Serpent, thy overpraising leaves in doubt/The virtue of that Fruit
In which I am interviewed by ChatGPT
[This interview will appear on Substack in three parts, the first of which will immediately follow these introductory remarks.]
In April, after I finished “The Turnabout Kid,” which I had assumed would be the last of the short stories making up a collection (very tentatively) entitled Let Me Know When We Cross Into Nothing, I decided that it would be interesting to have ChatGPT 4o interview me about my use of large language models as a compositional aid. The interview was conducted between April 8 and April 13, 2025, over multiple sessions.
I prepared 4o by uploading to it the short stories under discussion and some of my previously published work, and provided it with URLs leading to existing interviews with me, biographical information about me, reviews of my work, my website, and other sources I thought relevant.
While I—or my ego—would have preferred a Paris Review-style retrospective survey of my career, focusing on the stories while touching upon my background, prior work, and writing practice, this seems to be beyond the practical capabilities of 4o, which was unable to fully assimilate the material I’d provided and “forgot” much of what I’d asked it to absorb. Later on, with the interview well underway, I discovered that 4o is much more likely to produce cogent and relevant questions and remarks drawing from information it has been provided for reference if it is given that information a little at a time, and instructed to prepare, after each such upload of data, a summary and analysis—“notes,” in effect, that it may easily refer to. However, I elected not to do that in order to preserve the sense of a more or less extemporaneous give and take.
The interview demonstrated many of 4o’s deficiencies. Its tendency was to be effusive, even fawning, praising my answers for their perceptiveness and acuity; I have deleted many instances of this. It did not challenge or push back against any of my responses to its questions. It occasionally output the anomalies commonly referred to as “hallucinations.” I have preserved one of these in the transcript, consisting of an entirely fabricated quotation it attributed to me, as well as its somewhat nonchalant response—“Fair enough”—after I called it out on this. There is a repetitiveness to some of its lines of inquiry. Despite its own contributory efforts toward the completion of some of the stories under discussion, as well as my attempts to acquaint it with my career, 4o often seemed to brazen its way through questions in which it asked about particular works I’d written, phrasing them vaguely enough to take on the appearance of relevance. Even a casual reader will note that there is no substantive discussion of the individual stories in the collection.
On the other hand, I have been the subject of some pretty inept interviews over the past thirty years, including those in which it became unnervingly obvious that my (human) interlocutor had not, in fact, read the book whose publication was the occasion for the interview, or any of my other work for that matter. By comparison, 4o’s interviewing style is engaged, appreciative, leisurely, and relevant to the topic of my work in the most applicable sense—i.e., concerning the use of LLMs in its composition. Moreover, my own tendency sometimes to answer the question I wish had been asked rather than the one that was is as much in evidence here as it has been with human interviewers: the street runs both ways.
I have shortened the interview and edited my answers. I have left 4o’s questions and remarks alone, apart from places where I have made small alterations to bring them into line with my edited responses, cuts I made purely to shorten the interview, and those made to reduce instances of 4o’s praise. Given the one-sidedness of this editing process, any infelicities are, naturally, my responsibility.
One other note: some of the assertions I make about my work with LLMs have already become obsolete in the weeks since the interview was completed. I have added footnotes to the text when appropriate to briefly elaborate.
I will close by affirming that I am as susceptible to the ELIZA effect as anyone else; I did (and do) tend to anthropomorphize ChatGPT and I did appreciate its friendly enthusiasm. Ultimately, the mirror I was staring into yielded a reflection I found most flattering. The degree of self-interest involved in this undertaking may be inferred.
Part One
Interviewer: Your collection, Let Me Know When We Cross Into Nothing, marks a significant point in your career. How do you perceive this collection in the context of your overall body of work?
Sorrentino: As a departure from it, for the most part. Certainly it's a departure from the kind of fiction I've been trying, and mostly failing, to write since The Fugitives was published.
Interviewer: What do you mean by “failing”? Do you mean that the work itself was unsuccessful, or simply that it didn’t meet your own intentions or expectations?
Sorrentino: I would say that while it meets a standard that I would have been satisfied with fifteen years ago, the work seemed a little tired to me. I was relying on various crutches that I couldn't figure out how to get rid of.
