It’s even worse. The model contains the information that claims are statistically likely to be followed by a citation, for example. So when the output produces

It’s even worse. The model contains the information that claims are statistically likely to be followed by a citation, for example. So when the output produces a claim, it follows up with a citation- and it completely makes that one up. It has no concept of what a citation is, or what purpose it serves or that a reader might actually go and validate that. It's just a specific sequence of words that follows a specific sequence of words.