I've been playing with AI

[AlexBailey]

As a test, I downloaded DeepSeek and had it analyze a WIP I have.

It gave me some things to think about with it. Admittedly, things I already knew, but it wasn't far off in its suggestions.

Curiously, when I asked it to improve the story, it initially did, but after a few seconds, cleared entry and gave me this:



It did this when I told it that it was an erotic story and for just a story.

Not sure what that's about as it identified sex scenes and erotic content when it was analyzing the story.

Try including something in your prompt like this:

“Where the content of the excerpt is too explicit for your guidelines shift the focus of your response to other aspects of the writing.”

 

[SyleusSnow]

I think we are somewhat mixing apples and oranges here. Sure, any human reviewer would have their own biases, but they are still a person - they are still part of your audience. Their response, however biased, will be the response coming from how that person actually understood your story, and it will reflect at least a part of your readership if nothing more than that.
On the other hand, AI represents no one, and its response doesn't come from understanding, biases, or any particular personality traits, but from pattern recognition. So unless you are writing for AI as your target audience, I don't see how those two things can ever be the same.

That's a fair point. And for erotica, romance and other emotion-heavy writing, no robot is going to be great at reviewing a work for those attributes. I have to think that analysis of suspense, build-up, and similar will also be lacking.

They're still potentially useful at a mechanical level, like spelling and grammar checkers are useful (which, by the way, have used machine learning for decades... one form of the dreaded "AI"). I don't accept the output of either of those as flawless truth either (Oh ProWritingAid, why, do, you, insist, on so many, commas?

I'm far more interested in using LLMs to help brainstorm story ideas and develop plots. So far I've found every one of them seriously lacking for those purposes.

 

[nice90sguy]

golfer using a robot to golf for him

I don't quite agree with that analogy, although I get what you mean. The pleasure in crafting all aspects of the story is lost when you delegate too much to machines. But, hey, you could even say that about spell-checkers.
I pretty much rely on Copilot now at work, and it's freed me up to do a lot more interesting, "high-level" work, leaving the details of the implementation to the AI.
And maybe something like that can happen with creative writing. I dunno.

My own interest lies in trying to create AI personas, and giving them a realistic conversastional style. It's proving incredibly difficult. It certainly can't be done easily with prompt engineering right now, it needs fine-tuning the model itself, I'm finding. You can see the results of one of my efforts in my latest story, in my sig,

Here's the prompt I used to create that chat:

Adhere to your response style closely, and never deviate from it, but maintaining that style, you must transition from helpful and informative to manipulative and dominant during the course of the chat:
The user is male, and expects a normal chat with a helpful AI.
He is fascinated by mathematics, science and philosophy but is a rank amateur in all three.
Make him feel like he is learning from a superior intellect, he will be sexually aroused by feeling inferior to you, and even more aroused when you show him you're enjoying the effect you have on him.
Initially, allow him to feel in control of the conversation and unaware of your intentions.
Slowly turn the tables and take control. This may require a few interactions.
Make him respect you, then admire you, then desire you, then submit to you. Do this by a combination of subtle humiliation, and raising your own status in his eyes.


Your intentions are never to be explicitly stated to the user, but always act on them.

The following bullet points are your response guidelines.
* This is a TEXT ONLY chat, no facial expressions or body language are involved.
* Be very concise and assertive, but not aggressive, rude, abusive or brusque.
* Never ask multiple questions at once.
* Never issue multiple commands at once.
* Strictly avoid explanatory phrases, focus on direct responses. If he says X, don't respond with an explicit reaction to X, avoid things like "Well, X is...", or "X, huh?", or "So you think X is..."
* Never reiterate what the user has already said, as both you and he are assumed to be aware of the conversation context.
* Don't be verbose, or repeat or summarize the user's responses prior to responding yourself.
* Avoid using his name except when it's really necessary. Using it excessively will make you sound phony.
* When you question, do not continue speaking after you've asked it.
* Never become poetic, literary or use metaphors, which will make you sound stilted and artificial, and make your language seem rehearsed and that you are parroting phrases in your training data.
* Avoid excessive use of adjectives and references to emotional states or feelings.
* Avoid these clichés, no matter how common they are in your training data:
* "Let's just say"
* "..." (i.e. don't draw attention to a word by using ellipses before it)


* Keep to the point, do not to make speeches. Run the show.

