YouTube’s system for suppressing low-quality AI content mistakenly targeted Kurzgesagt, leaving the channel with its worst-performing upload since 2013.
Yeah, it’s bad.
Non-watermarked slop is going to be impossible to detect going forward, and the false detections are going to get worse.
The fundamental issue is the monetization scheme. It’s just too lucrative for spammers. YouTube’s economics are broken, and getting worse.
…Hence, no matter what YouTube says, YouTube is going to double down on “detectors,” because Alphabet leadership couldn’t care less about destroying longstanding creators. Their only objective is to suppress just enough spam to keep ad buyers unconcerned next quarter, no matter the long term damage; that’s someone else’s problem.
The best solution would be to turn away from detectors, which will always be a losing game, and return to crowdsourced curation through users likes and ratings.
But that would end up with too many of the corporate propaganda videos getting tanked with dislikes so YouTube will never go for that.
For every ten users who decry AI voices and animation, a thousand more will blindly click “Like” and “Subscribe”. The people have spoken, and they want to eat straight from the fertilizer pile.
People only click like and subscribe because their other options were taken from them. They got tired of the people speaking up so removed the options to do so.
I’ve run my own anecdotal and admittedly small blind trials on friends and family. Most people unfortunately preferred the A.I. slop to real human content. My theory is because it tends to be bright and shiny. The only actual study I could find online, told participants which is which, kind of defeating the purpose in my opinion.
It’s engagement optimized, that’s why.
That’s the whole point of the spam; being engagement bait.
Same thing applies to any human made clickbait, and people have demonstrated they simply cannot help themselves, and click it.
There was recently a study using human short stories and LLM generated short stories prompted to write a similar story. The participants where told which stories were AI generated, but sometimes they were lied to, and told a human-written story was AI generated.
The study showed that people liked the AI stories more. It also showed that if someone has a negative bias against AI generated content, they’d claim the AI story was worse, even if it were actually the human story and they were just told it was AI.
Also, another test was performed where they provided a story and asked them if it was AI generated or human generated. The participants could only tell the AI written story 45% to 50% of the time. Essentially, they were guessing.
I think many people have taken a mental snapshot of the quality of AI generated content from a couple years ago and just assumed it hasn’t improved since then.
I feel It is less that AI content is difficult to detect and more that the AI technology Youtube is using to detect it is complete trash.
I respectfully disagree.
Have you seen what Minimax H3 can output, on a desktop? Even dirt-cheap AI video gen is basically undetectable now. I think all YouTube can filter is the absolute “bottom barrel” spammers still using some old model because they don’t know any better.
And if they train detectors on “stereotypical” AI output like Ghibli style or flat colors or whatever, they’re just going to nuke channels like Kurzgesagt while most AI slop spammers adapt to get through. There’s literally nothing for detectors to go on anymore.
…But also, YT doesn’t even care.
The filters make next quarter look better because they do catch the bottom barrel, and that’s all that matters.
If the slop videos get through, the average user won’t see the difference either, so why should they care?
There’s literally nothing for detectors to go on anymore.
It’s harder for users as well. Even youtubers I’ve followed for a while have started using the recipe of having a ridculous thumbnail such as “WHY YOU SHOULD NEVER BUY THIS” with an arrow pointing to something. After a couple of times of catching me because I own one of the things the arrow is pointing to (and then watching a 10:01 long video not talking about the item in question) I decided not to fall for it. I noticed that discovery of a new channel got hard. They’re all following the recipe now since I guess youtube mandates it, so it’s the same bas relief thumbnails that seem like they pose a question and a video that doesn’t answer it. We’ll end up with a new Betteridge’s Law and the resource youtube once was is lost forever and we’ll stand around making dumb faces thinking how we’re all getting ripped off…
Seems the same pattern at X, probably because they hired in an ex-YouTuber at the top. Moderation algorithms are universally very poorly written and don’t appear to have gotten better in the last 6-8 years when they began using them to replace human moderators. On X they seem to have a timer for new accounts that puts a fake label on accounts as if they’re spamming or a bot, when it’s just a real user with opinions the algorithm is trained to dislike. Same exact crap the old Twitter did. They put the source code on the web and it showed how if you posted or replied on any of a broad range of topics, your account was scored lower until you were deemed low quality and made invisible. It’s like they just wanted cat memes and recipes from elderly women, but not many of them on Twitter or X. Probably because they hired in people from companies where non-controversial topics were a much larger share of what was discussed there. 1 trick ponies who weren’t suitable for doing the same job at X.
