The Agent Framework for Shaping Preference and Personality (AFSPP) is introduced as a new approach to shape personality and preference in Large Language Models (LLMs). It explores the impact of social networks and subjective consciousness on LLM-based agents’ preference and personality formation. AFSPP has successfully replicated several key findings from human personality experiments. This framework can enhance the efficiency and scope of psychological experiments and provide valuable insights for Trustworthy Artificial Intelligence research. It helps in formulating strategies to prevent undesirable preference and personality development.
Publication date: 8 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.02870
