I’m trying to understand how ChatGPT decides what to carry from one conversation into another. If a saved memory points one way but my custom instructions say something different, which one is more likely to shape the response?
It would also help to know how this works in practice when memory is turned off, or when I want to remove one remembered detail without changing the rest of my setup.