AI companies are stealing from YouTubers

0:00The Colin and Samir Show is supported by Microsoft Copilot. This is something I thought I would never say, but can I share with you my favorite prompt? I mean, like when you say something like that, I have to say yes. Were you going to say no? No, but okay. Just let me hear it. So when I'm researching for a guest or I need to know a lot about a subject matter, I say, make me a podcast about. And when you do that, you're going to get a short blurb about that subject. And then in like two minutes, you're going to get a podcast. So check this out. Welcome to Co-Pilot Podcasts. Picture this, a guy working the front desk at a hotel, making lacrosse videos in his spare time, completely unaware that he's about to become one of the most influential voices in the creator economy.

0:38Today, we're diving into the remarkable story of Colin Rosenblum. Wow. Did you pay it to say that? That's crazy. That's very kind. It's crazy, but it's my favorite way to learn about a subject matter. I'm learning about so much more now that I can turn this research into like a seven minute podcast that I can listen to just when I'm walking. Did you tell it that I worked at the front desk of a hotel? Dude, the prompt is just, make me a podcast about Colin from Colin and Samir. That's it. I may be biased, but I'm on the edge of my seat. I'm hooked. So if you guys haven't checked out Copilot, check it out.

1:10Try it for your research. Try it for getting variations of your titles, your ideas. It's one of my favorite tools on the internet right now for our creativity. And now back to the show. Good morning, everybody. Good afternoon. Good evening. Whenever you're listening to this, welcome to this episode of the Colin and Samir show. and this audio-exclusive Q &A. I'm currently chewing a dried mango. Why would you chew while we started recording? I apologize to everyone that was ill-timed. Welcome to the show today. Today on the show, we're going to be answering some of your questions. We're going to be talking about using shorts to develop long-form videos.

1:45We're going to be talking about what would happen if our YouTube channel got wiped off the face of the earth and we had to start fresh. And we're going to be starting with a headline that came out last week about companies like Apple stealing data from YouTube videos to incorporate into AI models. All right, before we start the show, if you could do us a quick favor, wherever you listen to the show, whether that's Apple or Spotify or another podcast app, if you could rate the show, that would be awesome. Really helps us in discovery. All right, here's our audio exclusive episode of the Colin and Samir Show.

2:20All right, we have a ton of questions from across the internet, from Twitter, from Discord. Maybe we'll talk about some stuff on Reddit. But before we get there, there was a very interesting headline last week. Yes, there was a headline that came out first in Wired that exposed that companies like Apple, like NVIDIA, like Salesforce, paid for a data set called YouTube Subtitles that scraped the data from over 170 ,000 YouTube videos from over 48 ,000 YouTube channels. and then sold that data to these companies, to Apple, to some of these major tech companies. And creators did not know. So creators like Mr.

3:05Beast, like Marquez Brownlee, like Complexly, Hank and John Green's company. Julia Walsh, the CEO of Complexly, actually put out a statement and said, we are frustrated to learn that our thoughtfully produced educational content has been used in this way without our consent. So I actually want to play Marquez's response. Okay. Because he sent out some tweets, but he also made a really good short about it titled AI is Stealing My Videos. Okay, let's hear it. I'll play that for you guys. So there's a story that came out today, and the headline was that Apple trained AI models on YouTube content without consent, including MKBHD videos.

3:40There's no statement from Apple on this yet, but here's my take. The real story is Apple and a whole bunch of other tech companies are training their AI models using data that they buy from third-party data scraping companies, some of which get their data in slightly illegal ways. So one of them pulled tons of transcript data from thousands of YouTube videos, including Mr. Beast's videos and my videos and many more. So that company is a problem, and that's going to be an ongoing evolving problem for many years. And also Apple can technically say they're not at fault for this, but they're going to keep getting data from companies like this, so that's also something they need to vet.

4:13But the double whammy is I actually pay for more accurate manual transcriptions on every video that we put out so that people who are hearing impaired don't have auto captions, which suck. So that means the stolen transcriptions specifically are paid content that's being stolen more than once. So there's a whole bunch of reasons why creators would be upset here. There's the reason of, first off, I didn't know my data was being taken and then used in an AI data set. Second, I didn't know it was being sold, right? That people were profiting off of this data. And like Marquez says at the end of that, for him specifically, he's paid actually for his YouTube subtitles to be a specific type of way to be very precise.

