That's generally my schema for the world, but the article raises a very good and important point, which is that those probabilities are sort of fictions at some

That's generally my schema for the world, but the article raises a very good and important point, which is that those probabilities are sort of fictions at some level, because they're associated with a model that itself can be, or almost certainly is sometimes, incorrect. In Bayes land, that's because basically P(interesting thing|obs) = P(obs|interesting thing) * P(interesting thing | model) * P(model) ... but we forget the last part. What they're pointing to is the fact we could assign a probability to each of those probabilities we hear, which is more accurate, but we generally don't, and it can have fundamental, wide-ranging implications. I guess it raises the question of whether it's better to not put numbers on things at all sometimes, or to put numbers on things but pretend that they're infallible. I don't know -- I guess I thought it made a good point about what to make of probability statements based on made-up models of reality rather than reality itself, about changing targets.