this has interesting implications if you take a Platonic morality seriously) conceptually, i would suggest you never really believe in it)
ponderlike on the one hand, yes a model is basically "frozen in time" given that its training data has a temporal cutoff. but i don't think that alone is sufficient to account for the "future date -> fictional scenario" inference