Saved in:
Bibliographic Details
Main Authors: Potertì, Daniele, Seveso, Andrea, Mercorio, Fabio
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.12055
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912234094985216
author Potertì, Daniele
Seveso, Andrea
Mercorio, Fabio
author_facet Potertì, Daniele
Seveso, Andrea
Mercorio, Fabio
contents The influence of personas on Large Language Models (LLMs) has been widely studied, yet their direct impact on performance remains uncertain. This work explores a novel approach to guiding LLM behaviour through role vectors, an alternative to persona-based prompting. We construct 29 role vectors derived from model activations and evaluate their impact on benchmark performance across multiple domains. Our analysis investigates whether these vectors can effectively steer models toward domain-specific expertise. We measure two key interventions: (i) activation addition, which reinforces role-specific directions, and (ii) directional ablation, which removes them. Results on well-established benchmarks indicate that role vectors do, in fact, influence model behaviour, improving task performance in relevant domains while marginally affecting unrelated tasks. This, in turn, suggests that manipulating internal model representations has a greater impact on outcomes than persona-based prompting.
format Preprint
id arxiv_https___arxiv_org_abs_2502_12055
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Designing Role Vectors to Improve LLM Inference Behaviour
Potertì, Daniele
Seveso, Andrea
Mercorio, Fabio
Computation and Language
The influence of personas on Large Language Models (LLMs) has been widely studied, yet their direct impact on performance remains uncertain. This work explores a novel approach to guiding LLM behaviour through role vectors, an alternative to persona-based prompting. We construct 29 role vectors derived from model activations and evaluate their impact on benchmark performance across multiple domains. Our analysis investigates whether these vectors can effectively steer models toward domain-specific expertise. We measure two key interventions: (i) activation addition, which reinforces role-specific directions, and (ii) directional ablation, which removes them. Results on well-established benchmarks indicate that role vectors do, in fact, influence model behaviour, improving task performance in relevant domains while marginally affecting unrelated tasks. This, in turn, suggests that manipulating internal model representations has a greater impact on outcomes than persona-based prompting.
title Designing Role Vectors to Improve LLM Inference Behaviour
topic Computation and Language
url https://arxiv.org/abs/2502.12055