Activation Steering for Bias Mitigation: An Interpretable Approach to Safer LLMs

Shivam Dubey

As large language models (LLMs) become more integrated into societal systems, the risk of them perpetuating and amplifying harmful biases becomes a critical safety concern. Traditi…