I think people read novels because they want to connect with the author on some level. LLM generated text is actually more logically airtight than human writing. For non fiction, I think the information dense LLM style is often better. But human writing leaves space for the reader to breathe. There are gaps scattered throughout.
In that sense, I sometimes feel like logical flaws might actually be a signal of humanity
Afaik I have yet to encounter LLM fiction, but I have seen plenty of AI generated music and it truly disgusts me.
Each to their own but to me a large part of music was always the life experience of the person making it. I'm listening to their life, their love, their pain in music form.
Without an interesting human behind it, art loses all magic to me.
That's why it's beyond annoying that my own brother absolutely loves generated music. He just wants the general ambient groove of reggae or blues, he doesn't care one bit for the artists.
That is a perfectly valid opinion and POV to have. While I don't share it personally, I can totally understand why one has it. That does not change, that I personally actually do not care the slightest about the question of who or what wrote a novel.
But that is just me, having studied literatur and fallen into a school of literary studies that held Roland Barthes' claim of the death of the author [1] in high regard.
That might just be a German thing, but in school I hated the question of teachers asking: "What does the author wanted to tell us with this story." Because -> The author never wrote that down. And deriving their intentions from a fictional text felt just like making shit up. You either hit the interpretation of your teacher or received a bad grade.
So when entering university and learning, that to analyse a text, to understand how structures within a text enable us humans to interpret and generate meaning, regardless of what the author originally might have intended, was freedom to me.
To make a long story even longer:
As long as the story is good - I don't give a rat's ass if an LLM wrote this, a monkey on a typewriter or a human being (or a group of authors writing under a pseudonym).
English teachers around the world have loathed that "The curtains were blue" meme, and are absolutely right to do so, but what happens when you ask what an author meant by the blue curtains and the answer is a fairly definitive "it's just the probabilistic output of the model the author used, it doesn't mean anything"?
Sure some people are just piggies lining up at the trough to consume content like it's the custard machine from tellitubbies, but the greatest works will be human written, and I don't see that changing. What will change is the discovery mechanism for a book or any art is shot to bits right now, I could see curation of books and reviews becoming much more important.
I concur with the author. Personally I think I just won't read fiction produced after 2023. There's so much good literature already produced, why bother with the new stuff?
I feel very sorry to any budding new authors. Maybe we are entering a literary dark age?
I can't tell if I'm getting old, or if this is a reaction purely to people making blog-posts saying they don't like AI, but, I am filled with "who cares?".
Unambiguously the point of view is valid. I don't share it, but whatever.
But why are you writing this, and why is it getting upvoted? We're several years in. Is there new ground being covered here? Is there something special or unique to this take that we haven't read 1,000 times before? Are we doomed to these identical "I don't like it and I need to tell someone" posts forever? Yes, it's a valid view point, but so is not liking Meta, and yet we don't get a "I don't like Meta as a company!!" post upvoted every single day.
I would like to read "actual" LLM fiction in the same sense I would like to read fiction from a lion. I wonder if that kind of alien writing is even possible with this kind of technology. The RL process would have to be quite different at least and even then it is still using human content as the base. Anyone seen any good or even decent or OK fiction of this type?
One of my favorite childhoods stories was told from a perspective of an arctic fox. It wasn’t a cartoonish “Lion King” type of story but pretty serious and mature stuff.
worth noting you don't usually read _that_ many books in your lifetime. e.g. 20 a books a year for 40 years is just 800 books
side note: i've noticed people writing with no caps, unusual grammar, splling mistakes, etc. which does differentiate their output from typical llm output - although ofc llms can probably imitate that
I won't (if I know) either, but for me it's because a very big part of reading fiction or enjoying any art is that you have a relationship of a kind with its creator. When a passage surprises me or a song lifts me or a painting inspires me it's partly because I'm amazed that someone was able to do that, and that they were able to reach me like that.
Everyone borrows or steals from prior art but I'm still amazed by the people that build on what came before to create something new. I just don't have that experience if I know it's generated by a LLM. I appreciate that it will be harder to know the difference as time goes on, and I'll need to be careful not to cast a side-eye at a lot of legitimate stuff unfairly in the future because of that.
