Is Your Neighborhood Safe? Place-based Stigma in Large Language Models' Urban Safety Judgments

Explain Like I'm Normal

Researchers found that most AI models can't judge neighborhood safety using coordinates alone, relying instead on neighborhood names to make predictions. This creates a feedback loop where models parrot historical stigmas rather than actual crime data, showing a lack of spatial reasoning at the 'frontier' level. Only the largest models showed any ability to understand geographic location without the name prompt.