AI is already baked into most every platform we use on the net—maps, search, translation, rideshare, etc. It’s an unavoidable part of present-day life. And LLMs have countless uses beyond generating unimpressive text: I watched in November, for instance, as the audio version of ChatGTP live translated a conversation between a group of English speakers and a Spanish speaker. It went on for 20 min. In The Hitchhiker’s Guide to the Galaxy, the Babel fish is “probably the oddest thing in the Universe,” but there it is, in everyone’s back pocket—right now. And it’s worth saying: We’re always interacting with the worst version of these models that exists; they will only get better from here—as the DeepSeek reveal shows. These tools are part of life today, and we might as well figure out how to use them wisely and talk about them sensibly.
I take your point about not living in denial about AI - have found it already very useful myself in helping with research, I'm with you on that.
What I'm unclear on is why anyone would want to read AI. Or at least, why the sort of people who are interested in reading Faulkner or Woolf would be interested in reading anything by a machine. We keep going back to the argument of, "yes of COURSE it's not great now, but imagine the masterpieces it'll be able to create in 5/10/50 years' time!", and I do have to keep asking, what does anyone even mean by 'masterpiece' in this context? What would that look like? If Hamlet had never been written and tomorrow a machine wrote it for us, it'd be completely meaningless. Wouldn't it?
(Schrodinger's Hamlet - both dead and alive until observed!)
This was a great morning read. I am looking forward to your next piece on AI and literary criticism.
I have many conversations about literature with ChatGPT. I certainly feel less guilty pestering it with questions than I do about e-mailing my Russian lit professor from 38 years ago.
For now at least you need to exercise a lot of caution though. I test out ChatGPT very often with questions about early modern literature in various languages and the only consistent thing about its answers is that it's plausible but wrong. (I mean, demonstrably, factually wrong, and in the most basic ways such as getting the author wrong, not matter-of-opinion wrong.) I share Henry's belief that it's very important to engage with this technology from a cultural perspective, so I'm not hostile to it at all. I actually find it very surprising how inaccurate it is, since many of the mistakes it makes are easily checkable. But perhaps DeepSeek is much better, I haven't tried it yet.
someone sent me a list of “top 50 poets” from GPT4 and said “this is awful”, which it was. I asked o1 and the list was exactly what you would expect. So I think sometimes the new models are quite a lot better. But I do agree about accuracy, especially as you get more abstruse or less mainstream… but these problems are being solved so rapidly!
What bothers me most is how plausible it sounds while making things up entirely! I don't understand why they do that. But yes, I'm sure accuracy will improve rapidly.
I do not disagree with you about a degree of caution.
I would not want my 15 year old son relying on it for literary criticism at the moment, However, I suspect it will be considerably improved by the time he goes off to university. I am fairly certain that his future children will benefit from it greatly.
Obviously, I feel more confident engaging with it over texts I am very familiar with.
I am a complete AI novice, I am fascinated by it and its potential but I do not have a clue how it works. For right now I just enjoy learning what I can about it.
Since there is a limit to what people like Henry and others can publish and I can't give them a call (I am thinking of Holden Caulfield in Catcher in the Rye who wishes you could call the author of a book you just finished whenever you feel like it) I do appreciate being able to engage AI with my endless queries whenever I want.
I am of the same opinion that @listentothesirens expressed on his last EconTalk podcast with @reidhoffman when speaking about Claude, "Claude doesn't get annoyed at my stupidity, doesn't roll his eyebrows, he doesn't get tired."
This is a massive blindspot for those who believe AI is an unalloyed good. You have no way of knowing whether information supplied by AI is accurate. Wikimedia has problems but it also has citations. AI is known to create events, documents and history that is completely false. Without redoing the research you just avoided by having AI do it for you, you’ll never know if you are badly misleading others. The major AI engines all relied on IP theft so asking AI to provide citations ain’t going to happen. Scam artists, not real artists, will profit from AI.
A realistic look at the potential future impact of AI on literature.
I use ChatGPT daily. It helps with aspects of life I find boring ("make this email to my colleague sound less passive-aggressive") and has been a crutch to support me when it comes to stress-testing my own ideas (both in terms of writing and world domination, neither of which are a good advertisement for success!).
Would I let it write for me? Would I publish anything written by an LLM? Would I knowingly read anything written by AI? No, no and certainly not.
