Response drift across frontier large language models

Mohammed Aledhari, Ali Aledhari, Fatimah Aledhari +2

All frontier large language models (LLMs) exhibit response drift -- producing outputs that deviate from expert-validated references -- yet the magnitude and structure of this drift…