Also, the longer the context window, the more likely the LLM derangement/ignoring safety. Frequently, those with questionable dependence on AI stay in the same chat indefinitely, because that's where the LLM has developed the ideosyncracies the user prefers.