Telling stories is the oldest human pastime (well, one of two). It builds connections and communities between people. The internet has broken down geographical barriers to forming those communities, but the same connection is maintained in some manner.
I view the idea of "AI slop" as a short-term issue, and it is already being maligned for its quality. However, I do see its benefits as a tool, especially when it comes to crafting ideas, asking questions and working as a sounding board.
Ugh. I fear you're onto something re AI and contemporary writing sometimes being a continuation of X/Twitter in different form. People are lazy. AI provides an antidote. If they can sit back and let The Machine do a lot of the work then, hell, many will take it. I never will. At least I tell myself that now.
yes but many will collaborate with the AI, and, honestly, a lot of writing is cliched anyway, perhaps the collaboration will be better in some ways, worse in others... who knows?
I enjoy how bullish you are on AI even if I hate it myself. It's like—well, somebody I know needs to be.… if only to put in a good word for me in the Skynet future…
Lol. The funny thing is I’m pretty bullish but not *that* bullish! I just don’t see how we ignore it and I find it incredible the answers you can get. The worries are real, but I’m not sure the anti position will work for us as a group.
despite my Skynet joke I don't actually _fear_ AI tbh—at least not in the realm of writing. It just doesn't interest me really. There are some realms where I worry about it being used in its present, "hallucinating the answer" form (mostly medicine).
It performs very well when tested in medicine! And that’s one reason I think it will work with people at least for the foreseeable future. I do think it’s a real challenge for writers, and I use it more than I would because I think I need to know, but I use it least than everyone I know who loves it, I think.
My first instinct was to lean away from LLMs, but it’s been pleasantly surprising how you can use them to be really helpful to you. I prompt GPT to prompt me with questions so I can get at and properly articulate what I like about a text or think about a topic when it’s all vague notions, “vibes” and half-formed opinions in my head. Then I might ask it to refute my points which makes me think of more dimensions to what I’m thinking—mostly, I use it to arrive at and deepen subjective understanding; still wary about relying on it for factual information
one thing I'm grateful to A.I. for is the chance to ask it the kind of embarrassing questions I may feel disinclined to ask of even the most supportive tutor, or group. I've found Deepseek to be (at least so far) infinitely patient with my misunderstandings, my repeated questions, and the huge gaps in my education that I'm still only really feeling my way through.
I share an instinct against a.i. generated content - by definition it will lack humanity - but I'm also mindful of change, endeared (perhaps irrationally) towards the a.i. for its support, and nonetheless keen to lean into my resistance and revulsions against any of its creations. and I'm also fascinated by it - what an incredible machine!
Perhaps I may never be truly, deeply moved by the work of an a.i., but I'm increasingly sure that, by at least engaging with it, I may well learn something unexpectedly useful.
I don't know of course, but i think it might just find certain kinds of literary creations trivially easy, and we may well be incidentally moved them.
it can spit out pretty deep sounding aphorisms already, and the grammar for those kinds of language objects is not *that* hard to understand. feel properly conflicted, though - one wants to sense a soul beneath the syntax!
The popularity of AI is inevitable, and literature needs to adapt. Art has never created society- it’s always been mimetic. To believe lit can continue without AI influence is kinda mistaken tbf
It always comes down to: “AI is coming/already here, stop complaining and take it.”
Sometimes it gets mixed in with some “just try it, it’s a tool,” and at some point the Buggy Whips are paraded.
This article was different for acknowledging that downsides and tradeoffs might exist (without dwelling on them too much, because we mustn’t be Luddites, must we?), so that’s something.
If a machine can make a “great” novel, then there’s no reason to be a novelist. There’s no reason to learn how to write, barely any to learn how to read. You have turned our major communication tool into a shiny red button that anyone can smash without any understanding of the result. Such understanding will be unnecessary.
So no, I won’t be joining your experiment, just because our Masters have decreed it. I would rather cut my fingers off than ever type a syllable into an LLM.
You have everything you need to write, inside your own mind. You are betraying yourself uploading this into the Machine.
Henry, you're like the cute boy on the other side of the class over there with your Middlemarch, so we're happy you're noticing The Metropolitan Review, even if it's a neg!
This is a smart post. Totally. Understanding what AI will do to writing is beyond our capacity - although we agree that the effect can't be anything else than seismic.
