Shifting Perspectives: Steering Vectors for Robust Bias Mitigation in LLMs
This addresses bias in LLMs for AI safety, presenting a novel method that is computationally efficient but incremental in scope.
The paper tackles bias mitigation in large language models by applying steering vectors to modify activations, achieving average improvements of 12.8% on BBQ, 8.3% on CLEAR-Bias, and 1% on StereoSet, and showing better performance than prompting and Self-Debias in all cases.
We present a novel approach to bias mitigation in large language models (LLMs) by applying steering vectors to modify model activations in forward passes. We compute 8 steering vectors, each corresponding to a different social bias axis, such as age, gender, or race, on a training subset of the BBQ dataset and compare the effectiveness of these to 3 additional bias mitigation methods across 4 datasets. When optimized on the BBQ dataset, our individually tuned steering vectors achieve average improvements of 12.8% on BBQ, 8.3% on CLEAR-Bias, and 1% on StereoSet, and show improvements over prompting and Self-Debias in all cases, and improvements over fine-tuning in 12 out of 17 evaluations. In addition, steering vectors showed the lowest impact on MMLU scores of the four bias mitigation methods tested. The work presents the first systematic investigation of steering vectors for bias mitigation, and we demonstrate that they are a powerful and computationally efficient strategy for reducing bias in LLMs, with broader implications for enhancing AI safety.