Source code size vs learned model size in ML and in humans? — LessWrong