Okay, judging from the lack of upvotes, this was pretty poorly received. I think I have plausible hypotheses of why, but would like to check.
I’m guessing it’s because I was treating “LLMs can actually write fiction pretty well” as a given and didn’t invest any effort in trying to prove it. Also most of my samples of LLM writing weren’t particularly interesting by themselves, suffering from a combination of 1) being generated for demonstration purposes rather than for a story I was personally excited about and 2) existing primarily for establishing character traits rather than being gripping prose. (I personally find that a slice-of-life-ish depiction of characters doing ordinary things can already get me to care about them, but not everyone is like this.)
So if you’re not someone who’s already bought into the idea of LLMs being decent writers as long as you prompt them right and is just looking for ideas on how to prompt them better, probably my post wasn’t particularly useful. Especially since the lack of examples of genuinely good writing would have made it easy to pattern-match me to the kinds of people who get enthused about LLM writing due to sycophancy and then fail to notice how bad it is.
I think the main “hook” of your post was this: “There is a preconception that LLM-written prose is bad… But if you do X, Y and Z, it can become much better”. When people read this kind of hook and then don’t see any convincing examples, they will be disappointed, even you weren’t about to go in that direction and really wanted to show something else.
Thanks, that’s a good point. I’d intended that line more as a disclaimer that “yeah I know that LLM writing tends bad by default, I’m not crazy enough to claim otherwise” rather than as a hook, but re-reading, it sure comes off as a hook.
I guess upon reflection I was a bit muddled on what exactly my intent with this post was in the first place, it’s somewhat a personal sharing of what’s been going on with me and somewhat a sharing of tricks but I didn’t have a very clear target audience in mind.
Especially since the lack of examples of genuinely good writing would have made it easy to pattern-match me to the kinds of people who get enthused about LLM writing due to sycophancy and then fail to notice how bad it is.
I mean, why wouldn’t people think that? You spend a while describing your setup and work as purely a hobby to cope with depression and about pleasure. Then to describe the results of this intensive many-months-long work with extreme prompt engineering, you describe most of it as being unpublishable trash—not fit for even a quick throwaway blog post:
I have mostly been writing for my own pleasure. The things I’ve been writing have been quite idiosyncratic to my interests and preferences, and I don’t expect many people to necessarily appreciate them. That said, I do have at least one story that I’ve been starting to feel has enough literary merit that I might want to share it more widely someday. I’ve been editing it as I write, but for now, that’s out of a pure love of the craft—I just enjoy working on it for its own sake, regardless of whether I’ll ever publish it.
Nor do any of your examples particularly change the reader’s mind.
You also don’t establish any particular bona fides for you having any literary talent or taste, and you spend time describing how you are a bad writer:
It used to be that I really liked writing fiction, but couldn’t write longer stories myself. I’d manage a couple of scenes and then just run out of ideas, or feel like my characters didn’t come properly alive. My style of writing tends to involve immersing myself in the head of one character and writing from their perspective. This has the problem that it makes it hard to jump into the heads of the other characters at the same time, so they end up flat and lifeless. This problem can be avoided if I have a co-writer. They write one character, I write another, and then we have those characters interact. I also love the interplay of ideas when writing with somebody—I introduce an element into the story and see how the other person builds on it, and then they introduce an element of their own and I build on it. It’s amazingly fun to come up with something and see what someone else does with it...For me, co-writing works off momentum. It takes some investment to drop into that character headspace, and once I’m there, I don’t want to leave. I’ve had several brief attempts at co-writing together with someone that ended because one of us couldn’t commit enough time and energy to make it work. Or worse, it did work and I got obsessed with it, but then my partner couldn’t invest a corresponding amount of time into it and the collaboration sputtered away just as I’d gotten excited about it.
If people write it off as a lengthy description of one person’s masturbatory LLM use generating midbrow (at best) fiction of no value to anyone else, they are only taking away what you seemed to want them to take away. Why should anyone read the post in full?
And why argue with you about it? Certainly I do not enjoy the many comments I have been leaving all other the Internet (not just LW) pointing out LLM confabulations, or blatant tells of ChatGPT use, only to be greeted by downvotes, mockery, lies, denial, or long-drawn-out paltering before finally acknowledging that yeah I was right all along, and that’s the easiest case for arguing with LLM-corrupted users (a mere factual matter of ‘did ChatGPT write this?’).
I missed this post when it was first published, but didn’t upvote because I went in hoping for very accessible ways to use LLMs for my own writing (which is fairly similar to yours- it’s a hobby I enjoy for its own sake rather than output). But it seemed more about the journey than actionable tips (I’m hopeful the new post might do that).
But it seemed more about the journey than actionable tips
Actually, I’m curious about this bit since I thought that this one did have actionable tips. For instance, the thing about “if you want the AI to write a realistic depiction of someone who knows X rather than just writing a popular stereotype of someone who knows X, reference specific things that such a person would know in the prompt rather than just asking for a character who knows X”. To me, that has felt like an important insight that wasn’t initially obvious when I started using LLMs, that I assumed would also help transform other people’s AI-generated content away from “oh this is just dumb stereotypes and bad writing”.
Was that too obvious to count as an actionable tip for you? Or just not relevant for the way you use LLMs for writing?
