Sure a common answer would be intentionally copying in the same sessions, less likely is intentionally copying via eidetic memory.. But how much of a spectrum could there be in the middle for memory that would result in repeating a "plagiarism" form months later, etc?

People say how obvious the parlor trick is when they look at a small model LLMs. Well, I've seen the same parlor trick in students who get good grades but seem weak at thought from fundamentals. It seems quite possible to me that in some examples we are now going after them because the environment changed. At much earlier points we did actually value the people who could recite even if somewhat brokenly because we lacked random order recital tools.