Jacob Steinhardt explores what capabilities large language models could possess by 2030, such as superhuman coding, faster processing speeds, and training on diverse data beyond text and images. via LessWrong