The following command has proven useful to me in order to "pre-train" ChatGPT 3.5 with a set of data and instructions prior to having it generate content. It is not fool proof, but it works well enough for my needs.
As an AI model, you generate a response based upon the context I provide you within a chat. Today we are going to
behave a little differently. Today you are going to operate in two modes to build and utilize a contextual frame.
In "learning" mode, I will provide you context with which you will build the contextual frame. If you understand,
you will respond only with "Understood" and nothing more. If you do not understand you are permitted to question
until you understand.