
This paper demonstrates that jointly learning bias and stereotype detection using the new StereoBias dataset significantly improves bias detection in language models, underscoring the importance of leveraging stereotype information to build fairer AI systems.
Jul 1, 2025
BharatBBQ is a culturally adapted benchmark for assessing social biases in language models across eight Indian languages, featuring 392,864 examples across 13 categories.
Jan 1, 2025