Prompting for Fairness: Mitigating Gender Bias in Large Language Models with Self-Debiasing Prompting
Keywords: natural language processing, large language models, gender bias, prompt engineering
Abstract: Large Language Models (LLMs) have transformed natural language processing, showcasing remarkable skill in language generation and comprehension. However, these models often exhibit gender biases inherited from the vast datasets used for training, which can lead to the perpetuation and amplification of societal stereotypes. Addressing gender bias in LLMs is critical to ensuring that these models contribute constructively across diverse fields without reinforcing inequities. This work proposes prompt-based techniques to mitigate gender bias in LLM outputs. We introduce custom zero-shot, zero-shot chain-of-thought (CoT), few-shot, and few-shot chain-of-thought (CoT) prompting methods designed to discourage biased responses and promote fairness and inclusivity. Our prompt debiasing approach leverages guiding prompts that explicitly direct the model to avoid stereotypes or engage in step-by-step reasoning, fostering more equitable language generation. Through experimental evaluation, we demonstrate the potential of prompt-based debiasing to reduce gender bias, paving the way for more responsible and inclusive applications of LLMs.
Archival Option: Yes
Submission Number: 3
Loading