Stuart_Armstrong comments on Learning (meta-)preferences