Jan Betley comments on Localizing goal misgeneralization in a maze-solving policy network