Reproduce “SESGO: Spanish Evaluation of Stereotypical Generative Outputs” 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

Melissa Robles, Catalina Bernal, Denniss Raigoso, Mateo Dulce Rubio

SESGO evaluates whether models respond to Latin American stereotypes through 4,156 Spanish prompts. It adapts BBQ’s ambiguous and disambiguated design, combining culturally situated and adapted material. For Llama 3.1 Instruct, matched prompts yielded Spanish accuracy of 0.449 versus 0.648 in English, with bias scores of 0.573 versus 0.369. These fixed-choice results do not establish causes or generalize to Spanish-speaking settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.