Curious to see if this will apply to music. YouTube seems to be filled with AI music these days - just do a search for "focus music" or the like, and you'll see creators pushing new 1-hr tracks every few days with no mention of where the music came from or the fact it is AI generated. People praising it in the comments seem none the wiser (or perhaps they're also bots).
I have a hard time believing that AI can be used to label AI-generated videos without there being a significant number of false positives/negatives. I think back to ZeroGPT and it labeling the Declaration of Independence as AI-generated.
When I used to use YT, i used https://untrap.app/, it was a great improvement.
If you think you can't quit youtube, I used to think that way, and then i did it by deleting my account and using libredirect for invidious. my usage went to just a few minutes or 0 per day.
Likewise. The page is youtube.com and then just /feed/ without anything else there. That's the blank page, thank goodness they've not ruined that yet :)
I hope their detector is better than the typical 'AI detection in text' services. False negatives are bad, false positives are worse as some creators could lose their source of income.
It must be a tricky problem to balance. On the one hand, you as Google want people to create 30 seconds of video per month with your cool Omni, Flow, Gemini, etc. tools.
On the other hand, as soon as people share those things on the logical platform for sharing videos, they'll be branded with the scarlet letter.
I wonder what Google is thinking - that people won't mind? That it won't matter? That Omni is just marketing and they don't actually want people to use it?
Google is not a monolith. For all intents and purposes YouTube might as well be a totally different company than Deepmind. Everyone in there own respective google fiefdom is trying to maximize their own metrics.
Maybe they're just going for "disclosure" as in people understanding it's AI, and hopefully mitigating fake news. Don't know if it impacts monetization?
If the video is entertainig I doubt people will mind it's AI. Let's wait and see.
Why are they saying to not distribute on YouTube they just want to give an indicator. Same with labeling if a video is an AD. I find some of the obvious AI content to be funny or informative .
No joke, I would pay for this more than i do for premium.
Half my feed or suggestions are AI crap.
Sadly I have lots of niche interests (various history topics, or engineering) and there's an endless amount of channels and videos that are entirely AI generated.
And it takes you a while to catch them, because they are well done, even the narrator is fake. But then I realize I'm just watching an AI dramatized summary and I get mad at Youtube.
The filter is what is key. If they label all AI videos but still serve me AI slop as the first response, then it doesn't matter if it's labeled at all.
One field I was wondering about. There are a lot of channels/videos where they take movie summaries, feed it into an AI to generate TTS, graphics... I hate these videos but I'm also like damn good job trying to capitalize on that, why don't I do it kind of thing. I don't have that money making drive/hustle. I need to.
The idea that you can automatically detect AI generated content seems misguided. It will make mistakes. I think I've heard of things being wrongfully tagged as AI generated on other platforms.
_before_? youtube is like the top 1000 creators / influencers / celebrities, 5% actual videos, and the rest is slop of various types for me.
The search is entirely gone and will straight up not list regular content that was made by humans, but it will absolutely surface the cheap AI-voices that can't say HVAC fluently.
Makes sense for youtube, too. For the average person, that seems to be enough, the ads are the same value, and people who dislike that probably have a large overlap with people who use adblockers and don't impulsively spend money on the latest browser-game, miracle cure for cancer, or financial advice from random scammers.
If you only visit youtube.com logged out in a private window, obviously it's going to show you what's the most popular. What else should it be doing?
Personally I never see those top 1000 creators / influencers / celebrities. If anything I find the recommendation algorithm too tuned to my topics of interest so it rarely surfaces new things.
Agreed on the search being less than useless, though.
You and I must be watching a very different YouTube. I don't see a lot of AI generated stuff in my recommendations or search.
My teenage daughter gets served a lot of iffy stuff with the ElevenLabs Adam voice though. Though sometimes I suspect some of that content is written manually by people who may not have great English speaking voices, so they add generated audio.
So far all it does to the video is add a small tag in the corner. It doesn't affect rankings or monetization. A false positive might annoy some subscribers at worst.
I know they can identify them because if I click on one by mistake that's all I get until I go to about:blank, close YT tab, clear cache, close browser, run bleachbit and start browser. I never log into their site.
I don’t care about gen AI video content. That’s fine. Saves creators from having to buy b-roll. I appreciate cinematography, but it’s not what I come to YouTube for.
What I absolutely loathe and instantly block is AI narration. That’s an instant deal breaker for me. And it’s gotten to the point that without a shot of the creator or obvious humanisms like microphone sounds, I assume a new creator is AI tts reading an LLM generated script. There are thousands of these channels.
