RESEARCH · RESEARCH · #1324
Demographic Pluralism: inference-time framework for modeling pluralistic human preferences
The paper introduces Demographic Pluralism, an inference-time method that estimates population-level opinion distributions without requiring opinion-distribution training data or task-specific fine-tuning by generating multiple perspectives within demographically grounded groups. Evaluated across four backbone LLMs on GlobalOpinionQA and VITAL, it reduces Jensen–Shannon distance by 8.4%–26.4% versus a Modular Pluralism baseline, with equal-weighted aggregation outperforming weighted variants and group-level error rising with group weight.
KEY POINTS
- The paper introduces Demographic Pluralism, an inference-time method that estimates population-level opinion distributions without requiring opinion-distribution training data or task-specific fine-tuning by generating multiple perspectives within demographically grounded groups.
- Evaluated across four backbone LLMs on GlobalOpinionQA and VITAL, it reduces Jensen–Shannon distance by 8.4%–26.4% versus a Modular Pluralism baseline, with equal-weighted aggregation outperforming weighted variants and group-level error rising with group weight.
- This provides a practical, no-fine-tuning way for LLMs to represent within-group preference variation, which matters for culturally sensitive alignment and fairness.
WHY IT MATTERS
This provides a practical, no-fine-tuning way for LLMs to represent within-group preference variation, which matters for culturally sensitive alignment and fairness.