The onboarding flow for #ChatGPT should require new users to view a response to an obscure question about which they are a subject-matter expert before they can ask the model any other questions.

Until you’ve sat aghast at the sight of a confident, detailed, but completely wrong answer, you will have no understanding of the skepticism you need to apply to the guidance it provides. Already losing track of the number of engineers I’ve seen apply ChatGPT advice that turns out to be terrible.