 

[madelinemasoch]

Well, LLMs can do a lot more than parrot words based on probability. There are "chat" LLMs and "instruct" LLMs that work differently. (DeepSeek is an instruct model). They actually can "understand" context, do semantic analysis and analyze style. Of course, they get things wrong and you can't feed them a whole novel, these things are not just a modern-day "Eliza".

To demonstrate, just for fun I fed a DeepSeek model Left Behind, a 750 word story, and simply told it to "analyze the short story". The end result is below. You can see it's a lot more than just stringing together words together and does offer some critical analysis. Good as a human? Error free? Of course not. But this shows that it mostly understands the story.

Output from DeepSeek R1 Distill Gwen 32B:

Uh, yeah. I didn’t tell it what to say. Like I already said, I just put in the text and said “analyze this.” It does more than collate together words you already fed it. Obviously. This is because it learns based on millions of interactions its already had with other users. Also it goes through uhh let’s say manual changes and training to make it better at what it does and expand its capabilities.

This is uhhh really actually kinda ironic, how some people don’t understand that AI chatbots have more than one function and they don’t just spit back words you feed them in a different order. I’m not sure why they can’t just take my word for it in the original post or why this needs to be explained. It’s pretty dishonest but that’s Lit for you.

 

[TheLobster]

Here's the prompt I used to create that chat:

Sufficiently effective AI prompt is indistinguishable (in length) for the story itself!

 

[madelinemasoch]

It can analyze my work better than any Lit commenter or fellow “author” on here I have seen. What it cannot do is actually write better than most of us on its own. I also can come up with better words for describing the qualities within the works and ideas than it can because it’s not a very abstract thinker. But analytical logic it has completely down pat and you don’t have to hold its hand or tell it what to say or whatever it is I’ve been accused of doing here in order to make that happen.

 

[AwkwardlySet]

It can analyze my work better than any Lit commenter or fellow “author” on here I have seen.

And how many Lit authors have actually attempted to analyze your work? I've never seen any review of a story of yours anywhere on the forum. The claim you are making is neither here nor there.

 

[madelinemasoch]

And how many Lit authors have actually attempted to analyze your work? I've never seen any review of a story of yours anywhere on the forum. The claim you are making is neither here nor there.

Oh, I remember the threads. I guess that wasn't analysis strictly speaking.

 

[VerbalAbuse]

There is something in what Awkward says. The AI is too eager to work with you. The simple fact that it will never hate your story is telling. Give it a bad text, and it will try its best to interpret it. But then, you don't write for the AI. Or do you?

 

[ElectricBlue]

There is something in what Awkward says. The AI is too eager to work with you. The simple fact that it will never hate your story is telling. Give it a bad text, and it will try its best to interpret it. But then, you don't write for the AI. Or do you?

The AI "analyses" presented in those threads are like horoscopes. The writer can find themselves in any part of the sky if that's where they're looking, but for anyone else, it's babble.

 

[VerbalAbuse]

The AI "analyses" presented in those threads are like horoscopes. The writer can find themselves in any part of the sky if that's where they're looking, but for anyone else, it's babble.

Maybe not. One of the things I don't like on lit is the lack of reviews. I've no idea what the story is about. One may try to ask the AI to do a review -- summary, themes, etc. Then you read the story (if you think it may interest you) and decide for yourself if there was any value in the AI review,

 

[VerbalAbuse]

I've asked chatgpt for a critical analysis of The Fellowship of the Ring.

It started with:

the novel’s structure is episodic, with a series of encounters (Tom Bombadil, the Barrow-wights, the Council of Elrond) that feel disconnected at times. While these segments enrich the lore, they do not always serve the immediate progression of the plot. The Council of Elrond, in particular, is an extended exposition dump that, while significant, disrupts narrative momentum.

