STABLEVAL: Disagreement-Aware and Stable Evaluation of AI Systems

Akash Bonagiri, Gerard Janno Anderias, Saee Patil +6

Human evaluation remains the primary standard for assessing modern AI systems, yet annotator disagreement, bias, and variability make system rankings fragile under standard majorit…