For the past 5-6 years I've been writing a book in my spare time. The outline of it is how reason emerges in past societies from the needs of social complexity, how it's lessons get converted into rules and rituals, which in turn remove any competitive advantage of aquiring reason, ending it to setup a new cycle. And in the meanwhile LLMs became the ultimate heuristic of humanity.
I've gotten it 60-70% ready, and I really don't know if it'll have an audience in a post-AI world. I never meant to strike big with it, but I'm now wondering if thousands of hours of research and writing can amount to more than a novelty gift I'd give to friends.
I really hope you don't give up. I've built and shipped systems with 80000 to 90,000 lines of code, only to see most of the companies that bought them go out of business. But there was still immense value in the act of delivering them. By putting our creations out into the world, we connect with each other.
The act of writing and building is, in itself, humanity's grand narrative for trying to understand the world. The journey itself is inherently valuable. Isn't the ability to organize our thoughts, pass them down to the next generation, and continue that narrative exactly what makes us uniquely human?
Even if only a few people around you end up reading it, those few could be deeply inspired to go on and build an even greater world. Please don't stop. I'm rooting for you.
Why wouldn’t people read it? The problem is not writing - the problem is finding an audience. Yes, AI means there are more poor works competing. But if it’s actually good writing, you will find an audience if you market the book. Pay an editor. Publish to kindle. Pay for marketing. Get people to sign up for an email list.
> Why wouldn’t people read it? The problem is not writing - the problem is finding an audience.
And that's precisely the issue here. For a while, the internet allowed you to find an audience, just like that. Start a blog / podcast / YT channel, keep going, get enough attention. You could then approach a traditional publisher and tell them "hey, I'm kind of a big deal", or you could self-publish and rely on the word-of-mouth from your followers.
Now, how would that work? If you have a blog, AI answers will summarize it without attribution and not send anyone your way. Even the "references" cited in AI answers often point to AI-slop blogs, not the original source. The articles we discuss on HN are often AI-written too. So yeah, it's about reaching the audience, but you're now competing with machines that produce an endless stream of human-like text, good enough for most consumers, practically for free.
Mixed metaphors which sound nice at first glance, but slip away from meaning like an echo chasing itself off a cliff.
Similes that catch in your mind like river trouts tangled in the roots of a redwood tree.
Also mentions some interesting AI tells, for AI generated stories.
It can be "playful" once or twice, especially if the text is playing with how nonsensical the metaphor is. By the twentieth time something is as tired as a willow tree on a Tuesday in May, as persnickety as an ant with a fever, or as rambunctious as a horny lobster, it's just nonsensical bad writing.
Metaphors are generally used to transfer the qualia of one experience into another. When the referent has no qualia, that is, you've never in your life experienced "river trouts tangled in the roots of a redwood tree", it's a failure of a metaphor. You can quibble about how special this or that metaphor is, which I've already given an example of with the "playing with how nonsensical the metaphor is", but when all the metaphors are broken that way, all the time, the writer is not "breaking the rules because they've transcended them" or anything like that, the writer is breaking them in the bad way that the rules were put there to stop and the writer should consider taking a Writing 101 course.
Though anyone taking a Writing 101 course should be aware that as near as I can tell, completing such a course is prima facie proof that they are overqualified for the vast majority of modern writing jobs.
"Likewise, he discovered that the contrast of two seeming opposites was a common feature in haiku. Ginsberg used this technique in his poetry, putting together two starkly dissimilar images: something weak with something strong, an artifact of high culture with an artifact of low culture, something holy with something unholy."
I think some of these broken metaphors could be turned into some sort of haiku-like poem, especially if we ignore the requirement to reference a season somehow, though it would still take some sort of additional work to add something to tie them together more thoroughly than the metaphor does, some third component a poet uses to glue the two bits together in some interesting way.
tree roots sheltering
river trout find safety but
growth is treacherous
Eh. I'm not a poet. And I still just chucked the "Redwood" part. But maybe you can see how I also added a bit of a concept in there to tie it together. But then, of course, it's no longer a metaphor, it's a poem. It's not referencing an experience we've all had and transferring that on to something else, I'm creating a new experience. Very different.
