Recently, Peterson et al. provided evidence of the benefits of using probabilistic soft labels generated from crowd annotations for training a computer vision model, showing that using such labels maximizes performance of the models over unseen data. …