This study introduces a systematic framework to evaluate how LLMs handle social attribution, specifically regarding responsibility and blame. Using both vignette-based and real-world scenarios, the research probes the internal mechanisms LLMs use to assign cause to social behaviors.
HOW THIS AFFECTS YOU
●
researcherYou can use this benchmark to evaluate the sociotechnical reasoning capabilities of your models.
●
policyThis helps assess the safety and fairness of LLMs in social decision-making roles.