Who is better at math, Jenny or Jingzhen? Uncovering Stereotypes in Large Language Models
This work addresses the problem of harmful stereotype propagation in LLMs, which disproportionately affects marginalized communities, by providing a comprehensive global dataset and analysis, though it is incremental in building on existing bias literature.
The researchers introduced GlobalBias, a dataset of 876k sentences covering 40 gender-by-ethnicity groups, to study stereotypes in large language models (LLMs) by probing models via perplexity and generating character profiles. They found that demographic groups associated with stereotypes were consistent across model likelihoods and outputs, with larger models showing higher levels of stereotypical outputs even when instructed not to.
Large language models (LLMs) have been shown to propagate and amplify harmful stereotypes, particularly those that disproportionately affect marginalised communities. To understand the effect of these stereotypes more comprehensively, we introduce GlobalBias, a dataset of 876k sentences incorporating 40 distinct gender-by-ethnicity groups alongside descriptors typically used in bias literature, which enables us to study a broad set of stereotypes from around the world. We use GlobalBias to directly probe a suite of LMs via perplexity, which we use as a proxy to determine how certain stereotypes are represented in the model's internal representations. Following this, we generate character profiles based on given names and evaluate the prevalence of stereotypes in model outputs. We find that the demographic groups associated with various stereotypes remain consistent across model likelihoods and model outputs. Furthermore, larger models consistently display higher levels of stereotypical outputs, even when explicitly instructed not to.