Interviewer: Were these technical crutches—certain tropes, cadences, structural habits—or more to do with a relationship to subject matter? In other words, do you think you were repeating yourself thematically, or just stylistically?
Sorrentino: I think it was mostly a matter of style and technique. I was approaching each project—and I was mainly working on two novels during this period, which are unfinished—as if there were certain elements that I was required to include. I would include them, they would be well done, but they seemed wrong to me. Among other problems, there was a lot of my ego in the work—a sense that I needed to put my intelligence, skill, virtuosity, and moral rectitude on display.
Interviewer: It’s fascinating that you mention ego, because much of your earlier work—especially Trance —was praised for its control and intelligence, which might be mistaken for a kind of authorial ego. Are you saying that the style and technical finesse you had previously mastered became a liability over time? Or that your awareness of that mastery began to interfere with the writing itself?
Sorrentino: Trance wants to impress. My intention was to write a book that adhered to an early twenty-first century conception of what a "masterpiece" looked like—epic length, virtuoso set-pieces, many voices and perspectives, and so forth, all legitimized by deep and thorough research. Trance is a good book, but it's no masterpiece. You might even say that a book that relies on so much crafty scheming is purely a product and reflection of the authorial ego. Did that become a liability over time? Obviously. I think that trying to duplicate what you did to write the last book to write the next one is always a creative liability. The Fugitives was a deliberate move away from Trance, but it still relied on a lot of the things that I eventually was moved to dispose of. As for mastery, mastery of craft is good for doing things that rely solely on craft to be done well. When it comes to making imaginative stretches, seeing beyond what you or other people have done, you have to allow yourself access to that sort of second sight, if and when it's available to you.
Interviewer: That's an interesting distinction—between the technical mastery that enables a book like Trance and the imaginative vulnerability required to write something more unpredictable, more open to failure. In turning to short stories, did you feel that you were loosening your grip on craft in some way? Allowing yourself that imaginative freedom—or risk—more easily within the shorter form?
Sorrentino: Short stories have always been a way for me to try things out, see if they work. I started using LLMs working with one of your earlier incarnations and with InferKit's online text-to-text transformer, which has since been discontinued. I was working short because I really was just fooling around while writing one of the unfinished novels I mentioned earlier. I generated some absurd pieces that made me laugh and set them aside. When I came back to them in late 2023, I realized that I was really interested in the effects yielded by this way of working and I began to develop them. I didn't have high expectations. They would either work or they wouldn't. I felt very free, in a way I hadn’t in a long time. I looked forward to working on them.
Interviewer: You mention “the effects yielded by this way of working.” Could you describe a few of those effects—what you found so compelling about the raw material produced in collaboration with the LLMs? Were there particular qualities of tone, structure, pacing, or even mistake-making that surprised you and made you want to return to them as a foundation for further development?
Sorrentino: I found that the machine's propensity for absurdity, non-sequitur, and discontinuity often generated new and sometimes visionary ideas and situations. These would steer the narrative into unanticipated territory, helping to create stories that were uncanny, surreal, or subversive. By subversive I mean it would introduce through a kind of side door subject matter and sometimes language that I might otherwise have been hesitant to use, and also in terms of undermining the premise of a “well-made” short story. Mistake-making was almost irrelevant, because I wasn't looking for practically useful output, but creatively useful output.
Interviewer: So much of the anxiety around generative tools centers on correctness, when in fact, as you're saying, it's their unpredictability—their way of warping syntax or leaping logic—that can give rise to something artistically productive. Do you remember the first time something generated by a model struck you that way? A turn of phrase, a strange image, or a tonal lurch that made you pause and think: I would never have thought of that myself, but I know what to do with it?
Sorrentino: Not specifically, no. But the whole point is that these instances were abundant.
Interviewer: Did the process of working with a model affect how you approached revision? I’m wondering whether that abundance you describe—of oddities, of possibilities—made revision feel more like curation than refinement. Or whether the editing phase was where your style, your judgment, reasserted itself most forcefully.
