I am trying to create a GPT to write structured data code for web pages, but finding that ChatGPT keeps forgetting critical points, even when the commands have been explicitly included in the instructions. I find I can add these points and it dutifully modifies the instructions, so they are obeyed at first, but then forgotten after a couple attempts and it becomes progressively less reliable as I edge towards the 50 task limit.
I have also seen that it can analyse a web page one time, then tell me the very same feature is not available at the next attempt. It's all very promising in theory, but in practice I am losing a lot of time having to continually correct things.
Are other finding this? I experienced this previously with ChatGPT and my hope was that it would be easier if I created my own GPT to sidestep the preliminary stages.