And asking ChatGPT for help with tuning the instructions themselves seems particularly hard. Has anyone successfully used ChatGPT (4) to improve on its own instructions? This seems impossible to me.
I see using ChatGPT as alternating between providing instructions – describing the desired end goal, the constraints, giving examples – and providing the ‘content’ for its task. For more complex things, you need to spend more effort on the instructions.
And asking ChatGPT for help with tuning the instructions themselves seems particularly hard. Has anyone successfully used ChatGPT (4) to improve on its own instructions? This seems impossible to me.