Combating AI Slop with Stephanie Humphrey - DTNS Weekend

0:02Welcome, everyone, to DTNS Weekend Edition. I am Sarah Lane. This is the fun part of the week where we have a fun and enlightening interview with somebody who can help us all learn a little more. Today, it's Stephanie Humphrey. Hi, Steph. Hey, how's it going? It's going well. It's going well. How are you? I'm good. I'm good. Happy to be back on DTNS. Well, we're happy to have you. So I know today we were going to talk about AI slop. Now, that is a term that gets thrown around a lot for a variety of reasons. But we're talking specifically about how AI can confuse people on social media. So what have you been looking at?

0:51So it's interesting because it's a couple different things, right? And when we talk about slop, I think it's more the idea of careless information, carelessly generated information. because I don't think the large majority of the things that we consider slop are intentionally malicious and intentionally are trying to mislead or be propaganda or anything like that. We got plenty of real creators doing that kind of thing. But I think a lot of the AI slop, if you will, is more careless generated information. So there might be an error in something or it might have hallucinated and made up something entirely.

1:35But the point is that it seems very real because it's presented with a lot of confidence and a lot of certainty as if it was factual information. So that, I think, is a little bit of a distinction between, you know, AI slop and just general misinformation or disinformation on the web. It's just careless. People just did it and they didn't think about whether or not it was right. They didn't care to check whether or not it was right. They just created it, generated it and put it out there. Yeah. Or how it affects somebody who sees that and goes, oh, that's real. Right. Even if it isn't. Exactly.

2:10It kind of reminds me of the olden days of people saying like, oh, that's Photoshopped. Back in my day. Yeah. We're all kind of used to the fact that something that seems real isn't always real. And some people just are going to take it, you know, willy nilly. And, you know, those of us who are kind of like, well, hold on a second. You know, is this real? You know, let's do some journalistic approaches. Right. You know, that has been going on for some time. I wonder how much the AI stuff, though, is hurting that journalistic approach. I think very much so right now, at least anyway, because there was a recent study that came out.

2:59SEO company called Graphite did a study and we are past the tipping point now where just over 50 percent of everything on the Internet is AI generated. And a lot of it is the slot, but we're at 52 percent. And I think that's all AI generated. And Pew Research did a study that said about a third of adult Americans can't distinguish between AI generated content and human generated content and real content. And I think that number is probably super low. So we're in a very precarious place with respect to how we consume content on the Internet. So thinking of X, formerly Twitter, as an example, I see all the time, it's like somebody will post something and the next response is, grok, is this real?

3:52Grok, is this real? Or you're part of the community who's supposed to add community notes, which I am a part of. So I'm kind of – I'm into that in general. There's a lot of stuff where it's like if it's supposed to be entertainment, you got the Soura watermark. I mean, guys, we know what this is. But so many people don't. It's all happening so fast. That's the thing. And you're – I think it's a younger audience as well. So you have that demographic that doesn't have the fully developed prefrontal cortex and all that good stuff. And you also have older adults as well who are flocking to the internet more than ever right now.

4:36I think about my mom who is always bringing me something on her phone and I'm like, that's not real. That's not real. It's okay. That's not real. You know what I mean? You're just reassuring her that the stuff that she's seeing is AI generated. So we are really in a very sort of tenuous place as it relates to content online and news especially because you don't have a lot of news outlets or quasi -news outlets disclosing that a lot of their stuff is AI generated. And the video stuff is still, I'm not going to say easy to tell whether it's real or fake because some of it is not very easy, but there are some signs.

5:15But you think about the AI slop text online. You have whole blog posts and whole articles that are written as if they are presented from some legitimate journalistic outlet, but are completely AI generated and are rife with errors and hallucinations and stuff that's just made up. So, you know, we got to kind of we're battling on a couple of different fronts as it relates to the slop. Yeah, I mean, I guess when it comes to video, it's like, well, maybe the solution is always have that watermark, you know, watermark never goes away. That's tough, too, because you there are now watermark removers out there, you know, people are cropping the video so that the watermark is not visible.