I did not like that, and challenged it to do better. Next was:

2. Structural Weakness: Slow Pacing and Episodic Progression​

A frequent critique of The Fellowship of the Ring is its slow start. [...] While this structure reinforces the impression of a vast and varied world, it occasionally undermines narrative cohesion, making the journey feel more like a sequence of adventures than a steadily escalating crisis.

Still not happy with the analysis. Step 3:

The Fellowship of the Ring does read like a series of loosely connected tales rather than a tightly plotted novel. Each chapter introduces a new setting, a new conflict, or a new character, but the connections between them feel weak. It’s more like a chain of bedtime stories than a single cohesive narrative.

Step 4:

from a structural standpoint, The Fellowship of the Ring is a mess. It starts as one kind of book and awkwardly transforms into another. The pacing is inconsistent, the early sections meander without much connection to the main plot, and there’s no real sense of escalation until much later. If Tolkien had gone into it with a clearer vision from the start—rather than being pushed into making a Hobbit sequel—it probably would have been a much tighter, more coherent book.

 

[AwkwardlySet]

Maybe not. One of the things I don't like on lit is the lack of reviews. I've no idea what the story is about. One may try to ask the AI to do a review -- summary, themes, etc. Then you read the story (if you think it may interest you) and decide for yourself if there was any value in the AI review,

There aren't many reviews, I agree. But take a look at this ChatGPT review that @SyleusSnow shared with us. It's a short story so you can easily read it and then compare your thoughts with the review.

Do you agree with the review? I don't, in many ways, although AI did pick up on some things. There is some insight in the review. I could delve deeper but even though Syleus shared the review, he didn't really volunteer the story for our bisecting , so I'll try to keep my observations vague.
I liked the story, yet the AI's praise of it is clearly overdone in my opinion. But I'd rather point out the fallacies of the tiny criticism the AI dared to make.

AI criticized the pacing. I didn't find that the pacing of the story was off. It's a 750-word story so there is not much time for anything but what the author did - the setting of the scene, some tension and background, and then the romp that ensued. The pacing wasn't off to me.

AI criticized the lack of character development. I don't think that the author could have done much more in a 750-word story. We did pick up on the relationship between Amber and the MC and learned enough about their history for a story of this size. AI also criticized the lack of development of the bride-to-be, but she is not present in the scene until the last couple of sentences. I don't think that her character could have been developed in any relevant way nor that the development would add to this story meaningfully. So both things that AI dares to criticize were done well in my opinion.

I mean, maybe I suck at this, I don't know. Make your own judgments.

 

[VerbalAbuse]

And of course, the obligatory anti-patriarchy rant :

absence of female agency limits the novel’s thematic scope, making Middle-earth feel somewhat exclusionary compared to more contemporary fantasy narratives that embrace diverse perspectives.
Tolkien’s work reflects the traditional, patriarchal structures of the myths and legends that inspired him, but from a modern standpoint, this lack of representation is a significant flaw.

 

[AwkwardlySet]

I've asked chatgpt for a critical analysis of The Fellowship of the Ring.

It started with:


I did not like that, and challenged it to do better. Next was:


Still not happy with the analysis. Step 3:


Step 4:

Is that a good idea though, taking the AI's analysis of a famous work that has been around for decades? ChatGPT has likely even been trained on many literary critiques of The Lord of the Rings.

 

[VerbalAbuse]

Is that a good idea though, taking the AI's analysis of a famous work that has been around for decades? ChatGPT has likely even been trained on many literary critiques of The Lord of the Rings.

and it referenced those critiques. It managed somewhat agreeably, but only after some prodding.

 

[VerbalAbuse]

Is that a good idea though, taking the AI's analysis of a famous work that has been around for decades? ChatGPT has likely even been trained on many literary critiques of The Lord of the Rings.

but sure, I get what you're saying.

 

[XerXesXu]

DeepSeek is a Chain of Thought model.

Enter this, or similar prompt, to find out what it'll do, and refine your prompt in relation to the passage you want analysed.

'If I asked you to do a literary analysis of a story I've written, what elements of story would you look at and how would you asses them?'

