill of writing decent prose, your use of a tool that helps you produce decent prose disqualifies your work from consideration. However great your ideas may be, use of that tool disqualifies them from consideration. And no one would consider them if you wrote them without a tool, since you can't write decent prose.'

And it's also gatekeeping with an additional prejudical slap at the author. Not only are they excluded, they are also 'lacking in effort' and 'uncaring.'

Repeating myself: I'm not in the least defending 'slop,' and there's an awful lot of slop out there. The vast majority of LLM-generated stories are, almost certainly, 'slop.'

Does that mean the tool itself is incapable of being used to produce anything other than 'slop?' In my view, no.

The question is whether a Turing machine can have qualia and process them. I can't find any reason why they can't, unless qualia are supernatural and not processed by the physical brain, or unless the physical brain corresponds to a model more powerful than Turing machines

Or consciousness and creativity are quantum effects which cannot be modeled in refined, processed sand that is very good at math.

That's absolutely possible, I agree. But quantum computers (not the same thing, I know!) do not exceed the computational abilities of a Turing machine by current theory, so it's not exactly that. It could be quantum randomness. Or it could be that 'consciousness' has an impact on the collapse of the waveform. But that begs the question: where does 'consciousness' lie, and what prevents it from lying in a sufficiently complicated set of refined, processed sand?

I'm not certain that it changes the discussion enough. We're still stuck with figuring out if biology can do things that silicon can't and, if so, if this is just human biology or if any sufficiently complex biology is sufficient (in which case we move to biological computers).

You seem focused on 'writing' meaning a completely individual activity without assistance from tools (and, apparently, without significant editing) - and that's fine. I disagree, but that's a personal-choice thing. I think 'decent prose' is 'decent prose' and can be judged as a black box, without any information whatsoever on who wrote said prose or what tools they used.

Jumping back to the Photoshop analogy, if one is focused on 'painting', someone who uses Photoshop has not 'painted a decent work.' If that's the yardstick, they fail.

Going back to the Stephen King quote: 'It is the tale, not he who tells it.' One could also quote D.H. Lawrence: "Never trust the teller, trust the tale." What you're concerned with is who tells it. Again, that's fine - nothing wrong with that. You're judging the quality of the tale based on who tells it.

Now, my original reason for jumping in was your statement that one can judge the teller 'uncaring' and 'lacking effort' based on their choice of tools, not how they use those tools. That seems quite prejudicial, but people are often prejudiced.

For me, I'm interested in reading good stories about interesting things that are written with sufficiently good prose that the execution doesn't throw me out of the tale. The tale is what's being judged, not the teller.

Quoting my previous example (and it's an extreme case, obviously): if the infinite monkeys hand over a perfectly executed copy of 'Romeo and Juliet', it's still a great story, well worth the reading, even if the monkeys never did anything even vaguely qualifying as 'writing.' And, if said monkeys were to fork over a perfectly executed play Shakespeare died before writing, but would have written had he had another decade of productive writing time in good health, it might well still be a masterwork, notwithstanding their lack of 'writing.'

As for the distinction between 'editing' versus 'writing', that strikes me as a strawman. Unless one is perfect, 'writing' involves 'editing.' Some people are terrible 'writers.' Their first draft is slapdash, with broken sentences, misspellings, logic faults, poor dialogue, and so forth. The actual story emerges during 'editing'.

The resulting work is entirely theirs. But, by the criteria you're espousing, they're terrible writers and thus not worth any consideration.

You're interested in reading stories written by hand by good writers, and (apparently) in judging anyone who publishes any other sort of story as 'uncaring' and 'lacking effort.' I'm interested in reading good stories, and trying not to be prejudiced about the creative process the author uses to produce said story. That, in a nutshell, feels like the root of the disconnect.

-- Additional thought below, not really core to the discussion.

There are obvious points of debate, and there's a continuum here. If one takes 'Romeo and Juliet' and uses find and replace to make it Ron and Jenny, that's uninteresting, clearly, notwithstanding that the story itself is no less brilliant than Shaakespeare. But, if one takes 'Romeo and Juliet' and makes it 'Tony' and 'Maria', with significant creative work, that's interesting.