However, we do think there are a couple of lines in the sand that can be drawn. One is that we believe that the writing of the future must be more democratic. There's just too much gatekeeping in the literary world; too much expression gets stifled. This platform opens up the possibility of people writing - and finding readers - on a scale that has never existed before. When we talk about the glorious future, that's what we mean.
The other line in the sand is that we feel pretty comfortable rejecting AI in toto. AI can become technically proficient, produce gazillions of great novels, and from our perspective that doesn't matter at all. The sole purpose of writing is human expression - people communicating their truth of being in the world. Technical proficiency is a very distant runner-up to that cry for help; or celebration of the fact of life; or whatever it is that people choose to say. Since AI is completely unperturbed by its own existence, what AI produces is, from our perspective, ipso facto uninteresting. We can keep an eye on it like we might the town circus producing a new trick, but we think it's just not where it's at.
I'm with Ted Gioia and Freddie DeBoer on A.I., sorry. It also doesn't have a business model. Maybe one day it will find one, or be subsidized so much by the government and Big Tech that it doesn't matter. The latter may be the reality. It's also abysmal for the environment.
I don't understand handwaving away the impact of conglomeration on literature. That's a huge story of the late 20th century.
AI is already baked into most every platform we use on the net—maps, search, translation, rideshare, etc. It’s an unavoidable part of present-day life. And LLMs have countless uses beyond generating unimpressive text: I watched in November, for instance, as the audio version of ChatGTP live translated a conversation between a group of English speakers and a Spanish speaker. It went on for 20 min. In The Hitchhiker’s Guide to the Galaxy, the Babel fish is “probably the oddest thing in the Universe,” but there it is, in everyone’s back pocket—right now. And it’s worth saying: We’re always interacting with the worst version of these models that exists; they will only get better from here—as the DeepSeek reveal shows. These tools are part of life today, and we might as well figure out how to use them wisely and talk about them sensibly.
yes exactly so, we don't know what's coming, but we do now it's going to be here before we quite understand what it is
I take your point about not living in denial about AI - have found it already very useful myself in helping with research, I'm with you on that.
What I'm unclear on is why anyone would want to read AI. Or at least, why the sort of people who are interested in reading Faulkner or Woolf would be interested in reading anything by a machine. We keep going back to the argument of, "yes of COURSE it's not great now, but imagine the masterpieces it'll be able to create in 5/10/50 years' time!", and I do have to keep asking, what does anyone even mean by 'masterpiece' in this context? What would that look like? If Hamlet had never been written and tomorrow a machine wrote it for us, it'd be completely meaningless. Wouldn't it?
(Schrodinger's Hamlet - both dead and alive until observed!)
This was a great morning read. I am looking forward to your next piece on AI and literary criticism.
I have many conversations about literature with ChatGPT. I certainly feel less guilty pestering it with questions than I do about e-mailing my Russian lit professor from 38 years ago.
footnotes on demand!!
For now at least you need to exercise a lot of caution though. I test out ChatGPT very often with questions about early modern literature in various languages and the only consistent thing about its answers is that it's plausible but wrong. (I mean, demonstrably, factually wrong, and in the most basic ways such as getting the author wrong, not matter-of-opinion wrong.) I share Henry's belief that it's very important to engage with this technology from a cultural perspective, so I'm not hostile to it at all. I actually find it very surprising how inaccurate it is, since many of the mistakes it makes are easily checkable. But perhaps DeepSeek is much better, I haven't tried it yet.
which model are you using?
Good question! Not sure. I have an account though.
someone sent me a list of “top 50 poets” from GPT4 and said “this is awful”, which it was. I asked o1 and the list was exactly what you would expect. So I think sometimes the new models are quite a lot better. But I do agree about accuracy, especially as you get more abstruse or less mainstream… but these problems are being solved so rapidly!
What bothers me most is how plausible it sounds while making things up entirely! I don't understand why they do that. But yes, I'm sure accuracy will improve rapidly.
they do it because they were trained on humans!
I do not disagree with you about a degree of caution.
I would not want my 15 year old son relying on it for literary criticism at the moment, However, I suspect it will be considerably improved by the time he goes off to university. I am fairly certain that his future children will benefit from it greatly.
Obviously, I feel more confident engaging with it over texts I am very familiar with.