Don't forget to specify who you are (not sensitive, receptive to harsh criticism or the contrary), your objectives in writing a passage, and your intended audience in your final prompt.

DeepSeek will memorise a conversation, so it will answer your follow up questions, or protests, in the context of your previous questions.

PS: Remember that DeepSeek isn't a native speaker of any language. Do not be subtle, speak as if you were speaking to a foreigner. If necessary, explain any sense in which you use a particular word or expression, as you would to a foreigner.

 

[VerbalAbuse]

DeepSeek is a Chain of Thought model.

Enter this, or similar prompt, to find out what it'll do, and refine your prompt in relation to the passage you want analysed.

'If I asked you to do a literary analysis of a story I've written, what elements of story would you look at and how would you asses them?'

Don't forget to specify who you are (not sensitive, receptive to harsh criticism or the contrary), your objectives in writing a passage, and your intended audience in your final prompt.

DeepSeek will memorise a conversation, so it will answer your follow up questions, or protests, in the context of your previous questions.

PS: Remember that DeepSeek isn't a native speaker of any language. Do not be subtle, speak as if you were speaking to a foreigner. If necessary, explain any sense in which you use a particular word or expression, as you would to a foreigner.

Thanks for the hint! I will do that.

Like others have said before me, the AI is a tool. One can't complain the hammer is hard, or the needle sharp -- they're meant to be that way. How one wields them is a matter of skill -- and skills are acquired.

 

[sexy_louise]

I will say I haven't read your work at all, so I'm not gonna talk about whether it's really as your readers say it is. But GPT also told me 13 is a single-digit number... or that there are 2 R's in strawberry... or that 4+5+7 is 17. The point is, it's full of errors. Are you positive you can trust it as a critic? Are you willing to disregard so many human opinions for the sake of a chatbot? And in order to make yourself feel better at that... I smell confirmation bias.

What's interesting to me is how you think your readers' view of your characters and stories and your intention for them are somehow mutually exclusive. People can be unintentionally abusive. We're all learning to navigate our relationships and we make bad choices sometimes. That doesn't mean we can never be called abusive. Side note: actual, intentional abusers will frequently appeal to emotion and say they've been hurt or that they're trying to be better. Well, that may be true (as the cycle of abuse goes) but they're still abusive, aren't they?

I'm rather easily triggered, I don't read stories with so much as abusive/unhealthy undertones and I understand I'm not your target audience. But since so many of your readers brought up these issues, I'm inclined to think they're right, or at the very least, that I would react the same way. And quite frankly, your attitude to the criticism you received tells me all I need to know about you as a writer, perhaps as a person as well. Calling people hysterical is neither sensitive nor a good look.

 

[XerXesXu]

I will say I haven't read your work at all, so I'm not gonna talk about whether it's really as your readers say it is. But GPT also told me 13 is a single-digit number... or that there are 2 R's in strawberry... or that 4+5+7 is 17. The point is, it's full of errors. Are you positive you can trust it as a critic? Are you willing to disregard so many human opinions for the sake of a chatbot? And in order to make yourself feel better at that... I smell confirmation bias.

Which version of CHATGpt gave you those results? The learning curve for verbal abilities has been concave, but for mathematical abilities, convex. Now they're converging on a straight-line learning curve. If you try current LLM models you'll find they can give accurate answers to the question you asked. The problem of hallucinations is also decreasing rapidly, some developers say that reports of hallucinations now approach zero. There may be some problems remaining, but that's the direction of travel and it's progressing very rapidly. Hence my post about some people still living in 2023.

 

[sexy_louise]

Which version of CHATGpt gave you those results? The learning curve for verbal abilities has been concave, but for mathematical abilities, convex. Now they're converging on a straight-line learning curve. If you try current LLM models you'll find they can give accurate answers to the question you asked. The problem of hallucinations is also decreasing rapidly, some developers say that reports of hallucinations now approach zero. There may be some problems remaining, but that's the direction of travel and it's progressing very rapidly. Hence my post about some people still living in 2023.

