How Are Large Language Models Built