"I was reading imaginary “analysis” of poems that did not exist. Hallucinated claims about non-existent poems." That's some Borges' nightmare sentence right there.
I wonder what you think about the entire publishing industry being opposed to LLMs as ghostwriters? I had an author openly (and dare I say proudly) inform me last week that her new manuscript was ghostwritten by AI and that the AI had captured her voice so well that she felt this book was better than her first book. Here's a conundrum. I published her first book, which she thinks is more poorly written than her second book, which I rejected on grounds that we will not publish ghostwritten-by-LLM books (or books with generative AI at all, though I know we and no publisher knows the extent to which we already are). I don't regret this decision because I fundamentally don't want to publish books written by machines on my press. But people have brought up interesting points to me: How is this different than collaborating with a human ghostwriter? How are genre writers churning out books with LLMs any different than a Patterson writer factory? Personally, I can't relate to this weird "ownership" claim you identify in this post, and which I've written about as well, but it seems for people whose primary identity is not author, who haven't toiled to become better at the craft, they're not grappling with the same ethical quandaries that the industry is tying itself in knots over. It's wild right now—and this author is going to self-publish and I imagine the book will do just fine. I also know for sure she will not give her LLM cover credit.
Yes I agree it's not all that different from a YA factory, James Patterson's brand name, or something like the Entangled "tailored to TikTok" books co-written by editors and agents to market trends (https://countercraft.substack.com/p/art-in-the-age-of-slop). But I don't really respect those books either, you know? People can do write and read what they want of course. But if an "author" hires a ghostwriter to write their book, I don't think are the author either. Like Trump did not write the Art of the Deal. The co-writer did. Trump has never been as articulate as that book.
I think LLMs bring up some other questions though. One is, why as a reader would you pay for an LLM novel if you could just... prompt the LLM to produce that fiction yourself? At least with traditional ghostwriters, it is still an exclusive product. James Patterson chooses his ghostwriters and is involved in the process. If you wanted to make your own Patterson book pre-LLM, you'd have to write it yourself or hire someone to write it. Prohibitively expensive. In the near future, you will perhaps be able to simply generate one with some careful prompts for free.
It's hard to know how that will all play out on the book market. I do think the self-publishing ebook world is likely going to get drowned in this stuff since there is no cost barrier to publish a book to the KU store. I also think readers do not want to be tricked by undisclosed LLM use, so the publishing industry is right to be cautious about this.
The idea that anyone else--machine or person--can "capture your voice" better than your own voice is very bizarre though. It's seems like saying a wax model looks more like you than you yourself look. In both cases, I guess it's the sense that you wish you were "better" than you are?
Totally agreed on this point. It speaks to a person who doesn't really know what their own voice is perhaps. If you tell it to write in your style, that's one thing. But I think a lot of non-writers haven't spent any time thinking about voice, so whatever the LLM spits out, as long as it sounds good—presto-voice-o.
I wish people would realize that. Kobo’s version doesn’t measure by pages read, it measures by time read. Harder to use LLMs to get around that.
Meanwhile, while my books are on Kindle, I don’t use KU and I don’t promote the Kindle link. It’s merely there for convenience. Once I get my post-D2D promotion figured out, Amazon isn’t gonna be in the fray at all.
But I wish people would wake up and realize that other distributors exist, and they have taken active measures that Amazon hasn’t.
I'm a human ghost writer (and I also write under my own name) and I can tell you that there is no comparison. Yes, we ghost writers do kinda churn and usually what we're paid to write is, uh, not that great. I wouldn't call it "slop," but we all know going into this gig that we're not being paid to produce great works of literature. "Fine" is "good enough." And honestly, it's nice to be able to pick up a few extra tens of thousands of dollars for a week's worth of relatively mindless writing, compared to the time and attention and artistry I put into my "real" work.