That example was specifically chosen to hit the fine line between 'plagiarism' and 'inspiration.' Some people find 'West Side Story' to simply be derivative, others find it very worthy art. But, however you answer, 'West Side Story' is demonstrably more derivative of a major work than most LLM output is of any particular work.

Going back to the James Patterson example, some people only want to read something Patterson actually wrote. The gaps filled in by a co-author grate on them.

I wouldn't call them gaps when they constitute virtually all the prose in the story.

A long while back (so I'm not sure what the current situation is), I read a few books by James Patterson's proteges. One was particularly dire, and I'm pretty sure it wouldn't have had a cat in hell's chance of being published without the 'James Patterson' on the cover. Others were better, and one was good enough to make me want to read more of her work if she ever went solo. Sadly I've since lost her name.

In my opinion, none of them were up to James Patterson's standard and none of them captured his voice, so it's understandable that readers might shun works not written by James Patterson himself.

AJ

LLMs are exactly as capable of 'creativity' as the human using it and prompting it to do things. The creativity of both is the same, and the creativity of resulting works spans the same range for the same reasons.

Given past revelations that LLMs don't necessarily choose the most popular token to follow the previous token, it probably has more creativity than a random number generator.

The chances of LLMs producing the complete works of Shakespeare are probably greater than monkeys with typewriters (not mentioning myself).

AJ

Star Wars is famously a mishmash of 1950s sci-fi tropes

Not really, but it is famous for borrowing heavily from Akira Kurosawa's The Hidden Fortress.

Had the comment been 1930s, it would have been accurate. Lucas was quite clear about having been influenced by Buck Rogers and Flash Gordon, and that's pretty visible in the final product.

It's also heavily based on Joseph Campbell's work, as well as Buddhism, the Tao, and the Vietnam War (Lucas saw the Rebellion as somewhat of an analogue of the Viet Cong, albeit perhaps a Viet Cong with purer motives).

The base point, that Star Wars is 'derivative', is correct. It's also, simultaneously, 'fresh and new.' Both can be true.

'West Side Story' can simultaneously be derivative and inventive, as can 'R+J'. Or 'Warm Bodies', for that matter. Three very different works (though one is word-for-word 'Romeo and Juliet'), each interesting and creative.

And the more AI I can see in anything, the more I believe the "creator" is just a talentless hack that refuses to take the effort to learn how to create real works and just wants everything done for them.

And it's an act of self-harm.

Creative writing is one of the best ways of staving off the effects of brain-ageing (albeit not as good as learning a musical instrument or a foreign language).

Even teenagers who depend on AI have been found to have less functional brains, with fewer connections, than their peers.

AJ

true

It is important when using a tool to know what its limitations are

If someone wrote a 'perfect' story, then submitted it to an AI for grammar, spelling and punctuation corrections, and better flow and word choice, the AI would make 'corrections' to the perfect story because it's been designed to please.

AJ

If someone wrote a 'perfect' story, then submitted it to an AI for grammar, spelling and punctuation corrections, and better flow and word choice, the AI would make 'corrections' to the perfect story because it's been designed to please.

My experience is actually the opposite. Grammarly most decidedly does NOT want to please me, and despite clicking on 'Dismiss', which is supposed to train it to not complain about certain things, it still does not work.

Two examples:

1. I turned off checking for split infinitives, and it refuses to stop flagging them, even when the 'to not' is used for emphasis (which is fairly common in English).

2. It has decided, in its more recent update, to insist that phrases such as 'you literally cannot do that' must be changed to 'you can not do that literally' in every single instance. It demands that 'literally' be moved to a place where the sentence no longer makes any sense.

I've reported these two things to Grammarly several times, and each time the request is dropped into a black hole and marked 'resolved' even though no resolution is available.

So, at least in Grammarly's case, it is not trying to please me; it's serving to p-ss me off!

In a similar vein, Grok returns different results for tasks than it does if you run a prompt interactively. I reported that, and had a conversation with an engineer who said the 'task' (i.e., background job) does not do well with real-time data. That kind of makes it useless for me to have a job that runs after the stock market closes with the day's market results.

