Full transcript
0:00There's this brilliant KFC ad.
0:03>> Welcome to Chicken Town,
0:06home of the imitators.
0:10Because when you're on top,
0:14everybody wants a piece of you.
0:17>> And lately, we've been feeling a lot
0:19like KFC.
0:21There's this thumbnail, this one, this.
0:23What even is this? Every thumbnail we
0:26invent, every title we construct, every
0:29animation we build, sometimes entire
0:31videos, countless AI slop channels
0:34relentlessly copy us. And one day we
0:37decided enough is enough. So we
0:41investigated. We looked into these
0:43channels and found who's likely behind
0:45them. We joined a Discord community,
0:48paid 75 bucks for coaching, and
0:50subjected our fact-checking team to an
0:52unhealthy amount of AI misinformation
0:54spread to hundreds of thousands of
0:56people on topics like, well, the
0:59Holocaust. Because all of this is
1:01endangering educational content on
1:03YouTube, it's damaging for everyone in
1:06our niche. It undermines trust in
1:08journalism. And really, it's just so
1:12annoying.
1:13>> They think that my voice is AI. They
1:15even think that my face is AI generated.
1:18>> The cost of production has gone down
1:20tremendously.
1:21>> This is a new form of industrialized
1:24plagiarism and creative theft.
1:26>> True crime, video essays, science. No
1:29space is safe. We're in the golden era
1:32of soulless slob.
1:34>> I think a couple of years ago, a 3D
1:36thumbnail was associated to good quality
1:38content, but now it's perceived as AI
1:40slop.
1:41>> There's no wish to make something
1:42creative. It's just a get-rich scheme.
2:12AI is undoubtedly useful in many aspects
2:14of life and it's here to stay. We use AI
2:17tools selectively in our own
2:18organization. For example, to organize
2:20and skim through our research. In rare
2:22cases, we use generated imagery as a
2:24basis for compositions in Photoshop if
2:26no comparable stock images are
2:28available. But the rise of specifically
2:31generative AI has also made it very easy
2:33to mass-roduce content. There's AI music
2:36on Spotify. Terrible AI recipes and home
2:38decor on Pinterest. Boomers are losing
2:40their mind on Facebook. This is another
2:42account that I found that posts images
2:44like this. It always seems to be a child
2:47sculpting something very impressive out
2:49of trash. And the people in the comments
2:51just eat this stuff up, dude. Great
2:53work. Keep it up. Great talent. Nice
2:55one. Thanks.
2:57>> But one platform especially has become a
2:59dumping ground for AI slop.
3:03More than 20% of the videos YouTube's
3:05algorithm shows to new users may now be
3:07slop.
3:08>> The cost of production has gone down
3:10tremendously. You can have content which
3:13is not so good or or or reasonable in
3:15quality which is now competing against
3:19something which is generated by a human
3:21who has spent a lot of time and
3:23investment and invested into acquiring
3:25this skill.
3:26>> This is Dr. Akil Bwash. He's an
3:28associate professor at the University of
3:30Bath School of Management.
3:32>> And then you have this thing which can
3:33generate precisely that at a very low
3:36cost and they both compete. There's
3:38something genuinely frustrating about
3:40clicking on a video and then realizing
3:42it's just another AI voice reading an AI
3:45script feeding you AI visuals. And we
3:48reached a full circle moment when we
3:50came across this video about the
3:51invasion of AI slop made with AI. The
3:55only comment reads, "Oh god, this video
3:58is AI too. The creator gave it a heart."
4:02And it's moments like these that make
4:03you believe the internet is really dead.
4:05Around 100 people work on Fern, our
4:08German channel and our German podcast. A
4:10dedicated team of journalists, art
4:12directors, producers, 2D and 3D
4:14animators, and fact checkers, all to
4:16bring you that weekly dose of something
4:18to watch during lunch. But now, thanks
4:21to the power of AI, you can be Fern,
4:23too. You can wear our skin and talk in
4:26our voice and walk in our shoes.
4:30So, I'll take screenshots from Fern's
4:32video, paste them into chat GPT, and
4:35simply say, "Regenerate this photo."
4:38Most people think you need to be
4:40professional 3D designer with years of
4:42experience to create videos like this.