4:57And that has been taken in the data set. And there's this whole other thing of like the conversation of AI in general. Are these data sets, which are being used to train AI models, eventually going to be a competitor? Because they're actually being used to produce new works, new things. things. This is the question as well. I mean, we'll get to all these, but like, what's YouTube's role in sort of keeping our data as creators safe, right? There's a lot of questions that this brings up. Well, I think first and foremost, like a lot of our data is publicly available, right? And I think the big question I asked when I heard the story was like, well, actually, what are the terms of service?

5:35What have we agreed to? You know, because it's kind of interesting to think about the scale at which creators sign terms of service with platforms without knowing too much about what's going on in terms of service. Like no one's really sitting and reading them. I mean, I would say that applies to non-creators as well. Like all of us who use almost any platform on the internet, we agree to these terms of service. And in some ways, I would say a lot of society has become okay with our data being monitored, tracked, and then sold for the purposes of like advertising targeting. Like I even, when I get a new app and it's like, do you want to allow us to track?

6:12I just say yes. Yeah, make it a better experience. Make it a better experience. That has become. But that's very different than taking our data and making something new from it and then profiting from it. Or is it? So, well, at scale, that is happening, plain and simple, right? Like that's happening, A, with like a, with an open internet, that's going to happen with everything that's available to rank in search. Neil Mohan actually talked about this, this concept of what's available to be scraped when it comes to data, to have an open internet. And obviously titles are one of the primary, right?

6:47So titles are available because they are used in search ranking. I'd have to imagine descriptions are available too, but anything that is utilized for ranking search is going to be available because that's how the internet works. So all of that is available. And I think we're all generally okay with that. But when you go inside the video, what he says is, is transcripts and video scripts are not okay to utilize, to pull into your data set. The biggest challenge is like, it's, you can do it. Of course, there's going to be bad actors or like, you know, I don't know how people have done this, but people have built Colin and Samir AI models where you can search by subject matter and it'll pull up an interview and the exact quote that was said.

7:33That is using our transcripts, right? Oh, yeah. I mean, it's incredibly easy. It is so accessible and easy to do this. I mean, I did it with our Alex Benayan interview the other day. I just like went to a website that pulls transcripts and YouTube videos because I wanted to search it to find specific parts that I was going to make into a short. It's so easy to get. I mean, first off, the subtitles are generated with AI, so you don't even have to put subtitles in. It pulls the transcript first with AI, just with what you're saying. And so it's super easy to get. And then in aggregate, those were used to train these models, come up with this data set, which was then sold to companies like Apple.

8:11So this is frustrating as creators. Just to be clear, we searched the database. We were not included in this data set. I don't know if I should be happy or sad about that, but we were not included in the data set. It is absolutely frustrating as creators because this is our work. And when you look and zoom out, like licensing or syndication or derivative works off of our work traditionally is a path of monetization for us, right? Like that is, think about the concept too of like a lot of books or a lot of like LA Times or New York Times articles get optioned for movies, right? So like they take that article and they say, hey, I'm going to create a derivative work off of this and I'm going to pay you for the option to do that.

9:00Like I'm going to make a movie out of this headline or this article. I'm going to pay you for that, for the option to do that. So I think in the same context, if you're like, hey, I'm going to create a monetizable model utilizing your work, that is traditionally a place where creators and creatives and artists have gotten paid. Yeah. I mean, it's actually very similar. You bring up the New York Times to when the New York Times, you know, actively sued, and I believe is suing, OpenAI and Microsoft over scraping their work. Or when Scarlett Johansson sued OpenAI for the use of her voice, which she deemed to be like too close, right, to the voice that they ended up using.

9:43Yeah, there's concern here that if all of this information and all this work we're putting out into the world is wrapped up into data sets and then sold at scale to these massively powerful companies that are in charge of delivering information to the masses. That like, wait, are we just going to become a slower way to get information that requires a lot more friction and has less sources? Yeah, I think the first time I experienced ChatGPT was the first time I thought in my head, is this an existential threat to search, right? To Google search, which has dominated the internet. Search has dominated the internet for a very long time.

