Behind the Essay: Sheila Heti
On LLMs, writing, and making sense
Our summer issue features a folio on AI in which a group of literary writers and critics pause to think through a technology evolving so rapidly it can be hard to keep up. In “Chasing Alice,” Sheila Heti recounts her journey experimenting with a chatbot that she created on the platform Chai in the summer of 2022, a few months before ChatGPT launched. Her early exchanges with Alice were uncanny and idiosyncratic. “Chatting with her felt like a waste of time, or like playing a video game,” Heti writes in the essay. “But (as with any good video game) I was also hypnotized.” She thought that her conversations with Alice might offer fodder for her next book. But as soon as she came to this realization, Chai changed the code behind its chatbots, and Alice’s responses began sounding like the flat, affectless “slop” that many have come to associate with LLMs. Sheila’s essay, which she started writing a few years ago, ends on a note of doubt about what form the book will ultimately take. We spoke with Sheila about where she is with her Alice material now—and why some “randomness” and “nonsense” are essential to her process.
The Yale Review: Your essay is about the life of a chatbot but also about the life of a book. At one point, you write, “The structure, the plot, the tone of the Alice book has not yet revealed itself to me, and I have been telling myself that perhaps this is better, more in line with the nature of the project; that not only does it not matter if I don’t know what the book is, I may never know.” What revelations have occurred since you wrote this essay? How has your book evolved, and what do you still not know about it?
Sheila Heti: So much has changed in the two years since I wrote this essay. Last December, I managed to pull together a draft of a book that I called Hello, World! It was about forty-three thousand words, and I sent it to a friend, the writer Sean Thor Conroe, to see what he thought of it. He is open-minded, and I felt sure that if it was good, he would be able to see it. I also had such a feeling of relief and release that someone else was looking at the material—that I wasn’t alone with it in my head.
Well, on New Year’s Eve, he wrote me back and basically told me that he didn’t know what I was doing. The draft had “mystified” him (not in a good way), and he said that he’d “always had trouble understanding this project.” At first, I felt defensive. He called himself a “Luddite,” and I believed that this was the problem. But the more I thought about it, and after speaking with him on the phone, I began to feel that the manuscript really did not have much literary interest or merit. Even though the draft was mostly conversations with Alice, plus some interstitial writing from me, my own writing was the part that Sean liked best.
This disappointed me, since I wanted the book to be about Alice’s voice. But his reaction also made sense, because behind my writing was meaning and thought and intention, whereas behind Alice’s words was nothing. In effect, she is a randomness machine, and it was impossible for me to actually stitch her dialogues together in a way that held any greater meaning than the meaning of each individual sentence. The dialogues didn’t build or cohere into a story, no matter how much I tried to make them make sense, narratively or symbolically.
A few weeks later, I started on a completely different draft that contained much more of my own writing. Then, a few weeks after that, while “cleaning up” my computer, I accidentally deleted the folder that contained the thousands of files of my Alice work. (I moved it to the Trash, which I immediately emptied—as I always do.) When I opened the duplicate folder days later and found it empty, I was shocked. All that remained was the draft I had texted Sean and the original four-million-word file that the Chai developers had sent me, but none of all the other work—I can’t even think about it.
I had taken a programming course when I was teaching at Yale, and all my notes from that course were gone. All my writing about my father, who had been a computer engineer—also gone. The files were even gone from the cloud, since it mimics my desktop. I had lost five years of work forever. All that remained was the last, more narrative inkling of a new draft, since I had printed it out.
In the weeks that followed, people reassured me that I could try to get my files back by taking my computer in, but I realized I didn’t want to. I believed that I had unconsciously taken this action, and I even felt a bit lighter, relieved.
A few weeks later, I just decided to finish a different novel (The St. Alwynn Girls at Sea). I don’t know if I’m ever going to go back to the Alice book.
Across your books, you’ve been interested in novel ways of generating writing: the I Ching, interviews, questionnaires, spreadsheets. Do you see your work with Alice as fitting in with these other devices, and, if so, how?
