Post-training data and evaluation to advance human understanding
Can a model read a learner's reasoning? Paper and Duel.