And as much as there are still people on HN who insist that AI text can't be detected algorithmically, it's worth noting that the original story is marked 100% AI by Pangram. So it's not just this person seeing things.
It's a bad short story. Whether it was generated at the hands of humans or a computer isn't really relevant. Speaking in riddles is cheap now, so authors better learn how to surprise the reader in simpler, more readable words. It seems to be something that LLMs are quite bad at.
Yeah, I found this extremely unpleasant to read. I wouldn't have pegged it as LLM-assisted, but I would have pegged the author as pretentious and bad at coming up with compelling analogies.
I recently built and delivered an AI-driven novel-writing program. The architecture involved chaining the Claude, GPT, and Gemini APIs together so they could cross-critique and iteratively revise the text, while systematically saving key plot points and lore chapter-by-chapter. (Serialized 'web novels' published on a per-chapter basis are a massive industry here in Korea). AI has already heavily infiltrated the fiction space in Korea, and it looks like the exact same trend is hitting the US.
Personally, I find it incredibly easy to spot AI-generated text in Korean, but catching it in English is much harder for me. That being said, they still have very distinct, overused patterns. You constantly see words like 'ultimately' or 'structural,' and they rely heavily on highly formulaic 'X is Y' sentence structures.
On the point of the impermeancy of digital media, I will say that actual traditional podcasts with distributed by rss can and are downloaded by design which means just like with books, a publisher can only destroy their own copies not listener copies. Also even if we pretend that YouTube podcasts can’t be deleted, most of the major ones also publish traditional podcast feeds too.
I wondering about other thing - I'm assuming that publishers are now using LLM's en-masse to "proofread", do initial evaluation, do editorial work etc.
So if an author abstained from using LLM in the writing process - isn't then new, original, not yet on the market book ending up in the LLM training data corpus even before it hits the market?
The worst thing about this is not really that somebody might have had AI help writing a story, but that an editor thought they could get any kind of useful information from asking an AI whether the story was written by AI. If there's any hope of editors staying ahead of this phenomenon, they will need to educate themselves a lot better about how it actually works.
For the past 5-6 years I've been writing a book in my spare time. The outline of it is how reason emerges in past societies from the needs of social complexity, how it's lessons get converted into rules and rituals, which in turn remove any competitive advantage of aquiring reason, ending it to setup a new cycle. And in the meanwhile LLMs became the ultimate heuristic of humanity.
I've gotten it 60-70% ready, and I really don't know if it'll have an audience in a post-AI world. I never meant to strike big with it, but I'm now wondering if thousands of hours of research and writing can amount to more than a novelty gift I'd give to friends.
I really hope you don't give up. I've built and shipped systems with 80000 to 90,000 lines of code, only to see most of the companies that bought them go out of business. But there was still immense value in the act of delivering them. By putting our creations out into the world, we connect with each other.
The act of writing and building is, in itself, humanity's grand narrative for trying to understand the world. The journey itself is inherently valuable. Isn't the ability to organize our thoughts, pass them down to the next generation, and continue that narrative exactly what makes us uniquely human?
Even if only a few people around you end up reading it, those few could be deeply inspired to go on and build an even greater world. Please don't stop. I'm rooting for you.
For what it’s worth, that sounds like a very interesting premise. N=1, but you have a potential reader here!
Why wouldn’t people read it? The problem is not writing - the problem is finding an audience. Yes, AI means there are more poor works competing. But if it’s actually good writing, you will find an audience if you market the book. Pay an editor. Publish to kindle. Pay for marketing. Get people to sign up for an email list.
> Why wouldn’t people read it? The problem is not writing - the problem is finding an audience.
And that's precisely the issue here. For a while, the internet allowed you to find an audience, just like that. Start a blog / podcast / YT channel, keep going, get enough attention. You could then approach a traditional publisher and tell them "hey, I'm kind of a big deal", or you could self-publish and rely on the word-of-mouth from your followers.
Now, how would that work? If you have a blog, AI answers will summarize it without attribution and not send anyone your way. Even the "references" cited in AI answers often point to AI-slop blogs, not the original source. The articles we discuss on HN are often AI-written too. So yeah, it's about reaching the audience, but you're now competing with machines that produce an endless stream of human-like text, good enough for most consumers, practically for free.