Yeah, not seeing the 'designed to please' there. 😜

AI doesn't necessarily make a poor story, but like any tool, if you don't know how to use it and accommodate its limitations, you don't get a worthwhile end product.

While reading a story yesterday, I realised there's another AI tell.

People use different language in different circumstances eg formal English for scientific papers, legal documents etc, florid language for rhetoric and descriptions, simple language for conversations. It appears LLMs don't make record this distinction, so AI can generate florid language, full of 100$ words, for occasions when real people would use simple language - for example gun safety instructions, where maximum comprehension is paramount.

AJ

I agree on the difficulty finding stories to engage me lately. I am hoping that its just a lag while some the better authors work on new postings.

It is also likely that a great many stories I am seeing now tend to be more and more extreme in concepts.

I do not joke when I say that I am sick of all the "motherfucking incest stories" in here. It seems like over half of them or more are based primarily around incest. And often take the "kitchen sink" approach of putting in as many tags as possible in what should be a 4 or 5 page story.

Honestly, I would rather read a "Mad Libs" story than most of the AI crap I see.

Maybe it's just me, though.

Nah, mate, it's not just you. There's been a marked decline in the quality of stories recently, even in just the few short years I've been active here. Between the accelerating die-off of OG contributors, the rampant and unchecked use of AI to wholesale generate slop content, and the apparent trending away from erotica in general toward derivative, unoriginal fantasy/sci-fi and RPG/game lit, the future is looking mighty dim. Fully one quarter of recently-posted content on this site is tagged no sex. When did SOL become Wattpad or AO3? Aren't there two sister sites specifically for this kind of content? I frankly have no interest in reading some kid's shitty, AI-generated fan fiction, yet here we are.

Fully one quarter of recently-posted content on this site is tagged no sex.

Don't abandon the ship just yet. I'm writing a lighthearted short story with sex for the fun of it. It'll soon be posted.

Fully one quarter of recently-posted content on this site is tagged no sex.

If you check the "All Time Classic" story lists, almost all of them are no-sex, including the highest-rated ones in SOL history.

A couple of my lowest-rated stories are no-sex, so even after 20 years, I obviously still don't know what I'm doing.

If you check the "All Time Classic" story lists, almost all of them are no-sex, including the highest-rated ones in SOL history.

This does apply to the "All-Time Short Classic" stories, but the highest-rated ones are mostly memorials and tributes. They are expressions of the authors' deep emotions, and connect with readers on a deeper level than the more erotic ones. Only a few are stories.

For the Long Classics, 20 of the top 25 have at least "Some Sex".

And you know this, how? If you believe that, why stay here?
I have strong feelings against AI-created/-polished text, but I see plenty of stories in the Updated Serials page by authors who have a long history of being able to write without AI assistance. So I'm calling bullshit on your offensive post.

just about everything here now is AI generated.

Do you have data to back up this assertion, or is this just as a case of making an outrageous statement and expecting others to believe you?

Anything before 2020 is likely to be written by a human.

I would say that's a bit conservative. Anything before 2023 is fairly safe.

Equally, like how AI imagery was really rubbish a year ago, it's now getting to the stage where it's almost lifelike. And as for moving images, they were terrible with no lip sync six months ago, now they are actually really good. Albeit for only ten seconds in length. Give it another year and they will have sorted the length problem. I reckon in a year or so, AI written works will start to become really good. It's just a matter a of time.

It's going to arrive a lot sooner than people are expecting and lot of people are going to be out of work. Look at Only Fans, more and more of the 'models' on it are AI constructs and that's only going to increase as the months pass.

AI is also getting easier to use, so you are going to see a lot more content of...everything... as people find they can 'create' with just a few text/voice prompts. Given the depravity of some peoples minds, LEA are in for a wild ride!

Equally, like how AI imagery was really rubbish a year ago, it's now getting to the stage where it's almost lifelike.

