Re:Don't Fall for AI: Reasons for Writers to Reject Slop
by dsign on Thursday July 17, 2025 @07:53PM
(#44599657)
The article’s reasons are valid for the most part, specially everything they said about indie writers, which you should be reading more of, but there is one point I feel is missing: AI’s are bad for writing because they don’t experience being human, but the consequences of it are very subtle.
For example, if I throw a hard science-fiction short story to ChatGPT and ask something specific about it, chances are ChatGPT will produce a broadly correct answer way more often than an average human reader (and oh boy, that’s a boomer!), simply because it has more raw knowledge and more sensible “attention”. So, from a certain, limited definition, LLMs understand some stories better than average readers[^1]. And I suspect that LLMs’ parrotting powers could be sufficient, if not now, at some point in the future, to create “derivative entertainment”.
However, LLMs can’t create stories where the author, a human, is trying to connect in some specific way with the reader, expected to be another human. One way of putting is by saying that LLMs don’t have human agendas. A person may decide to write as an act of self-reflection, as a thought experiment expressed in story, or as a way to share a personal experience or world outlook in a way other people can/want-to digest, or even as a way to share a personal fantasy. By the way they work, AI-as-LLMs won’t do that[^2].
[^1]: This is less true for long stories, where LLM skills seem to get “diluted”, which I attribute to their “one-shot” nature.
[^2]: But woe what an advanced AGI with an agenda may do, even a long-gone AI, even an AI whose agenda was “to preserve the human species as the top indisputable species on the planet, and allow them to survive climate change and other pressing problems.” Go and check my wrote-during-commute novella for a more fun and sexy take on the paperclip maximizer problem: https://www.amazon.com/When-Rows-through-Gates-Duat-ebook/dp...
For example, if I throw a hard science-fiction short story to ChatGPT and ask something specific about it, chances are ChatGPT will produce a broadly correct answer way more often than an average human reader (and oh boy, that’s a boomer!), simply because it has more raw knowledge and more sensible “attention”. So, from a certain, limited definition, LLMs understand some stories better than average readers[^1]. And I suspect that LLMs’ parrotting powers could be sufficient, if not now, at some point in the future, to create “derivative entertainment”.
However, LLMs can’t create stories where the author, a human, is trying to connect in some specific way with the reader, expected to be another human. One way of putting is by saying that LLMs don’t have human agendas. A person may decide to write as an act of self-reflection, as a thought experiment expressed in story, or as a way to share a personal experience or world outlook in a way other people can/want-to digest, or even as a way to share a personal fantasy. By the way they work, AI-as-LLMs won’t do that[^2].
[^1]: This is less true for long stories, where LLM skills seem to get “diluted”, which I attribute to their “one-shot” nature.
[^2]: But woe what an advanced AGI with an agenda may do, even a long-gone AI, even an AI whose agenda was “to preserve the human species as the top indisputable species on the planet, and allow them to survive climate change and other pressing problems.” Go and check my wrote-during-commute novella for a more fun and sexy take on the paperclip maximizer problem: https://www.amazon.com/When-Rows-through-Gates-Duat-ebook/dp...