Less consideration of "the best we have", more consideration of "the best we had three years ago", relatively speaking, and how this might change as time goes on.

The range of AIs that are obviously scary but not terminally dangerous is quite short.