My own take on this article was that the beginning was about my journey, but then everything from the “basics of getting good writing” heading on was meant to be actionable tips:
“Opus is best model for writing fiction in my experience”
“if a human would struggle to produce a good story from your prompt, probably so would an LLM, so try to think of prompts that would inspire a human”
“you can immediately get more complicated psychology if you just tell the LLM you want it”
“it’s useful to start brainstorming the story beforehand, both because it gives you more ideas and it guides the LLM in what to write afterward”
“it’s worth explicitly trying out different frames in the initial brainstorming, such as a narrative lens and a psychological lens”
“what you get out reflects what you put in”
the bit about knowing X
“if you still get stereotyped characters even after doing the above, bring in some other sides of the character so the LLM knows that you don’t want a one-dimensional character”
“if the LLM gives you details that don’t make sense, see if you can spin them into inspiration”
Thanks for the feedback! I’d say the second post is more about the journey too, it has some tips but if you’ve already done something similar then you might very well already have figured them out on your own.
Okay, judging from the lack of upvotes, this was pretty poorly received. I think I have plausible hypotheses of why, but would like to check.
I’m guessing it’s because I was treating “LLMs can actually write fiction pretty well” as a given and didn’t invest any effort in trying to prove it. Also most of my samples of LLM writing weren’t particularly interesting by themselves, suffering from a combination of 1) being generated for demonstration purposes rather than for a story I was personally excited about and 2) existing primarily for establishing character traits rather than being gripping prose. (I personally find that a slice-of-life-ish depiction of characters doing ordinary things can already get me to care about them, but not everyone is like this.)
So if you’re not someone who’s already bought into the idea of LLMs being decent writers as long as you prompt them right and is just looking for ideas on how to prompt them better, probably my post wasn’t particularly useful. Especially since the lack of examples of genuinely good writing would have made it easy to pattern-match me to the kinds of people who get enthused about LLM writing due to sycophancy and then fail to notice how bad it is.
I think the main “hook” of your post was this: “There is a preconception that LLM-written prose is bad… But if you do X, Y and Z, it can become much better”. When people read this kind of hook and then don’t see any convincing examples, they will be disappointed, even you weren’t about to go in that direction and really wanted to show something else.
Thanks, that’s a good point. I’d intended that line more as a disclaimer that “yeah I know that LLM writing tends bad by default, I’m not crazy enough to claim otherwise” rather than as a hook, but re-reading, it sure comes off as a hook.
I guess upon reflection I was a bit muddled on what exactly my intent with this post was in the first place, it’s somewhat a personal sharing of what’s been going on with me and somewhat a sharing of tricks but I didn’t have a very clear target audience in mind.
I mean, why wouldn’t people think that? You spend a while describing your setup and work as purely a hobby to cope with depression and about pleasure. Then to describe the results of this intensive many-months-long work with extreme prompt engineering, you describe most of it as being unpublishable trash—not fit for even a quick throwaway blog post:
Nor do any of your examples particularly change the reader’s mind.
You also don’t establish any particular bona fides for you having any literary talent or taste, and you spend time describing how you are a bad writer:
If people write it off as a lengthy description of one person’s masturbatory LLM use generating midbrow (at best) fiction of no value to anyone else, they are only taking away what you seemed to want them to take away. Why should anyone read the post in full?
And why argue with you about it? Certainly I do not enjoy the many comments I have been leaving all other the Internet (not just LW) pointing out LLM confabulations, or blatant tells of ChatGPT use, only to be greeted by downvotes, mockery, lies, denial, or long-drawn-out paltering before finally acknowledging that yeah I was right all along, and that’s the easiest case for arguing with LLM-corrupted users (a mere factual matter of ‘did ChatGPT write this?’).
That’s fair and very useful, thanks!
I missed this post when it was first published, but didn’t upvote because I went in hoping for very accessible ways to use LLMs for my own writing (which is fairly similar to yours- it’s a hobby I enjoy for its own sake rather than output). But it seemed more about the journey than actionable tips (I’m hopeful the new post might do that).
Actually, I’m curious about this bit since I thought that this one did have actionable tips. For instance, the thing about “if you want the AI to write a realistic depiction of someone who knows X rather than just writing a popular stereotype of someone who knows X, reference specific things that such a person would know in the prompt rather than just asking for a character who knows X”. To me, that has felt like an important insight that wasn’t initially obvious when I started using LLMs, that I assumed would also help transform other people’s AI-generated content away from “oh this is just dumb stereotypes and bad writing”.
Was that too obvious to count as an actionable tip for you? Or just not relevant for the way you use LLMs for writing?
My own take on this article was that the beginning was about my journey, but then everything from the “basics of getting good writing” heading on was meant to be actionable tips:
“Opus is best model for writing fiction in my experience”
“if a human would struggle to produce a good story from your prompt, probably so would an LLM, so try to think of prompts that would inspire a human”
“you can immediately get more complicated psychology if you just tell the LLM you want it”
“it’s useful to start brainstorming the story beforehand, both because it gives you more ideas and it guides the LLM in what to write afterward”
“it’s worth explicitly trying out different frames in the initial brainstorming, such as a narrative lens and a psychological lens”
“what you get out reflects what you put in”
the bit about knowing X
“if you still get stereotyped characters even after doing the above, bring in some other sides of the character so the LLM knows that you don’t want a one-dimensional character”
“if the LLM gives you details that don’t make sense, see if you can spin them into inspiration”
Thanks for the feedback! I’d say the second post is more about the journey too, it has some tips but if you’ve already done something similar then you might very well already have figured them out on your own.