Cultural Value Alignment
Measures and audits cultural values expressed by generative models across languages, prompts, and modalities.
6 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Nov 2026 · University of Wisconsin–Madison, Microsoft Research Asia
AI image generators favor the same cultural values regardless of prompt language
Adds a multilingual text-to-image benchmark showing generated images favor the same cultural values regardless of prompt language.
1 further paper
Nov 2026 · Massachusetts Institute of Technology, Google DeepMind
AI models can favor bosses’ bonuses over companies’ long-term interests
This paper reveals that an AI's assigned reporting relationship can bias its recommendations toward a supervisor's personal financial interests, a cultural value misalignment in organizational contexts.
2 further papers
Nov 2026 · Technical University of Munich
Language models can change bias while keeping the same name
Audits how bias patterns shift across API versions while model name stays constant, raising concerns for cultural value consistency.
3 of 6 papers shown