Antoni Czolgowski

Cultural Misalignment in Large Language Models: Detection, Measurement, and Mitigation Through Targeted Fine-Tuning

This is the first study to target worst-case demographic personas with LoRA fine-tuning for cross-cultural bias mitigation, and country-level decomposition reveals that fine-tuning redistributes rather than removes bias.