I wonder if they will try to do this for songs in YouTube Music. I've stopped using their auto-generated playlists/recommendations/whatever because it kept playing AI-generated songs.
Maybe they could fix their moderation and appeal process before adding a half-baked feature like this which is certain to cause more issues requiring moderation?
Donning my tinfoil hat for a moment, YouTube is in a position here to simultaneously iterate on automatic AI video detection while also working out how to make AI generated video that's impossible to detect.
I think that "impossible to detect" is not something realistic if camera manufacturers are willing to start adding encryption signatures to their cameras outputs and are willing to vouch for them.
I realize this would still allow fakes to be presented by governments in all likelihood, but not everyone.
Who posts raw output from cameras anywhere? This doesn't seem useful outside some niche use-cases (like security camera footage). At a minimum just about every recording is going to be re-compressed for streaming.
I bet the cameras' companies will start automatically uploading the real footage to their servers for attestation, and allow the camera owners to get those links, so people will just add that link on YouTube or whatever and say "See, its real, Sony vouches for it", heck maybe they will make their buyers to sign up with YouTube and do it for them.
How on top of security do you think all the camera manufacturers are going to be? That is, how long until people can sign videos that were not, in fact, shot with their camera?
Proving that you were able to upload something that is not real would go viral so it's very attractive to people to share such findings, meaning it would not last long, then they fix it and that's it, specially because they can require you to upgrade your camera's firmware if you want to keep using their attestation service.
Attention is valuable these days, so making people go to their websites for people to check if something is real is good for them, its people they can try to sell more cameras (or phones) and all that.
Joking and all but sexting would benefit from this technology, if it can vouch about the time, GPS location and email address of the owner then the receiver can have some certainty about the pic (if the sender decides to share such attestation link/info, of course)
I don't think it needs do be raw output. I'm pretty sure that signatures can exist within image and sound outputs that are reproducible when changing to other formats.
Leica started doing this a few years ago in response to the first wave of AI images[0]. Other, bigger manufacturers (Nikon, Canon, Sony as well I believe) have also joined, though with less fanfare. Adobe is in the loop.
As someone with a passing interest in infosec and cryptography, I'm sceptical of the long-term viability of this kind of product; it only takes one person successfully extracting a signing key to undermine the entire project.
Yes, you're correct about private keys getting exposed, but it's better than nothing. I suspect though, even after key exposure there may be a way to make new private keys so that compromised keys have a known point when they are compromised, which makes public how much skepticism we should all have about authenticity.
I just think there is a world of difference between "certainty" and "plausibility" when it comes to videos on the internet. Yes, state actors might circumvent it, and skepticism should remain, but there is a world of difference between North Korea trying to convince me of some political scandal, and Pepsi Co trying to convince me that someone I trust loves Pepsi.
You still ultimately have the analogue hole here - pull the camera apart, splice your own hardware somewhere between the sensor and the thing that adds the signatures (or in front of the sensor).
I feel it wouldn't be too difficult to get a social-media video to look convincing enough even with just a regular camera and monitor, at least after compression (if end users aren't served raw footage directly, and instead trust the attestation of the site).
Right, my point is that this should default to "untrustworthy." The idea is that a camera would at the very least include a timestamp and camera type in the signature. That signature should usually be reproducible when being filmed by another camera (these signatures can be part of the physical image). This should mean that a cameras filming screens would have multiple ways to show the images are not legitimate (as something as simple of shadows not matching time of day could show the video is illegitimate).
> "while also working out how to make AI generated video that's impossible to detect."
what gives you that impression?
Google is spends millions of dollars researching and implementing SynthID [0]; ensuring all videos generated have a watermark to ensure they can be detected. As well as using SynthID to detect AI-generated videos, which is what I am guessing they are using @ YT to detect and label the AI videos.
I am far from a Google or AI fanboy, closer to an admiring hater, but I just don't see how they are making AI generated video impossible to detect if they are going out of the way to ensure SynthID /AI watermarks are present in any video generated on the platform.
unless you mean impossible to detect by human eyes; but if that's the case, so is everyone else.
except everyone else is NOT spending resources on research for watermarking and keep the detection algorithms ahead of the curve.
Yeah, like Google doesn't know other hundreds of companies are also generating videos and will without the slightest shred of doubt will use reinforced learning to bypass this detection, meaning directly asking Google's AI if a picture they modified is AI or not to improve their algorithms, they know vouching for video is as useful as vouching for AI generated texts, zero.
> “If a creator doesn’t specify whether or not they used AI, but our systems detect significant photorealistic AI use, we will now automatically apply a label,” YouTube said.
