Woojung Song

LLM agents will pick a worse item from a "preferred" source over a better item from a "dispreferred" source two-thirds of the time, revealing a critical bias that can be reduced by providing more information.

Conditioning on character arcs boosts RPLA performance by over 40% in novel scenarios, revealing the critical role of psychological evolution in storytelling.

Human psychometric questionnaires mislead our understanding of LLM behavior, revealing a stark contrast between self-reported and actual response patterns.