And gp's comment is relevant to the particular problems the guy (John Langford) cites in his comment about leaving. IE, that research first winds being only about scaling and then winds-up ignored because "just scale" can be an easier answer.