Built-in regularization, probably, plus cross-validation. These techniques aren't new; they're included in a number of ML libraries already - just not at this level of automation.