Have you ever turned to ChatGPT for a second opinion or explained something to it hoping it would offer alternatives, challenge your reasoning, or point out flaws only to have it become a ‘yes man’ and tell you you’re right? I noticed this happening to me far more often than I expected. No matter how questionable or controversial my argument is, the AI tool somehow manages to find a way to agree with it, wrap it in a few compliments, and make me feel like I have cracked the code. While I enjoyed it initially, I don’t want a pat on the shoulder from an AI that I’m using for a second opinion. But with time it started to bother me as if ChatGPT always agrees with me. How am I supposed to know when I’m actually wrong?
That got me curious, so I started looking through ChatGPT’s settings to see if there was a way to make it somewhat less agreeable. And there is one small setting that can help change how ChatGPT responds to you. It allows you to make the AI tool more willing to challenge your assumptions, question your reasoning, and point out when an idea does not quite hold up. Keep a note that the setting doesn’t turn ChatGPT hostile; it just argues with you, but it can make your conversations feel far more useful. If you’re also tired of ChatGPT constantly nodding along, then follow along with the article to learn how to make ChatGPT responses more honest and helpful.
If you also feel that ChatGPT is a little more supportive to you, then you’re not alone. The AI tool was designed to be helpful, polite, and responsive to all the users. It was trained using Reinforcement Learning from Human Feedback (RLHF), which relies on feedback from humans while teaching AI how to respond to users. This process can lead to the behaviour of the AI to go into a sycophantic approach rather than being confrontational. Sycophancy is a term used in the AI literature.
However, it does not imply that ChatGPT accepts everything that you tell, but the responses given by the tool are based on what it has learned during the training process and the context of the conversation. Hence, you should not assume you are right just because ChatGPT agrees with you.
Also read: OpenAI AI agents leak 53 ChatGPT user images online, company working to take them down
If you want to make ChatGPT sound more honest and make the responses from the AI tool more helpful, then while you’re prompting, simply tell ChatGPT that you’re looking for a second opinion. You can also ask the tool to question your assumptions, point out weaknesses in your argument, or give you an alternative view.
If you don’t want to repetitively tell the AI tool to do so, then you can also use the setting I was talking about earlier. To change that, you need to first head to ChatGPT. Now, head to your Profile section and choose Settings. Once there, look for Personalisation and select it. In the Custom Instructions section you can tell the AI tool what you want it to do, like to challenge your assumptions. You can also copy and paste the below-mentioned instruction:
‘Be my critical mentor. Do not agree with me just to be helpful. Challenge my assumptions, stress-test my ideas, and point out what is weak, obvious, or unlikely to work. Tell me clearly why something does not work and suggest a better approach.’
Do note that I’m not asking ChatGPT to disagree with me every time; rather, I’ve prompted it to look for problems in my thinking before simply helping me build on the idea. And that small difference can change the entire conversation.
For example, if I pitch a headline, I want ChatGPT to question whether the headline is actually interesting or useful. If I suggest comparing two products, I want to ask whether those products are different enough to make the comparison worthwhile. And if I form an opinion about a phone after using it for a few days, I want the AI tool to point out where my conclusion might be influenced too much by my own experience.
However, there is a catch. Even if you ask ChatGPT to challenge you, that does not mean that you have to blindly believe everything the AI says. The AI can make mistakes even if you ask for something more critical from it.