TikTok Gives Power to Limit AI Memes in Feed  ## Metadata - Author: [[Hard Fork AI]] - Full Title: TikTok Gives Power to Limit AI Memes in Feed - Category: #podcasts - URL: https://share.snipd.com/episode/a59fae84-5f1e-43a9-94a5-365a296795c2 ## Document Notes ## Highlights - **TikTok Feed Is Shifting Toward AI Content** - TikTok is shifting from pure UGC to feeds that include a growing amount of AI-generated content. - This change alters what users see and challenges assumptions about authentic, low-polish video culture. Transcript: Jaeden Schafer This is obviously an app that used to just be 100% user generated content. In fact, it kind of spurred this whole industry of UGC user generated content where you made these really realistic, even companies trying to do things would make these really realistic Ads, where it's just a user sitting in their bedroom talking about a product. And it's kind of gone are the days of these like really polished videos, TikTok really pushed that forward. All of a sudden, we have AI generated contentated content that is taking over the For You page. And because of this, TikTok has introduced a bunch of new advanced labeling technologies for AI-generated content. So the new AI-generated content, it's called the AIGC, so AI-generated content control. It's coming out in the Manage Topics tool. It essentially is going to let you choose what you see on TikTok. This is what they said about it. They said Manage Topic already enables people to adjust how often they see content related to over 10 categories like dance, sports, food and drinks. Like these controls, the AIGC setting is intended to help people tailor the diverse range of content in their feed rather than removing or replacing content in feeds entirely. I personally think this is a fantastic move if you can actually go and completely ban certain categories. I might, you know, people might hate me, but like, I personally don't want to see any dance videos in my TikTok feed. It's just not what I'm interested in. But there's other areas like technology and AI that I'm deeply interested in, I would like to see more of. ([Time 0:01:15](https://share.snipd.com/snip/613ab6e1-2ed0-4002-b9fc-db7f5c88351d)) - **Engagement Drives What You See** - Engagement historically trains TikTok's algorithm, so commenting or reacting amplifies content you dislike. - Hard-coding preferences counters that feedback loop and can suppress unwanted topics regardless of past actions. Transcript: Jaeden Schafer Now, up until up until point, you and I both know the best way to fine tune your TikTok algorithm is to go and engage with a lot of content. I talked to people and they're like, I hate TikTok because I just see like a bunch of spam and all this stupid stuff on there. And it's like, well, you probably commented on it. And even you know, you'll talk to people that are like, man, did you hear what this person said? Like, absolutely hate them. They're so dumb. I was commenting on their posts the other day about blah, blah, blah. You know, I'll hear this about like political figures. And I'm like, okay, fantastic. You are now going to see a ton more content from that person because that is what the algorithm thinks you engage with. So whatever you engage with grows. ([Time 0:02:42](https://share.snipd.com/snip/c0d16235-cd2b-43c8-a7c4-b7e1467f6039)) - **Viral AI Clip Fooled Many Viewers** - Jaeden describes a viral AI video of a man, deer, and bear that fooled many viewers before odd artifacts gave it away. - He notes the clip spread across platforms, showing how AI content migrates beyond its origin app. Transcript: Jaeden Schafer Seen this you know video or I don't know okay I'll explain it someone was like you think your day is bad until you see this guy and it was like a guy on the toilet on his phone in like a lodge And a deer jumped through the window at the back of the bathroom and then a bear jumped in and he's in this tiny port-a with a bear and a deer and they're all wrestling around and it's like Insanity and like for the first two seconds you just think oh my gosh that's crazy and then you're like how can a bear jump through that window you start looking and you notice like the Guy is sitting on the toilet but he's also just like wearing his underwear and there's like all these weird things and like the toilet paper roll like disappears and then all of a sudden The deer disappears and the guy disappears and like it's just for two seconds you might think it's real and then you realize it's not but it's going on every single platform is my point It is on linkedin it is everywhere i saw so many people liking and commenting on that video on LinkedIn, and most of them did not realize it was AI generated. And of course, over on Facebook, there's the entire Facebook page, which is AI generated images people think are real. That was kind of the OG, I think, and we've moved beyond AI generated images. Now it is AI generated videos. That group in particular literally just documented any time they saw AI generated images on a specific Facebook group, and people were commenting on it, usually elderly people that Thought it was a real image. This is going to happen with video, like at an insane rate, you just see the most crazy unhinged things. And people think they're real. ([Time 0:03:56](https://share.snipd.com/snip/6f3417b6-89af-4628-a09a-338fae5e8222)) - **Video Is The New Frontier For Deepfakes** - AI-generated videos are escalating beyond images, producing ultra-realistic, shareable clips that can fool