← 研踪 PaperTrailarXiv:2609.25010 · v2
Do Synthetic Personas Predict Real Audience Response? A Headline-Ranking Study with a Generic-Reader Baseline
首次提交 2026-07-28 01:03
版本更新 2026-11-03 09:11
雷达观测 2026-11-04 23:40
摘要
Do demographic personas improve an LLM's prediction of audience engagement? We compare a fixed ten-persona panel with a generic-reader baseline on Upworthy headlines, with supplementary news and Reddit comparisons. The primary experiment collects Gemini 3.1 Flash-Lite and GPT-4.1 observations on the same 190-package exploratory cohort, selected with simultaneous winner screening. It allocates thirty scores per arm and variant within model, retains exact raw responses, and reports metric-specific valid and paired counts. Complete-package coverage is 188 for Gemini and 188 for GPT-4.1. The baseline-minus-panel Kendall $τ_b$ gap is +0.222 (CI [+0.061, +0.385]) for Gemini and +0.257 (CI [+0.119, +0.391]) for GPT-4.1, on respectively 136 and 188 valid rank pairs. Expected top-1 is 39.1% versus 54.8% for Gemini and 42.0% versus 61.7% for GPT-4.1 (panel versus baseline). These intervals condition on the selected cohort and recorded generations and do not adjust for dependence between related experiments. The Upworthy archive documents a randomization-risk period. Excluding that period in a post-hoc sensitivity retains positive expected top-1 and lift gaps in both configurations, while Gemini tau-b and Spearman intervals include zero. Supplementary Reddit top-1 and captured lift favor the panel; these aggregate comparisons use unreconciled collection protocols and do not isolate dataset effects. A raw-replay artifact supports offline reproduction of the primary experiment. The results concern this panel, cohort, and protocol; they are not a held-out evaluation or evidence of universal persona harm. Provider request protocols differ across models, and no causal mechanism is isolated.
作者备注：14 pages, 3 figures, 11 tables
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-03 09:11 · 最新观测版本
Do Synthetic Personas Predict Real Audience Response? A Headline-Ranking Study with a Generic-Reader Baseline
Do demographic personas improve an LLM's prediction of audience engagement? We compare a fixed ten-persona panel with a generic-reader baseline on Upworthy headlines, with supplementary news and Reddit comparisons. The primary experiment collects Gemini 3.1 Flash-Lite and GPT-4.1 observations on the same 190-package exploratory cohort, selected with simultaneous winner screening. It allocates thirty scores per arm and variant within model, retains exact raw responses, and reports metric-specific valid and paired counts. Complete-package coverage is 188 for Gemini and 188 for GPT-4.1. The baseline-minus-panel Kendall $τ_b$ gap is +0.222 (CI [+0.061, +0.385]) for Gemini and +0.257 (CI [+0.119, +0.391]) for GPT-4.1, on respectively 136 and 188 valid rank pairs. Expected top-1 is 39.1% versus 54.8% for Gemini and 42.0% versus 61.7% for GPT-4.1 (panel versus baseline). These intervals condition on the selected cohort and recorded generations and do not adjust for dependence between related experiments. The Upworthy archive documents a randomization-risk period. Excluding that period in a post-hoc sensitivity retains positive expected top-1 and lift gaps in both configurations, while Gemini tau-b and Spearman intervals include zero. Supplementary Reddit top-1 and captured lift favor the panel; these aggregate comparisons use unreconciled collection protocols and do not isolate dataset effects. A raw-replay artifact supports offline reproduction of the primary experiment. The results concern this panel, cohort, and protocol; they are not a held-out evaluation or evidence of universal persona harm. Provider request protocols differ across models, and no causal mechanism is isolated.
查看此版本解读 →查看此版本原文 ↗