Harm Laundering in GPT Models: Evidence That Gender Discrimination Is Transformed Rather Than Reduced Across Safety-Trained Generations
The study shows that safety training in successive GPT models transforms rather than eliminates gender discrimination, raising concerns about the effectiveness of current alignment methods.
The study shows that safety training in successive GPT models transforms rather than eliminates gender discrimination, raising concerns about the effectiveness of current alignment methods.
Auto-scored: The paper provides evidence of persistent bias in AI models, indicating a setback for safety training effectiveness.