Just hit play!
This is an example of a sparse quadtree, a method of reducing the number of voxels required to represent a given shape. At this resolution, there would be 4096 voxels to store without using a quadtree, a typical list would require just over 4kB to store that information. A bitstream would only require .5kB, but only allow two types of voxel, lacking interesting features. This method, which is a further improvement on a dense quadtree (look inside to generate that one instead), allowing nodes to have variable numbers of children, instead of a constant 4. The sparse quadtree only needs to store 320 voxels, though the process generates 597 nodes. These 320 voxels would need to include some spatial data, bringing their required storage down to .64kB. Despite being worse than the bitstream in for storing two types of voxels, the major benefits for a quadtree come at runtime. Quadtrees, and their 3D counterparts Octrees, are much for efficient for collision detection or mesh generation than scanning the entire 4kB of voxels each time an update has to be made.
