It's definitely asymmetrical though, once figured out: Garbage in : garbage out turns into.. Gold in : Magic out.

This barrier may also form the core of GPT-3 powered apps: The real value won't be in the coding, but in the hardcoded prompt structures hidden within that app that are used to get the very best performance from GPT-3.

This. Also note @sh_reya's and @notsleepingturk's awesome gpt-sandbox library!

It's like a whole different form of computational thinking. I also wonder if people "Prime" their prompts differently. I can't necessarily tell if the future of GPT is going to require this sort of Priming approach, but if it does... it will definitely need some intro course.

Maybe when gpt-4 comes around, it will learn what are good prompts for gpt-3. So you will just ask it to give you a good prompt to try on gpt-3.