AI RESEARCH
STABLEVAL: Disagreement-Aware and Stable Evaluation of AI Systems
arXiv CS.LG
•
ArXi:2605.02122v1 Announce Type: new Human evaluation remains the primary standard for assessing modern AI systems, yet annotator disagreement, bias, and variability make system rankings fragile under standard majority vote aggregation. Majority vote discards annotator reliability and item-level ambiguity, often yielding unstable comparisons across annotator subsets. We