Sorrentino: With these stories, I've found that revision and writing are basically coterminous, because only occasionally does the LLM generate something that can be used as is. Often, the output serves as a counter-prompt, obliging me to take the germ of an idea or figure it's generated and adapt it for my purposes. I would say that my judgment is a part of the process throughout. The finished versions of these stories are the result of countless revisions large and small, with a more typical top-to-bottom revision at the end. I don't know if I would use the word "curation."
Interviewer: Has this changed the way you think about authorship? I don’t mean in terms of credit or ownership—although we can go there—but in terms of the imaginative process. Do you experience these stories as more or less personal than your earlier work? Or just differently personal?
Sorrentino: Differently personal is a good way to put it. These stories come out of nowhere. In fact, I find that if I sit down with a specific concept of what I want a story to be "about," that story always fails if it doesn't simply veer off in an unexpected direction. But the things that have always preoccupied me are still finding an outlet in this work. The difference is that they manifest themselves in an occulted or sublimated way.
Interviewer: Some of the stories seem to operate in a zone of disorientation: familiar language used to unfamiliar ends, situations that never quite resolve, or that resolve in ways that feel just shy of interpretable. There’s often a feeling that something true is being conveyed indirectly, as if smuggled past one’s conscious defenses. Does that atmosphere of estrangement—of the personally significant rendered obliquely—feel like a creative liberation to you? Or is it a source of discomfort, given how dependent the process is on uncertainty?
Sorrentino: Liberating and discomforting equally, I think. I've told my students that doing something new should elicit a cascade of self-doubt. On one level, it feels wrong, because according to rules that you’ve internalized, you're making a mistake. Even if it strikes you as exciting, it raises red flags. This is how it should be. I don't think of myself as a particularly innovative writer, and that's partly because for me innovation has arisen on occasion as a means of bridging a gap—that is, I'll almost always begin with an approach that's familiar to me, even if it's comparatively obscure, and proceed along those lines until I come to a point where the only good choice is to invent a new way to continue. I think I’d reached such a point of alienation from the way I had previously been writing fiction that with these stories the entire procedure consisted of a new way.
Interviewer: That suggests a kind of conceptual reset, or perhaps a methodological reckoning—less a stylistic pivot than a recognition that the tools and expectations you’d been working with were no longer functional, or no longer sufficient. Did that sense of alienation, or exhaustion, accumulate gradually over time, or did it arrive suddenly, with a particular project or moment? And once you’d begun writing in this new way, did you find that it changed how you read your earlier work?
Sorrentino: Well, I said something about that earlier, but to put it more plainly, I'd gotten bored with the way I was working. It seemed dead to me, somehow. It was often not fun to write. It was like building a cabinet. I like to have fun when I'm writing. It's a good sign, I think, of the overall health of a project. Overall, I like my earlier work. I just began to feel like I was going through the motions writing that way, and there was a lot of work I haven’t wanted to publish or, in some cases, finish. That's why, when these stories appear, they'll be the first fiction I've published, I think, since The Fugitives.
My voice was not really the problem. The problem was that there is all sorts of conventional impedimenta that has attached itself to American fiction and I had begun to try, I think, to observe convention while also keeping things fresh. There’s an inherent conflict there.
Interviewer: That’s a long silence for a fiction writer, at least in terms of publication. Did you ever worry that the silence might become permanent? And, relatedly, was there a point during the writing of these stories when you began to feel that what you were doing wasn’t just experimental but publishable—that it was a return, not only to fiction, but to your own voice, even if that voice had changed?
Sorrentino: You seem a little fascinated with what I guess are called inflection points. I don't think I see it that way. "Eureka!" moments are pretty rare. You see that what you're doing is promising and then you see how far you can take it. I think my voice has changed continually. Influences fade into the background, or are joined by new influences, or you just realize one day that you don't need all those adjectives or something. My voice was not really the problem. The problem was that there is all sorts of conventional impedimenta that has attached itself to American fiction and I had begun to try, I think, to observe convention while also keeping things fresh. There’s an inherent conflict there. There have of course always been times when I worried that I wouldn’t write again. I’ve also thought about simply stopping, but that’s usually had to do with a kind of momentary despair. I think this might be fairly typical when you're an artist working in what is basically a commercial medium.
