I noticed that my custom GPT works really well when I send a prompt in the preview tab, extracting the correct values from my file, and performing correct pattern recognition. Then I start an actual chat with the GPT, and it doesn’t behave similar to the preview. Is there a fix to this or do i just have to accept it? Does anyone have a workaround?
4 Likes
First make sure the update went through.
If it has, you need to check instructions for ambiguous bugs. Note the responses and logic you should be able to find error in your code and never let a GPT write anything complicated if you have custom novel functions . But long story short ambiguous instructions will cause that and in chaotic systems like GPT errant behavior rise from it. I write logic like I am making a wish and the genie is a butthead 
I also put redundant refreshers in instructions for my structure I’ll repeat important instructions as refreshers at top middle and end of instructions
1 Like
Welcome to the forum!
This is an amazing place. 
1 Like