David Hartmann

New BAFA method drastically cuts LLM fairness audit queries

Researchers have developed BAFA, a new method for efficiently auditing the fairness of large language models (LLMs). This approach treats auditing as an uncertainty estimation problem, using a version space of surrogate…