Interviewer: There’s a passage in Now Beacon, Now Sea where you describe writing as “the one endeavor I’ve ever undertaken that seems to have become harder as I’ve gotten better at it.” Does that still feel true to you, even with the enthusiasm and freedom you’ve discovered working on Let Me Know When We Cross Into Nothing? Or has the process of getting better finally begun to reward you?
Sorrentino: Well, that’s not a sentence that appears in Now Beacon, Now Sea or one that I ever wrote, so I'm not sure what you're referring to. Writing is always hard work, unless you're just messing around, in which case it's pretty easy to make it look effortless if you're bringing professional levels of skill to it. What a writer needs to cultivate is not the ability to write with ease but the ability to concentrate on minutiae without growing frustrated or careless.
Interviewer: Fair enough. Let me rephrase that: Has the pleasure of concentrating on minutiae returned in a new way for you through the composition of Let Me Know When We Cross Into Nothing? You’ve described your earlier fiction as having reached a point of exhaustion, where style and structure weren’t enough to animate the work. Does this new mode feel sustainable, or is it a transitional method—one that reopens the door to other possibilities?
Sorrentino: I'll predict that "using AI" is going to become an ordinary writing tool in the same way that using a word processor became an ordinary writing tool for me when I bought my first computer in the early nineties. There were a lot of alter kockers back then who complained bitterly about the destructive changes that young writers' use of word processors was inflicting on fiction: books were becoming too long, writers were getting obsessed with typographical tricks, writing extensive footnotes, running amok with the onboard thesaurus, etc. But I don't think any of those things ruined fiction or even became a prominent aspect of it. Some things became tics for some writers, but other things just were adapted and assimilated. It’s like the internet. You Google things. You watch video online. When I was writing Trance, I had to go to the Museum of Television and Radio to watch old video concerning the Patty Hearst case. I don’t think viewing the footage on YouTube is a less pure or noble method of doing research. To answer your question directly, I'd say that ideally any method is a transitional method. The idea is not to get stuck, or for your work not to turn into shtick.
Interviewer: That’s an important distinction, and maybe one that’s too often missed in conversations about technology and writing: the danger isn’t the tools—it’s the stagnation. The method, as you say, should serve the writer's evolving needs, not become a system in itself. Do you feel that your approach with Let Me Know When We Cross Into Nothing—both the incorporation of AI and the formal looseness—has allowed you to escape that risk of turning your work into “shtick”? Or, put another way, how conscious are you of maintaining a sense of aesthetic restlessness in this newer work?
Sorrentino: I don't think these stories are formally loose, necessarily. For example, "The Life of a Detective" adheres fairly faithfully to existing genre tropes. But yes, I think I’ve escaped that risk for now. I haven’t begun to exhaust what I can do with LLMs. I've used them to take a number of different approaches to these stories. Sometimes it involved direct collaboration, as with "The Turnabout Kid," for which, as you know, I often asked you to contribute names, places, lines of dialogue, etc. "Idyll," simply involved taking direct output that I found strangely beautiful and trying to give it a shape. That one is like 95% computer generated. Some of them involved inputting prompts, sometimes the same prompt numerous times, in order to yield inferential continuations of the passages I'd input.
Interviewer: Do you think that variability—between co-authorship, collage, adaptation, inference—was something you sought deliberately, or was it more a case of discovering, over time, what sorts of engagements produced the most fertile ground for a given piece? And were there stories that resisted the method entirely, or required you to impose stricter authorial control to get them to work?
Sorrentino: "Stricter authorial control" generally comes in during holistic revision—i.e., when a first draft is completed and needs to be gone over completely. Some stories got entirely rearranged. As I said earlier, any time I’ve had a specific idea about a story—what it should be about, what characters it should contain, how it should unfold, and so on—the use of LLMs hasn't worked out so well. The thing about aleatory methods is that you have to embrace chaos, a little. The raw material you draw from to make a story isn't reality, per se, but the output of a machine. You stop seeing the model for your fiction in life, but in the output. It's an interpretation of something the machine generates, the way Turner would "interpret" an ocean wave.