But there are still so many important decisions about plot, pacing, character work, etc. that need to be made in these projects. Even a book that doesn't aspire to anything particularly lofty still needs to hold together and make some kind of sense. I've tried out prompting AIs to write entire novels before, just to see what I was up against as a writer-for-hire, and the results were... bad. LLMs don't seem to understand how stories work beyond the most bare-bones concept of "story." Inconsistencies abound, and the cliches are too much even for a ghost-written project. It still ends up reading like bad LinkedIn posts (the more I see of AI writing, the more convinced I become that LinkedIn is its preferred source for examples of the written word... ugh.)
The fact that your author felt AI duplicated her voice well just doesn't say anything good about her taste level or her voice. LLMs can produce coherent sentences but so far, not coherent narratives that actually work the way stories are supposed to work. The more I see of AI writing, the less threatened I feel by it, too. Haha.
I should be clear that in my case I would not compare the act of ghostwriting to LLMs--I have also done ghostwriting, although not for novels--but rather as the "author" on the cover there is something similar in outsourcing the writing of "your work" to someone/something other than yourself that is similar.
The Rosenbaum stuff is shocking and weird! Utterly baffling that a man who spent four decades writing without AI now says he will never…write without AI. So he stopped being a writer, became a hack, and is doubling down on it
The Rosenbaum thing is so freaking weird. It really does make me feel like, for a certain portion of nonfiction authors (probably a large portion), the necessity of actually WRITING a book gets in the way of their ultimate goal, which is to direct more people to their "platform," whatever it may be. Books are not ends or goals unto themselves for these people. They are chores they must complete in order to be considered relevant within their own ecosystems of authority and power.
Bizarre, for a fiction writer observing all this. Fiction is typically much more about the BOOK for the people who write it.
I remain grateful that there are still some good writers in the nonfiction realm whose work I actually enjoy reading. But they do seem to be few and far between.
Many years ago, when the new technology was just beginning to infiltrate the book publishing business, an editor friend said, “I’m happy I’ll be retired when the shit gets really bad.” I haven’t talked to him for a long while but I learned that he retired last year.
Thanks Lincoln as usual excellent and comprehensive. I especially appreciate the links. There's gonna be plenty of AI scandals and I hope you keep writing about them. The good ones will involve plagiarism and, since you're an author, I'll bet there will be some good stuff for ya.
I don't use AI for my writing. IMHO, it's unethical because it was built on stolen, uncompensated work, and because it's an environmental catastrophe. And, I have no interest in reading works generated by LLMs.
At the same time, I wonder (assuming my objections were addressed) if there might not be some ethical and appropriate uses for LLMs (apart from other aspects of AI that are closer to expert systems.) A member of my writing group tends to see it as unmitigated evil and of no value to writers. Their so strident in their opinion, there's no discussion to be had.
When ChatGPT first popped up, I played with it just a bit. (And discovered that it thought I died in 2011, which was news to me.) In particular, I asked an editor with some built in AI to recast something I'd written (heavy on telling, weak on showing) to address the problem. I'm still learning the craft, and getting thought some specific feedback on my work (rather than abtract theory) would be useful. It was (to some extent), and I didn't have to wait a few weeks for my writing group to give me feedback.
The use case I see here is that of an instructor or editor. The tool would not rewrite the piece, but would point out some specific edits to illustrate the problem. Your article, " The View from Inside: On Adding Interiority to Your Fiction," is one of the most useful pieces of craft I've read. I re-read it on a regular basis to keep the ideas fresh. But what I wouldn't give to have specific feedback, grounded in that article, on something I'd written with suggestions on how to be better!
Some other potential uses that I think are reasonable because they're focused on the mechanics of a piece (that's already written). With the caveat, depending on how the tool is created, functions, and is used.
Reverse engineer an outline of the novel. Which I'm terrible at, because I keep falling into the weeds.
Generate an encyclopedia of the novel, a list of characters and places with a short description.
Check for inconsistencies and continuity errors.
