Maybe it doesn't 'understand' in the experiential, qualia way that a human does. Sure. But it's still a valid and useful simile to use with these models because they emulate something close enough to understanding; so much so now that when they stop doing it, that's the point of conversation, not the other way around.