Sam Cole (9:23)

So anyways, what do you think? Yeah, no, but thank you guys for staying.

Jason Kebler (9:29)

Yeah, thank you for staying. Thank you for not running away in horror.

Sam Cole (9:32)

You're very brave. Yeah, they're sweating, I can tell, but it's fine.

Jason Kebler (9:35)

But Sam, you've written a lot about AI Slop as well and how it can be dangerous. You know, field guides to eating mushrooms generated by AI, things like this.

Sam Cole (9:46)

Yeah, I mean, I feel like it's really funny to look at the slop. First of all, I love that slop is the term for this. I think maybe Jason coined that. Someone coined it. Possibly you, probably someone here.

Jason Kebler (9:57)

Possibly. I don't know.

Sam Cole (9:58)

But yeah, I mean, the AI slop stuff is really funny. But it's also, I mean, we're looking at this as journalists, obviously. So we're like, why is this happening? What is the kind of the ecosystem, the perverse economy that's behind this? And so there's the visual examples, but then there's also LLMs and AI slop of like a text based form. So we had, when we first started 404, one of our first stories is about AI generated books on Amazon for mushroom foraging, which is obviously a very niche hobby that can get you in a lot of trouble if you do it wrong. It's something that the New York Mycological Society had flagged as being life or death. Essentially it's, you know, if you, if you eat the wrong mushroom in the wild, it could be really serious. And these books were saying, you know, like, it's no problem, just cook it and eat it. It's like no human is involved in the expertise that went into this book. But obviously those books sell and you know, people buy them and Amazon makes money and the sellers make money. And I think we're going to get into a little bit about why these make money and how these kind of images make money. But yeah, I mean, it's scary in a way that is so absurd you have to laugh. But you know, it's.

Jason Kebler (11:22)

Brian, you write a lot about AI empowered automation, but have you thought about all the jobs that AI Slop is creating for teenagers who want to get rich quick on Instagram?

Brian Merchant (11:36)

Will anyone think of the teenage Content farmers spamming AI content slop algorithms. Yes. No, I mean, I think that is a really good point and that it's easy to scoff at this stuff, it's easy to write it off. But just like Sam was talking about, if even a few of these guides to consuming mushrooms or foraging for mushrooms make it into the public hands and they don't understand what it is properly that it was created by these algorithms, it can be very dangerous. And on those edge cases, it is having sort of outsized effects on the broader economy. Right. Like so by flooding especially creative adjacent or creative fields. Properly, with all of this AI slop both image and text, it is slowly eroding the market for this stuff. And, you know, it's not really anything new. This is a process that began, you could argue back when people started, you know, blogging or aggregating, you know, Internet content and sort of capturing some of the value that was put forward by reporters or writers or people sharing images online. But by being able to spam this out at such great sort of scale and with such ease right now, it's very cheap, it's very easy to do this. We can't really take for granted some of the potential long term impacts of this. So I wrote a book called Blood in the Machine. It's also the name of my newsletter that I kind of document a lot of this stuff going on now. And you can draw a lot of parallels to what's happening with sort of creative labor now. Back with what happened to the artisan workers, the cloth workers in the industrial revolution, where these sort of technologies of mass production and automation enabled the cheap production of goods, the eschewing of the skilled worker. So there's quite a few parallels. And the stuff doesn't need to be good. Right. It just has to be good enough a lot of times. So there are different categories of slop. The stuff that Jason just showed us, I'm glad I didn't have to watch that. Again, some of it's pretty grotesque. But then there's the stuff that you probably see more of that's just slowly being substituted for other purposes. Maybe it's on somebody's newsletter substack, maybe it's being used on social media, maybe it's making its way into a corporate presentation. So I think we have to both recognize that this stuff is ridiculous. And I think most people object to its artistic quality most of the time, but it can still have real impacts as it does sort of gain steam.

Jason Kebler (14:19)

Yeah, I think that's the biggest thing I Want y'all to understand is that as I've been writing about this, a lot of people say, do people really think that this is real? Are the people engaging with it real? And I spent a long time scrolling through Facebook comments and trying to determine whether the people commenting on these things were real or whether they were bots themselves, whether there was some sort of, like, engagement mechanism here that made these things go viral. But there wasn't really that many human beings in the loop looking at it. And I've come to the conclusion that it's a mix. I mean, there's definitely some automation happening here where maybe there's 1000 likes at the beginning that gets it going in the algorithm. But I think that the target of this AI slob is not other human beings. The people who make it say exactly that. The audience is not human beings. The audience is the algorithm itself. And the audience is figuring out how to game the algorithm so that it does show up in the feeds of human beings. And if something is shown to 100 million people and a few of them like it and comment on it, like, that is enough. And I think from what we know about Instagram algorithm, the Facebook algorithm, it's like opening up a reel and seeing a man turn into a spider in the first three frames of the clip will make you stay on that, because you might say, hey, what the fuck is this? And that's a signal to the algorithm that this is interesting content. And I think that as writers, and we're not artists, but we work with a lot of artists, it's like, we do have to think sometimes about, will this perform well on social media? Will this perform well in the algorithm? And let's say I spend a week working on a story that I report very deeply and put out there, and it doesn't get much engagement. It doesn't perform well in the algorithm. Well, you have these people making 1,000 images in three minutes, and they're able to iterate on that instantly. And so it has the effect of not just flooding out human content, but also they're able to A, B, C, D, E, test all of this stuff. And then when something is working, they can just spam more and more and more of that and collect the fractions of a cent.

