AI Engineering
Typicality Bias
- Definition
- The systematic human preference for 'typical' texts over unusual ones – a well-documented phenomenon in cognitive psychology. Measured at α = 0.57±0.07 in LLM alignment data. The main cause of Mode Collapse, as RLHF/DPO amplify this bias.
- Category
- AI Engineering
Deep Dive: Typicality Bias
The systematic human preference for 'typical' texts over unusual ones – a well-documented phenomenon in cognitive psychology. Measured at α = 0.57±0.07 in LLM alignment data. The main cause of Mode Collapse, as RLHF/DPO amplify this bias.
Business Value & ROI
Why it matters for 2026
Explains why AI outputs often feel 'generic' or 'safe'. Understanding Typicality Bias helps you design prompts and systems that encourage more diverse, creative outputs.
Typicality Bias is why your AI sounds like everyone else's. We design systems that consciously counteract this bias when originality is valued over safety.
Production-Ready Guardrails