While I understand the hesitancy for LLM fiction, I do look forward to LLM powered games, *IF DONE RIGHT*. I imagine a game like Skyrim, where each character has handwritten backstory and dialogue, with an LLM extending that.
Having the city guard dynamically react to what you do, stay in character, and even point you to new POI from those conversations seems like a very cool new dimension in storytelling.
I took a creative writing class back in the day, and we talked about how fiction writers often have a massive "canon" backstory to many characters and worlds, and draw from that to create the story. That same backstory details can be use to flesh out a very immersive world. You still have guidance and handwritten essential lines, just use LLMs to enhance the world.
Now, will there be lots of projects that use LLMs to take shortcuts and be far from this idealized narrative, yes. But not all of them. Some will stand out for getting it right.
Definitely agree with LLM texts being boring and lacking personality. Which for technical text is fine, but devastating for fiction. I wonder if you could coax it into writing less average formulations with some clever prompting?
I don't oppose LLM fiction on principle, and I don't think McCormick does either, but it's just too damn boring to be worthwhile, and I think McCormick hits the nail on the head as to why. This seems like it'll be the last frontier for LLMs, good fiction requires a departure from LLM nature, and that'll inherently be difficult to solve.
I do wonder if you could do something like style transfer on texts? I'm not a very good author either, but occasionally I'll get an idea for a story and wish I had the skill to write it in a way that'd be worth reading, but my poor prose tends to get in the way. Would be cool if I could feed it my poorly written attempts and say basically "this story but in the style of [insert great author]" the way you can snap a photo and get it in the style of a Van Gogh painting.
If nothing else I think I'd enjoy a Tolkien sci fi story, or a Asimov high fantasy.
I have come across a lot of AI-generated youtube videos lately, most of them about trivialities and clearly aimed at a not very critical audience. A few of them also contained hallucinations.
I think a lot of people might be creating AI generated content, books, videos, etc as a kind of side hustle unfortunately.
Whenever I read something that's LLM generated without knowing it, and it's actually interesting, I have questions for the author. After they read my emails and reply back, I realize that they have no answers for me. It's all because both me and the author are reading the article/story/code for the first time.
This effect will disappear when, instead of having 4-6 big models that everyone uses, we'll have trillions of models with memory / continual learning, each independently developing its own voice / personality, drowning all human expression.
It's not impossible to skew an LLM into a specific style of writing, one to reflect the author's intentions.
I think that LLMs made everything easier, as in, lowered the barrier. With that, comes a natual inscentive to be lazy, and not implement your style of writing into it. This will mean the production of more "slop" articles but also a chance for great novels with the help of LLMs. As I say, both human and AI slop exists, as well as truly great things from both. Similarly to how before AI, we had "little to no code!" promises, "make your app with no experience", as well as the endless WordPress/"PHP scripts" promising "SEO optimization" and likes from everyone.
Except all these pre LLM methods were limited and could easily be ignored; now the machine can truly get "smarter" and output new content to the point where beginners can truly become something better (or output beginner quality work at mass) with little to no work, exactly as promised before.
What strikes me about (some, or most of, as I see nowadays) Anti AI is that you never really cared about the human before AI, it was all about quality. But only now do you decide to care about the human? If you truly did, won't you have cared about the human before?
I feel like LLMs could produce some pretty cool fiction - like cranking up the temperature setting like you can with many models to produce nonsensical garbage that isn't even real sentences during a planning stage to end up with some ideas that human writers wouldn't commonly come up with, and then letting a model that is less lobotomized do the actual writing.
I think many people forget that the vast majority of books aren't high art.
Your mind goes to Heinlein, Arthur C. Clarke, N.K. Jemsin, or Pratchett.
My mind goes to Ice Planet Barbarians[0] or Crimson Worlds[1] (both written by actual humans, but I think they are good representations of their genres)
There are so many books in the romantasy genre specifically that could be just mad libs templates for AI Generated books.
