Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.10843 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913139732250624 |
|---|---|
| author | Kiet, Huynh Trung Minh, Dao Sy Duy Nguyen, Tuan Tran, Chi-Nguyen Pham, Phu-Hoa Quy, Nguyen Lam Phu Han, The Anh Tran-Thanh, Long |
| author_facet | Kiet, Huynh Trung Minh, Dao Sy Duy Nguyen, Tuan Tran, Chi-Nguyen Pham, Phu-Hoa Quy, Nguyen Lam Phu Han, The Anh Tran-Thanh, Long |
| contents | Large language models increasingly mediate decisions that turn on moral judgement, yet a growing body of evidence shows that their implicit preferences are not culturally neutral. Existing cultural alignment methods either require per-country preference data and fine-tuning budgets or assume white-box access to model internals that commercial APIs do not expose. In this work, we focus on this realistic black-box, public-data-only regime and observe that within-country sociodemographic disagreement, not consensus, is the primary steering signal. We introduce DISCA (Disagreement-Informed Steering for Cultural Alignment), an inference-time method that instantiates each country as a panel of World-Values-Survey-grounded persona agents and converts their disagreement into a bounded, loss-averse logit correction. Across 20 countries and 7 open-weight backbones (2B--70B), DISCA reduces cultural misalignment on MultiTP by 10--24% on the six backbones >=3.8B, and 2--7% on open-ended scenarios, without changing any weights. Our results suggest that inference-time calibration is a scalable alternative to fine-tuning for serving the long tail of global moral preferences. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_10843 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Training-Free Cultural Alignment of Large Language Models via Persona Disagreement Kiet, Huynh Trung Minh, Dao Sy Duy Nguyen, Tuan Tran, Chi-Nguyen Pham, Phu-Hoa Quy, Nguyen Lam Phu Han, The Anh Tran-Thanh, Long Computation and Language Artificial Intelligence Computers and Society Large language models increasingly mediate decisions that turn on moral judgement, yet a growing body of evidence shows that their implicit preferences are not culturally neutral. Existing cultural alignment methods either require per-country preference data and fine-tuning budgets or assume white-box access to model internals that commercial APIs do not expose. In this work, we focus on this realistic black-box, public-data-only regime and observe that within-country sociodemographic disagreement, not consensus, is the primary steering signal. We introduce DISCA (Disagreement-Informed Steering for Cultural Alignment), an inference-time method that instantiates each country as a panel of World-Values-Survey-grounded persona agents and converts their disagreement into a bounded, loss-averse logit correction. Across 20 countries and 7 open-weight backbones (2B--70B), DISCA reduces cultural misalignment on MultiTP by 10--24% on the six backbones >=3.8B, and 2--7% on open-ended scenarios, without changing any weights. Our results suggest that inference-time calibration is a scalable alternative to fine-tuning for serving the long tail of global moral preferences. |
| title | Training-Free Cultural Alignment of Large Language Models via Persona Disagreement |
| topic | Computation and Language Artificial Intelligence Computers and Society |
| url | https://arxiv.org/abs/2605.10843 |