Brian Merchant (16:41)

Yeah. And importantly. Right. The platforms don't care, or they're happy to have this happen because it's more content and more engagement when it does fire off. Right. So there's Facebook doesn't really care if the content that's going onto it is AI Generated or human generated. In fact, it might preference the latter. So, I mean, I'm curious because you guys have reported on this a bunch and when you first did, I honestly, bless my naive soul, I thought that Facebook would respond to the outrage by sort of clamping down on it. But that hasn't happened really, at all?

Jason Kebler (17:19)

No, not at all. In fact, Facebook has tried to create its own AI agents. They've talked about allowing users to create AI profiles. Mark Zuckerberg has talked about this future where you might have your own Facebook profile, but then you'll have seven others that are your AI avatars that are out there spamming Facebook for you. It's interesting because people ask me, what is the end game of this? Why is Facebook? And I think from what we know about Facebook, and I always say Facebook because that's where it's worst. But the other platforms are the same, more or less. They're all filled with this stuff. We know that they're trying to advertise to people in as specific a way as possible. They're trying to create psychographic profiles of you, get your demographic information, and then provide you the most optimized advertisement to you. And I think what they are hoping is that AI content will be able to provide you extraordinarily specific content that is tailored to what Facebook thinks you are. And so I don't know if that's like weird crabs turning into Jesus and you're really into that. There will be a never ending feed of AI generated crab Jesus that you can scroll through and they can target ads to you based on that. The other thing that I wanted to say about how Facebook enables this and is profiting off of it is that they're an advertising business. And for a long time you used to be able to make several ads on Facebook. So if you were a brand, you could make like three or four different versions of an ad and you could test different messages and then you could put money behind the one that is performing the best. And very recently they announced this program called Advantage plus, where instead of making two or three ads and a B testing different messages to different audiences, maybe the call to action is slightly different on one, maybe the image is slightly different on another. You can now do this with AI, generative AI, and you can make 1000 ads and target it to 1000 different segments and only put money behind the ones that are performing. And so when I talk about like a brute force attack on reality, it's like it's not just the people who are spamming Facebook, it's that the companies themselves are making and enabling these tools for, you know, advertisers to just hit you over the head with this stuff constantly. I don't think that's good, personally.

Sam Cole (19:55)

Yeah, I mean, I guess something we've talked about in private, but I would love to kind of work through it here too, is why this is bad for everyone. It's like if you don't care about AI and you think, oh well, the stuff is whatever, or if you're like, AI is good in certain use cases. But I think we also, we've kind of gone back and forth on that. But like, why does this kind of trash the whole economy of creativity at scale? I guess if you're working toward an algorithm, appeasing the algorithm and getting engagement for an algorithm, how does that affect everyone? And I know Brian's written quite a bit about this. It's like it's not just, oh, this stuff is ugly to look at. It's actually affecting how we use these platforms and the way that we're contributing to them.

Brian Merchant (20:46)

Yeah, I mean, it is. And it does really seem like, I mean, there's a couple, and there's a couple caveats too, and we could talk about in a second too, because I think one of the things that's enabling this right now is the fact that like the cost of, of, of running and like an AI generated search or producing an image, you know, the, like, sort of like the compute costs and the, and the energy costs of doing that right now are really priced into a lot of models. Or if they are, it's, if it's something, if it is like a model that's being run by a big tech company, then, you know, they have such a huge war chest of capital that they can do that for a long time. But an interesting question is like, is eventually, you know, if those costs are factored in, does it make sense anymore to continue to do this? To sort of, if you have to pay to generate images, even, you know, just a small threshold more, then does that sort of ruin sort of the slop economy for those worst offenders that are spamming it out there? I'd be curious to hear what you think.

