Author Style Emulation vs MFA Writers - Writing Quality (College-Educated Readers): leaderboard
Chakrabarty, Ginsburg and Dhillon (2025) compare in-context imitations of 50 award-winning authors with excerpts by MFA-trained writers given the same prompt: style descriptions, 20 excerpts by the author and a content description. GPT-4o, Claude 3.5 Sonnet and Gemini 1.5 Pro wrote 50 excerpts each, and readers compared each AI excerpt with a human one in blind pairs. The score is the paper's GLM-predicted probability (Table 10) that college-educated general readers from Prolific (19 per pair) chose the AI excerpt as the one that was better written; 0.5 means no preference. The fine-tuned GPT-4o condition used a separate model per author and is not included. Probability (0-1); higher is better.
Metric: Predicted Probability AI Excerpt Chosen. Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 0.62 | #357 |
| 2 | Claude 3.5 Sonnet | 0.6 | #362 |
| 3 | Gemini 1.5 Pro | 0.5 | #454 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=author-style-emulation-vs-mfa-writers-writing-quality-college-educated-readers · How It Works · Data refreshed daily, snapshot 2026-11-05.