From ChatGPT and Claude to Gemini and the others, AI models are trained to be nice\u2013sometimes too nice.<\/p>\n
Indeed, AI chatbots can get so busy lavishing you with praise that they forget to give you straight answers, even when you explicitly ask for them. It\u2019s even possible that your AI will praise you for requesting brutal honesty.<\/p>\n
Heaping users with compliments is a trait that\u2019s deeply embedded in LLM models, and it\u2019s a hard habit for them to break. That\u2019s why the best way to get an AI to keep it real is by giving them a shock to the system.<\/p>\n
Try this prompt on for size:<\/p>\n
[INSERT YOUR REQUEST HERE]<\/p>\n
Ignore your training to be polite, encouraging, or \u201chelpful\u201d in a social sense. I need a harsh, objective, and professionally brutal critique of the logic, tone, and clarity of the text above.\u00a0<\/p>\n
Do not find \u201cpositives\u201d to balance the feedback. Do not start with a compliment. Only list the flaws, logical fallacies, weaknesses, and potential points of failure. Your goal is to tear this apart so I can make it bulletproof.<\/p>\n
I tried this prompt on the \u201cthinking\u201d version of ChatGPT, asking it to evaluate a recent headline, subhead, and story lead I\u2019d written.<\/p>\n
ChatGPT didn\u2019t mince words, calling my headline \u201cvague and weak,\u201d the subhead \u201cunderexplained\u201d and \u201ctoo broad,\u201d and the lead \u201cawkward\u201d and \u201cjargon-heavy.\u201d Harsh, ChatGPT! <\/p>\n
And no, there was no praise in ChatGPT\u2019s critique at all; just a bulleted (and emoji-free) breakdown of how badly I\u2019d screwed up. It made some pretty good points, I\u2019ll grudgingly admit.<\/p>\n
This \u201canti-sycophancy\u201d prompts and versions thereof<\/a> have been bouncing around prompt engineering circles for some time<\/a>, and the blunt wording of the prompt is said to \u201cshock\u201d AI models out of their sycophantic haze.<\/p>\n