GitHub Repository

Business insights on AI scaling inspired by biological principles (based on PNAS 2007 research).

0 starsPython

3,225 trials of LLMs guessing author age – confident and wrong

Name: 3,225 trials of LLMs guessing author age – confident and wrong
Availability: InStock
Author: BrianneLee011

by BrianneLee011·Jul 15, 2026·1 point·0 comments

AI Analysis

●●SolidDark Horse

3,225 trials prove frontier models confidently misattribute teen writing to adults.

Strengths

•Uses ecologically real, consented handwritten text instead of scraped social media data.
•Measures model behavior after revealing ground truth, capturing confession and blame patterns.
•Isolates specific context framing effects that trigger or reduce age attribution errors.

Weaknesses

•Sample size of three authors limits the generalizability of the specific findings.
•More of a behavioral case study than a new tool or dataset for the community.