All of you, who think it looks bad or it repeats, the answer is simple. Read the description.
• Tessellation enables a continuous ramp on the level of detail, more than doubling the total triangle count and providing up to a 500x increase in detail on the nearest objects.
• Three dimensional displacement maps allow greater detail than simple height maps.
• The demo positions and renders over 600 million triangles per second with a single GPU.
• The city is procedurally generated. The buildings are constructed from a small set of objects into a city of complex structures using a set of rules called an L-system.
• In the area surrounding the viewer, over a million meshes are arranged by the L-system, including roughly 500,000 light sources.
• As the viewer moves, the demo endlessly (hence the name) generates more of the city to keep the viewer surrounded by buildings.
• The L-system randomly selects elements to assemble so each block looks different from the last but if you return to a place you saw before, it’ll regenerate the same buildings you saw previously.
• Split screen rendering modes operate in eye space so they look correct in 3D Vision.
• Screen-space ambient occlusion enables high quality shading on randomly generated and dynamic geometry which cannot be accomplished with precomputed texture maps.
• Dynamic lighting on everything; can shift from day to night.
Also, it's pretty obvious, that the small objects used in the demo to build the structures have low details for a reason. Tessellation looks more impressive that way. The tech demo is all about the future, where developers don't have to spend so many resources on the details, because tessellation will do it for them, thus making everything cheaper.