Good rebuses are hard, it should be interesting to see whether VLMs can be trained to judge them