Jason Kebler (21:49)

Yeah, I mean, right now, all of the guides to doing this, because there's always a guide, there's always like a guy at the top of the pyramid, if you will, who is teaching the people under Neith how to do this. And for a small fee you can purchase their prompts and their tips and things like that. Almost everyone is using free tools at the moment. And by free, I mean free trials of generative AI tools or things like Bing Image Creator, which is free. And when you run out of generations on the free account, you can just make another account and have seven different accounts and do that. And so it's kind of the Marc Andreessen Uber effect, where Uber was cheap for a very long time because it was being subsidized by the venture capitalists that were investing in it. And that's the period that we're in with artificial intelligence right now, where the people investing in and building this technology are taking a huge loss on this stuff and presumably they don't want to do that forever. So there may come a day when it's not profitable to do this. I think the big question for me is whether social media will be completely destroyed before that happens. And what I just showed you where I was scrolling my feed and that's what I'm seeing. I don't think that's most people's experiences at this very moment. I think that I've given enough signals that as a journalist who writes about this show me the weird AI stuff, that's what I've told the algorithm that I want. But as you saw, like a lot of these videos have hundreds of thousands, millions of likes. And so that is starting to happen on Instagram and Facebook at scale where people's feeds are being taken over by this stuff. And I wonder when the collapse point comes where it is so hard to find other human generated content that these social media platforms are just all AI, like more or less. And it's happening on search as well. I mean, Google is being flooded. And of course Google itself is putting AI overviews at the top of things too. But the entire SEO industry is AI focused at this point too.

Brian Merchant (24:03)

And I mean, it's just a reflection of the problem when you have just like such highly monopolized sort of tech infrastructures and platforms that we rely on that maybe Facebook would be more responsive to sort of the abuse of being flooded by AI images for a little bit of profit for some content farmers, maybe Google would be more receptive to improving its product and not serving us, you know, results that are going to tell you to eat poisonous mushrooms or whatever if, you know, there, if there were more viable alternatives or an ecosystem in which they didn't dominate 90% of the market share. And I do think that, like I do, I do really ultimately think this is a really toxic thing because I kind of see, as you call it brute forcing, which I like. I also think about sort of AI slop almost as kind of like the shock troops, right? They're just kind of like these are the most egregious examples of sort of AI generated stuff. And then it kind of moves the Overton window and makes other AI generated stuff seem a little bit more reasonable by comparison. Here's a fun and topical example. So I was the tech columnist at the LA Times until last year when I was part of the mass layoffs along with like 120 of my colleagues. And then at the end of that year, the owner, Patrick Soon Shong, who has not managed to keep himself out of the news much these days for a number of reasons, but he announced a new feature called AI Insights. And it's a new feature that's on an opinion piece or they call them voices now. But if somebody writes an opinion piece now, it features an AI that will determine or attempt to determine the political content or positioning of this piece. So it'll say this is a moderate liberal piece or something like that. And then it will take the initiative to offer some, some counter examples and some yeah, buts, right? Some both sides ing, which is what all journalists love is when they have to both sides their argument. Anyways, this thing recently got in trouble for somebody wrote a column, one of my former colleagues wrote a column saying, my hometown of Anaheim has this checkered past because at one point the city council was dominated by the KKK and we need to remember this sordid history so that it doesn't happen again. And of course the AI bot comes out and says, well, maybe the KKK wasn't that bad. Maybe it had some points. Some people have said it wasn't a totally hate driven movement and obviously this caused some backlash. They had to get rid of the AI thing for now, at least on that article. But this is just an example. Like here's some AI generated content that is, I think we could call it sl. It's garbage. Nobody really wants it. And yet the management of a company is trying to use it to sort of create more value. In my last newsletter, I kind of being cheeky, said they've replaced me with an AI that defends a kkk. But it's not quite that one to one. But you can look at the value propositions here, what the tech companies are trying to do to sort of plug the gap or to see how what they can get away with by generating content as cheaply as possible or getting people to engage with content as cheaply as possible. And it's not just Big Tec. Right. We've seen this at Sports Illustrated using AI slop. We've seen it at some portions of USA Today and seen at the old tech company. So it is like creeping up buzzfeed. Buzzfeed fired all of its reporters and then said we're going to try making some AI quizzes instead again. So it's just this rise of this force that is, you know, I do think that it has this destructive capacity that needs to be recognized.

Jason Kebler (28:07)

Yeah, I think that. So Sam was the first person who ever wrote about deepfakes ever, and she totally owned that story. You know, 2018, something like this many years ago. And when deepfakes came out, like face swapping technology, Sam was reporting on how it was being used to make non consensual pornography of celebrities and women and you know, essentially do sex crimes to people. And the immediate response from like politicians was one of this is going to be really bad for national security because what if someone makes Donald Trump saying that there's going to be a launch a nuke or something. And so I want to ask Sam, because I think to Brian's point, there are, there's the intended use of this stuff which is like use it to make yourself more efficient, use it to make your workers more efficient, so on and so forth. But it's like every time one of these things gets released, there's like the quote unquote utopian productive uses of this thing, which we could argue about ad nauseam anyway. And then there's the way that people are actually using this. And to this day, the dominant use of artificial intelligence is to make porn of non consensual porn of women. And Sam has done amazing reporting on that and just wanted to hear more.

