Developing A Framework to Support Human Evaluation of Bias in Generated Free Response Text

LLM evaluation is challenging even the case of base models. In real world deployments, evaluation is further complicated by th e interplay of task specific prompts and experiential context. A t scale, bias evaluation is often based on short context, fixed choicebench-marks that can be rapidly evaluated, however, these can lose validity when the LLMs' deployed context differs. Large scale h u-man evaluation is often seen as too intractable and costly. H ere we present our journey towards developing a semi-automatedbias evaluation framework for free text responses that has human insights at its core. We discuss how we developed an operational definition of bias that helped us automate our pipeline and a methodology for classifying bias beyond multiple choice. We additionally comment on how human evaluation helped us uncover problematic templates in a bias benchmark.