Study: Conversational context biases LLM answers in high-stakes scenarios

Arxiv CS.CL (NLP/language models)

Even one conversation history can cause outcome differences between users in legal, medical, and financial advice, finds a new paper. The study shows conversational topics act as proxies for sociodemographics, influencing LLM answers.

llm

Arxiv CS.CL (NLP/language models)·Vera NeplenbroekGabriele SartiArianna BisazzaRaquel Fern\'andez

8 days ago