4:46Well, they're dead wrong. I've cracked
4:48the code. He uses a Fern idea generator,
4:51a Fern script writer, and a Fern image
4:54generator. It's literally identical to
4:57what Fern uses. And I would dare to say
5:00that ours could be even better than what
5:03Fern is using here as an example. And
5:05the best part, you don't have to be good
5:07at literally anything anymore.
5:10>> You see, I don't have that budget
5:11available to me. Neither do I have the
5:13skills to animate, do academic research,
5:16or I have that smooth voice over. But I
5:18do know how to use AI.
5:20>> Some of it is just outrageous
5:22misinformation. You've probably seen
5:24channels like Fern on YouTube where
5:26everything looks like it came out of
5:27Blender. And when you first see it, you
5:29probably think that everyone spend weeks
5:31building it inside Blender and they
5:34actually didn't. Um, we actually did and
5:37it just doesn't end.
5:40Yikes.
5:42Imitation is flattery. We get that.
5:44We've been inspired by the grades on
5:46this platform all our career. But to us,
5:48there's a difference between countless
5:50hours in After Effects to recreate Vox
5:52style maps and asking a machine to just
5:55clone something. We get why they choose
5:57to copy our work. Edutainment is a good
6:00niche to target for slop gods. Channels
6:02are often faceless, a version of the
6:04animations are now replicable with AI,
6:06and people like to learn a little
6:08something while binging down a hot
6:09pocket. The thumbnails don't rely on the
6:11creator's face, so you can fool people
6:13into thinking any considerable effort
6:15went into whatever your chain of tools
6:17spit out. So, of course, we're not the
6:18only entertainment channel having to
6:20deal with this. So, when I started
6:23making videos, I very much built my
6:25channel around 3D animations and
6:27highquality visuals. And a few years
6:29ago, I think that was very unique and
6:31there wasn't a lot of that out there.
6:33But now, that very style that I based my
6:37channel off has become kind of
6:38synonymous with AI slop, and that's a
6:42challenge for me. What I've noticed is
6:44that several people in my audience look
6:46at my 3D style of thumbnails and they
6:48associate them to loweffort content and
6:51my videos take a couple months to
6:52research and produce. So, I'm struggling
6:54a bit to convey that in the thumbnail uh
6:57while keeping it attractive enough to
6:58get clicks. I've seen videos take
7:00inspiration from my visual style before,
7:02which is fine, by the way, but this was
7:05very different. These visuals weren't
7:07just recreated in some animation
7:09software to look like mine. They were
7:11literally the same scenes with just a
7:13few things changed. There's even math
7:16slop which is just oh god. Imagine
7:19trying to actually learn something on
7:21YouTube like it's 2017. Especially our
7:24thumbnails get ripped a lot and slightly
7:26changed by slop channels likely using
7:28AI. So we thought it would be funny to
7:30use AI to find matches. Yanik, our data
7:33guy, because we actually have talented
7:34humans working here, set up a program to
7:36identify cases in which our thumbnails
7:38get lifted. And there are so so many.
7:44We confronted several of the creators of
7:46these tutorials. Some took down their
7:47videos afterward. Others explained that
7:49they didn't earn money from their
7:50tutorials. And others considered the
7:52tutorials free marketing for us. One
7:54creator stated clearly he just wanted to
7:56make money and had only briefly used
7:57Fern as an example. With regards to we
8:00have Fern at home, one AI channel stands
8:02out in particular, Loom.
8:05It's part of Loom Studio, a network of
8:07three channels. Loom Rocks and Ago. In
8:10total, they have more than 300,000
8:12subscribers. Ago and Loom seem to
8:14generate quite some revenue. These
8:17channels regularly appropriate our
8:18thumbnails. They rip them and then
8:20likely let AI change or add elements,
8:23but sometimes they also seem to simply
8:25copy and recreate full videos of ours
8:27like this one. Some of their content
8:29gets over a million views. They also
8:31seem to take from others, like Mr.
8:33Beast. In many thumbnails, Mr. Beast
8:35gets replaced with a 3D character. In
8:37one instance, a video with such a
8:39thumbnail got over two million views.
