Examining Human-Like Behaviors in LLMs: A Multi-Dimensional Analysis of Model Behaviors, User Factors, and System Prompts

Apple researchers have published an analysis examining how large language models display human-like behaviors such as emotional expression, relationship building, and boundary setting. This study identifies a need for empirical methods to measure these traits, as current systems increasingly mirror human interactions. By investigating the influence of system prompts and user factors on these behaviors, the findings aim to help developers better understand and control how models simulate social engagement during interactions.