Linear Personality Probing and Steering in LLMs: A Big Five Study

TL;DR

This paper explores using linear directions aligned with the Big Five personality traits to probe and steer large language models' behavior, demonstrating effective personality detection but limited control in open-ended tasks.

Contribution

It introduces a method for using linear directions in activation space aligned with personality traits for probing and steering LLMs, providing a cost-effective alternative to existing approaches.

Findings

Linear directions effectively probe personality traits in LLMs.

Steering capabilities are context-dependent, working well in forced-choice but limited in open-ended tasks.

Trait-aligned directions can detect personality with high accuracy.

Abstract

Large language models (LLMs) exhibit distinct and consistent personalities that greatly impact trust and engagement. While this means that personality frameworks would be highly valuable tools to characterize and control LLMs' behavior, current approaches remain either costly (post-training) or brittle (prompt engineering). Probing and steering via linear directions has recently emerged as a cheap and efficient alternative. In this paper, we investigate whether linear directions aligned with the Big Five personality traits can be used for probing and steering model behavior. Using Llama 3.3 70B, we generate descriptions of 406 fictional characters and their Big Five trait scores. We then prompt the model with these descriptions and questions from the Alpaca questionnaire, allowing us to sample hidden activations that vary along personality traits in known, quantifiable ways. Using linear regression, we learn a set of per-layer directions in activation space, and test their effectiveness for probing and steering model behavior. Our results suggest that linear directions aligned with trait-scores are effective probes for personality detection, while their steering capabilities strongly depend on context, producing reliable effects in forced-choice tasks but limited influence in open-ended generation or when additional context is present in the prompt.