Donald Hobson comments on The “best predictor is malicious optimiser” problem