Sam Cole (29:30)

Yeah, yeah. I mean, I don't know if it's. I feel like deepfakes are not quite in the slop bucket, but slop is kind of like an outcropping of the deepfake phenomenon. Yeah, I mean that's the story with the Internet too.

Jason Kebler (29:47)

Right.

Sam Cole (29:47)

It's like people had this utopian idea where it was going to be like an amazing thing for humanity and it was going to be nothing but good and egalitarian and beautiful. And then people started using the Internet in a really big way and you know, hundreds and then thousands of people started getting online and actually interacting with each other and it kind of turned crazy quickly. So I think the same thing happened with AI and with this technology someone found a way to face swap really easily, which was Something people were already doing using AI and it just exploded from there. I think the experts that I talked to at the time said, oh, give it, you know, this will be a huge thing within a few months. And it was maybe three weeks where people were making non consensual intimate images of celebrities en masse. It was a huge thing. And now I think we're seeing it get more awareness. I think because it's happening in schools, kids are dealing with this now. And I think that's something that we're going to have to kind of, we're going to have to reckon with legally and as a society and as far as like Internet regulations and things like that. But also just, I think there needs to be some kind of introspection about how we got here, why we, why we let it get this far. Who kind of let it go to the point where 13 year olds are making porn of their classmates using AI. I think that's something that the adults in the room need to sit down and introspect about a little bit. And I think that's something that we try to do when we write about this stuff in a really early stage. It's like, let's talk about it now, let's kind of get the lay of the land now and figure out what's going on before it becomes so commonplace that a 12 year old is getting arrested for it. You know, that's, it's such a, such an explosion so quickly. So yeah, and I mean the KKK example is really crazy. But also it's just showing up in consumer ways too, I think. Like the Amazon reviews that we wrote about recently, where reviews of Mein Kampf were getting five stars because people were like, it's a nice print of a very concerning book. It's not five stars on the ideology within it. But then the AI said, oh, people say this is an easy and interesting read and a true work of art. And it's like, I don't think, I think AI is kind of slipping in, in these ways that it's now cracking open where lots of people are realizing it. So you aren't in the depths of the Internet anymore looking at deepfakes or looking at, you know, weird stuff that it's on. It's in the LA Times.

Brian Merchant (32:19)

You know, has the character, since you have been on the deepfake beat for so long, has the character changed of like, of the way that people are producing or engaging with deepfakes changed sort of materially since this, since the sort of reinvigoration of the, you know, midjourney chatgpt, the more recent AI boom.

Sam Cole (32:39)

It's definitely easy. It's so much easier and it's really realistic. I think looking back at how it was done in 2019, you needed hundreds or thousands of pictures of someone. That's why celebrities were getting de faked. Now you need one picture and you can make a full video. It's not even just make an image. This is something that our colleague Emmanuel.

Jason Kebler (32:58)

Writes about you even with a text prompt at this point too, which is something that everyone in this, I hate to call it community, but the community of people who make porn of Taylor Swift will try immediately. Whenever a new AI model is dropped, they will see if there are guardrails on this new tool that can be circumvented in some way. Often they can be, unfortunately, yeah, it's.

Sam Cole (33:26)

Hard to guardrail around the human mind. I feel there's just no way.

Jason Kebler (33:43)

Our event at south by Southwest was sponsored by Deleteme, which has been a sponsor of 404 Media for a long time. Delete Me is one of the few products that I've used for years and it's made a meaningful difference in my life. Delete Me makes getting your personal information deleted from data broker websites really easy. As someone who's been doxxed multiple times, that's really, really important. With data brokers and bad actors increasingly selling your information and trying to dox you, it's great for you, your family, or your employees. Delete Me makes it harder for these bad actors by scrubbing your employees details regularly. It's simple. Attackers are lazy. If it's too hard to find your contact info, they'll move on to easier targets. Deleteme takes care of this for you, doing the heavy lifting so you don't have to worry about it. They keep removing the information so it stays down, protecting you and your team from constant exposure. That's what I personally like best about DeleteMe. When I used to try to get my address deleted from different data broker websites, I would spend hours filling out different forms and trying to figure out how to get myself removed from any individual data broker. I'd succeed, but then another data broker would just pop up and I'd have to start the entire process all over again. With Delete Me, it's set it and forget it. So go to JoinDeleteMe.com and use code 404Media for 20% off of DeleteMe today. That's JoinDeleteMe.com promo code 404Media now back to the event.

Brian Merchant (35:17)

Let's open it up for some questions for a few minutes. If anyone wants to talk AI Slop.

Jason Kebler (35:23)

Yeah, one. One second. Before we do questions, I wanted to thank all of you for coming and I wanted to thank Flipboard for letting us do this is very, very, very cool.

Brian Merchant (35:33)

Round of applause for Flipboard. You guys are the best.

Jason Kebler (35:37)

Surf is amazing. So check out their app store. And then also we got some help from Deleteme for this as well, which is a product that I actually use. So we have like discount codes for that. We have discount codes for subscription to 404 Media as well. We'll talk more about that later. But thank y'all for coming. I'll give you the mic so that we can hear it.

