Comment by samuelknight

That's half true. A very smart model should be able make good explanations, which include simple understandable prose. That can should be possible even as its thought process gets more alien.