LLM-generated survey data shows promise for population synthesis

Researchers have explored the use of synthetic data generated by zero-shot large language models (LLMs) for population synthesis. A study using GPT-4.1 and Gemini-2.5-Pro to create health survey data for Colorado and Mississippi showed that LLMs can produce geographically differentiated data. While the synthetic populations reproduced some state-level contrasts and census tract-level patterns, the performance was variable and not yet a replacement for real survey data. AI

IMPACT LLMs can generate geographically differentiated synthetic data, showing potential for supplementary use in population synthesis, though not yet a replacement for real survey data.

RANK_REASON Academic paper detailing a novel application of LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Behavioral Risk Factor Surveillance System (BRFSS)
• Colorado
• Gemini-2.5-Pro
• GPT-4.1
• Mississippi

• paper
• other

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