I think a good yardstick for this is something that is generative; so, for instance, can the system generate a good programming challenge question? This is still a no.