8:41That could equate to $11,000 in revenue.
8:44Mr. Beast even reacted to it on X. Did
8:47they literally just swap me on my
8:48thumbnail for a red guy and get 1.8
8:50million views? What the hell, lol? But
8:53instead of being embarrassed, Loom
8:55Studio seems to be so proud of it, they
8:57even put it on their website. Mr. Beast
9:00tweeted, "We just kept going." We
9:02confronted Loom about this and they told
9:04us they've never reused another
9:06creator's thumbnail. Their content is
9:08supposedly created by their team from
9:10scratch. But as if all that wasn't
9:12cringe enough, Loom Studio also offers a
9:15coaching program. Their VIP course costs
9:18a casual $1,499
9:21included a blueprint plus weekly calls
9:24with our founders. We review your
9:26channel, fix what's wrong, and push you
9:27to your first 100K. They seem to be
9:30pretty experienced. They say they've
9:32built more than 100 faceless channels
9:34and generated over 100 million views.
9:37And we can see why. This company is run
9:39by a team of professionals, a lion, a
9:42wolf, and a tiger. Nothing screams men
9:45building their AI company more than
9:47three predator mammals with sunglasses.
9:49The wolf conjures AI pipelines from thin
9:52air. The one who makes the machine
9:54think, create, and scale so humans don't
9:57have to. Oh, how I hate having to be a
10:00human that thinks and creates. Finally,
10:03the wolf can make it stop. We dug deeper
10:06and found that Loom sits at the center
10:08of a small network of AI channels
10:10supporting each other. Yanni created
10:12this map to show how these channels are
10:14linked together. Whenever two channels
10:16collaborated on a video, a line connects
10:18them. True Crime Vault pumps out
10:20fictional true crime stories, which is
10:22just a fantastic oxymoron. It
10:25occasionally collaborates with other
10:26channels of the Loom network, so we
10:28suspect it could be a client of Loom
10:30Studio. We asked the operator via email,
10:32but didn't receive a reply. The channels
10:34Ago, Loom, Rocks, and Lumicus all linked
10:37to the same Instagram account of the
10:39lion, Ardom. Just like in nature, the
10:41lion seems to be the real king behind
10:43this company. On Insta, we find a
10:45profile with the same name, Ardom I,
10:47which has a link to Loom in his profile.
10:51Ardom often posts pictures of his
10:53vacations. For example, in Dubai, there
10:55are luxury cars, fancy hotels, a big
10:58yard.
11:02>> Chad, how do I say something is a fat
11:04cliche without saying it?
11:06>> You could try something like, "We're
11:08wandering straight into a well-worn
11:09narrative."
11:10>> Okay, got it. Got it. Got it. Yeah.
11:11Yeah. Yeah. Wow. Yeah. We're wandering
11:14straight into a wellworn narrative here.
11:16Our research team found out that Ardm is
11:19probably from Ukraine. We found an
11:21address in the north of the country. He
11:23also seems to run a Ukrainian Telegram
11:24channel about AI. In the past, he
11:26seemingly used a different last name. Of
11:29course, Lum are not the only ones doing
11:31business like that. We want to dive
11:33deeper into the scene, so we join a
11:35Discord channel. YouTube Faceless
11:38generally promotes a kind of hustle
11:39culture. Not all of it is about AI, but
11:41quite a big part. I've tried doing AI
11:44generated videos, but I think it's just
11:45easier to steal people's content as long
11:48as you react to the video, aka using AI
11:51voice over. I think you can still get
11:53monetized directly under that comment.
11:55Another user links our fentinel video.
11:58Where do you get the vids from? Fern
12:00channel. Can someone tell me how to get
12:02Fern? Let me know. Cipher AI voice. Hey,
12:04anyone here in the Fern disrupt niche? I
12:06feel bad for Fern and the Fern
12:08animators. Lol. Why is that? Cuz they
12:12spend hours animating 10 to 30 seconds
12:13of footage and some random kid with AI
12:16can replicate the quality. But can they
12:18really replicate our quality? We decide
12:20to take a course to find out. We aren't
12:22ready to spend $15,000 to take the alpha
12:24animal coaching. Luckily, AI with Gary's
12:27course comes at a bargain price of $75.
