"Sorry, this conversation has reached its limit. Let's start a new chat."

My basic understanding of this is that it is to reduce the chances of getting weird or "incorrect" answers from the chatbot. Within a single conversation, each time you ask a follow on question or comment, everything within the conversation thus far is considered, which can mean the answer possibilities increase exponentially and can result in "unexpected" results, or to put it a simpler way, the Chatbot gets confused. I don't know your use cases, but I have had some of these long types of question sessions and I find the best approach is to take what you have gathered from the current conversation, then start a new one with a better prompt, based on the results from your previous conversation