Evidence against Learned Search in a Chess-Playing Neural Network — LessWrong