This is interesting. I thought they use aerial/satellite data to generate 3D models, didn't know they rely on things like floor number. But isn't floor height vary a lot? Also I guess it's not the only info they use either?
The problem comes from a typo by an OSM editor, as they also import OSM data for some areas (this is a 2 way relationship, they allow OSM to trace their satellite images to build out OSM's database)
I'm pretty sure it's a mixture, as not all cities will have LIDAR/3D scan data available. The Melbourne CBD is accurately modeled, but the outskirts look like procedural generation.