My Sunday newspaper, which I buy for its once-decent football coverage, likes to adorn the rest of its content with female celebs wearing not a lot (but no nudes). It's getting harder and harder to detect where the images have been enhanced but one weak area is thumbs, presumably because AI doesn't grok them. In several cases it seems to have admitted defeat and left the celeb with four fingers but no thumbs.

There seems to be a sweet spot for AI-generated chapters of 1,300-1,600 words. If you see a new story on SOL's home page with a first chapter of that length, it should scream, "Danger, Will Robinson!"

AJ

There seems to be a sweet spot for AI-generated chapters of 1,300-1,600 words. If you see a new story on SOL's home page with a first chapter of that length, it should scream, "Danger,

Whoa! That chapter length could be one of my stories. I may not be the most intellectual, but what intelligence I have is not artificial.

So short chapters and em-dashes define artificial intelligence? OMG! I'm an AI.

I may not be the most intellectual, but what intelligence I have is not artificial

But that's what an AI would say... 🤔 🤪

Whoa! That chapter length could be one of my stories

Me too!

Actually, looking at today's AI offerings, perhaps that should be 3000 words.

AJ

but one weak area is thumbs,

That used to be the case.

Take for instance the following Youtube video. Totally AI generated. All of it, the music, lyrics and video.

https://youtu.be/63vgx6Mq88g?si=2qglu2pTcfY1w5Tb

It's weak in places, most notably at 23 seconds where the couple 'try' to hold hands. What is notable about the video, is that AI used to be abysmal with tattoos. Now it can keep them 'on the flesh' and keep the form/shape of them without distortion as limbs moves.

The other issue was lip sync. However they are getting on top of that as well. Take the following;

https://youtu.be/iM6XBNOkKQg?si=lT83PsAz7t7SZxU1

For most people, especially those who consume media on their phones, that is more than decent enough to watch and is almost lifelike in quality. Would I watch a TV series or film with AI characters of that visual or vocal quality? Yes, yes I would.

And you will note AJ, that the thumbs are pretty decent on both.

All that's holding them back, is that it all goes to shit after about ten seconds (I'm pretty sure I read somewhere that the processing power increases exponentially which is why clips/takes are restricted to ten seconds (ish)). Given the speed at which things are progressing (This was all a pipe dream five years ago), I would not be surprised if they fix the time issue in two to three years.

I'm actually looking forward to it. Can you imagine the ability to upload any book ever written and have it made into a film a few hours later...

It cost studios millions and months of work to create the likes of Oliver Reed for just a minute or so screen time for Gladiator, 25 years ago. Now, someone in their bedroom can have Oliver Reed at that same quality, do and say pretty much what they want, for forty odd pounds a month.

Would I watch a TV series or film with AI characters of that visual or vocal quality?

Allegedly the Wallace and Gromit films were all illegally scraped, so now anyone can use AI to produce their own Wallace and Gromit films without manipulating plasticine figures for a year.

AJ

You do realize that you can split these ~500 token outputs to get smaller chapters, and you can paste multiple chapters into one file to get a longer file.

Some authors do that, and integrate AI-generated stuff into their stories quite seamlessly. It's the beginners who reveal themselves with chapter lengths, em-dashes etc.

AJ

I hope he doesn't - otherwise people will adjust the AI output to hide.

A number of authors already do that, integrating AI-generated descriptions with enough genuine story to keep the AI proportion below Lazeez's 50%.

I'd be interested in the results if every chapter had to stay below 50%. I was reading an admittedly good story which has avoided the AI tag when I hit a chapter that was pure slop. I guess the author was having an off day and dialled in that chapter, but I found it very jarring.

AJ

I've finished a grand total of 1 story to date

Impressive. I haven't even managed to finish one yet...

'Hot with a fiery lust burning like Mount Vesuvius she screamed, "play Raid: Shadow Legends the hit mobile..."

The problem people will run into with these is that they aren't free to use.

I hope that's to pay royalties to the authors whose in-copyright works they used for training ;-)

AJ

I just wish there was an easy way to simply discard all of them so I never see them in searches.

Use the Category Exclusion Preferences. Check the "AI Generated" tag. It'll exclude those that have the tag.

