is it still a useless model if, say, you can run it at (prompt+output)*24/s and use it to make executive function decisions?