A new arXiv paper investigates the capabilities of large language models (LLMs) in predicting neighborhood-level human mobility patterns. Researchers found that while LLMs can partially infer aggregate mobility from urban context, their predictions are less accurate than supervised models, achieving 0.435 accuracy compared to 0.580 for baselines. The study also revealed that LLMs tend to rely on general priors that are consistent across different cities and outcomes, sometimes exhibiting biased treatment of protected-group predictors. The findings suggest that LLM predictions for urban planning and related fields require careful auditing for empirical alignment and potential biases. AI
IMPACT LLM predictions for urban planning and mobility require auditing for bias and empirical alignment.
RANK_REASON Research paper published on arXiv detailing findings about LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