12:30This video totally using AI generated
12:32video and YouTube automation. If you
12:34already know, there's a channel called
12:36Fern. And this YouTube channel is
12:38blowing up with cinematic 3D animations.
12:40In their videos, they're using a lot of
12:42mannequin figures, and it's really
12:44effective for this kind of content. So,
12:46if you're a beginner or still stuck on
12:48creating highquality YouTube content,
12:50you need Gary Studio to speed up your
12:52workflow. You also need someone with
12:53experience to teach and guide you in
12:55detail. Just join my community. So,
12:57using a fake identity, we join. First,
13:00we receive an email that the payment was
13:02successful. A few hours later, we get
13:04another email from Gary. He writes that
13:06we can now join school with a monthly
13:08subscription. We thought the $75 would
13:10unlock all materials, but the 75 bucks
13:13is for content access. The school
13:15community requires a separate
13:17subscription, so both payments are
13:19required to join. We reluctantly pay
13:21another $67 and finally get access.
13:24We're in and uncovered a recipe for
13:27successful and captivating AI
13:29documentaries on YouTube. Multiple
13:31shitty exists.
13:34On top, AI with Gary provides a basic
13:36tutorial on how to produce AI videos.
13:38from finding your niche and choosing a
13:40style to word for word prompts for
13:42writing AI scripts or generating
13:44animations.
13:46So far, we just looked at the process,
13:48but what about the results? It's a
13:50pretty crucial aspect of edutainment
13:52after all. How much does it eat you? We
13:54bring in our head of facteing, Pascal,
13:56because we do have a team of fact
13:58checkers who go through every single one
13:59of our scripts and look for mistakes.
14:01Pascal has years of experience in
14:03investigative journalism, but now she
14:05faces the ultimate challenge.
14:08sit through hours of slop and fact check
14:11it.
14:12We published this video on Fern in
14:14November 2025. We sent a production crew
14:17to Scott Payne to let him tell his
14:18story. Just boring all human activity.
14:21Just weeks later, this video went online
14:24on a channel called Black Files. Only
14:26seconds into the video, and the opening
14:28scene is already misleading. In the
14:30scene the video describes, the clansmen
14:32did not wear these robes. In this scene,
14:34the ropes are incorrect. They feature a
14:36red cross within a circle, not just a
14:38red cross. And Pascal suspects that this
14:40entire scene is just made up by AI. It
14:44directly contradicts the detailed
14:46sources she could find. She also tried
14:48to fact check this video, but just
14:50concede. Quote, "The language is so
14:52weird, it is hard to fact check. Most
14:54sentences are not proper sentences. It's
14:57hard to follow." And yeah, yeah, we get
15:00it. Opens fire. Six bullets. Head seal
15:03sees it coming. puts head down on
15:06steering column. Too late, falls dead
15:08almost instantly. Velas runs to waiting
15:11Buick. This last example might be funny
15:13because it's bad, but factual errors and
15:15apparent hallucinations and content
15:17covering hate groups like the KKK are
15:19highly problematic. We reached out to
15:21Blackf and they reply the channel
15:23defense production process, saying its
15:25videos are created by a team using AI
15:26alongside in-house technology and are
15:28reviewed for accuracy and originality.
15:31They deny copying or infringing on third
15:33party thumbnails, but acknowledge
15:34inaccuracies in the video we referenced
15:36and say they have since updated it.
15:38Anyways, we can't help but wonder how
15:40these AI channels cover the Holocaust.
15:43This video by Ago reached more than half
15:45a million people. A true historical
15:47documentary about courage, moral
15:49failure, and one escape that changed
15:51history forever. Let's start with a
15:53visual highlight. prisoners inexplicably
15:56collecting, shoveling, and throwing a
15:58mountain of toys into a fireplace. We
16:01asked a Holocaust expert about this, and
16:03she told us the scene is obviously
16:04inaccurate. The victim's belongings,
16:07like toys, were taken and collected, but
16:09not piled up and burned in the
16:10crerematoria. So, someone just made this
16:13up. Maybe the creator, maybe the AI, who
16:16knows? At one point, the AI voice says,
16:19"By 1944, his figure was staggering.
