I don't even see how using AI to upscale the video could possibly save them any money.
It will still use the same amount of storage and bandwidth, actually more since the AI adds fake detail to lower quality source material, and now they have to spend an enormous amount of energy on the upscaling process.
They're also not increasing quality since most uploads are high quality and they're upscaling even the highest quality footage which is actually losing detail due to the shitty AI upscaling.
Given that it makes no sense from a quality, money, or energy perspective... you have to ask yourself, why exactly would they decide to do this?
and now they have to spent an enormous amount of energy on the upscaling process.
I don't know this part is true. Not all AI is equal, and AI upscaling and denoising is sufficiently lightweight that it's used as a post-processing step to make raytracing more practical.
Probably doing it because they think people want higher quality videos, but you're right it'll cost them a lot of money.
The 4k version of OP video is using 3.5 GiB. Maybe ten copies with redundancy and in different datacenters, that's maybe $0.30 in hard drive space. That's going to add up if they upscale lots of their 1080p to 4k. Also electricity and upscale AI isn't free, and they'll be redoing this over and over as the upscale models improve.
It's probably not actually a nefarious master plan to make people doubt everything. They're just looking at micro metric A/B testing saying "which vid do you prefer" and the mechanical turks they ask say the upscale because less blurry.
...but still hilarious that in the longer term they're undermining themselves because nobody asked, "yeah but should we do this?"
Market share for people searching for high quality video could be quite high, if the demand outstrips the supply then upscaling content that isn't in "27k 2700fps!!" might make it worth the advertising bucks.
But if they can ultimately achieve AI up-scaling on the client side then it would pay off infinitely.
I don't even see how using AI to upscale the video could possibly save them any money.
It will still use the same amount of storage and bandwidth, actually more since the AI adds fake detail to lower quality source material, and now they have to spend an enormous amount of energy on the upscaling process.
They're also not increasing quality since most uploads are high quality and they're upscaling even the highest quality footage which is actually losing detail due to the shitty AI upscaling.
Given that it makes no sense from a quality, money, or energy perspective... you have to ask yourself, why exactly would they decide to do this?
I don't know this part is true. Not all AI is equal, and AI upscaling and denoising is sufficiently lightweight that it's used as a post-processing step to make raytracing more practical.
Probably doing it because they think people want higher quality videos, but you're right it'll cost them a lot of money.
The 4k version of OP video is using 3.5 GiB. Maybe ten copies with redundancy and in different datacenters, that's maybe $0.30 in hard drive space. That's going to add up if they upscale lots of their 1080p to 4k. Also electricity and upscale AI isn't free, and they'll be redoing this over and over as the upscale models improve.
It's probably not actually a nefarious master plan to make people doubt everything. They're just looking at micro metric A/B testing saying "which vid do you prefer" and the mechanical turks they ask say the upscale because less blurry.
...but still hilarious that in the longer term they're undermining themselves because nobody asked, "yeah but should we do this?"
Market share for people searching for high quality video could be quite high, if the demand outstrips the supply then upscaling content that isn't in "27k 2700fps!!" might make it worth the advertising bucks.
But if they can ultimately achieve AI up-scaling on the client side then it would pay off infinitely.