What can they do though?
Either turn it off or reduce the resolution until it looks more like NY
Compare those shots with these
To get better visuals the spatial resolution needs to be even higher. I guess they can dd more detail levels at the source, so it loads quicker when you stay at the same altitude, optimized meshes by altitude and optimized meshes for glide paths.
Handcrafting all that is simply not possible. Another solution is improving the AI buildings with the PG data, ie more AI placed models to better match the actual data. That still means a lot of handcrafting for all the building types and styles and a reliable way to detect what goes where.
For ‘fun’ very crude estimate for Earth at London PG resolution. London PG area is about 42.5 GiB of data (I estimated it between 41 and 44 based on my grid search in the second link). The area the PG data covers is between 115km2 and 145 km2 (measured using the map created in lnm, it’s roughly square minus a rough square on the lower right side, but I was flying outside the edges as well)
Total land area of Earth is about 148 326 000 km2, which would be between 41.5 petabytes and 52.3 petabytes. And it’s not detailed enough yet!