I have no idea if any of those are possible. I have no expectation that they'd be perfect, and assume that I (or more likely an editor) would have the final say to catch what such a tool might miss. Spellcheck is far from perfect, but it's good for a first pass. But they could potentially save me a lot of time and grunt work so that I can focus on creative aspects.
Again, thank you for all your essays. I very much appreciate that you so generously share your insights and knowledge.
If we could separate 1) outsourcing your actual first-drafting of text to a large language model and 2) the very valuable things that an LLM can do in copyediting text, we might have some sort of a productive discussion. We could simply agree on some reasonable standards and move forward into the AI-assisted editing and writing era. Actually we are already there.
*But it must be disclosed.* The problem now is if you disclose any level of AI use you get this shit kicked out of you. That just isn’t necessary.
I used to flip through my printed synonym finder for untold hours looking for exactly the right word. I didn’t generate any of those lists of different synonyms. If an AI can give me a list of 5 to 10 alternate words for a word I’m struggling with what’s the problem? (Often all of the suggestions are bad by the way.)
You could say that it makes me a poorer thinker but I would not agree with that. Often the synonym finder listed 50 or more words. Using that resource didn’t make me a worse writer. It actually taught me a lot of new words. AI could easily be doing that too.
Again we need to set some reasonable standards and then be transparent.
In a foreshadowing of the Shy Girl scandal, I had a student tell me a couple years ago that her smart cousin had written an essay that was clearly written by AI. I ended up being more concerned that she didn't understand that the excuse of having a human ghostwriter wasn't any better than the digital one. The only way to teach is to find out what your students don't know, and the only way to find out what they don't know is to watch them make mistakes.
"I was reading imaginary “analysis” of poems that did not exist. Hallucinated claims about non-existent poems." That's some Borges' nightmare sentence right there.
It seems we're all living in the Borges Extended Literary Universe, like it or not.
I wonder what you think about the entire publishing industry being opposed to LLMs as ghostwriters? I had an author openly (and dare I say proudly) inform me last week that her new manuscript was ghostwritten by AI and that the AI had captured her voice so well that she felt this book was better than her first book. Here's a conundrum. I published her first book, which she thinks is more poorly written than her second book, which I rejected on grounds that we will not publish ghostwritten-by-LLM books (or books with generative AI at all, though I know we and no publisher knows the extent to which we already are). I don't regret this decision because I fundamentally don't want to publish books written by machines on my press. But people have brought up interesting points to me: How is this different than collaborating with a human ghostwriter? How are genre writers churning out books with LLMs any different than a Patterson writer factory? Personally, I can't relate to this weird "ownership" claim you identify in this post, and which I've written about as well, but it seems for people whose primary identity is not author, who haven't toiled to become better at the craft, they're not grappling with the same ethical quandaries that the industry is tying itself in knots over. It's wild right now—and this author is going to self-publish and I imagine the book will do just fine. I also know for sure she will not give her LLM cover credit.
Yes I agree it's not all that different from a YA factory, James Patterson's brand name, or something like the Entangled "tailored to TikTok" books co-written by editors and agents to market trends (https://countercraft.substack.com/p/art-in-the-age-of-slop). But I don't really respect those books either, you know? People can do write and read what they want of course. But if an "author" hires a ghostwriter to write their book, I don't think are the author either. Like Trump did not write the Art of the Deal. The co-writer did. Trump has never been as articulate as that book.
I think LLMs bring up some other questions though. One is, why as a reader would you pay for an LLM novel if you could just... prompt the LLM to produce that fiction yourself? At least with traditional ghostwriters, it is still an exclusive product. James Patterson chooses his ghostwriters and is involved in the process. If you wanted to make your own Patterson book pre-LLM, you'd have to write it yourself or hire someone to write it. Prohibitively expensive. In the near future, you will perhaps be able to simply generate one with some careful prompts for free.
It's hard to know how that will all play out on the book market. I do think the self-publishing ebook world is likely going to get drowned in this stuff since there is no cost barrier to publish a book to the KU store. I also think readers do not want to be tricked by undisclosed LLM use, so the publishing industry is right to be cautious about this.
