But you've always got humans consuming it. When they evolve into an entirely new medium, that's a different problem, but there's a canonical 1-3 hour format in a medium-large room. DVDs now have bloopers and commentary, but they aren't anywhere nearly as valuable as the movie itself, so those are more error-resilient.
Incidentally, video understanding has been called the Inverse Hollywood problem, translating to scripts, and if you can just store the scripts that will generate the movie, you can do massively better video modeling and compression.
