Trained to Perceive

The chapter analyses the prejudice towards women in large language models (LLMs), including ChatGPT, Gemini, and Pi, as well as in your own AI. If you frequently utilise LLMs in your daily life, it is essential to recognise their potential biases. This chapter employs critical discourse analysis and intertextuality to examine model outputs across various tasks, including text generation, with an emphasis on creativity, gender, and the interplay between gender and family/career dynamics, as well as perpetrator/victim roles, pronoun usage and resolutions, and responses to gender stereotypes and humour related to gender. These trends illustrate the cultural patterns present in the training data and highlight the inadequacy of current filtering and training techniques. The chapter emphasises that significant enhancement and execution of LLMs are essential to foster, rather than hinder, gender equality.