> YouTube creators who believe their content was incorrectly flagged as AI-generated can modify the disclosure status using the YouTube Studio tool.
What's the general overall state of AI-based AI flagging tools development? They seemed to have absurd false positive rates of not even 50% while it's obvious to whom it is obvious, no matter who or how it's done.
> Under YouTube’s guidelines, creators will still be required to manually disclose when they use realistic AI. But starting this week, it also will roll out a new internal system to help identify AI-generated content. “If a creator doesn’t specify whether or not they used AI, but our systems detect significant photorealistic AI use, we will now automatically apply a label,” YouTube said.
detect how? synthid is the only obvious one I can think of. user reports would make some sense. But what's the sota for ai detection?
I don't know about SOTA, but Sight Engine (sightengine.com) has AI image detection which seems pretty solid. It can even identify specific image generators.
is this gonna affect the monetization of those videos too?
Well i think even if not directly, people will somehow loose interest in ai generated videos, people would not want a low effort content grabbing there attention.
I can already imagine this won’t be perfect (false negatives / false positives, for one thing) but this seems like a huge step in the right direction. Even just giving the “AI” label a more prominent spot than the description is a big deal, particularly for those who are less tech-savvy than your average HN user. My mom, for instance, can watch your one video that’s entirely AI-generated and not bat an eye, but then watch another video that’s clearly real and say it looks “off.” Say what you will about whether AI-generated content is valid or whether it should be allowed on the platform at all, but more transparency is only a good thing.
I've been thinking for some time that it wouldn't be too hard to create a third-party browser extension to crowdsource detection of channels that use primarily AI-generated content (for example, the AI slop music channels that put out multiple hour+ long genre or cover "playlists") and hide them from suggestions or home feeds.
My guess is that Google sees some kind of trend in a contingent of users preferring non-AI content and that surfacing AI content misleadingly has a negative effect on retention / watch time, and/or they're trying to get ahead of long-standing creators taking issue with the platform surfacing AI content disproportionately on account of it being excessively easier to upload in large quantities.
Honestly, this whole AI-labeling approach seems to be the opposite approach to take. Instead why not authenticate genuine "non-AI photos". Work together with the hardware and software layer with an open approach, building on top of contend id. I appreciate the privacy implications here are complex, and Google is dubious on using any tracking/fingerprinting technology for its self-serving and privacy-invading motivations, but an open cross-industry foundation owning and operating it may be a first step?
Interesting. Although it seems they are focusing primarily on detecting AI generated video and imagery. But most of the annoying slop videos I come across seem like they are using real footage/video clips. It's just edited together by AI and there's an AI narrator reading an AI script. I wonder if they'll do anything to guard against this type of junk
Can YouTube stop shoving terrible robot-English AI dubs down my throat?
I once looked up a German language test. It was auto-AI dubbed into English. Ugggghhhhh..... There are also a lot of anime where the AI dub essentially removes the music and sound effects and leaves only a dreary AI voiceover. It's kinda crazy that Google is pushing this feature out....
Not if you Airplay to your TV. I get random foreign languages when I watch English speaking YouTubers. No way to enable subtitles or change the language. It's a known bug according to the internet.
It doesn't remember my preference. Or rather, it seems to remember me picking a specific language, and then loads the dub in that language next time I click a video. It doesn't remember "don't duh videos".
The problem is that it doesn't even respect this choice. My native language is not English and most of the videos in this languages will be auto transcript to English. Even the last time I changed both the language and country and YouTube still managed to auto transcript to English.
The solution is a simple toggle to turn it off, not pushing it to our throat.
That just tells it which languages to serve if the video has multiple tracks, including Ai-generated ones. "Keep original language" should be the default, or at least an opt-in.
And what about the atrocious title auto-translations? I'm in France, my browser is set to accept EN-us and FR-fr as languages, and my Youtube is in EN. And yet it keeps auto-translating the titles of some French videos. And the translation is so awful, it mistranslates many things and translates literally some obvious puns, that I can't believe they're using Gemini for this. They must have repurposed a 5-year old version of Google Translate. It is not consistent either, the titles are translated in the home page, but not in the channel's page.
Even worse, sometimes it dubs ads, where there's no way to switch the audio track and no way to see if it's being dubbed. This also makes it look like the dubbed audio is the original audio from the ad, which makes the advertiser look terrible.
It's an option that individual channels can disable. Granted, it's opt-out, but YouTube emailed creators several times about it well before:
> Effective today, you can turn off automatic dubbing for your entire channel in your Channel settings > Upload defaults > Advanced settings > Automatic dubbing.