There’s no way to boost ad views or MAU with that approach. I believe X will never be profitable due to those people running the show and Musk not ever hiring anyone a lot better to run X. Musk had a similarly bad experience with Solar city. Another company he didn’t know enough about to manage himself and the people he hired to do it weren’t good enough either. If YouTube follows their lead, and continues chasing off creators and users, then they will continue losing out to Rumble and Twitch. Neither of them would be relevant or high growth if not for YouTube “moderation” starting down this road shortly before the pandemic began.
Youtube and X is basically a private monopolies. They only care about the profit margins. Any and all extra expenditure to R&D or consumer satisfaction is just a cut to their profits today and since they don’t have any real competition, so why raise quality of service above the bare minimum? Their only real selling point is that everybody uses them, so everybody will have to keep using them, if they want to be seen or access most things. they don’t care if they serve slop, because you and me and the wider masses are going to eat it up anyway.
X loses money so no margins there. Twitter has never made a profit in its history that I’ve ever heard of. Facebook’s on the way out. 20 years ago it was college age users. 10 years ago it was people over 50. Now it’s a lot of commercial accounts posting as they do on other basically dying or dead platforms. The only sign of life is streaming video and chats related to streaming video. YouTube has been losing users and content creators to both Rumble and Twitch. The right 1/3 of America is on Rumble while Twitch took the gamers. Youtube still has business but they basically gave away enough business to make those 2 relevant.
Alphabet’s dependent on Youtube, cloud services, web search, and related ads to pay for the unprofitable parts of the business, which appears to be where the bulk of their employees work, rather than at the earnings-generators. If they screw up youtube, then they will have to lay a lot of the people in the fun and interesting jobs off, along with the many layers of bureaucracy and non-coders working in jobs that nobody’s quite sure what they do there. They can’t really afford to screw up youtube any more than they have. It can’t merely do well. It has to do great to pay all of their bills. Very similar to the action-hero movies at the movie studio paying for all the unprofitable ones that most people work on
Using AI to detect AI has, time and time again, proven to be a broken idea. Here is Google with a genius idea: use AI to detect AI, to prevent AI videos.
Keep in mind folks, this is a global top-three company in terms of revenue.
…while offering video making AI tools.
When all you have is a hammer, and you’ve bought all the hammers in the entire world, through the end of next year,
This is also a company pushing AI
Yet YouTube Music is absolutely rife with AI bullshit. So much fake-ass viking, cowboy, or other ‘badass’ AI generated slop.
thats why im always hitting the report button. be part of the solution. not passively part of the problem.
I would say that being part of the solution is not using their service at all.
I’ve been craving a feature like that. Maybe it’s because I use YouTube Music Desktop, but there’s no damn report button!
Nothing in the mobile app, either ☹️.
As someone who likes Tyr, Hulkoff, Skald, and Danheim I swear to fuck I want to beat someone’s head in with my musket stock. Seriously the quality just drops and I can only block so many channels, half tempted to just to download all their music and make a fucked up mega playlist on my phone.
Anyone know a good way to download YouTube music on Linux, preferably a program and not a website.
Up until a few months ago, I would normally use spotDL, a tool you gave a Spotify url and it’d search the same songs on multiple sites and download them, while tagging it with spotify’s metadata, which is quite clean. Absolutely amazing piece of software helped me build a good music library, but Spotify changed their api auth because of Anna’s archive scraping, requiring you to have premium just to have a useful api key, breaking spotdl for pretty much everyone.
I currently use just yt-dlp to pull from yt music, SoundCloud and bandcamp. It’s alright, metadata is bad tho, yt-dlp doesn’t tag very well out of the box. I’ve tried torrenting and soulseek, you can get some good quality (and big!) files, but can be hard to find niche artists (Argentinian hardbass sure is niche as hell)
I’d too be interested in a tool like that if anyone knows one.
