How far can bias go? Tracing bias from pretraining data to alignment

ArXi:2411.19240v2 Announce Type: replace As LLMs are increasingly integrated into user-facing applications, addressing biases that perpetuate societal inequalities is crucial. While much work has gone into measuring or mitigating biases in these models, fewer studies have investigated their origins. Therefore, this study examines the correlation between gender-occupation bias in pre-