Answering your questions about business and technology
When I tell ChatGPT that I am thinking of doing something, it replies “That is a great idea”. When I come back the next day and tell it that I slept on it and no longer want to do that same thing, it replies “No, you shouldn’t, that wasn’t a good idea to begin with”
How can I get ChatGPT to be less of a cheerleader and actually challenge my assumptions?
~ Regina
Hi Regina,
That was a great question. Thank you for asking.
(Kidding)
So, the short answer is that you have to explicitly tell ChatGPT that you want it to challenge your thinking or ideas. This goes for most of the popular AI chatbots - Copilot, Perplexity, Claude, Gemini, etc.
All of the models in those chatbots have a system prompt that defines the chatbot’s behavior and response patterns. You can safely assume that all of their system prompts have instructions to be “a helpful assistant”.
This usually translates to the cheerleader-type behavior that you are describing. Unless you tell the model to challenge you, it will likely analyze challenging you as not being helpful and it won’t do it. You’ll have to play around with the wording until you get something that works. One thing that usually works for me is along the lines of -
It would be helpful if you challenged my ideas and gave me alternatives
Unfortunately, OpenAI doesn’t publish their system prompts (not that I’ve been able to find at least), but Anthropic does. If you want to get an idea of what goes in a system prompt, look at one of the models on this page - https://docs.anthropic.com/en/release-notes/system-prompts
I would post the details here but they are usually 2000+ words long.
Hope that helps
~ Damien