Research & Publications Hub

Large Language Models (LLMs) reflect human knowledge and social biases present in our society (Resnik, 2025; Gallegos et al., 2024). This paper proposes Bias Inversion Prompting (BIP) — a method that uses the model's disparities to converge biases toward a more accurate representation of the context with the purpose of inferring balanced responses. This method shifts the perspective intentionally, analyzing divergent outputs and guiding models to reconcile them. Bias Inversion Prompting forces model bias into alignment, helping users obtain balanced outputs.