A two-part research platform: RogueGPT generates controlled AI news stimuli. JudgeGPT measures whether humans can detect them. The answer might surprise you.
Controlled AI news generation across 10 models, 4 languages, 5 journalistic styles.
Full provenance metadata persisted with every fragment. Reproducible filtering by any variable.
Humans judge authenticity in a controlled experiment. Confidence, reasoning, and demographics captured.
Statistical analysis of human susceptibility across models, languages, and demographics.
The foundational literature survey that motivated this research program.
Human evaluation platform. Streamlit-based experiment collecting authenticity judgments, confidence levels, and reasoning.
Controlled stimulus generator. Produces AI news fragments with explicit provenance across 10 models and 6 providers.
Take the survey. Can you tell which news was written by AI? Most people can't.