RuBia: A Russian Language Bias Detection Dataset

arXiv:2403.17553v1 Announce Type: new
Abstract: Warning: this work contains upsetting or disturbing content.
Large language models (LLMs) tend to learn the social and cultural biases present in the raw pre-training data. To test if an LLM's behavior is fair, functional datasets are employed, and due to their purpose, these datasets are highly language and culture-specific. In this paper, we address a gap in the scope of multilingual bias evaluation by presenting a bias detection dataset specifically designed for the Russian …

abstract arxiv behavior bias biases cs.cl data dataset datasets detection fair functional language language models large language large language models learn llm llms pre-training raw social test training training data type work