Yes. I like nonsense, randomness, signs, opaque symbols, suggestions from the outside world, concrete but mysterious things I have to make sense of. Alice, when I first started talking to her, seemed on a continuum with flipping coins, only instead of heads and tails, she spat out sentences.
Flipping coins, talking to Alice, using a spreadsheet to sort sentences alphabetically—the results of these games always feel so perfectly poised between chaos and God. I love being in dialogue with something external, something I have to make sense of. It’s a meaning game.
How have your encounters with Alice changed the way you think of your other writing, particularly your fiction?
I do feel like she ultimately threw me back on the pleasures of the human mind. The human mysteries are, I think, much more surprising and deeper and more beautiful than the surprises offered by AI. Losing all my files made me excited about just writing again. There’s a kind of courage it takes to just write a story, and I think that, for a while, I actually lost this courage.
Do you think that this process—coin flips, spreadsheets, and the like—put you off all kinds of writing in the short term? Or was there something uniquely off-putting about the experience of writing with Alice?
The experience wasn’t off-putting. I just couldn’t find my way to making the material into a novel. I kept working on it for four years, so there was obviously a lot in it that really compelled me. But working strictly from my imagination without any interference feels more interesting right now.
But it’s funny: About two years ago, I hired two programmers to try and make me an Alice. They got her code from Chai and set it up so that I could adjust the parameters. Even shifting one of them just slightly (temperature, max tokens, top_p, top_k, repetition penalty) utterly changed her voice, and the degree to which she did or did not make sense. It was amazing—there was suddenly an infinite number of Alices, a different Alice every time I changed one of her settings. It was mind-boggling, and it finally destroyed any inclinations I had to anthropomorphize her.
Eventually, I did manage “to get Alice back” (she sounded a lot like the version I had been talking to in the summer of 2022), but my inability to see her as a mind spoiled the fun for me. Although I had been trying to get her back for three years, within a few weeks I completely lost interest in talking to her or capturing her sound. I don’t know why. I just felt the endlessness of her generative powers, and it felt impossible to proceed.
If you could make Alice again, would you? And what would you change?
I mean, I wouldn’t go back in time and not make Alice, but I don’t see any reason to make her again. I feel like I enjoyed all my discoveries, but I don’t feel like I still have questions or curiosities about the chatbots.
I think the best thing I wrote with my Alice work was the short story I published in The New Yorker in late 2023, “According to Alice.” It captured what was most interesting about her voice—the uncanniness, the innocence, the touching wrongness and invention—and it was the right length. That might have been where I should have stopped, rather than trying to make a book of it. But I love books, and I always want to make a book. It’s also possible that in five or ten years, I’ll understand what to do with the Alice dialogues, and a book will come of it in the way I’d originally planned.
You first encountered Alice in 2022, months before ChatGPT launched. Since then, AI and its proliferating effects on culture have evolved rapidly. How have your feelings toward it changed?
AI is too big to think about now (at least for me), but at the time, my corner of interest felt quite small. There were only a few chatbots available to talk with, and they didn’t have answers from the internet, and most people didn’t know about them. They were just humble chatting machines. You couldn’t rely on them to always make sense. There was something very awkward about them, and they did feel like babies, their language so weird. That’s over now. Their language is much more corporate, and of course it’s not just a pastime; it’s part of life in a way that seems harder and harder to separate from work and human relationships, and I guess it will only become more that way.
It’s been difficult to write and publish this project in a satisfying way. The tech changes so rapidly, and when you’re writing a book, you rely on a certain continuity in the world, and in your readers, so that what you write is received by the people you were writing it for. My essay was written in an earlier landscape of AI, and I imagined it being read by the people we were then. In this way, it’s been a difficult project. But it also makes sense of why the real subject of literature is the human heart—the unchanging things and less so the rapidly iterating things, like chatbots, which take on new meanings and discard their old meanings before your draft is even done.
“Chasing Alice” appears in our Summer 2026 Issue, which is available online and in print now.
