YouTube trying to combat fake/AI content?

[CentreMetre]

I've uploaded videos to YouTube as little as 2 months ago and didnt see this (image attached), so it must be recent. Anyone know when they did add it? I wonder if it was because of the recent AI video made by (i think) open AI. Also the wording 'seems real' can be kinda subjective, something can look real but *seem* off and not real if you get what i mean. They specify Real events, and video *and* sound so i think theyre trying to get ahead of any fake content, especially befor the US 2024 elections aswell.

 

[Kisai]

It's very recent, like in the last two weeks.

 

[podkall]

4 minutes ago, Kisai said:

It's very recent, like in the last two weeks.

 

so they're gonna combat AI content and not AI image using, identity-theft violating scam ads?

[CentreMetre]

Also just thought, i wonder if this will affect YouTubers like kwebblekop who use AI instead of actually being in a video? At least he did do that, not sure if he still does, but there are likely 'creators' that do.

[Mark Kaine]

On 3/28/2024 at 1:29 AM, CentreMetre said:

Also just thought, i wonder if this will affect YouTubers like kwebblekop who use AI instead of actually being in a video? At least he did do that, not sure if he still does, but there are likely 'creators' that do.

who?

 

On 3/28/2024 at 1:07 AM, Kisai said:

It's very recent, like in the last two weeks.

 

does it only come up with certain content?  because im not getting this for vidoegame uploads and as we all know video games look *extremely* realistic nowadays!

 

 

 

On 3/28/2024 at 1:05 AM, CentreMetre said:

They specify

reminds me of when i visited nyc, i had to fill out a form and one of the first questions was "do you plan a terrorist attack?"

y/n ...

 

being a kid i was really flabbergasted as to how and why anyone who does would admit this... followed by thinking "americans are really stupid apparently", "is this a joke...?" and "what happens if i say 'yes'?"  (my mom immediately pointed out that that wouldn't be a good idea tho )

[CentreMetre]

6 minutes ago, Mark Kaine said:

who?

https://www.youtube.com/@kwebbelkop

[Kisai]

45 minutes ago, CentreMetre said:

Also just thought, i wonder if this will affect YouTubers like kwebblekop who use AI instead of actually being in a video? At least he did do that, not sure if he still does, but there are likely 'creators' that do.

I think the intent is to prevent AI "misinformation", eg fake political ads, fake disasters, etc.

 

I'm OK with this, but this isn't nearly good enough. There should be an entire category of "AI Technology", and where the "game category" would normally exist under gaming, it would instead have tick boxes of:

(o) This video features AI generated material (100%)

( ) This video uses some AI generated material

[x] Visuals from (timestamp) to (timestamp) (add more)

[  ] Audio from (timestamp) to (timestamp) (add more)

[  ] Text/Script from (timestamp) to (timestamp) (add more)

[________] Upload metadata*

 

* metadata would include hashes of the model, inference script/program, and prompt data. This is typically the JSON file used to generate the AI result. This would allow youtube to identify what is being used. This would have the additional benefit of embarrassing the AI generated farmed content, because it would clearly show what media they've stolen to generate the output.

 

[Kisai]

7 minutes ago, Mark Kaine said:

does it only come up with certain content?  because im not getting this for vidoegame uploads and as we all know video games look *extremely* realistic nowadays!

 

 

It's definitely there. 

 

[OhioYJ]

I think this is just YouTube wanting to look like they are trying to do something about it. I highly doubt they care at all about it, as all they care about is views and advertising money. They don't care how it happens. 

[da na]

This is a change they've been rolling out over the past 2 weeks or so, requiring AI videos to be flagged. It is fantastic, but there seems to be no punishment for the creator for not complying. I don't "create" AI content so I am not 100% sure, but on a platform as large as YouTube, actually enforcing this rule will be very difficult.

[05032-Mendicant-Bias]

This is good.

 

Here Google is just making it look like it's doing something without actually do anything, but hopefully it heralds regulations will come in force to disclose manipulated content by any means, and in what amount, and prescribe penalty for non disclosure. just like for advertising, it's the non disclosure that's the big problem.

 

Not that it will stop channel being hijacked to run a musk livestream with a crypto scam... I can't fathom why the Content ID system can't spot the same livestream uploaded at infinitum...

[TetraSky]

How about all those youtube channel that takes content off Reddit and just have a TTS voice over to read it? There's tools to do all this automatically with little to no inputs from the channel owner. Would that be included since "realistic sound" could include the rudimentary AI powered TTS voice over we have these days or would it fall into the crack since it technically isn't "altered content" and is just stolen content?

[Needfuldoer]

11 hours ago, Kisai said:

I think the intent is to prevent AI "misinformation", eg fake political ads, fake disasters, etc.

I'm sure the people who upload said content will answer this form truthfully.

 

/s

[Kisai]

Just now, Needfuldoer said:

I'm sure the people who upload said content will answer this form truthfully.

 

/s

Far more likely that everyone will ignore it until there's some actual policy with teeth on it.

 

Cause I swear the amount of "elon shilling crypto" ads I've reported to youtube, along with the various other fake ads ain't bothering with it.

 

The real target is stuff these videos mention:

https://www.youtube.com/watch?v=qSS6t1ej1Jk

 

It's not a secret, and many of the "AI content farms" just steal a popular video from another site, let chatGPT rewrite the script, and then munge the visuals with AI.

[Spotty]

*tin foil hat*

YouTube is only introducing this so when they use YouTube videos to train Google's own video generation AI it allows them to easily filter out AI generated content that may be less desirable as training data.

[CentreMetre]

Also the option is kinda hidden in a sub menu, so im not sure if its even required that you fill it out. I assume it is set as not ai by default if not filled out.