ParentFull threadshikon7·It seems to me that thinking models are harder to decensor, as they are trained to think whether to accept your request.View on HN