I need help finding something, a couple of weeks ago I read a very good article that explained that if you added or spoke in a certain way (depending on the model) the results were much better.
It is a study that I believe was made by openai itself, but I don’t remember well… Can somebody help me? I wanted to analyse it in more detail now and I can’t find it 
Appreciate any help.
If you are in ChatGPT, the input you provide that will get the best results is simply the question or task you wish to have fulfilled, in clear structured language, separating the instructions to be followed from the data to be acted upon.
Trying to make the AI act differently, you should use custom instructions, as the memory that ChatGPT keeps is fleeting, and you will find within one turn it dramatically no longer barks like a dog in response but it has reverted back to being ChatGPT.
Yep I know that part.
But as I say, in the study I mentioned, it gave certain recommendations to send the questions in a certain specific way, because the answer will be more efficiency and also how correct the answer could be.