Hannah Cyberey

New white-box auditing method reveals hidden LLM biases

Researchers have developed a new framework for auditing large language models (LLMs) that goes beyond traditional black-box testing. This white-box approach utilizes activation steering to examine the model's internal w…