Running 139 Natural Language Processing before LLMs π 139 From n-gram counts to attention, learning by doing