I am a complete AI novice, I am fascinated by it and its potential but I do not have a clue how it works. For right now I just enjoy learning what I can about it.
Since there is a limit to what people like Henry and others can publish and I can't give them a call (I am thinking of Holden Caulfield in Catcher in the Rye who wishes you could call the author of a book you just finished whenever you feel like it) I do appreciate being able to engage AI with my endless queries whenever I want.
I am of the same opinion that @listentothesirens expressed on his last EconTalk podcast with @reidhoffman when speaking about Claude, "Claude doesn't get annoyed at my stupidity, doesn't roll his eyebrows, he doesn't get tired."
This is a massive blindspot for those who believe AI is an unalloyed good. You have no way of knowing whether information supplied by AI is accurate. Wikimedia has problems but it also has citations. AI is known to create events, documents and history that is completely false. Without redoing the research you just avoided by having AI do it for you, you’ll never know if you are badly misleading others. The major AI engines all relied on IP theft so asking AI to provide citations ain’t going to happen. Scam artists, not real artists, will profit from AI.
Well said
You Russian lit professor is an unequivocally better resource than chat GPT.
Can't?
Or shouldn't? Wouldn't? Ought not?
Can’t. It’s here.
Seems to me that literary culture is quite good at dismissing and ignoring a great many things in history, good and bad, no?
lol yes it can be, very much so, but in periods of flourishing it has a very responsive relationship to the modern, even if it dislikes it
So again — since they can be good at dismissing what is modern — is it an issue of ability, morality, volition...?
And related, sort of predicating that, are there times where it has both flourished and rejected the modern? Say in totalitarian states?
“The essence of poetry is invention.”
Invention — a word, as Heidegger said, whose root etymology means discovery. What if the invention in question neuters discovery?
What if it burns the very roots and salts the very ground of language per se through theft?
Would that not be worth ignoring, if indeed poetry — and all vocational fine art — opposes technocracy?
A realistic look at the potential future impact of AI on literature.
I use ChatGPT daily. It helps with aspects of life I find boring ("make this email to my colleague sound less passive-aggressive") and has been a crutch to support me when it comes to stress-testing my own ideas (both in terms of writing and world domination, neither of which are a good advertisement for success!).
Would I let it write for me? Would I publish anything written by an LLM? Would I knowingly read anything written by AI? No, no and certainly not.
Telling stories is the oldest human pastime (well, one of two). It builds connections and communities between people. The internet has broken down geographical barriers to forming those communities, but the same connection is maintained in some manner.
I view the idea of "AI slop" as a short-term issue, and it is already being maligned for its quality. However, I do see its benefits as a tool, especially when it comes to crafting ideas, asking questions and working as a sounding board.
Ugh. I fear you're onto something re AI and contemporary writing sometimes being a continuation of X/Twitter in different form. People are lazy. AI provides an antidote. If they can sit back and let The Machine do a lot of the work then, hell, many will take it. I never will. At least I tell myself that now.
yes but many will collaborate with the AI, and, honestly, a lot of writing is cliched anyway, perhaps the collaboration will be better in some ways, worse in others... who knows?
I enjoy how bullish you are on AI even if I hate it myself. It's like—well, somebody I know needs to be.… if only to put in a good word for me in the Skynet future…
Lol. The funny thing is I’m pretty bullish but not *that* bullish! I just don’t see how we ignore it and I find it incredible the answers you can get. The worries are real, but I’m not sure the anti position will work for us as a group.
despite my Skynet joke I don't actually _fear_ AI tbh—at least not in the realm of writing. It just doesn't interest me really. There are some realms where I worry about it being used in its present, "hallucinating the answer" form (mostly medicine).
It performs very well when tested in medicine! And that’s one reason I think it will work with people at least for the foreseeable future. I do think it’s a real challenge for writers, and I use it more than I would because I think I need to know, but I use it least than everyone I know who loves it, I think.
Enjoyed reading this!
