“Richard Sutton is not sure that Large Language Models are "bitter lesson pilled" because they are trained on finite, human-generated data.”