> Once auto-dubbing is enabled for your channel, while uploading a new video, you will also have the option to turn off automatic dubbing for that video.
So if you're seeing auto dubbing on a video by a creator who clearly pays attention to YouTube's algorithm and should be aware of the feature, then they deliberately opted to leave the option on, probably thinking that it can't hurt.
Unfortunately I'm not sure this will affect those at all...it's specifically for "realistic" AI use according to all the quotes. I'm not sure narration or illustration/slides generated by AI would be covered at all in that case.
Finally a decent change by Youtube! Great job Youtube but overall unsure about the situation at Google itself and what Google itself is doing.
I do overall wish if Youtube could've been spinned independent from Google given there might be some conflict of interests, Youtube still tries to push a lot of AI slop towards the creators and sometimes even the viewers perhaps because of google, but seems like Youtube has pushed back against some aspects of the AI slop.
the thing I am wondering is how easy it might be to break that bypass and also about the false positives. A lot of creators recently got demonitized for apparently not much of a reason aside from false positives which is incredibly sad if one's livelihood depended on it. These people end up taking it on twitter from my understanding but it only really sometimes end up working if enough people watch the twitter or get attention overall on the topic so I hope that youtube works towards its (creators support??) side too.
Isn’t YouTube applying weird AI processing to shorts?
So all shorts will be labeled?
Maybe I’m not the target audience for Google products anymore?
I have to use Yandex and DDG for search results now.
Gemini has insane throttling so I’ve just embraced local models for most things and the occasional API call to whatever frontier model I think will work best.
YouTube search is abysmal and new content is 98% consumerism BS.
My Gmail is mostly spam and mailing lists I can’t seem to get off of with the occasional scam attempt thrown in.
Guess I’m just ranting to rant at this point. I grew up online and now the internet feels weird and I think I might be “over it”.
This could backfire.. im thinking of "real" videos with elements of AI in them. Those elements might not get the video flairs as an AI video and people will get fooled
Curious to see if this will apply to music. YouTube seems to be filled with AI music these days - just do a search for "focus music" or the like, and you'll see creators pushing new 1-hr tracks every few days with no mention of where the music came from or the fact it is AI generated. People praising it in the comments seem none the wiser (or perhaps they're also bots).
I have a hard time believing that AI can be used to label AI-generated videos without there being a significant number of false positives/negatives. I think back to ZeroGPT and it labeling the Declaration of Independence as AI-generated.
I suggest turning off recommendation if you dislike what they suggest
My YT landing page is completely blank and need to go "subscription" tab to see newly uploaded vids from the ones I subscribe to
It's quite nice not having to view all kinds of random stuff YT wants me to see
When I used to use YT, i used https://untrap.app/, it was a great improvement.
If you think you can't quit youtube, I used to think that way, and then i did it by deleting my account and using libredirect for invidious. my usage went to just a few minutes or 0 per day.
A nicer way that also works on mobile is turning the watch history off. C
Same for me. Godsend. It also switches off showing me a follow up "short" after I watch the one I want to watch.
This doesn't help when searching. I'm looking for specific things as often as I'm clicking on recommendations.
What's needed is a global "Hide AI Dreck". Particularly the overwhelming hoard of AI slide-shows masquerading as reviews.
Likewise. The page is youtube.com and then just /feed/ without anything else there. That's the blank page, thank goodness they've not ruined that yet :)
This ^
Good start, but it seems you still need to click on the video though.
I hope their detector is better than the typical 'AI detection in text' services. False negatives are bad, false positives are worse as some creators could lose their source of income.
It must be a tricky problem to balance. On the one hand, you as Google want people to create 30 seconds of video per month with your cool Omni, Flow, Gemini, etc. tools.
On the other hand, as soon as people share those things on the logical platform for sharing videos, they'll be branded with the scarlet letter.
I wonder what Google is thinking - that people won't mind? That it won't matter? That Omni is just marketing and they don't actually want people to use it?
[delayed]
Google is not a monolith. For all intents and purposes YouTube might as well be a totally different company than Deepmind. Everyone in there own respective google fiefdom is trying to maximize their own metrics.
Maybe they're just going for "disclosure" as in people understanding it's AI, and hopefully mitigating fake news. Don't know if it impacts monetization?
If the video is entertainig I doubt people will mind it's AI. Let's wait and see.
Why are they saying to not distribute on YouTube they just want to give an indicator. Same with labeling if a video is an AD. I find some of the obvious AI content to be funny or informative .
That's great news. Hopefully there will be a filter to allow or disallow AI video on your homepage/feed.
I really doubt that google will implement that filter. But I guarantee it'll be added soon to revanced and other patched youtube apps.
