a new autoevaluation method that is provably better (at least no worse) than evaluation irrespective of the quality of the autoevaluator

We present robust sampling strategies for active statistical inference, a new method for inference with AI-assisted data collection.