The Social Worldview Taxonomy, grounded in Cultural Theory, operationalizes hierarchy, egalitarianism, individualism, and fatalism to examine the socio-cognitive orientations encoded in large language models. Evaluation across 28 models reveals distinct profiles, while experiments informed by Social Referencing Theory examine how explicit social cues modulate those profiles. The work explores the flexibility of model worldviews and its implications for socially responsible AI.
BibTeX citation
@misc{li2025socialworldview,
title = {Analyzing Cognitive Differences Among Large Language Models through the Lens of Social Worldview},
author = {Jiatao Li and Yanheng Li and Xiaojun Wan},
year = {2025},
eprint = {2505.01967},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url = {https://arxiv.org/abs/2505.01967}
}