# Values as Style: Disentangling Values from Semantics with One-Way Mixing for Low-Damage LLM Steering

arXiv: https://arxiv.org/abs/2609.39701 · HTML: https://mlsanity.com/paper/2609.39701/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 32 of its 35 references that are in the map.
- First posted on arXiv 30 Sep 2026.

## Builds on (most cited first, 30 of 32)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [A Style-Based Generator Architecture for Generative Adversarial Networks](https://mlsanity.com/paper/1812.04948/) (2018)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Squeeze-and-Excitation Networks](https://mlsanity.com/paper/1709.01507/) (2017)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Domain-Adversarial Training of Neural Networks](https://mlsanity.com/paper/1505.07818/) (2015)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [FiLM: Visual Reasoning with a General Conditioning Layer](https://mlsanity.com/paper/1709.07871/) (2017)
- [Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization](https://mlsanity.com/paper/1703.06868/) (2017)
- [Barlow Twins: Self-Supervised Learning via Redundancy Reduction](https://mlsanity.com/paper/2103.03230/) (2021)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning](https://mlsanity.com/paper/2105.04906/) (2021)
- [Mass-Editing Memory in a Transformer](https://mlsanity.com/paper/2210.07229/) (2022)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [Multimodal Unsupervised Image-to-Image Translation](https://mlsanity.com/paper/1804.04732/) (2018)
- [Improving Alignment and Robustness with Circuit Breakers](https://mlsanity.com/paper/2406.04313/) (2024)
- [Diverse Image-to-Image Translation via Disentangled Representations](https://mlsanity.com/paper/1808.00948/) (2018)
- [Swapping Autoencoder for Deep Image Manipulation](https://mlsanity.com/paper/2007.00653/) (2020)
- [Disentangled Representation Learning for Non-Parallel Text Style Transfer](https://mlsanity.com/paper/1808.04339/) (2018)
- [Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties](https://mlsanity.com/paper/2309.00779/) (2023)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [Improving Disentangled Text Representation Learning with Information-Theoretic Guidance](https://mlsanity.com/paper/2006.00693/) (2020)
- [ValueBench: Towards Comprehensively Evaluating Value Orientations and Understanding of Large Language Models](https://mlsanity.com/paper/2406.04214/) (2024)
- [Internal Value Alignment in Large Language Models through Controlled Value Vector Activation](https://mlsanity.com/paper/2507.11316/) (2025)
- [Measuring Human and AI Values Based on Generative Psychometrics with Large Language Models](https://mlsanity.com/paper/2409.12106/) (2024)
- [The Anatomy of Alignment: Decomposing Preference Optimization by Steering Sparse Features](https://mlsanity.com/paper/2509.12934/) (2025)
- [Resolving Lexical Bias in Model Editing](https://mlsanity.com/paper/2408.10411/) (2024)

## Abstract (arXiv)

Value steering should change an LLM's normative priorities while preserving the scenario, facts, and task constraints underlying its answer. Conventional activation edits often change both. We introduce an editable semantic-value interface on frozen residual states, with a one-way semantic-to-value pathway that grounds value recognition in context. Stop-gradient blocks feedback through this pathway; swap consistency, topic de-confounding, and decorrelation encourage selective codes. At inference, editing the value code produces a residual delta while holding the semantic code fixed. On two instruction-tuned backbones, this interface improves semantic preservation and reduces benign refusals at comparable value alignment. A matched mixing-by-gating ablation separates representation learning from selective edit activation, and dimension-matched probes establish improved code selectivity. Against validation-selected prompting on LLaMA-3.1-8B, the method achieves comparable alignment (0.750 vs. 0.748), higher BERTScore (0.938 vs. 0.923), and fewer contradictions (5.1% vs. 7.6%). Human ratings and cross-taxonomy controls provide complementary evidence for low-damage value steering.