Running 128 Natural Language Processing before LLMs π 128 From n-gram counts to attention, learning by doing