My first instinct was to lean away from LLMs, but it’s been pleasantly surprising how you can use them to be really helpful to you. I prompt GPT to prompt me with questions so I can get at and properly articulate what I like about a text or think about a topic when it’s all vague notions, “vibes” and half-formed opinions in my head. Then I might ask it to refute my points which makes me think of more dimensions to what I’m thinking—mostly, I use it to arrive at and deepen subjective understanding; still wary about relying on it for factual information
one of the best ways to use them!
one thing I'm grateful to A.I. for is the chance to ask it the kind of embarrassing questions I may feel disinclined to ask of even the most supportive tutor, or group. I've found Deepseek to be (at least so far) infinitely patient with my misunderstandings, my repeated questions, and the huge gaps in my education that I'm still only really feeling my way through.
I share an instinct against a.i. generated content - by definition it will lack humanity - but I'm also mindful of change, endeared (perhaps irrationally) towards the a.i. for its support, and nonetheless keen to lean into my resistance and revulsions against any of its creations. and I'm also fascinated by it - what an incredible machine!
Perhaps I may never be truly, deeply moved by the work of an a.i., but I'm increasingly sure that, by at least engaging with it, I may well learn something unexpectedly useful.
yeah I think this is right, if nothing else it is simply too early to know what sort of art AI might produce
I don't know of course, but i think it might just find certain kinds of literary creations trivially easy, and we may well be incidentally moved them.
it can spit out pretty deep sounding aphorisms already, and the grammar for those kinds of language objects is not *that* hard to understand. feel properly conflicted, though - one wants to sense a soul beneath the syntax!
I agree that people will find themselves moved by something AI wrote, probably without realising AI wrote it
The popularity of AI is inevitable, and literature needs to adapt. Art has never created society- it’s always been mimetic. To believe lit can continue without AI influence is kinda mistaken tbf
I suspect a new sort of writing will appear, perhaps written by people currently not part of the literary scene...
You get it right, as usual!
ah thanks!
It always comes down to: “AI is coming/already here, stop complaining and take it.”
Sometimes it gets mixed in with some “just try it, it’s a tool,” and at some point the Buggy Whips are paraded.
This article was different for acknowledging that downsides and tradeoffs might exist (without dwelling on them too much, because we mustn’t be Luddites, must we?), so that’s something.
If a machine can make a “great” novel, then there’s no reason to be a novelist. There’s no reason to learn how to write, barely any to learn how to read. You have turned our major communication tool into a shiny red button that anyone can smash without any understanding of the result. Such understanding will be unnecessary.
So no, I won’t be joining your experiment, just because our Masters have decreed it. I would rather cut my fingers off than ever type a syllable into an LLM.
You have everything you need to write, inside your own mind. You are betraying yourself uploading this into the Machine.
There’s some disturbing comments in here. ChatGPT is a shortcut to thinking and writing and creating.
Henry, you're like the cute boy on the other side of the class over there with your Middlemarch, so we're happy you're noticing The Metropolitan Review, even if it's a neg!
This is a smart post. Totally. Understanding what AI will do to writing is beyond our capacity - although we agree that the effect can't be anything else than seismic.
However, we do think there are a couple of lines in the sand that can be drawn. One is that we believe that the writing of the future must be more democratic. There's just too much gatekeeping in the literary world; too much expression gets stifled. This platform opens up the possibility of people writing - and finding readers - on a scale that has never existed before. When we talk about the glorious future, that's what we mean.
The other line in the sand is that we feel pretty comfortable rejecting AI in toto. AI can become technically proficient, produce gazillions of great novels, and from our perspective that doesn't matter at all. The sole purpose of writing is human expression - people communicating their truth of being in the world. Technical proficiency is a very distant runner-up to that cry for help; or celebration of the fact of life; or whatever it is that people choose to say. Since AI is completely unperturbed by its own existence, what AI produces is, from our perspective, ipso facto uninteresting. We can keep an eye on it like we might the town circus producing a new trick, but we think it's just not where it's at.
Happy to chat this out further!
- TMR
1) Gates and gatekeeping are good. Ask any sonnet writer. Good fences make good neighbors and great writing.
2) The sole purpose of writing is what?
did you guys use AI to write this lol
I'm with Ted Gioia and Freddie DeBoer on A.I., sorry. It also doesn't have a business model. Maybe one day it will find one, or be subsidized so much by the government and Big Tech that it doesn't matter. The latter may be the reality. It's also abysmal for the environment.
I don't understand handwaving away the impact of conglomeration on literature. That's a huge story of the late 20th century.
You lost me at Sally Rooney, who could well be a LLM.