That was a shit ton of paragraphs starting with I's 😂

Were you you the individual I noticed doing this before? Or was that someone else? I have no idea as to what conversation this was raised in before, as my memory is pretty much goosed after a week or two...

More to do with blows rather than age... 😂

YouTube

When a video is AI-generated, I click 'Do not suggest this channel in the future' after giving the video a thumbs down.

I've seen some of the most godawful crap imaginable, unwatchable for more than a few seconds. Other content is done reasonably well, but the dead giveaway is when it reads a date out as if it were a number ('in one thousand nine-hundred thirteen', rather than 'in nineteen-thirteen'), something no human narrator would do, and bang, into the AI garbage bin.

I've noticed (today in fact) some very similar You-tube channels all with the same visual AI narrator (It's very good, I didn't at first spot it until I noticed the eye movements are on a repeating cycle).

I've noticed all the video's are based on the same subject, (peddling financial falsehoods as fact), and I wonder if this is actually a Chinese Psyop to try and destabilise the American dollar via easily impressionable American citizens.

So far I have noticed the following; Econemy (sic) Meet History, Your John AG, The Archivist, Macro Finance Explained, Boring Historian, The Boring Currency, Coin Financials, channels are basically the same. There are probably more, but these are the ones I spotted tonight.

The worst one I saw was 'Data Guru'. It was, in fact, the only AI channel I watched, mainly because it was slick enough to draw people in and seem accurate and authoritative.

I debunked each video (easily, given my knowledge of history and finance) and was nearly always the only commenter.

After about three months, as views decreased, they deleted the entire channel and all the videos. 😎 I'd guess the monetization didn't even cover the cost of their AI slop!

but there's a big difference between an AI-written story, a human-written story

I believe several authors use AI to generate wallpaper to flesh out human-written plot and character passages. With practice, they can stay under management's 50% disclosure limit.

AJ

I believe several authors use AI to generate wallpaper to flesh out human-written plot and character passages. With practice, they can stay under management's 50% disclosure limit.

I've continued my testing of AI generation, and it is markedly improved, to the point where I could, if I chose, have it generate an entire chapter and then edit it to conform my closely to my style, correcting any egregious errors.

It won't be long before it could, based on my corpus of work, generate 90% or more in a form that's easily tweaked for inclusion. The key was feeding it my writing to use as a base. I'm actually impressed.

I only actually use AI for research (which saves me a TON of time, though I do have to verify sources). A perfect example was asking about the process of creating oil of vitriol in medieval times, which explained the step-by-step process, including the time required, the materials, and so on. It saved me at least half a day of research.

I think we're actually not far from the point where a good model with a body of work could write stories you couldn't distinguish from ones I wrote. That both impresses and depresses me.

Was this post written by you, or an AI trained on your posts?

An AI, obviously, because all the paragraphs start with 'I'.

(Sorry, Pixy. You snooze, you lose)

AJ

😂

(You know you can't not notice it now...)

Was this post written by you, or an AI trained on your posts? Forget that, how would I be able to tell if the response was generated by an AI purporting to be you?

Me, despite the comments about starting each paragraph with 'I'. Didnd't even use AI to spellcheck or proofread it. And no sentence in this response begins with that accursed letter! 🤪

And no sentence in this response begins with that accursed letter!

Totally convinced I am not that a succession of paragraphs starting with 'I' is a reliable indicator of AI-generation.

Female authors would probably generate more false positives from AI-detectors than male authors because it's one of the 'spots' for a female author.

Of course, it's always possible that all AIs are female: la AI, une AI as opposed to le AI, un AI ;-)

AJ

The more mistakes, the more likely it was written by a human.

Why waist time when their is an easy answer?

We're safe. This post was written my a human. LOL

(waist/waste, their/there)

I suspect that was done deliberately SB, so as to make a point.

I suspect that was done deliberately SB, so as to make a point.

My comment with the "LOL" was adding to the humor.

India is an Asian subcontinent. It is not part of Africa or Europe.

The 'Asian' label is stupid. It's supposed to refer to Chinese, Japanese, Korean, etc... (you know, the ones with that race). It's the 'Politically correct' version of Oriental (I don't know why that one is not kosher).

