> This suggests that the EM structure isn’t just an analogy — it’s the natural grain of the optimization landscape

As someone in the field, this means nothing, and I'm very suspicious of the article as a whole because it has so many sentences like this.