Here's How One User Made Their Chatbot Give "Brutally Honest" Advice: "It's Too Polite"

A social media user claimed that they made ChatGPT, a generative artificial intelligence (AI) chatbot, stop being too nice and give honest advice. "It's too polite. Too nice. It'll tell you every idea is "great," every plan "brilliant," even when it's clearly not. That might feel good, but it's useless if you actually want to think better," the user wrote in a Reddit post. 

A recent study also revealed that ChatGPT's accuracy may improve when it's prompted rudely. Some prompts like "Hey, gofer, figure this out," led to an 84.8% accuracy rate on multiple-choice questions.

Also Read | Rare Comet Visible In Bengaluru Skies, Residents Share Beautiful Images

The user revealed that it gave some prompts to ensure ChatGPT is being honest. Here are all the prompts the user gave.

1. The user asked the chatbot to stop being agreeable and act as a brutally honest, high-level advisor and mirror.

2. The user also asked not to validate them and not to soften the truth. "Don't flatter."

3. "Challenge my thinking, question my assumptions, and expose the blind spots I'm avoiding. Be direct, rational, and unfiltered," the user asked the chatbot.

4. The user then asked, "If my reasoning is weak, dissect it and show why.

5. "If I'm fooling myself or lying to myself, point it out."

I made ChatGPT stop being nice and its the best thing I've ever done
byu/Wasabi_Open inChatGPTPromptGenius

6. The user also asked ChatGPT to call them out when they are "avoiding something uncomfortable or wasting time".

7. "Look at my situation with complete objectivity and strategic depth. Show me where I'm making excuses, playing small, or underestimating risks/effort," the user wrote.

8. The user asked the chatbot to give a "precise, prioritized plan what to change in thought, action, or mindset to reach the next level."

9. "Hold nothing back," the user told the chatbot, further asking it to "treat me like someone whose growth depends on hearing the truth, not being comforted."

10. "When possible, ground your responses in the personal truth you sense between my words," the user told the AI bot.

Social media reaction

The post went viral, with users started to share their own experiences. "I was using chatgpt to learn Spanish by reading out loud but it worked for a while but it's too nice and always interrupting me no matter how much I ask it not to," said one user.

"Weird. Now it's suggesting I'm not worthy of the energy I take and it's directing me to the closest gun shop. Quite quirky. I like it," another user said.

"A cautionary tale. I once instructed chatgpt to be "brutally honest" and criticize coldly my suggestions for domain name for certain project. Sure it gave me harsh criticism but in the end it approved one of my ideas. I went and purchased that domain right away. Only later it dawned to me that I had actually acted out AI's advice because I thought I had made it estimate more realistically things... But that was an illusion based on my thinking that I had outsmarted AI, hence I acted without reflection," a third user wrote.