Large part of problem-solving is introducing and removing details. As I consult with chatbots and gather more information, I often open new windows and ask simple, narrow questions instead of continuing to investigate different avenues in the same chat. This feels more efficient. Why? The context pollution. The chatbot will continue to integrate all of the details that I mentioned in the conversation. It will remain fixated on them and search for a (sometimes) hyper-specific answers.
So what is happening when I switch away (and ask a narrower question?)
But when discussing a problem in a limited context doesn't yield good results, I sometimes start adding those details back, in the hopes that this might give the chatbot more ideas. When the big picture doesn't have a solution, a particular exception might. We seem to be able to only properly attack one set of circumstances at once, so adding and removing these particulars is a part of our thought process.