Reproduce “Do LLMs Have Values? A Quantitative Analysis and Alignment Framework for Values in Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Keqing Zhang, Jingyu Chen, Yufan Liu, Yongqiang Zhu, Nai Ding, Lai Jiang, Congyan Lang, Bing Li and 1 more
The authors measure LLMs’ value profiles by mapping responses to human survey items across Schwartz’s ten dimensions while varying prompt contexts and reasoning. They decompose influences into model parameters, prompt environment, and cognition, then prescribe dimension-specific interventions. Most models exceed the human distance benchmark; prescriptions match the lowest effective level in 14 of 20 held-out pairs, within this survey-based measure.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.