I hear you WoodenChair. For this particular task, the stakes are very low (really, zero). The model agrees with our human graders, quite well. In fact, statisti

I hear you WoodenChair. For this particular task, the stakes are very low (really, zero). The model agrees with our human graders, quite well. In fact, statistically speaking, the machine scoring is actually tightly correlated to the highest inter-rater reliability that we could achieve, among our human graders. You would be surprised.