Brian Merchant (35:59)

Hi, my name is Wesley and the seems the balance of the power overwhelmingly.

Jason Kebler (36:05)

Is actually the advertisers.

Brian Merchant (36:07)

So if they ever get sophisticated enough.

Jason Kebler (36:11)

To be able to detect and say that this is an issue for them.

Brian Merchant (36:15)

Because they're seeing maybe more views, but diluted conversions, I know in marketing there's a very strong metric of looking at the number go up, but when the views aren't translating to sales and they go to Meta and say we need to. To be taken care of. Does Meta make a new tier of verified humans viewers? Do they lower their cpm?

Jason Kebler (36:43)

How. How does.

Brian Merchant (36:45)

When. If this tide does turn and then the. Because it's all about money. And if they go to advertisers and.

Jason Kebler (36:51)

Saying we can sell you this tool.

Brian Merchant (36:53)

For you to get a score of.

Jason Kebler (36:55)

How many people are actually viewing this.

Brian Merchant (36:57)

They'Ll eat that up, whether it's accurate or not. So what happens? What does Meta do?

Jason Kebler (37:02)

What does CNN do when they hear this? I think that's a really amazing question and it's something I think about constantly because I think my opinion here. But there's a gigantic amount of fraud happening across not just social media, but also, you know, a lot. A lot of the AI slop we just looked at leads off platform to websites that are laden with ads and also like other additional AI slop. And a lot of the traffic to these websites are bots and a lot of the people engaging with them are bots. They're not people at all. And I think that this is probably a very massive scandal that people in the industry may understand, but that has not been reported properly to this point. And it's something I've tried to do a little bit of reporting on, but it's very difficult. It's just the Internet is very vast and getting access to whether, you know, Facebook's views are real and the likes and comments are real and things like that is really tough. And then at the same time, I think the ad industry feels like the advanced targeting that AI in general provides them and then the customization that generative AI provides them, where instead of showing one type of ad to 100 different types of people, they can show 100 types of ads to three different types of people and then put money behind the one that's working. I think that on balance, the advertising industry complicit in all of this as well. And so I think that I'd like to say that maybe they could hold. Hold them accountable if they're not happy. But I think on balance, they may be getting better performance because this is one of the few ways that AI is actually working. Like if you look at Meta's earnings reports and things like this, they say, actually people are spending more time on our platform now and ad performance is up and blah, blah, blah. And whether those numbers are real, I have no idea. But that's the narrative at the moment.

Brian Merchant (39:11)

Yeah, it is. It's a really interesting question, and one that theoretically, again, should, you know, should mean a great deal to advertisers trying to see a return on their spend. But once again, it's another case where sort of the dimensions of the monopoly, the platform are said, where else are you going to. Where else are you going to advertise? What are your other options? You're quite limited. And Jason's exactly right. Meta has been, you know, on a tear right now, and AI is one of the reasons that investors are citing as being behind its booming stock valuation, even though it hasn't sort of historically been thought of as one of the leading sort of AI companies in the AI space, even though it's got Llama and its open source project and all that. But precisely because advertisers and investors are just tantalized by the prospect of using AI to distribute ads and to do kind of a version of what we've been talking about, about today, it's really sort of rising Meta's stock price to this radical degree. And I think that just underlines the extent to which these AI technologies are useful or interesting or enthused by investors is that they are taking place on these monopoly platforms where the company can have so much control over them.

Jason Kebler (40:32)

There's two related points I want to make very quickly. One. One is, I don't think that people like generative AI, by and large, I don't think this is Something that people want, but because of the nature of monopolistic power, it's being shoved down people's throats, it's being surfaced in the algorithm and you cannot escape it. There was this video of a spider turning into a man in an airport that I saw earlier today that had been viewed according to Instagram, 330 million times. And we've been doing 404 media for 18 months and we've been pretty successful. That is like 10 times, 15, 20 times as many views as all of our articles combined in that 18 months. And you know, the person making that spent one half of a second making it. And that is, it's pretty demoralizing to be honest. It's see that sort of thing. But these platforms have a scale of such that it's mind boggling. And so I think that we can siphon people often and bring them to our platforms. And I think that increasingly we need to reach people through group texts and word of mouth and newsletters and the Fediverse and decentralized social media that's not dominated by algorithms. But when you're talking about how many people are on the Internet, it's pretty tough. And then the other thing, no, I said it'd be quick, but it wasn't. People have their guard up about artificial intelligence now. My partner's aunt reads 404 Media and she sent me some AI generated content that she saw on Facebook the other day and said, I don't want to get tricked by this AI. What she sent me was a video of a girl drawing a, doing a painting. And it was real. It was like, it was real. This woman was like painstakingly documenting what she was doing. And I think that beyond just flooding out everything else, it's like people don't want to get tricked by this stuff. And so they'll see real things and they will say, well that's bullshit, it's AI. And that that's become like a mechanism for saying you're fake news or you're making shit up. And I think that that is very, very alarming. So with the idea that people are creating more and more with AI or companies or whatever, and then the AI.

