DPN-LE: Dual Personality Neuron Localization and Editing for Large Language Models

2026-04-30

AI Summary

This work proposes a novel sparse linear intervention paradigm for personality editing in large language models, addressing the limitations of existing methods that require modifying numerous neurons—often impairing general capabilities and struggling to precisely identify personality-relevant units. By comparing MLP activations between samples exhibiting high and low levels of target personality traits, the approach leverages a dual criterion combining Cohen’s d effect size and activation magnitude to select mutually exclusive subsets of neurons. Remarkably, editing only ~0.5% of neurons enables precise personality modulation. The method reveals, for the first time, the dual functionality and mutually exclusive representational patterns of personality-related neurons. Evaluated on LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct, it significantly outperforms current techniques using merely 1,000 sample pairs while better preserving the model’s reasoning abilities during efficient personality control.

Abstract

With the widespread adoption of large language models (LLMs), understanding their personality representation mechanisms has become critical. As a novel paradigm in Personality Editing, most existing methods employ neuron-editing to locate and modify LLM neurons, requiring changes to numerous neurons and leading to significant performance degradation. This raises a fundamental question: Are all modified neurons directly related to personality representation? In this work, we investigate and quantify this specificity through assessments of general capability impact and representation-level patterns. We find that: 1) Current methods can change personalities but reduce overall performance. 2) Neurons are multifunctional, connecting personality traits and general knowledge. 3) Opposing personality traits demonstrate distinctly mutually exclusive representation patterns. Motivated by these findings, we propose DPN-LE (Dual Personality Neuron Localization and Editing), which identifies personality-specific neurons by contrasting MLP activations between high-trait and low-trait samples. DPN-LE constructs layer-wise steering vectors and applies dual-criterion filtering based on Cohen's $d$ effect size and activation magnitude to isolate mutually exclusive neuron subsets. Sparse linear intervention on these neurons enables precise personality control at inference time. Using only 1,000 contrastive sample pairs per trait, DPN-LE intervenes on $\sim$0.5\% of neurons while achieving competitive personality control and substantially better capability preservation across reasoning tasks. Experiments on LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct demonstrate the effectiveness and generalizability of our approach.

Problem

• personality editing
• neuron localization
• large language models
• representation specificity
• multifunctional neurons

Innovation

• personality editing
• neuron localization
• sparse intervention
• mutually exclusive representations
• large language models