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 liked the playfulness of:

  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'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.

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 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.

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.