Prevalence of Problematic Content
Comprehensive evaluation of AI systems in real-world deployment contexts
What this is
A research direction studying what problematic content looks like at scale, how often it appears, and how platform and model design choices shape exposure and harm.
What I focus on
- Defining “problematic” with clear operational criteria
- Measuring prevalence under realistic conditions and user profiles
- Understanding how system incentives and UI decisions shape outcomes
Methods
- Audits and measurement studies (platform or model behavior)
- Annotation schemes and reliability checks
- Comparative evaluation across conditions and policies
Output
- Prevalence estimates with transparent uncertainty
- Taxonomies of problematic content types
- Evidence-backed recommendations for safer deployment