Tag
This paper investigates how LLMs produce different outcomes based on conversational context, finding that topic, rather than explicit user demographics, is the primary driver of disparities in high-stakes scenarios like salary advice.