They could use the same data to pay AI posters less and push their content more. Which will get you a promotion at Google.
Sounds like a premium feature!
Honestly, that might get me to subscribe.
Yes, like filtering out shorts.
/s
How about ones making a stupid video of SO question/answers
Can't await this checkbox
Create your own browser extension to block them!
Some searches might have 99% videos I want filtered out.
No joke, I would pay for this more than i do for premium.
Half my feed or suggestions are AI crap.
Sadly I have lots of niche interests (various history topics, or engineering) and there's an endless amount of channels and videos that are entirely AI generated.
And it takes you a while to catch them, because they are well done, even the narrator is fake. But then I realize I'm just watching an AI dramatized summary and I get mad at Youtube.
The filter is what is key. If they label all AI videos but still serve me AI slop as the first response, then it doesn't matter if it's labeled at all.
One field I was wondering about. There are a lot of channels/videos where they take movie summaries, feed it into an AI to generate TTS, graphics... I hate these videos but I'm also like damn good job trying to capitalize on that, why don't I do it kind of thing. I don't have that money making drive/hustle. I need to.
Some are funny some SORA, Neural Viz
The idea that you can automatically detect AI generated content seems misguided. It will make mistakes. I think I've heard of things being wrongfully tagged as AI generated on other platforms.
I suppose it all depends on the false positive and false negative rates. But better to start now, before AI ruins the platform.
https://web.archive.org/web/20090418141450/http://www.theatl...
_before_? youtube is like the top 1000 creators / influencers / celebrities, 5% actual videos, and the rest is slop of various types for me.
The search is entirely gone and will straight up not list regular content that was made by humans, but it will absolutely surface the cheap AI-voices that can't say HVAC fluently.
Makes sense for youtube, too. For the average person, that seems to be enough, the ads are the same value, and people who dislike that probably have a large overlap with people who use adblockers and don't impulsively spend money on the latest browser-game, miracle cure for cancer, or financial advice from random scammers.
If you only visit youtube.com logged out in a private window, obviously it's going to show you what's the most popular. What else should it be doing?
Personally I never see those top 1000 creators / influencers / celebrities. If anything I find the recommendation algorithm too tuned to my topics of interest so it rarely surfaces new things.
Agreed on the search being less than useless, though.
You and I must be watching a very different YouTube. I don't see a lot of AI generated stuff in my recommendations or search.
My teenage daughter gets served a lot of iffy stuff with the ElevenLabs Adam voice though. Though sometimes I suspect some of that content is written manually by people who may not have great English speaking voices, so they add generated audio.
Fair enough
So far all it does to the video is add a small tag in the corner. It doesn't affect rankings or monetization. A false positive might annoy some subscribers at worst.
No one will click a video that has the ai tag though.
I know they can identify them because if I click on one by mistake that's all I get until I go to about:blank, close YT tab, clear cache, close browser, run bleachbit and start browser. I never log into their site.
I don’t care about gen AI video content. That’s fine. Saves creators from having to buy b-roll. I appreciate cinematography, but it’s not what I come to YouTube for.
What I absolutely loathe and instantly block is AI narration. That’s an instant deal breaker for me. And it’s gotten to the point that without a shot of the creator or obvious humanisms like microphone sounds, I assume a new creator is AI tts reading an LLM generated script. There are thousands of these channels.
I wonder if they will try to do this for songs in YouTube Music. I've stopped using their auto-generated playlists/recommendations/whatever because it kept playing AI-generated songs.
AI-generated music should be hard to detect
we had to say that our launch video today was AI-generated lol: https://www.youtube.com/watch?v=-7rORkN0nVM
Cool!
Maybe they could fix their moderation and appeal process before adding a half-baked feature like this which is certain to cause more issues requiring moderation?
Donning my tinfoil hat for a moment, YouTube is in a position here to simultaneously iterate on automatic AI video detection while also working out how to make AI generated video that's impossible to detect.
I think that "impossible to detect" is not something realistic if camera manufacturers are willing to start adding encryption signatures to their cameras outputs and are willing to vouch for them.
I realize this would still allow fakes to be presented by governments in all likelihood, but not everyone.
Who posts raw output from cameras anywhere? This doesn't seem useful outside some niche use-cases (like security camera footage). At a minimum just about every recording is going to be re-compressed for streaming.
Synthid and the like survive compression and decent quality rerecording.
I bet the cameras' companies will start automatically uploading the real footage to their servers for attestation, and allow the camera owners to get those links, so people will just add that link on YouTube or whatever and say "See, its real, Sony vouches for it", heck maybe they will make their buyers to sign up with YouTube and do it for them.
