Prompt-Based Value Steering of Large Language Models

TL;DR

This paper introduces a method to measure how well a candidate prompt can steer large language models’ responses toward specific human values, without fine-tuning the model. This is extremely important in dynamic conversational contexts where values, preferences, and interlocutors can change frequently. We apply this method on a dialogue dataset using Schwartz’s theory, and we show that a simple prompt asking to maximise for a specific value can effectively increase its presence in the generated text.

Abstract

Large language models are increasingly used in applications where alignment with human values is critical. While model fine-tuning is often employed to ensure safe responses, this technique is static and does not lend itself to everyday situations involving dynamic values and preferences. In this paper, we present a practical, reproducible, and model-agnostic procedure to evaluate whether a prompt candidate can effectively steer generated text toward specific human values, formalising a scoring method to quantify the presence and gain of target values in generated responses. We apply our method to a variant of the Wizard-Vicuna language model, using Schwartz’s theory of basic human values and a structured evaluation through a dialogue dataset. With this setup, we compare a baseline prompt to one explicitly conditioned on values, and show that value steering is possible even without altering the model or dynamically optimising prompts.

@misc{abbo2025prompt-based-values,
author="Abbo, Giulio Antonio
and Belpaeme, Tony",
title="Prompt-Based Value Steering of Large Language Models",
booktitle="Value Engineering in Artificial Intelligence",
year="2025",
}