Are We Prompting New AI Models the Wrong Way?