Role Playing
Role-playing, using artificial intelligence to simulate characters and interactions, is a rapidly evolving field focused on improving the realism, consistency, and safety of AI agents in conversational settings. Current research emphasizes developing robust evaluation benchmarks, mitigating biases in model outputs, and enhancing the ability of large language models (LLMs) to maintain character consistency and appropriately refuse inappropriate requests, often employing techniques like representation editing and narrative context generation. This work has implications for various applications, including interactive entertainment, personalized assistants, and sociological research, by advancing the development of more engaging and ethically sound AI systems.
Papers
Tell Me What You Don't Know: Enhancing Refusal Capabilities of Role-Playing Agents via Representation Space Analysis and Editing
Wenhao Liu, Siyu An, Junru Lu, Muling Wu, Tianlong Li, Xiaohua Wang, Xiaoqing Zheng, Di Yin, Xing Sun, Xuanjing Huang
RoleBreak: Character Hallucination as a Jailbreak Attack in Role-Playing Systems
Yihong Tang, Bo Wang, Xu Wang, Dongming Zhao, Jing Liu, Jijun Zhang, Ruifang He, Yuexian Hou