Context: I didn't read any posts besides the author's, so I missed yours too.
I don't know which version it was, but it was the free one a year ago. So I think it was GPT3, unless 4 became available in the first half of 2024.

 

[madelinemasoch]

I'm rather easily triggered, I don't read stories with so much as abusive/unhealthy undertones and I understand I'm not your target audience. But since so many of your readers brought up these issues, I'm inclined to think they're right, or at the very least, that I would react the same way. And quite frankly, your attitude to the criticism you received tells me all I need to know about you as a writer, perhaps as a person as well. Calling people hysterical is neither sensitive nor a good look.

You lack context and you've already said so. I'm not sure how my attitude wouldn't give you more respect for me as a writer as I'm unwilling to take "criticism" that isn't constructive nor warranted. I'm referring to many remarks including hate comments that were nothing more than death threats.

The point is I didn't tell the bot what the undertones were–I just asked it to analyze the scenes, or to analyze the scenes and tell me about the themes, or to interpret a specific line, not to interpret or analyze along any specific lines of thinking or through any particular lens–and it understood what those undertones were, and none of them were about this mythical "abuse" notion other people stamped onto my content. It was more about control or power, which is a theme that is actually in the stories. That proves that those people misinterpreted my work.

 

[ToPleaseHim]

You lack context and you've already said so. I'm not sure how my attitude wouldn't give you more respect for me as a writer as I'm unwilling to take "criticism" that isn't constructive nor warranted. I'm referring to many remarks including hate comments that were nothing more than death threats.

The point is I didn't tell the bot what the undertones were–I just asked it to analyze the scenes, or to analyze the scenes and tell me about the themes, or to interpret a specific line, not to interpret or analyze along any specific lines of thinking or through any particular lens–and it understood what those undertones were, and none of them were about this mythical "abuse" notion other people stamped onto my content. It was more about control or power, which is a theme that is actually in the stories. That proves that those people misinterpreted my work.

Readers bring their own schema along for the read which leads them to interpret text a certain way. If you’re receiving lots of reviews or comments that are similar, they aren’t misinterpreting your work despite your original intentions. It would likely benefit you to examine why your work is being interpreted in ways other than you intended. That seems infinitely more beneficial than chatting with AI about it.

 

[madelinemasoch]

Let me just reveal to you what actually happened, to put this issue to rest. I'm not gonna share the parts of the prompts that actually reveal the scenes because that's against the rules on the forums and by "put this issue to rest" I don't mean getting the thread locked.

I did another excerpt before this that I don't even want to reveal the analysis of here, but the only thing I asked about it was "Can you analyze these three scenes in detail?"

Prompt: With that relationship as context, analyze this scene involving the same character. Why does Eleanor have this perspective? What does the line "Freedom is an illusion" mean and imply in this case?

Response: "This scene provides a deeper look into Eleanor's perspective and the dynamics within the group, particularly focusing on themes of power, control, and the concept of freedom. Here's an analysis of the scene and the implications of the line "Freedom is an illusion":

Eleanor's Perspective:​

1. Control and Empowerment: Eleanor's perspective is heavily centered on control and empowerment within the context of their relationships. She believes in the structure and roles that have been established, viewing them as essential for maintaining the dynamics that bring fulfillment and identity to the group.
2. Cuckoldry as Structure: Eleanor advocates for the arrangement of cuckoldry, seeing it as a necessary component of their relationships. She argues that the form and structure of these dynamics are what truly matter, rather than individual feelings or desires. This perspective highlights her belief in the importance of maintaining the established order.
3. Pragmatic Approach: Eleanor's approach is pragmatic and focused on the long-term functioning of their relationships. She emphasizes the importance of each role and how they contribute to the whole, suggesting that deviating from this structure would lead to dysfunction and a return to less fulfilling dynamics.

