i've experimented with llm's for fiction writing at length, doing things like feeding various models a story i've written up to a certain point & then an outline and seeing what comes out, and:
– it's always awful cliché soup. just the worst reddit-grade 'and that man's name? albert einstein' dreg
– it cannot keep the details straight (not surprising since it doesn't have an actual internal model the story's world)
so the end result is nothing but stupid smarmy 'plot twists' that make no sense
@esvrld LLMs are trained to be the most common denominator of redditor so it makes sense. My pal was complaining about a predictable failure of an "ai support bot" and it made sense when I framed it as "imagine you put every bullshit reddit post into your keyboard prediction thing and got it to try to solve a task."