The people who read them don't care about the plot, they want specific kinds of tropes in the book. They look for enemies-to-lovers, 7 foot tall men, tiny women and other very focused traits (ages, races, LQBTQIA+) in the books. They are consumable fun, not something anyone expects to be stored in the Library of Congress for future generations because they were so earth-shatteringly insightful.
Yes, AI authored (or assisted) books won't bring us the next Culture books by Iain M. Banks, but the next generic romantasy or self-published military scifi books could very well be. And nobody will notice - or care. It's 200-300 pages of entertainment that hits just the specific itch you had.
Art requires intentionality, and this is most clear in long-form fiction. Without guiding intention, you can't have consistent characters, or satisfying plots, or meaningful themes. There has to be a mind guiding it, and LLMs don't have the intentionality or interiority to do so.
Theoretically, a sufficiently skilled and dedicated prompter could guide AI precisely enough to produce a good result, but
1. There are no examples of this.
2. Anyone with the focus & talent to do that would rather not.
As a result, AI produces--and is likely to continue to produce--extremely bad fiction (though bad in totally distinct ways from bad human writers).
Can LLMs distinguish between non-fiction and fiction, i.e., true and false
The literature on LLMs states that LLMs sometimes output falsehoods irrespective of whether the persons using them have prompted for non-fiction or fiction
The papers imply that it is up to the person using the LLM to distinguish what is true from what is false
26 comments
[ 2.6 ms ] story [ 20.5 ms ] threadIn that sense, I sometimes feel like logical flaws might actually be a signal of humanity
Each to their own but to me a large part of music was always the life experience of the person making it. I'm listening to their life, their love, their pain in music form.
Without an interesting human behind it, art loses all magic to me.
That's why it's beyond annoying that my own brother absolutely loves generated music. He just wants the general ambient groove of reggae or blues, he doesn't care one bit for the artists.
There are too many texts generated by LLM on Internet now.
The experience of exploring Internet is very bad now.
But that is just me, having studied literatur and fallen into a school of literary studies that held Roland Barthes' claim of the death of the author [1] in high regard.
That might just be a German thing, but in school I hated the question of teachers asking: "What does the author wanted to tell us with this story." Because -> The author never wrote that down. And deriving their intentions from a fictional text felt just like making shit up. You either hit the interpretation of your teacher or received a bad grade.
So when entering university and learning, that to analyse a text, to understand how structures within a text enable us humans to interpret and generate meaning, regardless of what the author originally might have intended, was freedom to me.
To make a long story even longer: As long as the story is good - I don't give a rat's ass if an LLM wrote this, a monkey on a typewriter or a human being (or a group of authors writing under a pseudonym).
[1]: https://en.wikipedia.org/wiki/The_Death_of_the_Author
Sure some people are just piggies lining up at the trough to consume content like it's the custard machine from tellitubbies, but the greatest works will be human written, and I don't see that changing. What will change is the discovery mechanism for a book or any art is shot to bits right now, I could see curation of books and reviews becoming much more important.
I feel very sorry to any budding new authors. Maybe we are entering a literary dark age?
Unambiguously the point of view is valid. I don't share it, but whatever.
But why are you writing this, and why is it getting upvoted? We're several years in. Is there new ground being covered here? Is there something special or unique to this take that we haven't read 1,000 times before? Are we doomed to these identical "I don't like it and I need to tell someone" posts forever? Yes, it's a valid view point, but so is not liking Meta, and yet we don't get a "I don't like Meta as a company!!" post upvoted every single day.
There will come a time, very soon, where you wont be able to tell the difference, and what then?
side note: i've noticed people writing with no caps, unusual grammar, splling mistakes, etc. which does differentiate their output from typical llm output - although ofc llms can probably imitate that
Everyone borrows or steals from prior art but I'm still amazed by the people that build on what came before to create something new. I just don't have that experience if I know it's generated by a LLM. I appreciate that it will be harder to know the difference as time goes on, and I'll need to be careful not to cast a side-eye at a lot of legitimate stuff unfairly in the future because of that.
Having the city guard dynamically react to what you do, stay in character, and even point you to new POI from those conversations seems like a very cool new dimension in storytelling.
