New LLM framework injects social diversity into opinion simulations

Researchers have developed a new framework called Parametric Social Identity Injection (PSII) to address the issue of diversity collapse in large language models used for public opinion simulation. Current LLM simulations often produce overly homogeneous responses, failing to capture social diversity. PSII injects explicit demographic attributes and value orientations directly into the LLM's intermediate hidden states, enabling fine-grained control over identity representation. Experiments using the World Values Survey demonstrated that PSII significantly improves the fidelity and diversity of simulated public opinion data compared to real-world survey results. AI

IMPACT Enhances the ability of LLMs to simulate diverse public opinion, potentially improving the accuracy and representativeness of AI-driven social science research.

RANK_REASON Academic paper detailing a new method for LLM agent simulation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Hexi Wang
• Large language models
• Parametric Social Identity Injection
• World Values Survey

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →