←back to thread

296 points jakub_g | 1 comments | | HN request time: 0.216s | source
Show context
crazygringo ◴[] No.45013548[source]
From the linked tweet from YouTube's head of editorial:

"No GenAI, no upscaling. We're running an experiment on select YouTube Shorts that uses traditional machine learning technology to unblur, denoise, and improve clarity in videos during processing (similar to what a modern smartphone does when you record a video)"

https://x.com/youtubeinsider/status/1958199532363317467?s=46

Considering how aggressive YouTube is with video compression anyways (which smooths your face and makes it blocky), this doesn't seem like a big deal. Maybe it overprocesses in some cases, but it's also an "experiment" they're testing on only a fraction of videos.

I watched the comparisons from the first video and the only difference I see is in resolution -- he compares the guitar video uploaded to YT vs IG, and the YT one is sharper. But for all we know the IG one is lower resolution, that's all it looks like to me.

replies(3): >>45013678 #>>45014070 #>>45014469 #
CodingJeebus ◴[] No.45013678[source]
This is an absolutely huge deal. It doesn't matter how small the scope of the change is, they thought it was a good idea to apply mandatory AI post-processing to user content without consent or acknowledgement.

Secret experiments are never meant to be little one-offs, they're always carried out with the goal of executing a larger vision. If they cared about user input, they'd make this a configurable setting.

replies(1): >>45013917 #
crazygringo ◴[] No.45013917[source]
Again, this isn't GenAI.

The idea of it being "without consent" is absurd. Your phone doesn't ask you for consent to apply smoothing to the Bayer filter, or denoising to your zoom. Sites don't ask you for consent to recompress your video.

This is just computational image processing. Phones have been doing this stuff for many years now.

This isn't adding new elements to a video. It's not adding body parts or changing people's words or inventing backgrounds or anything.

And "experiments" are just A/B testing. If it increases engagement, they roll it out more broadly. If it doesn't, they get rid of it.

replies(2): >>45015118 #>>45018568 #
giantrobot ◴[] No.45015118[source]
> Again, this isn't GenAI.

Yet.

replies(1): >>45016636 #
crazygringo ◴[] No.45016636[source]
How about we stick to the facts of what is actually happening?

I mean, I'm also not Brad Pitt. "Yet."

replies(1): >>45020510 #
1. giantrobot ◴[] No.45020510[source]
Given Google's history and the fact they rolled this out without notice or consent makes me feel comfortable saying "yet". If YouTube can get away with making GenAI YouTubers (via some likeness sign off buried in the T&C) without paying the originals I'm sure they'd love to do so. All the ad impressions with none of the payout to creators.

Their AI answers box (and old quick answer box) has already affected traffic to outside sites with answers scraped from those sites. Why wouldn't they make fake YouTubers?

> I mean, I'm also not Brad Pitt. "Yet."

Not with that attitude!