I took a creative writing class back in the day, and we talked about how fiction writers often have a massive "canon" backstory to many characters and worlds, and draw from that to create the story. That same backstory details can be use to flesh out a very immersive world. You still have guidance and handwritten essential lines, just use LLMs to enhance the world.
Now, will there be lots of projects that use LLMs to take shortcuts and be far from this idealized narrative, yes. But not all of them. Some will stand out for getting it right.
I don't oppose LLM fiction on principle, and I don't think McCormick does either, but it's just too damn boring to be worthwhile, and I think McCormick hits the nail on the head as to why. This seems like it'll be the last frontier for LLMs, good fiction requires a departure from LLM nature, and that'll inherently be difficult to solve.
I do wonder if you could do something like style transfer on texts? I'm not a very good author either, but occasionally I'll get an idea for a story and wish I had the skill to write it in a way that'd be worth reading, but my poor prose tends to get in the way. Would be cool if I could feed it my poorly written attempts and say basically "this story but in the style of [insert great author]" the way you can snap a photo and get it in the style of a Van Gogh painting.
If nothing else I think I'd enjoy a Tolkien sci fi story, or a Asimov high fantasy.
I think a lot of people might be creating AI generated content, books, videos, etc as a kind of side hustle unfortunately.
Enjoy this "boring LLM" era while it lasts.
I think that LLMs made everything easier, as in, lowered the barrier. With that, comes a natual inscentive to be lazy, and not implement your style of writing into it. This will mean the production of more "slop" articles but also a chance for great novels with the help of LLMs. As I say, both human and AI slop exists, as well as truly great things from both. Similarly to how before AI, we had "little to no code!" promises, "make your app with no experience", as well as the endless WordPress/"PHP scripts" promising "SEO optimization" and likes from everyone.
Except all these pre LLM methods were limited and could easily be ignored; now the machine can truly get "smarter" and output new content to the point where beginners can truly become something better (or output beginner quality work at mass) with little to no work, exactly as promised before.
What strikes me about (some, or most of, as I see nowadays) Anti AI is that you never really cared about the human before AI, it was all about quality. But only now do you decide to care about the human? If you truly did, won't you have cared about the human before?
The problem is that regardless of all of that RLHF and positivity bias and slop has essentially ruined any and all LLM writing for the foreseeable future: https://blog.kronis.dev/blog/ai-slop-is-a-self-inflicted-tra...
In short, the writing could be good if it wasn't fucking bullshit.
Your mind goes to Heinlein, Arthur C. Clarke, N.K. Jemsin, or Pratchett.
My mind goes to Ice Planet Barbarians[0] or Crimson Worlds[1] (both written by actual humans, but I think they are good representations of their genres)
There are so many books in the romantasy genre specifically that could be just mad libs templates for AI Generated books.
The people who read them don't care about the plot, they want specific kinds of tropes in the book. They look for enemies-to-lovers, 7 foot tall men, tiny women and other very focused traits (ages, races, LQBTQIA+) in the books. They are consumable fun, not something anyone expects to be stored in the Library of Congress for future generations because they were so earth-shatteringly insightful.
Yes, AI authored (or assisted) books won't bring us the next Culture books by Iain M. Banks, but the next generic romantasy or self-published military scifi books could very well be. And nobody will notice - or care. It's 200-300 pages of entertainment that hits just the specific itch you had.
[0] https://www.goodreads.com/series/151379-ice-planet-barbarian...
[1] https://www.goodreads.com/series/97149-crimson-worlds
Theoretically, a sufficiently skilled and dedicated prompter could guide AI precisely enough to produce a good result, but
1. There are no examples of this.
2. Anyone with the focus & talent to do that would rather not.
As a result, AI produces--and is likely to continue to produce--extremely bad fiction (though bad in totally distinct ways from bad human writers).
Can LLMs distinguish between non-fiction and fiction, i.e., true and false
The literature on LLMs states that LLMs sometimes output falsehoods irrespective of whether the persons using them have prompted for non-fiction or fiction
The papers imply that it is up to the person using the LLM to distinguish what is true from what is false