viewers. - As quality improves, distinguishing real from synthetic will become increasingly difficult across platforms. Transcript: Jaeden Schafer Now it is AI generated videos. That group in particular literally just documented any time they saw AI generated images on a specific Facebook group, and people were commenting on it, usually elderly people that Thought it was a real image. This is going to happen with video, like at an insane rate, you just see the most crazy unhinged things. And people think they're real. And the problem is, right now, I think there's lots of things. It's like, Oh, you know, the toilet paper roll disappeared, or why is this thing funny? Or that's kind of low quality, or there's a watermark but all those things are going to disappear the quality is going to get better and better things are going to look ultra realistic And it's going to be really difficult to tell what is a real video and what is not and this is what I think a lot of people are complaining about a lot of TikTok users are using AI to create Visuals for posts about a lot of different topics like history and celebrities, which some people think is a good use of that. ([Time 0:04:59](https://share.snipd.com/snip/b1ccaee6-5608-42a3-b62e-14e34920d379)) - **Invisible Watermarks Complement Metadata** - TikTok pairs labeled metadata (C2PA content credentials) with a new invisible watermark to track AI-made videos. - Invisible watermarking embeds pixels in the video itself so labels persist even if metadata is stripped or edited. Transcript: Jaeden Schafer And TikTok is now testing a technology called invisible watermarking. TikTok already requires people put a label on realistic AI generated content, and they use a cross industry technology called content credentials, which is from C2PA, which basically Embeds metadata into content to let other platforms know when something's AI generated. However, because you can actually generate AI stuff on TikTok itself. All that being said, TikTok does say that the labels can be removed when their content is re-uploaded or if you edit it on another platform, right? Like you download it, it's in the metadata, you go open it in Adobe Premiere, edit it, or just literally straight up export it immediately and change all of the metadata. So with the new quote unquote invisible watermark, TikTok is going to be adding this is another layer of safeguards. By using that, it can essentially detect when things have this invisible watermark, even if the metadata has been changed. That means it's going to be harder for it to remove because the watermark is in the video itself, you won't be able to see it. But there's little pixels in there that TikTok can detect. And they will know TikTok is going to start adding these invisible watermarks to AI generated content made with TikTok's tools. So they have their AI Editor Pro. It's also adding them to any content that's uploaded with the C2PA's content credentials. ([Time 0:06:26](https://share.snipd.com/snip/4af8ef6a-2abf-4569-9d03-53710a3fff3c)) - **Don't Rely Solely On Visible Labels** - Expect labels to be removed when content is re-uploaded or edited, so rely on TikTok's invisible watermark for detection. - Recognize that watermarks and metadata combined improve labeling reliability but aren't foolproof. Transcript: Jaeden Schafer However, because you can actually generate AI stuff on TikTok itself. All that being said, TikTok does say that the labels can be removed when their content is re-uploaded or if you edit it on another platform, right? Like you download it, it's in the metadata, you go open it in Adobe Premiere, edit it, or just literally straight up export it immediately and change all of the metadata. So with the new quote unquote invisible watermark, TikTok is going to be adding this is another layer of safeguards. By using that, it can essentially detect when things have this invisible watermark, even if the metadata has been changed. That means it's going to be harder for it to remove because the watermark is in the video itself, you won't be able to see it. But there's little pixels in there that TikTok can detect. And they will know TikTok is going to start adding these invisible watermarks to AI generated content made with TikTok's tools. So they have their AI Editor Pro. It's also adding them to any content that's uploaded with the C2PA's content credentials. And they said that all of these watermarks are going to essentially help them label content more reliably. ([Time 0:06:45](https://share.snipd.com/snip/6f8be8d9-89e8-4022-8069-d1cf50e7842f)) - **Education Is Part Of TikTok's Strategy** - TikTok pairs technical measures with education, funding nonprofits to boost AI literacy and safety. - The company sees awareness as essential since detection and obfuscation remain a cat-and-mouse game. Transcript: Jaeden Schafer They also announced that they are launching a $2 million AI literacy fund, which is aimed at experts like a bunch of nonprofits to create content that teaches people about AI literacy And safety. So they're trying to approach it from, yes, like they have these invisible watermarks, and they have this kind of metadata that they're going to detect things with. But they're also trying to approach it from an educational perspective, because at the end of the day, it's always a cat and mouse game where people are trying to hide that it's AI generated, Hide the watermarks, hide the metadata. ([Time 0:07:54](https://share.snipd.com/snip/b684a14a-058e-43a1-a88f-ec5e88f985e5))