Mechanistic Personality Analysis of LLMs Steering Personality via Latent Feature Interventions

2026-06-27

AI Summary

This work addresses the challenge of controllably modulating personality traits in large language models without compromising their language modeling performance. It introduces a novel approach that integrates sparse autoencoders (SAEs) with contrastive activation analysis to identify interpretable feature directions in the residual stream corresponding to the OCEAN personality dimensions. By applying small-magnitude additive interventions along these directions, the method enables targeted personality modulation. A grid-search-optimized linear weighting heuristic is employed to balance intervention strength and model fidelity. Experimental results demonstrate that this technique significantly enhances desired personality traits while preserving near-baseline performance on standard benchmark tasks, thereby establishing the feasibility and efficacy of mechanistic, fine-grained personality control in large language models.

Abstract

Large Language Models (LLMs) have demonstrated the ability to simulate human-like OCEAN personality traits in generated text. Previous efforts have focused on prompt engineering or fine-tuning to shape LLM personality. In this work, we propose a mechanistic interpretability approach that directly intervenes on the model's latent features. Our method identifies latent directions in the residual stream corresponding to a target OCEAN trait using sparse autoencoders (SAEs) and contrastive activation analysis. We formalize an additive steering vector in activation space and demonstrate how applying a small additive shift to the hidden states enhances the target trait while preserving overall language modeling performance. To determine the optimal combination of feature shifts, we explore a linear weighting heuristic with grid search optimization that balances personality expression with task performance. Our approach shows promise in controllably steering personality traits at the mechanistic level while maintaining high performance on standard benchmarks.

Problem

• personality steering
• Large Language Models
• OCEAN traits
• latent features
• mechanistic interpretability

Innovation

• mechanistic interpretability
• latent feature intervention
• personality steering
• sparse autoencoders
• activation space