Probably to reduce storage costs

But the upscaling isn't applied live/on viewing, right? The video being upscaled is still stored on their server and then streamed. How does it reduce storage costs?

Do you know that for a fact?

Maybe Google has done the math and realized it's cheaper to upscale in realtime than store videos at high resolution forever. Wouldn't surprise me considering the number of shorts is probably growing exponentially.

The economics don't make sense, each video is stored ~ once (+ replication etc. but let's say O(1)) but viewed n times, so server-side upscaling on the fly is way too costly and currently not good enough client-side.

Are you considering that the video needs to be stored for potentially decades?

Also shorts seem to be increasing exponentially... but Youtube viewership is not. So compute wouldn't need to increase as fast as storage.

I obviously don't know the numbers. Just saying that it could be a good reason why Youtube is doing this AI upscaling. I really don't see why otherwise. There's no improvement in image quality, quite the contrary.

Boost visual quality, which improve viewer retention. So, money. I've tried many times to get a short with retention > 90% that is, 90% of viewers watch all the way to end. That's the key to going super viral. Very hard to do. I've had many shorts get around 75% and about 1k views but then die. Maybe I need some AI!

Is the visual quality really boosted? It seems to give a very distinctive, almost uncanny-valley look to the video.

AI upscale does not improve quality imo, I'd much prefer to watch grainy vhs originals to AI upscaled ones that insert weird shapes in the image.

This is especially bad in animation, where the art gets visibly distorted.

There are people out there who can vote and can (sometimes) buy and drink alcohol and who never used VHS in any capacity.

And a new generation what is trained on a constantly enabled face filters and 'AI'-upscaled slop is already here.

> Boost visual quality

So to make edible stuff from shit.

Perceived quality? They tried to pull an "everything 4k@60Hz" for their 360p@30Hz low poly Stadia content as well.

To make people more accustomed to the AI generated look so that when they release their next Veo integration to YouTube content creator tools, these videos will stand out less as unnatural.

There's also the on-by default, can't be disabled, auto-dubbing YouTube performs on every video that's not in the single browser's language. The dubbing quality is poor for the same reason, to intentionally expose viewers to AI content.

It's 100% a push to remove human creators from the equation entirely.

That's exactly it. All social media platforms are experimenting with replacing humans with AI.

Sadly, this is a real possibility. I would even conjecture they are testing a new pipeline, in which the input is real videos and the output are AI-generated.

For now it's a kind of autoencoding, regenerating the same input video with minimal changes. They will refine the pipeline until the end video is indistinguishable from the original. Then, once that is perfected, they will offer famous content creators the chance to sell their "image" to other creators, so less popular underpaid creators can record videos and change their appearance to those of famous ones, making each content creator a brand to be sold. Eventually humans will get out of the pipeline and everything will be autogenerated, of course.

> Then, once that is perfected, they will offer famous content creators the chance to sell their "image" to other creators, so less popular underpaid creators can record videos and change their appearance to those of famous ones, making each content creator a brand to be sold.

I'm frightened by how realistic this sounds.

To me, this is the only thing that makes sense. Why else would you spend so much money doing this?

Maybe it's to make it more difficult to train AI video models from YouTube. Think about it, they have the raw footage so could use it if they want, but competitors using scrapers will have slightly distorted video sources.

Here's how I imagine it went:-

1. See that AI upscaling works kinda well on certain illustrations.

2. Start a project to see if you can do the same with video.

3. Develop 15 different quality metrics, trying to capture what it means when "it looks a bit fake"

4. Project's results aren't very good, but it's embarrassing to admit failure.

5. Choose a metric which went up, declare victory, put it live in production.

I can’t think of a more dislike-able company than YouTube. I used to love youtube and watch it everyday and it would make me a happier, smarter person. Now youtube’s impact on their users is entirely negative and really the company needs to be destroyed. But they won’t be because they are now evil, and evil is profitable.