Eigil Rischel comments on Has anybody used quantification over utility functions to define “how good a model is”?