Now imagine an adaptive system in a game, where the game seamlessly transitions between the original model on the left to the one on the right depending how many pixels each polygon takes up on screen, and the player could continue to zoom in to an infinite amount without seeing any hard polygonal edges. The actual modelled asset is far lower poly then we'd seen in a modern AAA game (look at that sphere that's literally represented by a simple cube), but when subdivided a few times, the actual form in revealed. Look at this example for Monsters University.