Asia is the largest continent. So when one says 'Asian' it can mean very blonde Russians, Brown Indians, Sirilankans , Middle Easterns, Arab Peninsula, Iranians, Pakistanis, Turkish, etc...

If we really need to replace 'Oriental' to designate those with distinctive eyes and yellowish skin, then we need to find a replacement for 'Asian' because 'Asian' can mean anything other than African black.

I'm Iraqi. Iraq is in Asia. Nobody would see me and refer to me as 'Asian'. Even though I'm from the Asian continent.

We should just abandon the idea of "race" completely and just use ethnicity.

That's a non-starter. How do you prioritise blacks for PSA testing, for example, if you're not allowed to specify their race?

AJ

I wish people would quit pretending Pakis and Indians are "Asian."

I used to correspond with a girl from India. She considered herself Asian.

I used to correspond with a girl from India. She considered herself Asian.

In the UK, the 'Asian grooming gangs' are principally of Pakistani origin.

AJ

It looks as if it's going to get worse…

If media reports are to be believed, none of the AI firms are actually making money yet. That's a situation impossible to sustain.

"Sex sells" exists for a reason because it's accurate. It's in almost all advertising, even if it's not immediately obvious. So it's hardly a revelation that firms are turning to porn in an effort to create a revenue stream.

At the moment, everyone has their knickers in a twist about the recent Tom Cruise Vs Brad Pitt scuffle. Inevitable and not surprising in the slightest. Whilst some are pointing out that one individual and an AI program has the potential to wipe out the entirety of Hollywood, the impact goes a lot further. Fight scenes are complex as limbs and body positions dynamically change and are difficult to render without limbs either going through the opponents body or generally merging with the scenery. They (AI) has obviously mastered that to a level that's believable (even if both of them did change their tops mid scuffle). So it's not going to be too much of a challenge to render something a lot slower and less limb excitable.

There is going to be an absolute torrent (in both meanings) of artificial porn, that can be created simply by word prompts. That's going to be very tempting to those with more extreme sexual fetishes. Are those individuals willing to pay to create their darkest desires? Of course they will. It will be like letting the proverbial fat kid loose in a sweet shop.

All the big AI firms desperately need funding and I seriously doubt they will have the moral character to turn down lucrative revenue streams. They might make public statements that they will restrict what AI renders to what's 'legal', but you can bet they will drag their feet for as long as they can get away with it, and once these 'checks and balances' are in place, there will suddenly be 'hacks' that can then render those 'checks and balances' obsolete.

It reminds me of when legislators suddenly became worried about the violence in video games, remember Carmageddon in the late 90's? In an attempt to appease various governments, the blood was changed from red to something else and 'pedestrians' became 'zombies' so the game could then be sold. Which slightly appeased some. As soon as their back was turned, patches were released that undid all the censorship, making the whole censorship thing a complete waste of time.

Ironically, the whole situation is not unlike American arms manufactures; "Yes, we make guns, but we are not responsible for what our customers do with our weapons after we sell them."

Legal process is notoriously slow in making legislature/legislation in pretty much all Western countries and I seriously doubt they understand what's about to hit them. Take for instance the UK. The Government and legal profession thought that age restricting porn would make their problems go away, even though every single IT professional and person with a bit of common sense said that it was going to be a waste of time. That people would either use VPN's or flirt with the Dark Web (and therefore find themselves faced with content of an extremity way beyond their comprehension) and thereby make the situation a whole lot worse. So typical of the UK Government, they took a small issue/fire and tried to extinguish it with petrol. Now, thousands of people who would never have been interested in the Dark Web, are stumbling around blindly in it, looking for their favourite 'Housewife calls the plumber' content and finding shit way more extreme.

At some point, in the near future, people will be able to run LLM's on their personal computers at home and all those legal 'checks and balances' that will have been created months, if not years, after the AI horse has bolted out of the digital stable, will have become useless and non-enforceable.

Interesting times and all that...