Brian Merchant (43:02)

Is also training on that data, I'm.

Jason Kebler (43:05)

Assuming, because so there's going to be.

Brian Merchant (43:10)

So much more AI data out there than there is actual data developed by humans from what I saw on the Internet.

Jason Kebler (43:17)

And I don't believe that everything's true because AI.

Brian Merchant (43:22)

It talked about how AI training on AI generated image actually degrades the images when it spits out new generated images.

Jason Kebler (43:32)

What are your ideas in regards to? Or any research that's out there in.

Sam Cole (43:37)

Regards to where that's headed with all.

Brian Merchant (43:40)

Of the proliferation of AI generated data or generated content.

Sam Cole (43:48)

I mean, I think the companies know that this is a problem. Right. Like, they kind of recognize the big companies, Microsoft and Meta and the ones making these tools. It scares them, I feel. I mean, I. Isn't this something that.

Jason Kebler (44:01)

There's a professor at, I believe ucla, Hani Farid. Is he at, at ucla?

Sam Cole (44:07)

He's at Berkeley.

Jason Kebler (44:08)

At Berkeley perhaps, who's done a paper about this. And you know, I have some hope that this, what does Jason call it? Habsburg AI. Basically, like human. Human Centipede AI, where it just degrades.

Sam Cole (44:23)

Over time, eating its own tail.

Jason Kebler (44:25)

I mean, I think that it's a good thought and I think that without the models being altered, you know, it's something that is happening and could happen. But at the same time, I think the AI companies realize that this is problem. And I believe the way that Deep Seq was trained, a lot of the training that it was done was based on data that itself had generated. And so that's one of the reasons it was able to be a bit more efficient. And so I don't think that is something that we can count on to save us, I guess, is what I would say as much as anything.

Sam Cole (44:58)

It's making it worse.

Jason Kebler (44:59)

Yeah, as much as I'd like to see it happen.

Sam Cole (45:01)

I mean, yeah, it's a nice thought, but like, I like, that's why they're so desperate for human. They're desperate for your data, they're desperate for your creative output.

Brian Merchant (45:08)

That's why the AI companies are hiring people to just sit in a cubicle and type out new.

Sam Cole (45:13)

Yeah, just gobble it all up.

Jason Kebler (45:16)

Are you seeing a push toward companies trying to use AI to curate your AI feeds?

Brian Merchant (45:23)

So, like the way that Google is searching the web and saying, okay, I'm going to pull the things that you want. And do you see an AI that wants to be your gateway to your Facebook feed and your Instagram feed and all that kind of stuff and pulling that one step closer to you.

Sam Cole (45:43)

Don't give them ideas.

Brian Merchant (45:46)

I mean, they have.

Sam Cole (45:47)

I mean, that's. It exists now.

Brian Merchant (45:49)

I mean, yeah, I mean, it's been a. I wouldn't say that that's like the generative AI is something a little bit different, but you've had algorithmically curated feeds for years and years and years now, and that's. That's why Kind of incidentally, why. So Jason has a dummy account where he just was actively looking for the AI and then it just serves him more and more. But I do think. I'm glad you bring it up because it does illustrate a good point where once sort of there's enough of the stuff that is palatable enough to the user and that they click on some of it, then we can only assume that it's going to serve more and more and then it's going to slowly sort of accelerate the process. Right. Or maybe not so slowly.

Jason Kebler (46:28)

Do this side of the room and then come back.

Sam Cole (46:31)

We see you, we'll get to you. I can't see your face though.

Jason Kebler (46:33)

You already started talking about this, but I was wondering how you see human created content adjusting to try and get.

Brian Merchant (46:40)

More attention and overcome all the AI slop.

Sam Cole (46:47)

Yeah. How humans are trying to adapt to the algorithm. Yeah. I mean that's something that I, I think of YouTube. Right. So like the shock stuff on YouTube, the Mr. Beasts of the world altering your thumbnail to make your face be like, you know, that classic people have been kind of tweaking their own behaviors to appease the algorithm for a minute, which I think is dark sided. I don't like it, obviously.

Brian Merchant (47:14)

Yeah. And one thing I bet we can expect to see more of is that are those demands changing more rapidly because as they get spammed, if you can create 1,000 images in a few minutes and then fulfill the current demands or requirements of that algorithm, then users are going to get tired of that pretty quickly. So it's going to have to shift. So we might not just have a more AI slop filled Internet, we might have a more volatile one where things are broken or glitchy or weird more often too.

Jason Kebler (47:47)

Can you pass the mic to someone else or do I have it?