The idea that anyone else--machine or person--can "capture your voice" better than your own voice is very bizarre though. It's seems like saying a wax model looks more like you than you yourself look. In both cases, I guess it's the sense that you wish you were "better" than you are?
Totally agreed on this point. It speaks to a person who doesn't really know what their own voice is perhaps. If you tell it to write in your style, that's one thing. But I think a lot of non-writers haven't spent any time thinking about voice, so whatever the LLM spits out, as long as it sounds good—presto-voice-o.
There’s more to selfpub than KU.
There’s more to selfpub than KU.
There’s more to selfpub than KU.
I wish people would realize that. Kobo’s version doesn’t measure by pages read, it measures by time read. Harder to use LLMs to get around that.
Meanwhile, while my books are on Kindle, I don’t use KU and I don’t promote the Kindle link. It’s merely there for convenience. Once I get my post-D2D promotion figured out, Amazon isn’t gonna be in the fray at all.
But I wish people would wake up and realize that other distributors exist, and they have taken active measures that Amazon hasn’t.
I appreciate your insights on this. I'm tracking all this stuff *almost* as closely as you are. :) Keep up the good work!
I'm a human ghost writer (and I also write under my own name) and I can tell you that there is no comparison. Yes, we ghost writers do kinda churn and usually what we're paid to write is, uh, not that great. I wouldn't call it "slop," but we all know going into this gig that we're not being paid to produce great works of literature. "Fine" is "good enough." And honestly, it's nice to be able to pick up a few extra tens of thousands of dollars for a week's worth of relatively mindless writing, compared to the time and attention and artistry I put into my "real" work.
But there are still so many important decisions about plot, pacing, character work, etc. that need to be made in these projects. Even a book that doesn't aspire to anything particularly lofty still needs to hold together and make some kind of sense. I've tried out prompting AIs to write entire novels before, just to see what I was up against as a writer-for-hire, and the results were... bad. LLMs don't seem to understand how stories work beyond the most bare-bones concept of "story." Inconsistencies abound, and the cliches are too much even for a ghost-written project. It still ends up reading like bad LinkedIn posts (the more I see of AI writing, the more convinced I become that LinkedIn is its preferred source for examples of the written word... ugh.)
The fact that your author felt AI duplicated her voice well just doesn't say anything good about her taste level or her voice. LLMs can produce coherent sentences but so far, not coherent narratives that actually work the way stories are supposed to work. The more I see of AI writing, the less threatened I feel by it, too. Haha.
I should be clear that in my case I would not compare the act of ghostwriting to LLMs--I have also done ghostwriting, although not for novels--but rather as the "author" on the cover there is something similar in outsourcing the writing of "your work" to someone/something other than yourself that is similar.
They have the same problems with developing arguments in non-fiction pieces - their impulse to equivocate and summarize is too strong.
The Rosenbaum stuff is shocking and weird! Utterly baffling that a man who spent four decades writing without AI now says he will never…write without AI. So he stopped being a writer, became a hack, and is doubling down on it
The Rosenbaum thing is so freaking weird. It really does make me feel like, for a certain portion of nonfiction authors (probably a large portion), the necessity of actually WRITING a book gets in the way of their ultimate goal, which is to direct more people to their "platform," whatever it may be. Books are not ends or goals unto themselves for these people. They are chores they must complete in order to be considered relevant within their own ecosystems of authority and power.
Bizarre, for a fiction writer observing all this. Fiction is typically much more about the BOOK for the people who write it.
I remain grateful that there are still some good writers in the nonfiction realm whose work I actually enjoy reading. But they do seem to be few and far between.
Yes the fact that Rosenbaum is a TED Talk guy probably has something to do with it. As you say, more about platform than writing per se.
Many years ago, when the new technology was just beginning to infiltrate the book publishing business, an editor friend said, “I’m happy I’ll be retired when the shit gets really bad.” I haven’t talked to him for a long while but I learned that he retired last year.
