The paper "Persona In-Context Learning (PICLe): Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning" introduces a novel framework for eliciting diverse behaviors and personas from large language models (LLMs). The authors formalize the persona elicitation task, aiming to customize LLM behaviors to align with a target persona. PICLe is grounded in Bayesian inference and introduces a new ICL example selection criterion based on likelihood ratio, designed to optimize the model's elicitation of a specific target persona. The effectiveness of PICLe is demonstrated through extensive comparisons against baseline methods across three contemporary LLMs: Llama-2, Vicuna, and GPT-J. The paper also provides a comprehensive evaluation of PICLe's performance and discusses its advantages, limitations, and potential applications.The paper "Persona In-Context Learning (PICLe): Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning" introduces a novel framework for eliciting diverse behaviors and personas from large language models (LLMs). The authors formalize the persona elicitation task, aiming to customize LLM behaviors to align with a target persona. PICLe is grounded in Bayesian inference and introduces a new ICL example selection criterion based on likelihood ratio, designed to optimize the model's elicitation of a specific target persona. The effectiveness of PICLe is demonstrated through extensive comparisons against baseline methods across three contemporary LLMs: Llama-2, Vicuna, and GPT-J. The paper also provides a comprehensive evaluation of PICLe's performance and discusses its advantages, limitations, and potential applications.