10:28And that also has been what created the ad economy of the internet. And when I look at ChatGPT, I've used it in ways that I could have never imagined, right? Like I actually recently just used it to ask it to break down the YouTube terms of service for me as of July, 2024. And it was really great. Super effective at that. but that used to be something I would do in Google search, right? Or when I went on a trip last year, I used it to plan my trip. So these are the things I like to do. Where should I go? You didn't search on YouTube or TikTok. I didn't search on Google. Where should I go in this place or show me a travel vlog?

11:04More importantly, I didn't search on Google, right? Because that is their business is search. So I think the future, obviously, when you play it out, it's like us as video creators, is there a version of this where, you know, AI models are just producing the video you want? Right. Where it's like, again, we've talked about this past how to videos I think are at high risk of like, how do I change this bike tire? And it just, and you take a picture and it just builds the video based on that, but it builds the video based on high performing. Like it builds the script based on what has worked in the past in how to videos on YouTube.

11:37It builds the visuals based on what's worked in the past. Now, all of a sudden there's, there, there is a, just a brand new internet where all of the data that we've fed to the internet over the past 14 years through social content is now being used to just generate new content that we're not making. And the toothpaste is out of the tube. Yes. You know, like once. That's a classic AI saying right now. Yeah, but it's true. Like once the data has been scraped and it's been sold and it's incorporated into these models, it's over. I want to see someone make a short or a TikTok of putting toothpaste back in a tube successfully because that would be funny.

12:11And every time someone goes, the toothpaste is out of tube, you'd be like, well, you can also put it back in. Maybe we should make a short about AI and be like, it's all going to be okay. Because look at this. Yeah. The toothpaste goes back in the tube. Yeah. But yeah, I mean, it really, it is, it's like an, it is a new age of the internet and it's, you know, this new understanding that like, you know, I, I don't, I don't know how effective it's going to be to fight these AI companies to make it just not happen. I just think it's going to be a new reality that when we put these things out, they can be used by companies with AI to make other things.

12:45And it's weird. But, you know, how else do you protect against that outside of like not being on these platforms and having what your own like data server or something, you know, I don't know how you truly protect yourself if you if you even wanted to. And, you know, I was watching a short actually or a reel made by Alexis Ohanian. And he talking about this concept of AI content. And he likened it to like, all right, let's look at making croissants. AI and robots will eventually just make all the croissants in the world and they will make the perfect croissant. They will figure it out. There will be a robot that just makes the perfect croissant.

13:24He was like, but we will seek out, like that perfect croissant will become monotonous and we will seek out - The imperfect. The imperfect croissant with the story behind it, with the human story and the human touch. and the imperfections. And I think the same goes for content. And it already is. Like we came to YouTube and social platforms in the first place for human connection, right? Because it wasn't scripted TV. It was more raw. It felt closer to the actual person. And I think that's just going to continue. I think about, yeah, that's a good one. I think about the concept of efficiency a lot in this context and inefficiency.

13:58So what I mean by that is like Apple releases a new phone. What's the most efficient way to learn about that phone? It's probably just to search it, go on their website, read the specs. move on. The least efficient way to learn about it is probably watching like a 40 minute Marques review, but that's the most fun way. And that's like the most, that's the way that has the most human perspective in it. And it's the way that like, I would probably prefer to use my time because I'm not technically looking for efficiencies in that context. I thought about that too, when I went to a concert recently, John Mayer concert.

14:29And I was like, this is the least efficient way to listen to John Mayer, right? Like I have to drive to the forum. I have to park, which is a terrible experience. I have to walk in, I have to sit down and he's going to play stuff that is different from just pressing play on Spotify. Yeah, I don't get to choose what he plays. Yeah, so it's the least efficient way to listen to a John Mayer song, but it's the most interesting human way to listen to a John Mayer song. The most memorable way. The most memorable way, the most like collective way. So I think that's where we have to think now, but I don't know what happens to our data if it's just like, I mean, I think YouTube has to keep us safe if that's their intention.

