Tag
This paper introduces Polistemics, a theory-grounded benchmark for evaluating how LLMs mediate political information in elections, and finds that while aggregate scores look good, models systematically fail under ambiguous or contradictory information.