home

beach reads

the stories [or jump to the story behind the stories]

known bug: some of the characters are mis-parsing

inspiring author title topics
Chimamanda Ngozi Adichie Midnight Tides skinny-dipping, beer, pop songs, sweat, and regret
Jane Austen A Most Uncomfortable Day at the Seaside swimsuits, constipation, high school, sand, and self-consciousness
Jane Austen A Midsummer's Awakening the Midwest, lust, ice cream, sweat, and ennui
William Faulkner Tides of Memory sweat, cookouts, sand, ice cream, and lust
Ernest Hemingway The Lake's Decision kissing, the Midwest, ambivalence, swimsuits, and privilege
James Joyce Tides of Ambivalence kissing, ice cream, ambivalence, sunscreen, and bug spray
Franz Kafka The Sweat Rebellion the Midwest, sweat, high school, ennui, and lust
Gabriel Garcia Marquez The Perfume of Forgotten Summers bug spray, constipation, cookouts, sunscreen, and ennui
Cormac McCarthy Tides of Regret beer, swimsuits, ennui, sand, and regret
Toni Morrison Saltwater Memories sunscreen, kissing, ambivalence, cookouts, and sand
Toni Morrison Sunscreen and Second Chances kissing, sunscreen, the Midwest, swimsuits, and regret
Haruki Murakami Midnight Swim in the Sea of Uncertainty skinny-dipping, bug spray, ambivalence, constipation, and ice cream
Vladimir Nabokov Sunburnt Desires: A Suburban Odyssey self-consciousness, lust, cookouts, sunscreen, and sweat
Vladimir Nabokov Heliopause pop songs, sunscreen, high school, sweat, and beer
J.K. Rowling The Sunscreen Sorcerer's Summertime Surprise sunscreen, the Midwest, sand, cookouts, and self-consciousness
Salman Rushdie Sunscreen for the Soul ennui, beer, ambivalence, self-consciousness, and sunscreen
Salman Rushdie Melting Rhythms on Saffron Sands pop songs, sunscreen, ice cream, lust, and ennui
Salman Rushdie The Sizzling Symphony of Summer's End high school, cookouts, privilege, pop songs, and swimsuits
Curtis Sittenfeld The Taste of Maybe the Midwest, self-consciousness, kissing, ambivalence, and regret
Curtis Sittenfeld Sunburnt Relevations bug spray, swimsuits, self-consciousness, ice cream, and constipation
Curtis Sittenfeld Skinny-Dipping at Sunset skinny-dipping, self-consciousness, sunscreen, kissing, and ennui
Curtis Sittenfeld Moonlit Inhibitions self-consciousness, the Midwest, kissing, beer, and skinny-dipping
Curtis Sittenfeld Mint Chip Rebellion ice cream, privilege, ennui, lust, and swimsuits
Zadie Smith The Weight of Paradise regret, privilege, self-consciousness, ambivalence, and ennui
Zadie Smith Fireflies and Footprints the Midwest, sand, cookouts, ice cream, and bug spray
Zadie Smith Melting Points ice cream, sand, constipation, the Midwest, and sunscreen
David Foster Wallace Midwest Impaction: A Gastrointestinal Odyssey ennui, constipation, pop songs, the Midwest, and regret
David Foster Wallace Saltwater Baptism: A Nocturnal Revelation in Four-Part Disharmony skinny-dipping, sand, beer, sweat, and kissing
Virginia Woolf Grains of Ambivalence ambivalence, sand, beer, privilege, and bug spray


the story behind the stories [or jump to the stories]

In her NYT guest essay ChatGPT vs. Me: Who Will Write a Better Beach Read?, author Curtis Sittenfeld proposes a head-to-head competition against ChatGPT ("in its current iteration"), to see who can write a better "beach read" vignette in Sittenfeld's own style. She asks her readers to vote on topics from a list that she provides,1 and then explains the parameters.

Both stories will be the same length (1,000 words), and both will incorporate the same five prompts. ChatGPT, which will be told to write in my style, will complete its story in a few seconds; I'll complete mine in a few weeks.
She explains her motivations as follows.
So it may go without saying that in this contest, I'm rooting for myself; I'm Team Human, and I'm hoping to honorably defend my species. But as with the steel-driving John Henry and the chess-playing Garry Kasparov before me, the outcome is unknown in advance. Maybe fiction writers' jobs are in danger, or maybe there's some ineffable quality, similar to the transcendent quality of a beach read, that still separates a story told by a person from a story told by a computer.
I appreciate her proposal, and am eager to learn whether readers can even tell the difference between Sittenfeld's and ChatGPT's stories. However, I disagree with her primary conclusion: Sittenfeld beating even just today's top AI chatbots (to say nothing of tomorrow's!) would give little evidence of such an ineffable quality of human writing. As is often touted, in its current form ChatGPT literally just "predicts the next word" (which is what allows it to write a 1,000-word story in just a few seconds). Said differently, the innate abilities of ChatGPT are deeply "hobbled", inasmuch as it is required to construct its responses one word at a time, without any possibility of higher-level ideation or revision.2 It does not require any new technological developments whatsoever for ChatGPT to become far better at writing: one only need appropriately structure its prompts to simulate such higher-order thinking.

To illustrate this point and enable a slightly fairer competition,3 I wrote a very simple Python script that prompts an AI chatbot to "think" before writing a 1,000-word story in the style of a chosen author; samples are linked above. Specifically, the chatbot is prompted first to describe the characteristics of the author's style and then to outline the story, and only then does it proceed to write the story itself. Here is an example of the complete conversation between the chatbot and the Python program (which plays the role of the chatbot user). This sequence of prompts seems to allow the chatbot to write somewhat more coherently, although the effect becomes much more pronounced for much longer stories. As a proof of concept, I generated two longer stories with the same prompting pattern: a ~5,000-word one here in the style of Gabriel Garcia Marquez and a ~10,000-word one here in the style of Franz Kafka.

To be clear, I'm rooting for Team Human too! But I think it's important to understand what exactly is being tested, and in what sorts of ways we humans might still hold an advantage. It's certainly not on speed or productivity: even with this additional prompting, on average each ~1,000-word story takes only about 57 seconds and $0.09 to write (using Claude Sonnet 3.5).4 On the other hand, it seems that there is a collective sense that human-generated art is somehow more important than AI-generated art, and I think the easiest and truest justification for this is that an essential purpose of art is to make us feel connected to artists and their lives, experiences, and intentions.

1 The full list of topics is: sunscreen, kissing, pop songs, regret, sand, ennui, cookouts, ambivalence, self-consciousness, ice cream, privilege, beer, high school, bug spray, swimsuits, skinny-dipping, constipation, the Midwest, sweat, lust.

2 The notion of the innate abilities of an AI being artificially "hobbled" is explored in Leopold Aschenbrenner's Situational Awareness essays (see here in particular).

3 Of course, another way of getting a fairer head-to-head competition would be to require the human competitor to narrate their entire story from start to finish as a single stream of consciousness!

4 On the other hand, with this prompting structure, the cost (which is charged per-token) grows roughly proportionally to the square of the length, and the time required also grows super-linearly.