New (old) tricks for online metalearning