My experience is that each question I ask or point I make produces an answer that validates my thinking. After two or three iterations in a row in this style I end up distrusting everything.
For example: - I implement something. - Then I ask it to review it and suggest alternatives. Where it will likely say my solution is the best. - Then I say something like "Isn't the other approach better for __reason__ ?". Where the approach might not even be something it suggested.
And it seems that sometimes it gives me some valid points.