PICLe: Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning
This work addresses the challenge of eliciting specific personality traits from LLMs for applications requiring tailored behavioral outputs, representing an incremental advancement in persona customization methods.
The paper tackles the problem of customizing large language model behaviors to align with a target persona by formalizing the persona elicitation task and proposing PICLe, a novel framework based on Bayesian inference with a likelihood ratio selection criterion, which demonstrates effectiveness through comparisons across three LLMs.
Large Language Models (LLMs) are trained on massive text corpora, which are encoded with diverse personality traits. This triggers an interesting goal of eliciting a desired personality trait from the LLM, and probing its behavioral preferences. Accordingly, we formalize the persona elicitation task, aiming to customize LLM behaviors to align with a target persona. We present Persona In-Context Learning (PICLe), a novel persona elicitation framework grounded in Bayesian inference. At the core, PICLe introduces a new ICL example selection criterion based on likelihood ratio, which is designed to optimally guide the model in eliciting a specific target persona. We demonstrate the effectiveness of PICLe through extensive comparisons against baseline methods across three contemporary LLMs. Code is available at https://github.com/deeplearning-wisc/picle.