At some point, in the near future, people will be able to run LLM's on their personal computers at home

I already do this using LMStudio and several standalone models. I'm working on an AI Agent that will run on my Mac Mini and perform portfolio analysis every night. I already have prompts for stock analysis, with the results displayed visually in tabular format and provided as a CSV.

I already do this

Well, there you go.... 😆

Fear not - soon AI "authors" will only have AI "readers" to please.

That's wishful thinking. Authors seem to be finding that using AI to generate large sections of their stories increases their 'productivity'. I'm not sure SOL readers recognise the slop, when the scores awarded to those stories hold up in the eights and nines.

AJ

I'm not sure SOL readers recognise the slop, when the scores awarded to those stories hold up in the eights and nines.

I've tried many non-AI stories with scores in the eights and nines that I thought were awful. So maybe AI is giving SOL readers what they want.

No one is forcing you to open and read it.

Besides, this is how normal conversations in the real world go. Someone says something, someone replies, someone replies to the reply and after a cup of coffee, someone asks "What were we talking about again?"

I read the story and enjoyed it but not enough to recommend it, as I have done with others in the past.

AJ

Now Megumi gets a clitoride award for best new author.

I have mixed feelings about that, but I can't point to any other new (in 2025) author who is clearly better.

Who the hell writes 21 stories, spanning 495,000+ words, over a span of eight weeks.

On my best days, I could generate about 20,000 words per day, and if I could stay in the zone, I could keep that up for maybe 3 days.

My usual output is around 6,000 words per day (for the past 10+ years). In 8 weeks, that would be about 335,000 words. One or two 3-4 day 'in the zone' periods would raise that above 400,000.

I only use machine assistance for grammar/spell checking.

I've only read her child-prodigy-defying-the-odds stories. Can you recommend any of her erotic stories to try? Mind you, IMO some of the clitoride-winning stories are stretching the definition of 'erotic' to the very limit.

AJ

Can someone with military experience please tell me if the above is correct? Drones couldn't detect a warm body on a cold rock?

In the vast majority of realistic scenarios, a drone equipped with a thermal infrared (IR) camera would reliably detect a living human body lying still on a cold rock face. The detection relies on fundamental physics of heat emission rather than movement, visual contrast, or ambient light, and a cold rock background typically provides excellent thermal contrast.

It might be true of a vehicle. A vehicle that had been shut down and unmoving for three days would have cooled to ambient temperature and could be difficult to detect on infrared.

Not true. The metal work on a vehicle absorbs AND loses heat faster than the surrounding ground. Partly caused by the insulation provided by the rubber of the wheels and the restricted air pocket beneath the vehicle.

What's the significance of 'partial shade'?

Could that disrupt the IR image so the done sees bits of a human body instead of a more recognisable whole?

AJ

I saw what you did.

Non-members can only see the first two replies, which pretty much echo the replies here. If any insightful revelations are divulged, I hope you'll pass them on.

AJ

I've moved this list to a blog post.

ETA I've added a poll to the blog.

AJ

To compete in either writing category (original or adapted screenplays), a script must have been written entirely by one or more humans who were living and breathing at the time they wrote it (read: no AI).

That would nuke some of the 2025 Clitoride winners. Would that make the results better or worse? The voters didn't seem to mind.

AJ

The voters didn't seem to mind.

Ah, but the voters for the Oscars are the people in the business. Remember the recent screenwriter's strike? They didn't want their jobs replaced by AI.

"Best NEW author of 2025" has been posting on this site since 2021,

Since the Clitorides isn't limited to SOL, maybe the author is new to another site and those readers nominated him as new author.

To compete in either writing category (original or adapted screenplays), a script must have been written entirely by one or more humans who were living and breathing at the time they wrote it (read: no AI).

That would seem to exclude the use of AI to improve the punctuation, but how would anyone, other than the author, actually know?

AJ

That would seem to exclude the use of AI to improve the punctuation, but how would anyone, other than the author, actually know?

How would they know if AI was used for more than punctuation? They are deathly afraid of AI so they put in the rule.

How would they know if AI was used for more than punctuation? They are deathly afraid of AI so they put in the rule.

And the accusations will fly when the 'wrong' winner is chosen.