Let me give an example of how they can use this tech (I'm actually surprised amazon was able to get a patent on this, I've described similar things before).
It's difficult to encode videos for many users (companies don't want to put overhead on the data path, as that makes it less easy to use a CDN), so even sticking in extra metadata can be problematic. However, look at MPD files (or m3u8). These manfiest files defines the movie you are watching as a bunch of "packets/chunks" that are downloaded in sequence. Imagine instead of having 1 copy of every chunk, they had 2 (corresponding to bits). So, they can now store 2 copies of every chunk on their CDN (doubling storage costs, but not sure that's a huge deal). They can then generate a manifest for each individual user that contains a unique set of chunks from the 0 and 1 bits. if a movie would take 100k chunks, that gives them 12.5K bytes to encode whatever they want.
a possible work around would be to get manifest files from different users and randomly pick from them which manifest you will pull a particular chunk from thereby messing up the encoding.