allenai/

persona-bias

Data accompanying the paper Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs at ICLR 2024. Paper || Code || Project website || License

This is a dataset of model outputs supporting our extensive study of biases in persona-assigned LLMs. These model outputs can be used for many purposes, for instance:

developing a deeper understanding of persona-induced biases, e.g. by analyzing the inhibiting assumptions underlying model… See the full description on the dataset page: https://huggingface.co/datasets/allenai/persona-bias.