Sam Cole (47:52)

Someone back there is. Someone back there is.

Jason Kebler (47:53)

Wait, I thought this was a fourth, A mysterious fourth mic.

Brian Merchant (47:58)

Please support independent media so next time we can have someone running the mic too.

Jason Kebler (48:04)

All right.

Brian Merchant (48:05)

I spent most of my career as.

Jason Kebler (48:06)

A journalist, which means I'm an optimist that's been continuously disappointed by life.

Brian Merchant (48:13)

And I've also came up through the.

Jason Kebler (48:15)

Social media growth and personally believe that it's objectively made the world a worse place.

Brian Merchant (48:22)

And I would like to poll you.

Jason Kebler (48:24)

And then maybe the audience from a scale of.

Brian Merchant (48:26)

It'll be okay to.

Jason Kebler (48:28)

No, we're completely fucked.

Brian Merchant (48:31)

Where do you feel that we are in regards specifically with generative AI or with media in general? Maybe we can convene another panel after this one. And I can hold forth. No. I am also always an optimist. I am perennially an optimist, even as conditions, I think are the worst that they've ever been for journalism in my lifetime, for American journalism. I think it's just an absolute train wreck right now. And this is before the LA Times, I was at Medium, which all the editorial work workers were laid off in one fell swoop there. Before that I was at Vice, which went bankrupt and is now owned in parts by a private equity firm. And that's just. Most journalists working these days have similar horror stories. And that's pretty reflective of the trajectory of the industry at large. And it's gotten to the point where if the trajectory continues unabated and we don't fight, we don't push for policy interventions, we don't push back against these countervailing trends, then yeah, I think we're in truly disastrous crisis mode right now. And I think journalists, one way to fight back is to do with what these fine folks have done and start an independent media company that can play by different rules and can adapt. I think local news is in such, is in truly, truly dire death spiral levels right now. And that's where it's not. You know, you need new tools.

Jason Kebler (50:14)

Ryan, did you see that one of the most popular local news outlets at this point is an AI generated newsletter that goes out to like thousands of towns in the United States.

Brian Merchant (50:26)

Awesome. Yeah, I did see that.

Sam Cole (50:29)

And meanwhile the fucked meter is now we've changed it.

Brian Merchant (50:32)

Yeah. Meanwhile, a city of hundreds, hundreds of thousands of people, Salinas in California recently is down to one editor. Their newspaper as one working editor, no full time reporters. And you can just kind of, you can imagine what's going on at the city council meetings and what's going on at the local real estate firm. Things that just nobody's knowing about now because so many of these sort of responsibilities have been. And they have to, we do have to be more critical of these tools. Media companies have had a bad track record of pairing up and partnering with the tech companies that have been eating their lunch for the last two decades. In my humble opinion, the whole pivot to video mentality and constantly trying to play by the rules, forming new licensing deals or partnerships with the AI companies. It's just going to lead us further down this spiral. So, you know, it's time to fight.

Sam Cole (51:34)

So yeah, yeah, I mean, I think we can pull this up out of that. I think we can get it. I mean, you're 100% correct. But I think. And this goes back to the question about, like, how are people adapting to algorithms? I think you can either try to adapt to the algorithm and try to appease the algorithm, which is the SEO game pivoting to video, whatever in journalism, or you can extract yourself from it for the most part, which I think is what people have started to do. It's what we're doing in a large way with our email newsletter. We want to talk to people directly. We don't want to talk to them through whatever algorithm decides to feed them our content. We want to talk to them because they care about what we're writing about. And we see it happen with people like Brian writing Blood in the Machine and people getting directly to their readers without the SEO game, the Google algorithm game. And I think that's a really hopeful thing that. I mean, I feel more hopeful about journalism than ever right now, ironically, even though the industry is in total chaos, because people are saying, you know what, let's do it a different way, I think we can do it differently.

Jason Kebler (52:37)

Everything that Brian and Sam said. And then also, it's just like, look around, there's, you know, 150 people here, 200, I don't know, a lot of people here. So a lot of people care about this. I think that you look at what Flipboard is doing, you look at what Ghost is doing, you look at the incredible energy around independent media, but also decentralized social media, and I think we're on the right path. We have some folks from Wikimedia here as well. One of the gives me more hope than anything. Just this. The greatest collaborative project humankind has ever done, perhaps. And so I think.

Brian Merchant (53:15)

Wow.

Jason Kebler (53:16)

Yeah, I think so. So I'm optimistic, despite all the stuff I just said and showed you.

Brian Merchant (53:26)

Hey, so, like, the. Put aside the automatic viewers of the AI slop for a second. What types of groups do we see being instrumentalized? Like, can you think of any more that I can't? So, for example, very young people, right? The children's content was one big one. And now they're talking about older people.

Jason Kebler (53:48)

Are being iPad addicted as well.

