In CustomGPTs, there is a limit of 8000 characters for the instructions. However, content from the knowledge files uploaded are pulled in depending on relevance and need, and they become part of the context window as you chat with the GPT. Doesn’t this mean we can effectively offload a bunch of instructions to the knowledge files? What is the advantage of having stuff in the instructions window, except for maybe guaranteeing that they are used?
For example, in my instructions to the CustomGPT, I can talk about the tone and writing style it should use. Or I can say “refer to the ‘style.doc’ document in the knowledge files for how to style your response”
Another example is, in the instructions I can say “if the user responds with X, then do this.” Or you can make a “reactions.doc” file and put all types of response-counter examples there.
From what I can tell, these work fine, so I am trying to understand the difference between the instructions and knowledge with respect to managing GPT’s behavior.