Reproduce “The Anatomy of Socio-Demographics in Pluralistic Offensiveness Modeling: When and Where It Helps, and the Role of Reasoning” 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
Yifan Wang, Leixin Zhang, Jinyi Mu, Yu Wang, Mayank Jobanputra, Isabel Valera, Vera Demberg
The authors model offensiveness judgments by separating a broad-pool estimate from a demographic-specific directional shift, evaluating prompted and trained models on three multi-annotator datasets. They find demographic gains are mixed for individual judgments but clearer after aggregation. For prompted models, DemoReason’s population-level Spearman correlation is 0.543 versus 0.434 for NoDemo, averaged across dataset–model pairs; this does not imply universal improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.