Google is using YouTube videos to train its AI video generator

[pier25]

To no one's surprise. If you're not the customer you're the product.

[emodendroket]

I think that phrase ought to be retired simply because even if you are paying money you often still are “the product.”

[echelon]

This should have resulted in an antitrust dismantlement by now. Google has every structural advantage in the world.

Years ago, Google would have been worth more if sold for parts. They were giving away far too much (and pissing on entire industries while doing so). Now they're activating all of those assets for strong, explosive incremental growth. It's hard to even call it incremental. More like checkmate world.

They're going to off so many businesses this decade and collect all the money.

They own the web, they own most of mobile, they control the other half of mobile, they own search, they own media, they own advertising. There's not a dollar that gets made that doesn't flow though Google somehow.

You can't even build a brand anymore without getting extorted by Google. You'll have your competitors paying to trademark squat you, and the browser itself defaults to Google search.

Google really needs to be split into about a half dozen companies. This is way bigger and way worse than Ma Bell.

[bgwalter]

Here is a free business idea: Create an agentic "AI" video watcher. "AI" YouTube creators can register with the service, which will then watch their videos, will generate click-throughs to the advertisers and interact with the advertiser's web pages. The service is financed by profit sharing.

This streamlines video watching, which humans are notoriously slow at. It could lead to efficiency gains in video and ad watching that are practically unlimited.

[kube-system]

I'm guessing that is a facetious response, but in case it isn't: this is just plain old fraud.

[gauku]

Almost sure that's a tongue-in-cheek response. Right?

[kunzhi]

Reading this article I couldn’t help but remember the Key & Peele skit about joke theft - “high on potenuse.” All this AI training feels similar to me on some level. Yeah, it’s “just making a copy” on the other hand the person who originated the idea doesn’t get to participate in the success.

Life is hard, but at least on the other hand, it’s also unfair.

[paxys]

Well, no shit.

Remember when OpenAI's CTO was asked to confirm that they don't use YouTube to train Sora and she evaded the question...?

Everyone is training on everything they can get their hands on, period.

[cavisne]

Hilariously there was a story how Google could not train on Youtube data due to their TOS, so they changed it for new videos. Meanwhile everyone else was scraping Youtube as much as they liked and training on it.

[hagbard_c]

This comes as no surprise as I suspect many if not most other 'AI video generator' projects are being fed 'content' from Youtube, Vimeo, Rumble and any and all other accessible video sites - where else would they get a wide spectrum of video material to train on?

[superkuh]

Additionally, no one blocks googlebot even though it's being used just as much for LLM/etc AI training as any web spider out there. Too big to block. Too big to not use.

[JohnFen]

genAI videos are already making YouTube worse than it was, and that trend is only starting. Maybe that, plus Google using user videos in this way, will finally allow one of their competitors to gain more traction.

[krunck]

... because Youtube videos meet some minimum level of content quality?

[add-sub-mul-div]

If low quality influencer garbage is what people are watching, they'll be happy to generate more of it and I don't think they'll lose sleep about the quality.

[adzm]

There is just so much of it, on so many different topics. Especially esoteric things that aren't popular "influencer" things that everyone is going to think of initially.

[kube-system]

Most of it is better than this: https://www.youtube.com/watch?v=XQr4Xklqzw8

[leumon]

Just like for llms for the base model you need quantity not quality. It just needs to learn how to correctly predict the next frame.

[greatgib]

It kind of does make sense, like a Library would use the books at its disposal.

But what is not normal is that they will easily block, ban and sue you if you try to do the same, like if the catalog of content was belonging to them.

[kube-system]

> It kind of does make sense, like a Library would use the books at its disposal.

Libraries don't really "use" books to produce anything, except to support accessibility like translations or indexing. Their lending of books is under the first-sale doctrine, which wouldn't be applicable to YouTube videos streamed under license.

> But what is not normal is that they will easily block, ban and sue you if you try to do the same, like if the catalog of content was belonging to them.

Because they do have rights to the content. All of the content on YouTube has been licensed to YouTube, and the licensor has assigned some rights to them.

[echoangle]

Does it not? Do you not give those rights to YouTube once you upload a video?