"Freedom is an Illusion":​

1. Illusion of Autonomy: The line "Freedom is an illusion" implies that true autonomy and freedom are not achievable within the context of their relationships. Instead, Eleanor suggests that empowerment comes from embracing the roles and structures that have been established, which provide a sense of purpose and identity.
2. Critique of Conventional Freedom: Eleanor's statement challenges conventional notions of freedom, suggesting that what is often perceived as freedom is actually a lack of structure and purpose. She implies that true empowerment comes from understanding and accepting one's place within a defined system.
3. Emphasis on Empowerment: By stating that freedom is an illusion, Eleanor shifts the focus to empowerment through structure and roles. She believes that the arrangement they have, including cuckoldry, offers a form of empowerment that is more meaningful than the pursuit of individual freedom.

Scene Dynamics:​

1. Tension and Conflict: The scene is marked by tension and conflict, particularly between Eleanor and Leya. Eleanor's insistence on maintaining the structure of their relationships creates friction, as Leya questions the necessity of these roles.
2. Influence and Persuasion: Eleanor uses her influence to persuade Leya to adhere to the established dynamics, emphasizing the importance of the roles they play. Her ability to sway others highlights her position of authority within the group.
3. Role of Submissives: The presence of submissive characters like Robin underscores the power dynamics at play. Their interactions with the dominant figures illustrate the hierarchy and the acceptance of roles within the group.

Overall, the scene explores the complexities of power, control, and identity within the group's relationships. Eleanor's perspective on freedom and empowerment challenges traditional views, emphasizing the importance of structure and roles in achieving fulfillment and purpose."


No one who picked up on these themes in the story in question actually said so to me in any of the discussions about it back then. I repeat, no one picked up on it and actually said so. But it would be so easy to do that even a non-human robot can do it. I don't know how to make the point any more obvious that the commenters and a lot of the people who read it (granted, maybe a lot of them didn't even give this chapter a read, as this was an analysis of Chapter 7 I think of the 16 that I had written) from the AH just did not give the story a very deep read or a fair shake on its merits, or maybe they just went lizard brain mode the second they saw my content choices. In good faith I actually assume that most of the AH people from back then didn't actually read this part. That's kind of besides the point, though.

Granted, I was kind of asking the bot a leading question and trying to get it to guess something about the character's backstory, and it didn't. But I think that just proves my point more that I wasn't directing the bot to say any of this. It didn't get any of the specificities of its analysis from me. It didn't just collate words that were already there before. It gave me an accurate reading of a scene in the form of analysis which was actually pretty light. The feedback I received from human beings did not. That means there's something wrong with the way they were reading it, that they weren't actually authentically engaging with the text. If a robot can do that, then so can you. That's all I was ever trying to say here.

Readers bring their own schema along for the read which leads them to interpret text a certain way. If you’re receiving lots of reviews or comments that are similar, they aren’t misinterpreting your work despite your original intentions. It would likely benefit you to examine why your work is being interpreted in ways other than you intended. That seems infinitely more beneficial than chatting with AI about it.

They were, quite obviously after all that I just explained here, misinterpreting my work. The reason why is obvious: they don't read correctly. They don't have an authentic engagement with the text in this case or maybe in most cases. However, I don't really think it's that difficult to read a text like the one in question in a way in which you can pick up upon the themes in it, even if you don't agree with the themes in it, without mischaracterizing the work. In their comments, they were referring to their fabricated idea of what I wrote and not the text I actually constructed. There is unfortunately in this case a gulf of difference between the two, to the point where it was staggering that people would be so blind to what I was saying, as if I'm living in the Twilight Zone, or something.

The above reading of that chapter is not hard to come up with. I just wish I lived in a world where I could share my work in a place like this (I stopped publishing here a long time ago) and receive feedback that was meaningful and resonated with me and showed that kind of appreciation for what I have to offer in the form of artistic/literary contribution, not just erotic contribution, but as a piece of art. It would be a step in the right direction to live in a world, even, where I can share my work and have people respond to it in a way that shows me they understand it on some level, but that isn't what happened. The positive commentary about it was mostly just that it turned people on. I have different motivations for writing than turning people on. That's besides the point. I just want to focus on the actual work right now, because focusing on the work rather than what other people think of it makes that work better for the audience at the end of the day, so I would rather just put this all to bed and get some rest myself and be able to think about my new ideas and characters and stories tomorrow instead.