I have set up the ChatGPT integration with messenger and it works well in terms of running conversation. In the Assistant Instructions I have put a detailed training prompt which in my understanding would be used as basis for the conversation.
The problem is that the answers I am getting from GPT are not aligned witht he guideline I have provided.
I have changed the training description 15 times but it still turns back to a specific option which I have clearly prohibited to even mention.