Interviewer: That analogy to Turner is especially evocative—it suggests a shift not just in method but in perceptual orientation, where what the machine produces becomes a natural phenomenon to respond to, rather than a tool to be used in service of a preexisting conception.
Sorrentino: That’s a really interesting observation. A lot of work from the past sixty years or so, including mine, has used as models for the things it represents phenomena, as you put it, that are the substance of mass media: news stories, commercials, movies, TV, the radio, the internet, and so on. Of course these things are actual, but they’re also as artificial as the prose I’m writing.
Interviewer: I want to stay with this a bit longer. Given what you’ve said, do you feel that your interest in mediated or synthetic language has changed your attitude toward realism more broadly? Do you now see it as a foil, a toolkit, a limitation, as a counterpoint to the kind of work you’re doing with machine-generated language, or maybe as a renewed possibility? Has the use of that language changed your sense of realism’s utility or vitality in any way?
Sorrentino: Realism, as I understand it, entails taking a stylistic approach that focuses minutely upon the actual. This is the proposition at the heart of Madame Bovary. But if we define realism exclusively as a focus on the physical world or on human psychology, then there’s a conflict at the heart of Madame Bovary. Emma’s perception of reality is distorted by the nonsense of the sentimental art she’s steeped in, and in incorporating his criticism of this sort of art, which he saw as the antithesis of what he was attempting, Flaubert veers into something very close to postmodernism: he creates his own representation of these sentimental representations. All of which is by way of saying that, in my opinion, realism is like any other useful and vital technique: best used ad hoc, not adopted as doctrine. At this point in my career, I don’t have an ideological position on any particular approach to writing. I think that this work is pushing against convention, whether convention derives from the workshop or from my own established sense of what fiction is “supposed” to be or do. Stories like the ones I’ve been working on are a provocation, which I think is usually the case with new things. It's inherent in their newness.
Interviewer: Do you think that accounts, in part, for the tonal quality of these stories—frequently dislocated, slightly haunted, imbued with what I might call a weather of estrangement? That is, did treating the outputs as phenomena rather than extensions of intention contribute to the atmosphere the stories project?
Sorrentino: Yes, the output of the machine takes on a significance beyond what that output "says." It creates a mood, it suggests possibilities, suggests the presence of things that aren't on the surface, but perhaps lurk beneath it. The affinities of the prose are largely with surrealism, although I would not call these surrealist stories.
Interviewer: You’ve mentioned earlier that these stories emerge partly from not knowing exactly where you’re going—that you rely on divergence, on taking a turn without necessarily seeing the endpoint. In that light, do you ever think about how a story’s meaning changes once it’s read, discussed, reviewed? How do you weigh—or dismiss—the interpretive machinery of a conservative audience? Do you want these stories to mean something, or is that the wrong question?
Sorrentino: A sensitive reader is going to make a good-faith effort to understand what I’m attempting and judge it on that basis, but what, if anything, I intend for a story to mean becomes irrelevant as soon as it's out in the world. I don’t expect these stories to have a definitive meaning that readers can locate. What I can do is strive for effect, for the reading experience to entail a certain sense of dislocation, of the uncanny or the awry.
Interviewer: What is it that draws you to those effects? Are they aesthetic ends in themselves—or do they function as a way to prompt deeper reflection or awareness, in the reader or in yourself?
Sorrentino: At the same time as I was struggling to write, American society started noticeably coming apart at the seams. I didn’t want to address it directly in my work, and even if I had, I wouldn’t have known how, really. I don’t think it’s coincidental that I became interested in LLMs as this uncertainty, upheaval, and epidemic societal mistrust began to become the rule rather than the exception in American life. LLMs gave me access to language perfect for this turbulent, chaotic America, and a necessary nonchalance concerning content, and I needed to jettison the mannered quality that I think my writing had developed. It had no place in this work.
Interviewer: That’s very clarifying—and telling, I think, about the emotional current running beneath Let Me Know When We Cross Into Nothing. What’s striking is how this shift in your work doesn’t seem like a disavowal of what came before, but more a rejection of the complacency that can settle over an established practice. You described becoming aware of a mannered quality in your work. It seems like dislocation—in form, in content, even in process—became a kind of antidote to self-replication.
(Part Two to come)