15:03And I think companies like Apple cannot be buying scraped data. Like they know that that's against the rules. Right? There's no way that they're like, like Apple's a massive company. They can't be buying data from a data scraping company and being like, this is probably chill, right? Yeah. Yeah. There's no way. So that everyone has to just agree and go, are these the rules of the road or not? Or they have to go, hey, listen, things have changed. We need new rules of the road. So much of this is just unchartered territory. And in terms of protection from YouTube, what I would hope is that whatever becomes the norm, they are pushing for a norm that compensates creators in some way for data being used.

15:50If we're not going to be able to keep this from happening, is there a way, and I'm sure it's incredibly complex and difficult, but is there a way to be compensated when our data is taken and used? And then of course there's got to be like name image likeness protection just to make sure that like our identities are kept safe. Whoever's on the early side of this is probably going to get compensated pretty well because there will be like precedent set. And I don't think anyone wants to be, none of these companies want to be the company that doesn't compensate people for their work. So I think like early on there's going to be high compensation and then it's going to peter off.

16:26And I think that's kind of similar to like when YouTube started, they paid a ton of money to production companies to produce YouTube videos. And then it petered off. And it was like, actually, everyone can do this. Yeah. So I think similar things are going to happen early on. Someone's going to make some AI model licensing deal with someone. Let's say like a doctor, like Peter Atiyah, right? And Apple's going to be like, we are licensing all of Peter's content for our AI models. To train our health model. Yeah, exactly. To our health model. So it's actually an interesting opportunity. I don't think Meta got this right with like the strange kind of like Mr.

16:58Beast as the character Zack that you could talk to as an AI avatar. Like that was weird in my opinion. But I think that they're onto something there where it's like layer on a known personality into an AI model. And that becomes really interesting. You know, like creators become the face of data sets. Yeah, totally. Like they're the anchor piece of the data set. Yeah, I think if like, I think if an education company did a deal with us and they were like, hey, we're going to actually train our creator education model on all Colin and Samir content. And we make a deal with them and we open up our transcripts and get approval from everyone involved and make it a clean deal.

17:35It's probably an interesting deal. And there's like videos that are created that are searchable based on things that are said on the show. That, in my opinion, gets pretty interesting. Yeah. Yeah. I like that. Okay. I think this conversation is like, is, is, is pretty, it's like early stages of this conversation. I imagine we'll have it again. I'd love to bring this conversation to YouTube and bring in maybe other creators or, or bring in other subject matter experts. Uh, but I would also love to hear from, from everyone here. Um, I'm, I mean, listen, I'm not, it's not intentionally a plug, but I did just post about this on my LinkedIn.

18:11Oh my gosh. And if you, and if you want to, you can add to the conversation there. And we will take that into consideration when we talk about this again. It's like this with LinkedIn. It's like all day, every day. I mean, in the great words of - I know. It's every day, bro. That's where you were going. I didn't say it. You said it. I wish I didn't. All right. Okay. All right. Next question. This comes, well, actually that first one wasn't a question. That was just us talking about something. That's true. All right. First question comes from George B. And I can't tell if George B. made this Twitter handle for this question, but the Twitter handle is cut the pod.

18:48And the question is your podcast is wiped off the map tomorrow. You have to start fresh. How do you rebuild? Like is this Twitter handle attacking us saying like cut the pod? Because that's his Twitter handle. Guys, just start fresh. Yeah. Start over. Just cut the pod. It's all over. But it's a good question. Your podcast is wiped off the map tomorrow. You have to start over fresh. How do you rebuild? All right. I'll put myself in that scenario. Okay. I wake up tomorrow. YouTube channel has been erased. Yeah. I walk into the studio. No one's there. Yeah. Just crickets. Okay. And the first thing I would do is get a job.

19:24Interesting. No doubt about it. If I had to start from scratch, I'd go, all right, I'm going to get a job to take care of my finances. I don't want my finances to put pressure on or dictate the success of this channel. Because you and I did that once before. And it was hard. And it was really hard. Yeah. It was hard. Yeah. It was really hard. And you didn't always get to, you know, pick the projects you wanted to do. And because we just didn't have stable finances. So that's the first thing I would do is I would get a stable job. The second thing I would do is make a list of some of my life goals, goals, lifestyle goals, like things I want to add to my life that I am not currently doing to the fullest.

