Memory and Custom Instructions

What is the point of these features if GPT do not follow them?

I give clear instructions in both of them, and when i start a chat, i give the same instructions again but GPT do not care, 75% of the time GPT do not follow them…

few exemple :

  • never say you’re sorry under any circumstances
  • never shows any empathy under any circumstances
  • never simulate any human feelings under any circumstances
  • your answer need to be short and concise
  • never give me any code or exemples unless a direct request

these are in my custom instructions, i asked GPT to put them in “his” memory ( and he did, i checked) and i repeat them at the beginning of every chat but as i said, GPT do not follow them…

worst to be worst, if he say "i’m sorry " and i remind him to follow my instruction about that, he answer back with something like “I am sorry that i said that i was sorry which is against your instruction”

this is killing me everytime…

so I am asking again, what is the point of having custom instructions and the memory features if GPT is not using them ??