The models might be so fast that they can autocomplete your prompt before you even finish it, and generate dozens of possible applications before you're even done asking.