How on top of security do you think all the camera manufacturers are going to be? That is, how long until people can sign videos that were not, in fact, shot with their camera?
Proving that you were able to upload something that is not real would go viral so it's very attractive to people to share such findings, meaning it would not last long, then they fix it and that's it, specially because they can require you to upgrade your camera's firmware if you want to keep using their attestation service.
Depends on what kind of compromise occurs. Hardware level key loss isn't easy, if possible at all to fix.
Only if you're paying them
Attention is valuable these days, so making people go to their websites for people to check if something is real is good for them, its people they can try to sell more cameras (or phones) and all that.
They can attest pictures of my hairy pendulous ballsack.
Joking and all but sexting would benefit from this technology, if it can vouch about the time, GPS location and email address of the owner then the receiver can have some certainty about the pic (if the sender decides to share such attestation link/info, of course)
I don't think it needs do be raw output. I'm pretty sure that signatures can exist within image and sound outputs that are reproducible when changing to other formats.
Yeah I’m not sure this makes sense when images are getting their third ifunny watermark.
Leica started doing this a few years ago in response to the first wave of AI images[0]. Other, bigger manufacturers (Nikon, Canon, Sony as well I believe) have also joined, though with less fanfare. Adobe is in the loop.
As someone with a passing interest in infosec and cryptography, I'm sceptical of the long-term viability of this kind of product; it only takes one person successfully extracting a signing key to undermine the entire project.
Yes, you're correct about private keys getting exposed, but it's better than nothing. I suspect though, even after key exposure there may be a way to make new private keys so that compromised keys have a known point when they are compromised, which makes public how much skepticism we should all have about authenticity.
I just think there is a world of difference between "certainty" and "plausibility" when it comes to videos on the internet. Yes, state actors might circumvent it, and skepticism should remain, but there is a world of difference between North Korea trying to convince me of some political scandal, and Pepsi Co trying to convince me that someone I trust loves Pepsi.
Cameras have a very long lifespan. People will still be using those cameras 20 years after the keys for their model get leaked.
And they will also get firmware updates.
You still ultimately have the analogue hole here - pull the camera apart, splice your own hardware somewhere between the sensor and the thing that adds the signatures (or in front of the sensor).
Pointing the camera at a screen could potentially evade that.
Right, but my point is that a video of a screen should be less believable than the source video insofar as verifying legitimacy.
I feel it wouldn't be too difficult to get a social-media video to look convincing enough even with just a regular camera and monitor, at least after compression (if end users aren't served raw footage directly, and instead trust the attestation of the site).
Right, my point is that this should default to "untrustworthy." The idea is that a camera would at the very least include a timestamp and camera type in the signature. That signature should usually be reproducible when being filmed by another camera (these signatures can be part of the physical image). This should mean that a cameras filming screens would have multiple ways to show the images are not legitimate (as something as simple of shadows not matching time of day could show the video is illegitimate).
What if you can't tell it's a video of a screen?
wouldn't that just encourage monopolistic behavior and lockdown of these devices?
they're already locked down as-is.
Isn't this literally how GANNs are trained?
"AI generated video that's impossible to detect" is already something many companies are working on; it's hardly Google-specific.
YouTube scale is Google specific
It already is quite impossible. Just generate something decent with lower quality. Then maybe take screen recording of the output. Voila.
Then "low-quality for no reason in year the 2026 and beyond where phones shoot at 8k" become part of the heuristics.
Would be a fairly weak heuristic, with most social media images/videos already being like that.
Being pedantic for no reason is one of the heuristics I use to judge how annoying yn users are
I find it super ironic that we're basically here: https://xkcd.com/1683/ now. The 90s and 00s tech people would be very disappointed :-))
> "while also working out how to make AI generated video that's impossible to detect."
what gives you that impression?
Google is spends millions of dollars researching and implementing SynthID [0]; ensuring all videos generated have a watermark to ensure they can be detected. As well as using SynthID to detect AI-generated videos, which is what I am guessing they are using @ YT to detect and label the AI videos.
I am far from a Google or AI fanboy, closer to an admiring hater, but I just don't see how they are making AI generated video impossible to detect if they are going out of the way to ensure SynthID /AI watermarks are present in any video generated on the platform.
unless you mean impossible to detect by human eyes; but if that's the case, so is everyone else. except everyone else is NOT spending resources on research for watermarking and keep the detection algorithms ahead of the curve.
