It literally cannot come up with novel solutions because it's goal is to regurgitate the most likely response to a question based on training data from the internet. Considering that the internet is often trash and getting trashier, I think LLMs will only get worse over time.