Running 138 Natural Language Processing before LLMs π 138 From n-gram counts to attention, learning by doing