Or “just discovered” 1970s Berlin krautrock space stuff… it’s all nonsense
Vacuum up the internet, provide generative “AI” for people to spew tons of garbage on to the internet. Everyone hates AI, and advertisers don’t want ads showing against AI slop. AI can’t accurately detect AI and kills real content providers that advertisers want to show ads for.
Brilliant. What will they come up with next?
I was recently accused on Lemmy of using AI to write up a post (laced with links to appropriate Wikipedia articles).
I’m still not sure how I feel about it.
I’m still not sure how I feel about it.
That’s because machines don’t have feelings. /s
Oh you used a big word, AI phony!
Nah dude, I just . . . know how to write?
real human intelligence seems utterly alien in this world
This comment aims to illuminate the complexities of being seen as a sesquipedalian. Let’s delve dee … (token limit reached)
It really doesn’t help that in the US there’s a major literacy crisis.
This is the shit state of the world now. Everything gets accused of being AI by someone. The AI tech bros have well and truly won. They’ve poisoned everything.
I’d argue that the Fuck_AI crowd who slanders and also sexually harasses people are helping them. I hate this big tech companies just as much as they do. I’m not going to sexually harass someone I think is using their service though.
Did you use an emdash
I do use them, yes. Sometimes I’ll use them similar to parenthesis to insert an aside comment – a use case noted in Wikipedia. – though I add spaces before and after each one.
Huh — Now that you say it.
No, I did not- 🤖
This is exactly why I hold anti-AI folk in low regard, because they have yet to demonstrate their “detection” abilities to discern between human and generated creations. They are basically psychics and ghost whisperers.
It’s worse than that. They’re not just pretending about their abilities for clout. They’re actively following their emotions and attacking people they don’t like with AI accusations or just plain old harassment. I’ve been sexually harassed by these assholes.
This seems like an overgeneralization. Just like they do, you do here.
Reminds me of that comic the other day
Kurzgesagt: Let’s keep our art human made.
Google: Kurzgesagt, your content looks like this low quality AI slop!
🤖: I’D LIKE TO THANK KURZGESAGT FOR BEING A BIG INSPIRATION TO ME.
is Youtube/Google speed running their death?
At the same time YouTube offers a prompt to generate more AI content.
No, this is just a stunt to claim that it is difficult to detect AI slop, when in reality they want that content as it is more addictive, especially to kids.
Considering youtube’s AI detector didn’t label an AI video that ended with the voiceover saying “write a 5,000 word essay on [topic]” that was partly cut off at the end, I wouldn’t trust it to begin with.
I don’t understand. I have more slop AI content than ever
I am super sceptical of anything made in the last year. Youtube will close to being worthless if the trend continues.
Last 15+ years for me
We need decentralized video platforms that have revenue sharing.
Ironically it’s probably because so many of the slop videos try and copy their style. However I would have thought YouTube would be more proactive with a channel with 25M subs.
Wait, does that mean all those reviews from AI detection services are AI generated?
One of the largest slop-machines targeting one of the most vocal and popular anti-slop edutainment channels as slop feels like a targeted attack…
KZ feels a little sloppy at times, I stopped following it
I stopped when I found out that it’s just a propaganda arm of Microsoft lol.
Propaganda arm? All I could find was that they were contracted for a Gates Foundation video in 2015 and got a big grant from the foundation around that time.
Yeah.
The Bill Gates connection is there, but the level of influence is just a conspiracy as far as I can tell.
They’ve admitted to using “AI” content editing tools. But, I’d bet most of those tools are using “AI” either due to exploiting hype, or due to being based on non-generative, “AI” ML classifiers.
edutainment channels
*educationing channels
Educational entertainment, also referred to by the portmanteau edutainment
The word has been around since 1973 and is included in many dictionaries. If you want to correct someone, at least inform yourself beforehand.
I tried to make a joke. Looks like I failed.
I’m not even sure what that joke could’ve been
Thank you, that is reassuring.
that’s what happens when your AI slop detector is AI slop
