I would rather just read the thought as it was originally expressed by a human somewhere in the AI's training data, rather than a version of it that's been laundered through AI and deployed according to the separate, hidden intent of the AI's operator.