VISPA: Pluralistic Alignment via Automatic Value Selection and Activation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911384248254464 |
|---|---|
| author | Zheng, Shenyan Zhong, Jiayou Shetty, Anudeex Ji, Heng Nakov, Preslav Naseem, Usman |
| author_facet | Zheng, Shenyan Zhong, Jiayou Shetty, Anudeex Ji, Heng Nakov, Preslav Naseem, Usman |
| contents | As large language models are increasingly used in high-stakes domains, it is essential that their outputs reflect not average} human preference, rather range of varying perspectives. Achieving such pluralism, however, remains challenging. Existing approaches consider limited values or rely on prompt-level interventions, lacking value control and representation. To address this, we introduce VISPA, a training-free pluralistic alignment framework, that enables direct control over value expression by dynamic selection and internal model activation steering. Across extensive empirical studies spanning multiple models and evaluation settings, we show VISPA is performant across all pluralistic alignment modes in healthcare and beyond. Further analysis reveals VISPA is adaptable with different steering initiations, model, and/or values. These results suggest that pluralistic alignment can be achieved through internal activation mechanisms, offering a scalable path toward language models that serves all. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_12758 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | VISPA: Pluralistic Alignment via Automatic Value Selection and Activation Zheng, Shenyan Zhong, Jiayou Shetty, Anudeex Ji, Heng Nakov, Preslav Naseem, Usman Computation and Language Artificial Intelligence Machine Learning As large language models are increasingly used in high-stakes domains, it is essential that their outputs reflect not average} human preference, rather range of varying perspectives. Achieving such pluralism, however, remains challenging. Existing approaches consider limited values or rely on prompt-level interventions, lacking value control and representation. To address this, we introduce VISPA, a training-free pluralistic alignment framework, that enables direct control over value expression by dynamic selection and internal model activation steering. Across extensive empirical studies spanning multiple models and evaluation settings, we show VISPA is performant across all pluralistic alignment modes in healthcare and beyond. Further analysis reveals VISPA is adaptable with different steering initiations, model, and/or values. These results suggest that pluralistic alignment can be achieved through internal activation mechanisms, offering a scalable path toward language models that serves all. |
| title | VISPA: Pluralistic Alignment via Automatic Value Selection and Activation |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2601.12758 |