16:251,750,000
16:28victims
16:31and the world did not know. That was the
16:34secret of the systems success. Secrecy,
16:37lies. No one outside these walls knew
16:40the truth. That figure of 1.75 million
16:42did circulate for a while, and it did
16:44come from Rudolph Verba himself, a man
16:46who escaped Avitz and whose story is
16:48covered in this true historical
16:50documentary. But the video fails to
16:52mention that it does not reflect the
16:54current historical consensus. Today,
16:56historians estimate that around 1.3
16:58million people were deported to Awitz
17:00between 1940 and 1945 and around 1.1
17:03million were murdered there. And it's so
17:06important to be accurate about this.
17:08Holocaust deniers often exploit outdated
17:10or imprecise numbers to cast doubt on
17:12the Holocaust as a whole. That's why
17:15videos covering this should clearly
17:17distinguish between historical estimates
17:18made at the time and the current
17:20academic consensus. The phrase the world
17:23did not know is also very misleading. In
17:261942, Allied nations had already
17:29publicly declared that Germany was
17:30carrying out Hitler's so-called final
17:32solution to the Jewish question, the
17:34mass murder of Jews. Historians also
17:37believe many Germans had at least a
17:38rough idea of what was happening as
17:40information and rumors about the crimes
17:42circulated. And last but not least, we
17:45again reach a full circle AI
17:47moment when we stumble across death
17:49scene.
17:50>> No one had ever escaped from Achvitz to
17:53tell the world the truth.
17:56>> That's incorrect, but also contradictory
17:58to Ago's own content. The channel had
18:01already published a video about Vl
18:03Pitky, another man who infiltrated and
18:05escaped Avitz. We covered that story
18:07ourselves. and Pitky as well as multiple
18:10of his comrades escaped or bought their
18:12freedom before Verba delivering reports
18:15about Avitz to the outside world. Maybe
18:18we shouldn't let the wolves machines do
18:20all the thinking and creating. I think
18:22with a lot of these educational AI
18:25channels, the visuals are really only
18:27half the problem. But a huge concern
18:30that I have is the quality of the
18:32information in those videos because you
18:34can be sure they're also using AI to
18:36generate all of their scripts. There's
18:38one channel that recently got very
18:40popular, Chloe Versus History, about a
18:42lady vlogging in the past and people
18:44love it. But if you actually look at
18:46what she's saying, if you look at the
18:47details, if you look at the whole story,
18:49almost everything is wrong. And
18:51especially with educational content, I
18:53think there's this idea that people
18:55trust the content and they walk away
18:58being informed, but actually they're
19:01walking away being less informed about
19:03certain topics. And I think that's
19:05really bad.
19:06>> These AI channels don't seem to care
19:08about the quality or accuracy of their
19:10content. Their goal seems to be simply
19:12to publish as many videos as possible to
19:15make as much money as possible. Oh, and
19:18I almost forgot. We actually spoke with
19:20a person familiar with the production
19:21process behind Loom's channels. They
19:23told us that they indeed use AI and that
19:25two people had to produce one script per
19:27day each. We confronted Loom and sent
19:29them many questions. In their response,
19:32they wrote, "Half of this is totally
19:33fictional and 30% wrong." They also told
19:36us that their audience doesn't watch
19:37Fern, which we are keen to believe. They
19:39claimed that 30 people work at their
19:41company with AI being used as a tool and
19:43lots of manual labor still involved.
19:44They were in disbelief of our apparent
19:46hate of them. Lastly, life is like a
19:49boomerang, okay? There's no wish to make
19:52something creative. It's just a get-rich
19:55scheme. And it's not just teenagers and
19:57Discord channels and cool guys who keep
19:59going even when Mr. Beast tweets driving
20:02the slop epidemic. Various AI companies
20:04seem to know exactly who their target
20:07audience is and how their products are
20:08used. Hakesfield offers a variety of
20:11generative AI tools and recently they
20:13uploaded this tutorial on their channel.
20:15I use Hakesfield AI and claw to build a
20:17$39,500
20:19a month faceless channel. I just found a
20:21faceless YouTube channel that's making
20:23almost $40,000 every single month. And
20:25it made me wonder, can I completely
20:27recreate this channel and build the
20:29exact same style of videos using AI? The
20:3140K a month educational channel he found
20:34is Bright Side. In case you're not
20:36familiar with Bright Side, the
20:37incredible irony might be lost on you.
