It's not a semantic difference. That's the problem. There is a difference between knowledge of protein folding (which is what the article is talking about), i.e. understanding proteins well enough to predict how they will fold, and modeling a folded structure (which is what AlphaFold is doing, based off of other pre-existing experimental structures).

AlphaFold isn't simulating folding at all. It's modeling what the final structure should look like based off other data, not knowledge of how a protein folds. That's what is frustrating me so much with this topic. It's being sold as "we can predict how proteins will fold" as opposed to "we can model a folded protein based on its sequence". There's a huge difference in our understanding of protein folding, between those two points.