What happened
Researchers have figured out how to use AI video generation models directly as compression tools instead of bolting them onto traditional codecs—the bitstream directly controls what the AI generates at each step. This means videos could be transmitted at much lower data rates while maintaining quality, since the AI model itself becomes the compression method rather than an add-on.
Why it matters
If this scales to real-world video systems, it could dramatically reduce bandwidth requirements for video streaming, which currently consumes about 80% of all internet traffic, by letting the receiving end reconstruct video using its own AI model rather than transmitting every pixel.