1073
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 18 Oct 2026
1073 points (99.1% liked)
Facepalm
3712 readers
3 users here now
founded 3 years ago
MODERATORS
Honestly AI writing is so fucking bad you can read a couple paragraphs and know its AI. Just makes going to a book store and opening the book up for a quick look before buying it worth it again.
technically, there are ways to make AI writing good and go beyond blunt imitation. However, to do so requires a very peculiar framework with rigid ruleset, specific sources and style guide and the progress will be extrememly incremental and you'll do much of the heavy lifting regardless - so it kinda defeats the idea of being fast and cheap and it won't handle much writing in a regular sense.
To expand on this because it's what I've been dicking around with for the past year or so:
the slop machine sucks at a lot of things. Particularly compartmentalizing. Character a knows something character b doesn't, but ai output has character b talking about it. Only way to prevent this is to ensure the ai doesn't know the secret information.
I could keep going. Will if anyone is interested.
Sounds like a lot of work when you can just, ya know, write the book yourself.
oh no, you got me with the sheer scale of your intelligence, and your ability to ascertain everything about me without any context whatsoever. What can I do to rebuke you when short lemmy comments don't leave room for nuance? I'm done for and withering
and I can't keep a straight face.
One of my ongoing projects is a detailed report on my opinions and experiences with ai, 100% human written. Bits of it are in my comment history if you care to look (cw: there's also horny shit and political shit in my comment history). There's space for a nuanced discussion if you actually care instead of wanting a simple "gottem".
Man, you took that really personally, huh?
Oh wow, some of your comments are…pretty sad.
once again, making assumptions based on limited information. I could copy pasta my previous comment and it would apply just the same.
You told me to look at your comments for proof and then said they were “limited information”.
Yikes-eroni.
You've just proved your own inability to read. I said I've posted parts of my opinions on the subject, with a content warning that there's a lot that's not relevant to the conversation at hand (with a short explanation why).
So what in my comment history is "sad"? My support of trans people? That I horny post on main? What's yikes? Capitalism could work if we killed the billionaires and enacted regulations and welfare programs? That I have no patience for morons? I'm open to actually discussing the topic at hand, and you're leaving me guessing what the hell you're talking about. If I must guess, then I'll assume the worst because you've already done the same of me.
I'm presuming you could actually write an entire plot structure with individual agents writing the actions of individual characters with one narrator deciding what to include then consume this structure incrementally having it preprompt each scene with the plot structure for that scene and what happened.
man, let's be friends! i used to work at a startup that researched generative AI and my field was generative literature!
My goal is to have the ai help, not have it do for me. Slop is for personal entertainment and inspiration purposes only.
Not that I haven't tried. I am currently without a keyboard so my responses will be limited.
genlit is basically a completely different field regardless - at this point it is more about tools and frameworks not really about content.
i.e. it works as long as the AI has a human ghost writer.
it will be just writing the whole setting and its rules and everything inside to the point you can train a model to generate bits in controlled environment. so basically a fancy text adventure. technically, it can be interesting but that's not the stuff you can do with regular chatGPT or Claude.
it's really bad writing, it tends to repeat a lot of the same ideas in rephrased terms while also accidently deconstructing meaning in narrative.
Yes, but right now it's also as bad as it's ever going to be again.
While I don't buy into the hype of the techbros, I do think it's something we're stuck with from now on, and they'll get better and better at hiding it.
True. I'm torn on where to rank it though. It's roughly B tier in "shitty use of AI", but I cannot place it in good conscience above coloring books made with AI. I think the coloring books subcategory puts books in A to A+.
Thoughts?
What?