A writer whose AI-assisted essay went viral reflects on testing GPT-3 early, Sudowrite's novel generator, and whether AI is good for writers and literature
Despite my success with AI-generated stories, I'm not sure they are good for writers—or writing itself. Mastodon: @AthenaHelivoy@mastodon.world . X: @englishoer , @srhm_ca , @dan_jeffries1 , @aldaily , @notoriousrel , @vauhinivara , @porcupinesquill , and @chameauleon See also Mediagazer Mastodon: Athena Andreadis, PhD / @AthenaHelivoy@mastodon.world : The dangers & risks of #LLM (routinely misnamed & discussed as if it were true #AI) are only too obvious. A Wired article discusses how #ChatGPT-assisted #fiction would further customize & atomize reader experience. Since isolation & loneliness have strong negative impact on our well-being, this is actually scary. … X: @englishoer : This looks like an honest, nuanced exploration of AI text in fiction and creative nonfiction. Setting it aside to spend time with it. https://www.wired.com/... Sil Hamilton / @srhm_ca : I'm quoted in this month's @Wired! In @vauhinivara's great piece on AI co-writing, wherein she argues GPT-3.5 has “the witless efficiency of a stapler.” I agree. For LLMs, words are concepts, and concepts are words—RLHF is linguistic relativity in action. https://www.wired.com/... Daniel Jeffries / @dan_jeffries1 : Nuance is hard to come by in the debate of AI versus art but this story delivers. We might debate whether we should or shouldn't use AI + what it all means but tomorrow's writers won't think twice about it + wonder what all the fuss was about. https://www.wired.com/... @aldaily : An essay about grief went viral in The Believer and was then adapted for This American Life. Does it matter that it was written by AI? https://www.wired.com/... Arielle Bernstein / @notoriousrel : I love this essay by @vauhinivara: “I have no doubt that AI will become more powerful in the coming decades...In the meantime, writers will still be here, searching for the words to describe what it felt like to be human through it all...” https://www.wired.com/... Vauhini Vara / @vauhinivara : The most-read and most-admired thing I've ever written is an essay I co-wrote with AI — ChatGPT's predecessor, GPT-3 — called “Ghosts,” about the death of my sister, which went viral. Not just that: I believe the AI wrote the essay's best lines. (1/x) https://www.wired.com/... @porcupinesquill : This is a fascinating article on AI and #literature because it (importantly) comes from someone who has actually used AI software in their #literary work. https://www.wired.com/... Camille Bromley / @chameauleon : I loved talking to @vauhinivara about her conflicted relationship with AI writing for her new piece in @WIRED. I've always been secretly, maybe unreasonably, fascinated by AI generated stories. Vauhini's essay thinks through the good and bad of all of it. https://www.wired.com/... See also Mediagazer
Context & Ripple Effects
Vauhini Vara’s viral GPT-3 collaboration, later adapted for This American Life, provides a concrete literary case study rather than a purely technical evaluation of text models. Her testing of Sudowrite extends that inquiry from one-off experimentation to tools designed for ongoing creative work.
The debate arrives as [[a:839755|text generators were expanding the supply of AI-written books and articles while reducing some demand for human writing work]]. This account matters because it separates a tool’s ability to produce publishable-seeming text from its effect on authorship and literature.
First-order effects
- Vara’s account gives writers considering GPT-3 or Sudowrite a practitioner-led warning that successful AI-assisted output does not settle whether the process improves their work or role.
- Sudowrite and similar writing tools are evaluated less as novelty generators and more on whether their assistance preserves meaningful authorial judgment.
Second-order effects
- Publishers, editors, and writing clients face a sharper distinction between inexpensive generated draft material and the human creative work they are willing to commission, amid growing AI-generated publishing output.
- Writing-tool vendors have incentive to demonstrate that their products support distinctive authorial work rather than merely automate prose generation.
Third-order effects
- If AI-assisted writing becomes routine, literary markets may increasingly compete on provenance, editorial selection, and a writer’s contribution—not only on the availability of text.
- The story is an early instance of generative editorial debt: efficiency gains may create longer-term questions about originality, voice, and the value assigned to human authorship.
The trend: Generative writing is moving from experimental collaboration toward commercialization, forcing creators and publishers to define what human authorship adds.