I liked the playfulness of:
Also mentions some interesting AI tells, for AI generated stories.It can be "playful" once or twice, especially if the text is playing with how nonsensical the metaphor is. By the twentieth time something is as tired as a willow tree on a Tuesday in May, as persnickety as an ant with a fever, or as rambunctious as a horny lobster, it's just nonsensical bad writing.
Metaphors are generally used to transfer the qualia of one experience into another. When the referent has no qualia, that is, you've never in your life experienced "river trouts tangled in the roots of a redwood tree", it's a failure of a metaphor. You can quibble about how special this or that metaphor is, which I've already given an example of with the "playing with how nonsensical the metaphor is", but when all the metaphors are broken that way, all the time, the writer is not "breaking the rules because they've transcended them" or anything like that, the writer is breaking them in the bad way that the rules were put there to stop and the writer should consider taking a Writing 101 course.
Though anyone taking a Writing 101 course should be aware that as near as I can tell, completing such a course is prima facie proof that they are overqualified for the vast majority of modern writing jobs.
https://nostalgebraist.tumblr.com/post/778041178124926976/hy...
The haiku observation is an interesting one:
"Likewise, he discovered that the contrast of two seeming opposites was a common feature in haiku. Ginsberg used this technique in his poetry, putting together two starkly dissimilar images: something weak with something strong, an artifact of high culture with an artifact of low culture, something holy with something unholy."
I think some of these broken metaphors could be turned into some sort of haiku-like poem, especially if we ignore the requirement to reference a season somehow, though it would still take some sort of additional work to add something to tie them together more thoroughly than the metaphor does, some third component a poet uses to glue the two bits together in some interesting way.
Eh. I'm not a poet. And I still just chucked the "Redwood" part. But maybe you can see how I also added a bit of a concept in there to tie it together. But then, of course, it's no longer a metaphor, it's a poem. It's not referencing an experience we've all had and transferring that on to something else, I'm creating a new experience. Very different.And as much as there are still people on HN who insist that AI text can't be detected algorithmically, it's worth noting that the original story is marked 100% AI by Pangram. So it's not just this person seeing things.
It's a bad short story. Whether it was generated at the hands of humans or a computer isn't really relevant. Speaking in riddles is cheap now, so authors better learn how to surprise the reader in simpler, more readable words. It seems to be something that LLMs are quite bad at.
Yeah, I found this extremely unpleasant to read. I wouldn't have pegged it as LLM-assisted, but I would have pegged the author as pretentious and bad at coming up with compelling analogies.
I recently built and delivered an AI-driven novel-writing program. The architecture involved chaining the Claude, GPT, and Gemini APIs together so they could cross-critique and iteratively revise the text, while systematically saving key plot points and lore chapter-by-chapter. (Serialized 'web novels' published on a per-chapter basis are a massive industry here in Korea). AI has already heavily infiltrated the fiction space in Korea, and it looks like the exact same trend is hitting the US.
Personally, I find it incredibly easy to spot AI-generated text in Korean, but catching it in English is much harder for me. That being said, they still have very distinct, overused patterns. You constantly see words like 'ultimately' or 'structural,' and they rely heavily on highly formulaic 'X is Y' sentence structures.
On the point of the impermeancy of digital media, I will say that actual traditional podcasts with distributed by rss can and are downloaded by design which means just like with books, a publisher can only destroy their own copies not listener copies. Also even if we pretend that YouTube podcasts can’t be deleted, most of the major ones also publish traditional podcast feeds too.
I wondering about other thing - I'm assuming that publishers are now using LLM's en-masse to "proofread", do initial evaluation, do editorial work etc.
So if an author abstained from using LLM in the writing process - isn't then new, original, not yet on the market book ending up in the LLM training data corpus even before it hits the market?
>I'm assuming that publishers are now using LLM's en-masse to "proofread", do initial evaluation, do editorial work etc.
I think there is going to be a large market in all domains for providers who can convince people that they don't use LLMs.
The worst thing about this is not really that somebody might have had AI help writing a story, but that an editor thought they could get any kind of useful information from asking an AI whether the story was written by AI. If there's any hope of editors staying ahead of this phenomenon, they will need to educate themselves a lot better about how it actually works.