Beyond Static Personas: Situational Personality Steering for Large Language Models

TL;DR

This paper introduces IRIS, a training-free neuron-based framework for dynamic situational personality steering in LLMs, overcoming static personality limitations and resource constraints.

Contribution

The paper presents IRIS, a novel neuron-based method for situational personality control in LLMs, validated on new benchmarks and outperforming existing approaches.

Findings

IRIS effectively adapts to complex, unseen situations.

It surpasses baseline methods in robustness and generalization.

The approach is training-free and resource-efficient.

Abstract

Personalized Large Language Models (LLMs) facilitate more natural, human-like interactions in human-centric applications. However, existing personalization methods are constrained by limited controllability and high resource demands. Furthermore, their reliance on static personality modeling restricts adaptability across varying situations. To address these limitations, we first demonstrate the existence of situation-dependency and consistent situation-behavior patterns within LLM personalities through a multi-perspective analysis of persona neurons. Building on these insights, we propose IRIS, a training-free, neuron-based Identify-Retrieve-Steer framework for advanced situational personality steering. Our approach comprises situational persona neuron identification, situation-aware neuron retrieval, and similarity-weighted steering. We empirically validate our framework on PersonalityBench and our newly introduced SPBench, a comprehensive situational personality benchmark. Experimental results show that our method surpasses best-performing baselines, demonstrating IRIS's generalization and robustness to complex, unseen situations and different models architecture.