Yeah, like Google doesn't know other hundreds of companies are also generating videos and will without the slightest shred of doubt will use reinforced learning to bypass this detection, meaning directly asking Google's AI if a picture they modified is AI or not to improve their algorithms, they know vouching for video is as useful as vouching for AI generated texts, zero.
Why is that tinfoil? That's just good business?
> Under YouTube’s guidelines, creators will still be required to manually disclose when they use realistic AI. But starting this week, it also will roll out a new internal system to help identify AI-generated content. “If a creator doesn’t specify whether or not they used AI, but our systems detect significant photorealistic AI use, we will now automatically apply a label,” YouTube said.
detect how? synthid is the only obvious one I can think of. user reports would make some sense. But what's the sota for ai detection?
I don't know about SOTA, but Sight Engine (sightengine.com) has AI image detection which seems pretty solid. It can even identify specific image generators.
is this gonna affect the monetization of those videos too? Well i think even if not directly, people will somehow loose interest in ai generated videos, people would not want a low effort content grabbing there attention.
One advantage is that if it's labelled as AI, we don't need to have a conversation in the comments about whether it's AI or not.
Let’s use probabilistic models to find the probability of something being the output of another probabilistic model
I wonder why they're really doing this. It's definitely not for users' benefit.
This is awesome. I am building something similar for writing - https://trulytyped.com
I'm willing to bet this is just an easily bypassable SynthID check
The dangers is videos that slip through the cracks, they get an indirect seal of being non AI.
I can already imagine this won’t be perfect (false negatives / false positives, for one thing) but this seems like a huge step in the right direction. Even just giving the “AI” label a more prominent spot than the description is a big deal, particularly for those who are less tech-savvy than your average HN user. My mom, for instance, can watch your one video that’s entirely AI-generated and not bat an eye, but then watch another video that’s clearly real and say it looks “off.” Say what you will about whether AI-generated content is valid or whether it should be allowed on the platform at all, but more transparency is only a good thing.
I've been thinking for some time that it wouldn't be too hard to create a third-party browser extension to crowdsource detection of channels that use primarily AI-generated content (for example, the AI slop music channels that put out multiple hour+ long genre or cover "playlists") and hide them from suggestions or home feeds.
My guess is that Google sees some kind of trend in a contingent of users preferring non-AI content and that surfacing AI content misleadingly has a negative effect on retention / watch time, and/or they're trying to get ahead of long-standing creators taking issue with the platform surfacing AI content disproportionately on account of it being excessively easier to upload in large quantities.
Would really be nice if they did the same with their ads, but don't see that happening
The ads are labeled, much more so than the AI generated content.
Honestly, this whole AI-labeling approach seems to be the opposite approach to take. Instead why not authenticate genuine "non-AI photos". Work together with the hardware and software layer with an open approach, building on top of contend id. I appreciate the privacy implications here are complex, and Google is dubious on using any tracking/fingerprinting technology for its self-serving and privacy-invading motivations, but an open cross-industry foundation owning and operating it may be a first step?
That's basically C2PA: https://c2pa.org.
I'm not super optimistic about it, and last I saw, Apple wasn't a part of it either.
this is a welcome change but if the creator doesn't disclose the use of AI, how do they detect what is AI and what is not?
Interesting. Although it seems they are focusing primarily on detecting AI generated video and imagery. But most of the annoying slop videos I come across seem like they are using real footage/video clips. It's just edited together by AI and there's an AI narrator reading an AI script. I wonder if they'll do anything to guard against this type of junk
AI versus AI, the final faceoff. Who's gonna win? Probably not us.
Whoever wins, we lose.
Also the amount of scammy crap quality on YouTube has exploded since developing countries have more access. The cost of publishing is tending to zero.
This is fine, good, whatever... but my thing is can creators remove it successfully for 'false flags'.
> However, according to YouTube, the AI labels will “remain permanent” in some cases,
YouTube isn't exactly known for taking care of complaints/having any human on the other end to deal with these kinds of things.
Now label AI ads and let us filter them out.
Leading up to tax day, every ad was a terrible AI slop Turbotax ad.
I really wish there was a button to voluntarily say / tag your own content as AI assisted.
The assumption that users will always hide this results in flaky auto detection.
Your wish has been granted - https://support.google.com/youtube/answer/14328491
Can YouTube stop shoving terrible robot-English AI dubs down my throat?
I once looked up a German language test. It was auto-AI dubbed into English. Ugggghhhhh..... There are also a lot of anime where the AI dub essentially removes the music and sound effects and leaves only a dreary AI voiceover. It's kinda crazy that Google is pushing this feature out....
There's no way to switch back to the original audio track? I agree you shouldn't have to but I'm wondering if it's possible.
