Toward Stable Value Alignment: Introducing Independent Modules for Consistent Value Guidance

TL;DR

This paper introduces SVGT, an independent module for LLMs that maintains stable, normative value representations separately from the backbone, improving alignment and safety without disrupting model internal processes.

Contribution

The paper proposes a novel architecture with independent value modeling and explicit guidance, enhancing value stability and safety in large language models.

Findings

SVGT reduces harmful scores by over 70% across benchmarks.

It maintains generation fluency while improving value alignment.

Experiments show robustness across multiple backbones.

Abstract

Aligning large language models (LLMs) with human values typically relies on post-training or inference-time steering that directly manipulates the backbone's parameters or representation space. However, a critical gap exists: the model's residual stream is highly dynamic, in which values exist as fragile, low-dimensional properties, inherently incompatible with the stability required for consistent value expression. In this paper, we propose the Stable Value Guidance Transformer (SVGT), which addresses this gap through an independent value module incorporating two key designs: (1) independent value modeling, maintaining normative representations in a dedicated value space isolated from the backbone, and (2) explicit behavioral guidance, transducing these stable signals into learnable latent Bridge Tokens. These tokens serve as dynamic value anchors to explicitly steer the generative trajectory, ensuring robust adherence across diverse contexts without disrupting the backbone's internal representations. Experiments across multiple backbones and safety benchmarks show that SVGT generally reduces harmful scores by over 70% while maintaining generation fluency, demonstrating the efficacy of architecturally grounded value modeling. Our code is available at https://github.com/Clervils/SVGT.git.