Apart from this apparently being an evidently unsafe application of LLMs, it strikes me as being an observation for "AGI is not lurking in the models' hyperplanes" insofar as training/fitting isn't leading to emergence of generalizable concepts like "quotes" and "citations."
Brendan KeoghSurely the solution here is just *don't use AI tools* for your writing and research and continue to do your writing and research in ALL the ways you were doing it before you started using AI instead of doing the work??