A new study published on arXiv investigates how large language models (LLMs) judge the safety of urban neighborhoods. The research found that LLMs rely heavily on neighborhood names, which carry demographic stereotypes, rather than objective geographical data or crime statistics. Models showed a tendency to lower safety ratings for neighborhoods with higher proportions of Black residents in Chicago and Hispanic residents in Los Angeles, even when controlling for crime rates. This suggests that LLMs may perpetuate place-based stigma, impacting their use in decision-support roles for urban planning and safety. AI
IMPACT LLM outputs may perpetuate demographic stereotypes, impacting their use in safety and urban planning decisions.
RANK_REASON Research paper published on arXiv detailing findings about LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]
- American Community Survey
- arXiv
- Black
- Chicago
- Hispanics
- Hugging Face
- large-language models
- Los Angeles
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