I ran into one of my old college professors a couple years ago and asked him how he was dealing with AI and he said, "I'm retiring next year!"
Almost every teacher I know is looking for an exit too :-/
Thanks Lincoln as usual excellent and comprehensive. I especially appreciate the links. There's gonna be plenty of AI scandals and I hope you keep writing about them. The good ones will involve plagiarism and, since you're an author, I'll bet there will be some good stuff for ya.
Thanks for a really thoughtful post.
I don't use AI for my writing. IMHO, it's unethical because it was built on stolen, uncompensated work, and because it's an environmental catastrophe. And, I have no interest in reading works generated by LLMs.
At the same time, I wonder (assuming my objections were addressed) if there might not be some ethical and appropriate uses for LLMs (apart from other aspects of AI that are closer to expert systems.) A member of my writing group tends to see it as unmitigated evil and of no value to writers. Their so strident in their opinion, there's no discussion to be had.
When ChatGPT first popped up, I played with it just a bit. (And discovered that it thought I died in 2011, which was news to me.) In particular, I asked an editor with some built in AI to recast something I'd written (heavy on telling, weak on showing) to address the problem. I'm still learning the craft, and getting thought some specific feedback on my work (rather than abtract theory) would be useful. It was (to some extent), and I didn't have to wait a few weeks for my writing group to give me feedback.
The use case I see here is that of an instructor or editor. The tool would not rewrite the piece, but would point out some specific edits to illustrate the problem. Your article, " The View from Inside: On Adding Interiority to Your Fiction," is one of the most useful pieces of craft I've read. I re-read it on a regular basis to keep the ideas fresh. But what I wouldn't give to have specific feedback, grounded in that article, on something I'd written with suggestions on how to be better!
Some other potential uses that I think are reasonable because they're focused on the mechanics of a piece (that's already written). With the caveat, depending on how the tool is created, functions, and is used.
Reverse engineer an outline of the novel. Which I'm terrible at, because I keep falling into the weeds.
Generate an encyclopedia of the novel, a list of characters and places with a short description.
Check for inconsistencies and continuity errors.
I have no idea if any of those are possible. I have no expectation that they'd be perfect, and assume that I (or more likely an editor) would have the final say to catch what such a tool might miss. Spellcheck is far from perfect, but it's good for a first pass. But they could potentially save me a lot of time and grunt work so that I can focus on creative aspects.
Again, thank you for all your essays. I very much appreciate that you so generously share your insights and knowledge.
Cordially
— Richard Pearce-Moses
If we could separate 1) outsourcing your actual first-drafting of text to a large language model and 2) the very valuable things that an LLM can do in copyediting text, we might have some sort of a productive discussion. We could simply agree on some reasonable standards and move forward into the AI-assisted editing and writing era. Actually we are already there.
*But it must be disclosed.* The problem now is if you disclose any level of AI use you get this shit kicked out of you. That just isn’t necessary.
I used to flip through my printed synonym finder for untold hours looking for exactly the right word. I didn’t generate any of those lists of different synonyms. If an AI can give me a list of 5 to 10 alternate words for a word I’m struggling with what’s the problem? (Often all of the suggestions are bad by the way.)
You could say that it makes me a poorer thinker but I would not agree with that. Often the synonym finder listed 50 or more words. Using that resource didn’t make me a worse writer. It actually taught me a lot of new words. AI could easily be doing that too.
Again we need to set some reasonable standards and then be transparent.
It’s chilling to think that a person can be one massive and lengthy power outage away from losing all capacity for creativity.
In a foreshadowing of the Shy Girl scandal, I had a student tell me a couple years ago that her smart cousin had written an essay that was clearly written by AI. I ended up being more concerned that she didn't understand that the excuse of having a human ghostwriter wasn't any better than the digital one. The only way to teach is to find out what your students don't know, and the only way to find out what they don't know is to watch them make mistakes.