Biases in Large Language Models: Origins, Inventory, and Discussion

In this article, we introduce and discuss the pervasive issue of bias in large language models that are currently at core mainstream approaches to Natural Language Processing (NLP). We first data selection bias, is, caused by choice texts make up a training corpus. Then, survey different types social evidenced text generated trained on such corpora, ranging from gender age, sexual orientation ethnicity, religion culture. conclude with directions focused measuring, reducing, tackling aforementioned bias.