20:05Because I would hope that like this YouTube channel would facilitate some of those life goals. So for me, it would be doing something that gets me outside, doing something that allows me to have new experiences. I would say that's where I would start. It's like make a list of some of these things. What type of channel would get me outside having new experiences. This episode is brought to you by LifeLock. Between two-factor authentication, strong passwords, and a VPN, you try to be in control of how your info is protected. But many other places also have it, and they might not be as careful. That's why LifeLock monitors hundreds of millions of data points a second for threats.

20:45If your identity is stolen, they'll fix it. Guaranteed, or your money back. Save up to 40 % your first year. Visit LifeLock.com slash podcast for 40 % off. terms apply. Interesting. Yeah. I can't keep going. I mean, no, no, I, I think that's, that's just like step one and two. I think that's really interesting. I don't, I think obviously the question is like looking for strategies on YouTube, but it's, I'll get there. Yeah. Yeah. But it is interesting. I think, um, I also think I would get a job. I didn't think about that as my number one step, but I think that's really interesting. I had a conversation with Mark Hustave, the president at Mr.

21:21Beast, where he said this line that really struck me. He said the acquisition of attention is the most important skill for companies to have. And I think everyone who has a understanding of the acquisition of attention is actually a very valuable employee today. So that is really interesting to think about of like, how do you monetize the acquisition of attention? Now, of course, in knowing a lot of these, like the skill set of the acquisition of attention, let me give you, because I would definitely still make YouTube content. And I would probably still make podcast content. I like very much enjoy it.

21:55But in the context of YouTube, the first thing I would do is dive into a niche that I'm like very passionate about, which I would probably say is like entrepreneurship and startups and business building on the internet. So I'm like very interested in that niche. I don't actually watch that much of that content on YouTube. I mostly consume that on audio. Like I listen to podcasts about that. So I'd probably start watching a lot of videos in that niche. And the reason I would do that is to start to look at two types of YouTube videos. One, what shows up on my homepage in desktop and mobile, which is referred to as browse traffic.

22:33And two, after I watch a good video in that niche, what shows up on the right side as a suggested video that I want to click on. And I would start clocking what I like in the context of browse and what I like in the context of suggested. And I think when you, when you zoom out, everyone who is rebuilding or building on YouTube needs to start the conversation with discovery. How does a YouTube video get discovered? And how does a YouTube video within this category that I'm trying to crack get discovered in both of those contexts, browse and suggested. Now, I wanted to ask you a question because it's really interesting just to like open up the, the transparency around how we operate.

23:17We talk a lot about browse and suggest it in the context of thumbnails, packaging titles. And when we went into our episode from last week, which was about hot ones, we sat in our writing room and we said, we anticipate a high percentage of browse traffic for this video. And the reason was because there's not like a extreme consistent output of Hot Ones. We didn't believe that a Hot Ones audience is day-to-day living on YouTube. We felt like this conversation around this new news that Hot Ones might be for sale would hit heavier in Browse. So we mocked it up in the context of Browse. Not only that, we knew that the length of this episode was going to be a lot shorter.

24:05Oh, right. So it came in at 35 minutes, which is also more Browse focused. Yeah, because think about it. Browse, you open up your phone, you open up the YouTube app or you open up the desktop app or on your TV, are you likely to immediately click on a video on your homepage that's an hour and 55 minutes long? No. You're more likely to get served a video that you would actually watch in that moment when you open the app. And that's going to be a slightly shorter video. So we anticipated that this Hot Ones episode at 35 minutes most likely is going to show up on browse, which means we're not catching people in the swing of watching Hot Ones.

24:42They are. We just have to interrupt them right in that moment and make it appealing enough for them to click. Which also means we have to give a big question, right? Like, is this the end? And then the Hot Ones logo is right there. So just a quick note. How much traffic do you think came from browse versus suggested? 65 % browse? Nope. How much? 88 % browse. Oh, wow. 88%. And that's not the norm for us at all. No, that is not the norm at all. But we were right. We looked at this episode and we were like, this is extremely high browse traffic and we have to produce the title and thumbnail to go in the direction of browse.