Bias in NLP
The discussion highlights the pervasive issue of bias in natural language processing, emphasizing that current models can encode harmful stereotypes. Safia's work illustrates how search terms related to marginalized identities, like "black girls," have historically led to inappropriate content due to underlying data biases. This raises critical questions about the responsibility of both data and model design in perpetuating these biases, suggesting that without systemic changes, these issues will continue to arise.In this clip
From this podcast
Gradient Dissent - A Machine Learning Podcast
Emily M. Bender — Language Models and Linguistics
Related Questions