PICLe: Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning

TL;DR

This paper introduces PICLe, a Bayesian inference-based framework for eliciting specific personality traits from large language models through optimized in-context example selection.

Contribution

It proposes a novel persona elicitation method using likelihood ratio-based example selection, improving the alignment of LLM behaviors with target personas.

Findings

PICLe outperforms baseline methods in eliciting desired personas

Effective across multiple large language models

Demonstrates significant improvement in behavioral alignment

Abstract

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.