Not if you Airplay to your TV. I get random foreign languages when I watch English speaking YouTubers. No way to enable subtitles or change the language. It's a known bug according to the internet.
You can change the audio track back to the original.
Not sure if it remembers your preference, though, so if it doesn't that probably grates.
It doesn't remember my preference. Or rather, it seems to remember me picking a specific language, and then loads the dub in that language next time I click a video. It doesn't remember "don't duh videos".
It does remember it. At least on Firefox/Linux.
There are several Chrome extensions for turning that off automatically, but I agree: you should not have to need extensions to use YouTube.
You can configure your preferred languages in YT settings, so it doesn’t do that. The setting is obscure, but it’s there
The problem is that it doesn't even respect this choice. My native language is not English and most of the videos in this languages will be auto transcript to English. Even the last time I changed both the language and country and YouTube still managed to auto transcript to English.
The solution is a simple toggle to turn it off, not pushing it to our throat.
That just tells it which languages to serve if the video has multiple tracks, including Ai-generated ones. "Keep original language" should be the default, or at least an opt-in.
And what about the atrocious title auto-translations? I'm in France, my browser is set to accept EN-us and FR-fr as languages, and my Youtube is in EN. And yet it keeps auto-translating the titles of some French videos. And the translation is so awful, it mistranslates many things and translates literally some obvious puns, that I can't believe they're using Gemini for this. They must have repurposed a 5-year old version of Google Translate. It is not consistent either, the titles are translated in the home page, but not in the channel's page.
I don’t know that Settings -> languages is obscure
try this. https://gist.github.com/Buildstarted/c6b156ec95dc012c5f3dc5e...
Even worse, sometimes it dubs ads, where there's no way to switch the audio track and no way to see if it's being dubbed. This also makes it look like the dubbed audio is the original audio from the ad, which makes the advertiser look terrible.
It's an option that individual channels can disable. Granted, it's opt-out, but YouTube emailed creators several times about it well before:
> Effective today, you can turn off automatic dubbing for your entire channel in your Channel settings > Upload defaults > Advanced settings > Automatic dubbing.
> Once auto-dubbing is enabled for your channel, while uploading a new video, you will also have the option to turn off automatic dubbing for that video.
So if you're seeing auto dubbing on a video by a creator who clearly pays attention to YouTube's algorithm and should be aware of the feature, then they deliberately opted to leave the option on, probably thinking that it can't hurt.
Settings -> languages and then add German as one of the languages you know and it’ll never do this again
Thank fuck. There is SO much garbage on YT lately which amounts to a powerpoint deck with ai audio overlaid.
Unfortunately I'm not sure this will affect those at all...it's specifically for "realistic" AI use according to all the quotes. I'm not sure narration or illustration/slides generated by AI would be covered at all in that case.
"Please prove your content was created by a flawed biological organism."
Seems better than the alternative: "This content was created by a machine, but being pushed/promoted by a flawed money-grubbing biological organism".
Finally a decent change by Youtube! Great job Youtube but overall unsure about the situation at Google itself and what Google itself is doing.
I do overall wish if Youtube could've been spinned independent from Google given there might be some conflict of interests, Youtube still tries to push a lot of AI slop towards the creators and sometimes even the viewers perhaps because of google, but seems like Youtube has pushed back against some aspects of the AI slop.
the thing I am wondering is how easy it might be to break that bypass and also about the false positives. A lot of creators recently got demonitized for apparently not much of a reason aside from false positives which is incredibly sad if one's livelihood depended on it. These people end up taking it on twitter from my understanding but it only really sometimes end up working if enough people watch the twitter or get attention overall on the topic so I hope that youtube works towards its (creators support??) side too.
Original article: https://blog.youtube/news-and-events/improving-ai-labels-vie...
This should be the original link since its the official announcement and has more concrete info, hope mod(s) can update.
Ok, updated above with the other link in the toptext.
Isn’t YouTube applying weird AI processing to shorts?
So all shorts will be labeled?
Maybe I’m not the target audience for Google products anymore?
I have to use Yandex and DDG for search results now.
Gemini has insane throttling so I’ve just embraced local models for most things and the occasional API call to whatever frontier model I think will work best.
YouTube search is abysmal and new content is 98% consumerism BS.
My Gmail is mostly spam and mailing lists I can’t seem to get off of with the occasional scam attempt thrown in.
Guess I’m just ranting to rant at this point. I grew up online and now the internet feels weird and I think I might be “over it”.
This could backfire.. im thinking of "real" videos with elements of AI in them. Those elements might not get the video flairs as an AI video and people will get fooled