20:39Bright Side is one of the OG slop
20:41factories before there was AI as we know
20:44it. terrible clickbait, unhinged content
20:46pumped out every few hours.
20:48>> It's just colorful thumbnails,
20:50interesting, clickable titles. It's They
20:53make things.
20:56They make things. That's about it. If
20:58you've never heard of Bright Side, good
20:59for you.
21:00>> Bright Side just straight up lies or
21:03just uses these like fear-mongering
21:05sensationalist tactics.
21:06>> Yet, this is the type of content
21:08Hicksfield presents as a great
21:09opportunity to make bank with their
21:11tools. Educational content appeals to
21:14everyone and has huge audience
21:15retention, which is exactly what drives
21:18you monetization.
21:18>> In this brave new world, educational
21:20videos obviously aren't made to educate.
21:23They're simply the most lucrative niche
21:25to make as much money as possible. And
21:27we can see how well that logic plays out
21:29just a few minutes later when he uses AI
21:31to create a video about Pompei. We could
21:33spend minutes explaining how much is
21:35wrong with these glass jars, but simply
21:37put, they just didn't exist in this form
21:38in Pompei in 79 AD. Now, you might
21:41wonder, who would possibly want to slurp
21:43up this type of slop? Well, the people
21:46at Hicksfield.
21:47>> I think this video is very interesting.
21:48This is the kind of stuff I watch right
21:50before going to bed. Really puts the old
21:52saying that you should make stuff you
21:53would want to watch yourself into
21:55perspective. Next, the tutorial
21:56demonstrates how to make even more
21:58videos fully generated with basically no
22:00human input. Okay, one video is cool,
22:03but we need more. So, I'm just going to
22:05type in make me two more videos. This is
22:08great news for anyone who wants to get
22:10paid by YouTube without actually making
22:12the content themselves. Let's launch our
22:13content machine that will bring in
22:15money. During a gold rush, sell shovels.
22:1911 Labs offers AI voice tools. And they
22:21have their very own tutorial on how to
22:23make AI B-roll. That's just like Fern.
22:25And what do you do once you have your
22:27slop visuals in place? Add a slop voice.
22:30>> As he began his speech, there was a
22:32shift in the room. You could feel it
22:35like the start of something important.
22:37During a gold rush, sell shovels. We
22:40sent both Hicksfield and 11 Labs a
22:42request for comment, but didn't receive
22:43a reply. Let's zoom out. In the best
22:46case scenario, these AI productions just
22:48result in a few ridiculous visual
22:50mistakes. Worst case, they lead to
22:52misinformation about some of humanity's
22:54darkest times. And how would viewers
22:56possibly know that what they're watching
22:58is wrong? Pascal, seasoned investigative
23:01journalist and our head of factchecking,
23:03wrote this after reviewing some of the
23:05videos. It is incredibly hard to verify
23:08things that never actually existed. Real
23:11people usually leave some kind of trace
23:13or reference point, but AI can just
23:15completely make things up. If it's
23:17already difficult for me, I don't think
23:19average users can realistically deal
23:21with this. After just checking half a
23:23video, my head was genuinely spinning.
23:26Luckily, many people instinctively
23:28distrust AI content, but that leads to a
23:31loss of trust in real creators, too. I
23:34get comments on my videos all the time
23:36where people think that the 3D
23:38animations are made using AI. They think
23:41that my voice is AI. They even think
23:43that my face is AI generated. People get
23:46so anxious about slop they start to
23:48suspect it everywhere. If YouTube lets
23:51this kind of content go unchecked and
23:52allows its creators to profit from it,
23:54in the long run, it could damage the
23:56platform's reputation and value
23:58proposition. So, what is YouTube doing
24:01about it? Already in 2025, the platform
24:05updated its channel monetization
24:06policies. They renamed repetitious
24:09content to inauthentic content.
