Joseph Bloom comments on Interpretability of SAE Features Representing Check in ChessGPT