Study: Readers Can't Tell AI Imitations From Real Writers, Sometimes Prefer AI
TuhinChakr · x · 2026-09-16
On Literary Hub's Fiction/Non/Fiction podcast, award-winning writers Vauhini Vara and Karan Mahajan discuss Vara's New Yorker essay "What If Readers Like AI-Generated Fiction?"
Vara describes scientist Tuhin Chakrabarty's research, which fine-tunes large language models on authors' entire oeuvres to produce better writing. When he showed the results to creative writing graduate students, they preferred AI imitations of writers like Junot Diaz, Sigrid Nunez, and Tony Tulathimutte — or could not tell the difference.
Vara and Mahajan also reflect on their decades-long friendship and familiarity with each other's writing, and muse on what it means when Vara convinced Chakrabarty to let her compete against a large language model.
More from AGI Musings
- AI's Jevons paradox of work: smart people are working more, not less — ayushtweetshere · 2026-09-17
- DeepMind Launches Institute and Working Paper on Economic Policy for AGI — soumitrashukla9 · 2026-09-17
- Median AI existential risk estimate hits 10%; pretraining down to 11% of compute — gleech · 2026-09-17
- Economist calls for boycott of top journal, blaming AI-fueled publishing rat race — soumitrashukla9 · 2026-09-17
- Tech CEOs' AI predictions are just "guys with domain knowledge talking" — thederbiedone · 2026-09-17
- Nicholas Thompson: AI extinction talk is overblown, but alignment research still deserves priority — GaryMarcus · 2026-09-17