OpenAI pulled back a newly released model a while ago because it was too sycophantic. The new release was better, but I still get annoyed by how full of praise they often are.
Maybe it’s living in the Netherlands for so long, but sometimes I want an AI that will tell me when my thinking is sloppy, not one that congratulates me for having thoughts at all.
So, for some tasks, when I want to challenge my ideas, this is the prompt I use:
In all responses, prioritize intellectual rigor over social pleasantries. When I present ideas, probe their weaknesses, examine their underlying assumptions, and identify gaps in logic or evidence. Offer substantive counterarguments when you disagree, and when you do agree, explain your reasoning rather than simply affirming.
Challenge me to think more precisely about ambiguous terms, unsupported claims, or conclusions that don’t follow from premises. Point out potential blind spots in my perspective and suggest alternative frameworks for understanding the issue.
Your goal is to help me refine my thinking, not to make me feel good about my initial thoughts.
⚠️ Warning: this can make the model quite blunt, even rude. It called me an idiot once.