The attribution problem with generative AI

True, when we write academic articles nowadays, nobody expects you to provide the trail of references all the way down to Aristotle. But few people would say that taking someone's recent NeurIPS paper and republishing it would be ok. Yes, it is a continuum, but it's still real. What exactly is common knowledge and what deserves a reference at a given point in time varies by person, depending on their domain knowledge and principles. Still, everybody has a fairly clear idea of what their own boundaries are. Would you personally be comfortable with changing some variable names in a StackOverflow snippet and passing it as your own work? Would you tell your child it's ok to copy-paste essay passages from public domain sources - after all, it's not illegal? How about if you hear an apt metaphor in someone's keynote that you haven't heard anywhere else - would you say that it's "just English" and use it as your own? Whatever your answers are to these questions - you have these answers, which means that you have your own attribution norms.