Brian Merchant (53:50)

And that's becoming an issue. And then maybe in the middle, like, hipsters and stuff, using AI, you know, ironically, blowing up Vance's head or whatever.

Jason Kebler (53:59)

You know, but then being consumed by.

Brian Merchant (54:01)

People who are not in on the joke is another one. Are there any other, like, human segments that are, like, amplifying the slop or using it for a different reason?

Jason Kebler (54:11)

So the question is just like, is there a demographic that's more responsible for amplifying this stuff. It is a lot of old people on Facebook. Not exclusively by any stretch, but I've scrolled through just like so, so, so, so many viral AI slob Facebook images. And, you know, it's people who spend a lot of time on Facebook. And I think that it's no surprise that Facebook has shown itself to be most susceptible to this in the early days because it's more popular with an older demographic. Which is not to say that younger people don't fall for it or don't promote it, but I do think that there's maybe an AI literacy aspect to this or like a news literacy aspect to this. And it's just like the Internet is a really, really complicated place to navigate at this point, and Facebook itself is a really bloated and tricky UI at this point. I don't know if you've been on Facebook lately, but it is a mess just to use at all. And so I'm not surprised that people can't navigate it. But I also think that because Facebook has been around for so long, it's almost like looking into the future for these other platforms because it has had a longer time to decay than the other social media platforms. And I think that that's not to say the others aren't susceptible to it. I think they very much are.

Brian Merchant (55:43)

Yeah, I think that's a good place. One more, one more, one more.

Sam Cole (55:50)

And then we're going to take a break and we'll be back.

Brian Merchant (55:52)

Two more.

Sam Cole (55:53)

We're going to take a break and set up our merch table, probably, so you find folks can make some purchases.

Brian Merchant (56:00)

Hi, I'm a social media founder myself.

Sam Cole (56:04)

And so I've seen a lot of.

Jason Kebler (56:07)

Evidence for how many social media accounts.

Sam Cole (56:10)

On mainstream platforms are bots. And not just bots, but really convincing bots.

Jason Kebler (56:15)

And so you've talked a lot about the real people behind this AI slob, but I'd love to know more about how you can even differentiate between those.

Brian Merchant (56:27)

Accounts and who's real and who's not.

Jason Kebler (56:29)

Just because they have become so convincing.

Brian Merchant (56:31)

Even though these AI videos in many cases are not. You want to ask your question too? We'll get those last two.

Jason Kebler (56:39)

Yeah, we're going to answer that one.

Sam Cole (56:43)

Mine is more of a cheer.

Brian Merchant (56:45)

So I created the Wikipedia article on AI Slop and it could use updating because that is how AI Slop learns about AI Slop, mostly through ingesting Wikipedia.

Sam Cole (56:55)

So if people want to have impact.

Brian Merchant (56:57)

On a conversation, the article is there.

Jason Kebler (57:02)

Thank you.

Sam Cole (57:03)

What was the question. I can't hear anything up here. I don't know why.

Jason Kebler (57:08)

The question was, how can you recognize AI Slop? More or less. I mean, more than that. But you're absolutely right.

Brian Merchant (57:17)

That and the accounts created.

Jason Kebler (57:18)

And the accounts creating it and the accounts engaging with it. I could talk about this for like a half hour, so I won't. But come find him after. Yeah, come find me after. So for a while, it was very easy to tell what was AI generated just because it looked bad. And there's this saying in AI world that this is the worst that it'll ever be because it's always improving. And I think that it is. I mean, it's getting harder and harder. As someone who's looked at, like, honestly, hundreds of thousands of AI Slop videos and images, it's like you can't just look at someone's hand and see if they have six fingers at this point. It's like they've gotten a lot better. There are some that still have watermarks on them, which I think suggests that people don't actually care if things are AI generated or not to some extent. And then as far as the accounts engaging with it, I wrote an article about why I think a lot of them are real. And it's because I would go into their Facebook profiles, see whether they were posting only about AI Slop or whether they had posted real images of themselves or had commented with other accounts that seemed real, things like that. That's obviously not doable at scale, but I was so obsessed with the idea of are these people real or are they not real that I spent couple weeks just clicking on a bunch of comments on AI generated content and trying to figure out whether the people were real or not. And the answer is, some of them are.

Brian Merchant (58:53)

All right, well, all this talk of AI Slop and the decline of journalism. I'm ready for a drink. I think we can call it here. Once again, 404 Media. They've done something really amazing. Support independent journalism. Subscribe to 404 Media. And since you've made it to the end, we can reveal now that every single article they've written has been produced by AI. So thank you. All right, open bar. Come find us if you have more questions. Say hi. We're all friendly people.

Jason Kebler (59:24)

Thanks so much. Yeah, thank you. It means the world that you're here. Thank you. If you like this panel, we're looking to do more of this sort of thing this year. So if you're throwing an event or have a conference or talk. You're interested in having us come out to. Reach out to me at Jason at 404 Media co. See you all next time.