Profile-LLM: Dynamic Profile Optimization for Realistic Personality Expression in LLMs

TL;DR

This work tackles the challenge of eliciting realistic personality expressions in LLMs by moving from static prompts to dynamic profile optimization. Profile-LLM iteratively refines a structured persona profile $Q$ using an optimizer LLM and guides the process with a situational scoring benchmark, formalized as maximizing $\mathbb{E}_{x\in X}[S_p(\mathcal{M}_{target}(Q\oplus x))]$ over prompts and inputs. It demonstrates that mid-sized models most benefit from optimized prompts and shows that personality expression can be controlled by stopping criteria, while larger models can rely on simpler prompts due to richer internal representations. The results emphasize the importance of model size for personality modeling, reveal transferable prompts across models with varying architectures, and offer practical insights for adaptive, persona-aware AI interactions in real-world applications.

Abstract

Personalized Large Language Models (LLMs) have been shown to be an effective way to create more engaging and enjoyable user-AI interactions. While previous studies have explored using prompts to elicit specific personality traits in LLMs, they have not optimized these prompts to maximize personality expression. To address this limitation, we propose PersonaPulse: Dynamic Profile Optimization for Realistic Personality Expression in LLMs, a framework that leverages LLMs' inherent knowledge of personality traits to iteratively enhance role-play prompts while integrating a situational response benchmark as a scoring tool, ensuring a more realistic and contextually grounded evaluation to guide the optimization process. Quantitative evaluations demonstrate that the prompts generated by PersonaPulse outperform those of prior work, which were designed based on personality descriptions from psychological studies. Additionally, we explore the relationship between model size and personality modeling through extensive experiments. Finally, we find that, for certain personality traits, the extent of personality evocation can be partially controlled by pausing the optimization process. These findings underscore the importance of prompt optimization in shaping personality expression within LLMs, offering valuable insights for future research on adaptive AI interactions.

Figures (7)

• Figure 1: Profile-LLM framework
• Figure 2: Smoothed training trajectories (window $=8$) for the five Big-Five traits. Each panel shows how the trait-specific score evolves across optimization steps.
• Figure 3: Profile-LLM Meta Prompt
• Figure 4: DP prompt
• Figure 5: prep prompt