24:11Loweffort mass-produced uploads
24:13shouldn't make money anymore. And in
24:152026, they updated them again with their
24:17unsatisfying or off-putting content
24:19policy. YouTube now targets content that
24:22feels interchangeable, emotionally
24:24manipulative, or mimics existing
24:26formats. Great news for us. Hopefully,
24:29YouTube also pushed towards disclosure
24:31and labeling. Synth ID is Google
24:33DeepMind's watermarking system. It
24:35embeds invisible, irreovable markers
24:37into AI generated content. Major
24:40generative AI companies such as OpenAI,
24:42Nvidia, Cacao, and 11 Labs have adopted
24:44the watermark. With 11 Labs on board,
24:47more AI generated voices can now be
24:49automatically detected by YouTube. In
24:51May, the platform announced they would
24:52start automatically tagging AI generated
24:55or AI edited videos. The labels will be
24:57clearly visible on their videos or on
24:59shorts. The goal is to make AI content
25:01easier to spot, even if creators don't
25:04disclose it. But of course, slop
25:06factories will try to get around the
25:07tech and demonetization. Hicksfield has
25:10them covered. And for those of you who
25:12are worried about AI content getting
25:14demonetized, YouTube has no problem with
25:16AI. It just blocks people who use it to
25:19make lazy, lowquality spam. This is an
25:22unironic statement in a video
25:23demonstrating how to automatically pump
25:25out Bright Side style content within 20
25:28minutes.
25:30We also asked YouTube about slop
25:31directly and shared our concerns and
25:34they actually got back to us saying that
25:36tackling slop is a key priority for
25:38them. They also said this YouTube's
25:41approach to information quality has
25:42always focused on promoting highquality
25:44content rather than fact-checking
25:46individual videos at scale. But under
25:48our misinformation policies, we remove
25:50misleading or deceptive content with
25:52serious risk of egregious harm. We do
25:54believe YouTube understands what's at
25:56stake and that the platform needs to
25:58crank up the heat even further. Many
26:01years ago, at the start of our career,
26:02YouTube was drowning in clickbait videos
26:04full of stock footage. That changed over
26:07time as the platform managed to measure
26:09and reward retention and watch time.
26:11Deceiving viewers didn't work that well
26:13anymore because YouTube would punish
26:14videos when people would click off fast.
26:17Optimizing for retention leads to its
26:18own problems, but generative AI feels
26:21far more threatening to overall content
26:23quality. At least for enterainment, this
26:26is the next thing harming the longevity
26:28of our ecosystem.
26:30Personally, I increasingly find myself
26:32scrolling through my recommendations,
26:34irritated at every piece of slop served
26:37to me. It makes me want to open the app
26:39less, and that can't be good.
26:46Contrary
26:50to some channels on YouTube, our 3D
26:52animations are made by humans. This is
26:54one of the artists who regularly puts in
26:56countless hours in creating them for
26:58you. Hi, I'm Leonard. When I'm creating
27:00scenes, I can easily spend the whole day
27:02sitting at my computer. So, shout out to
27:03Flexisport for sponsoring this video and
27:05sending me this. One button and I can
27:07switch from sitting to standing.
27:09Creating a scene like this roughly comes
27:11down to three steps. Step one, modeling.
27:14It's similar to the real thing. I stack
27:16simple cubes to block out the desk. Then
27:18dress the set with a plant computer and
27:20the display. The usual clutter. Step
27:22two, animating. This desk actually
27:25moves, so I have to animate the legs
27:26sliding into each other with a little
27:28ease out so it doesn't feel robotic.
27:30It's a three-stage frame, so it adjusts
27:32to you no matter if you're 5'2 or 6'5.
27:36Step three, shading. It looks a bit
27:38bland still. I've got the white oak. But
27:40if you want something even more premium,
27:42they have great options for 100% solid
27:44wood desktops. And it's not just a work
27:46desk. It can be a gaming station, a
27:48dining table, or where all your stuff
27:50ends up living. Mine's basically all
27:53three. And even when I completely load
27:54it up, it's still really solid and
27:56stable. If you want to upgrade your
27:58setup, use code Fern50 for $50 off the
28:02E7 Pro. And if you're not sure,
28:04Flexispot offers 30-day risk-free
28:06returns and a 15-year warranty. Link is
28:09in the description.