Evaluating alignment of behavioral dispositions in LLMs

Google Research introduces a systematic framework to evaluate behavioral dispositions in large language models. The approach transforms psychological questionnaires into situational judgment tests, quantifying how closely model outputs align with human social tendencies. The framework uses standardized measures like IRI for empathy and ERQ for emotion regulation.