6:03So they're already kind of finding ways to get around that as well. That's our best line of defense. But as you just said earlier, even with the watermark, some people aren't even going to pay attention. You know, I remember Sora does a decent job of making the watermark fairly prominent. But there are other tools that make it very hard to detect. It's almost transparent. You know, it's at the bottom corner. It's tiny. So, you know, a lot of people aren't doing a lot of the companies, the platforms aren't doing their due diligence as it relates to the watermark. But, you know, to your point, when we get to the to the print side, you know, who knows what's real or what's fake?

6:43They're snatching screenshots of random people or they're making AI generated little avatars to make a fake name up for a journalist. And so, you know, on the print side, I think it's even more important to be a little bit more discerning about what's real and what's not. You know, it's funny. I was using I use chat GPT. I use a variety of LLMs, you know, for reasons, but I use chat GPT sometimes to like format something for a sub stack post, right? You know, it was started in Google Sheets, you know, format this for a sub stack post. And the other day for the first time, it said to me, you know, because I was like, format this, you know, and here's all the text.

7:25and it was like, do you want to make it seem like it was you talking? I forget what the prompt was specifically, but I was like, whoa. Yeah, I was like, whoa. He can do that now? Yeah, I was like, well, I mean, I wrote all this. So, I mean, I feel pretty confident in that. But it was like it was taking it a step further, which is maybe sometimes that would be great. um you know maybe it would sound better than me at some time um but but does it need to though well and and and are your words this is these are your thoughts you know expressed on the page do we need to make it sound better you know better than what is right yeah it's like better than me right i'm like i mean you could always be better than that but you know it's it yeah it was it was It's something where I was like, huh, I mean, I guess I could say yes to this, but then I'm presenting this to the world as myself.

8:29And it's asking you, do you want it to take your voice away? Right. That's basically what it's asking you. And how do you answer that when that's what everybody's doing these days? You know, it's like, no, I don't want you to take my voice away. I want this presented the way it came out of me. And that's the risk we take with all of this slop because, you know, they talk about recursion and the fact that these LLMs are going to continue to be trained on slop. If 52 % of the Internet is AI generated and that's the stuff we're going to be feeding back into the models, you know, what is the output going to look like a year from now, five years from now, 10 years from now?

9:12And do we want our voices to be extinguished, basically? That is a very good question. I wonder, specifically, I don't have children of my own, but I wonder about the younger set, knowing kind of this is what they know. And younger than the least last younger set, which is internet is all we know. I remember a life when there was no internet. That's how old I am. Same. But, but you know, the, the, the youngest set, knowing that, you know, these are tools, use them in school. I know all schools are different. Classrooms are different. But, um, you know, that's, that's something that feels like you really gotta, you really gotta get a handle on that or, or else the kids aren't going to learn anything.

10:05They just know how to get around learning. Now, I say that with a caveat of, you know, when I was in like trigonometry, and I, you know, I had a calculator, it's like, could I do everything without the calculator? Of course not. That was my crutch. But we all had them. And so I'm going to stop you, though, because there's a difference. There's a really great article. I mean, I wrote it down. The Atlantic did a really great piece. It's called AI's Invasive Species, and it's by Charlie Warzel. And he talks about the idea of creativity without craft. There's a difference. You know, you can be as creative as you want with these AI tools and you can write a prompt that would output something that no one's ever seen before and is beautiful, but you have no idea how that happened and how that was done.

11:00Now, to your point about the trig, like, yes, you need the calculator to do some of the things, but you knew the formulas you were supposed to use. You knew the sort of methodology of getting through the problem to get to the answer. You had studied the craft, you know, and that's where we get in trouble, I think. And that's where we're going to get in trouble with AI. We got a lot of creativity, but no craft. And it really, you know, as as humanity, honestly, I think we need both. So a lot of people are, you know, wringing their hands about this a bit. And for good reason. What where do you stand on?

11:40What do we do going forward? These tools aren't going away. They're here to they're here to stay. So so what what do we do to not kind of lose sense of ourselves? Well, we're going to have to get very creative. I think educators and parents are going to have to get very creative in the way that they teach. You know, a lot of stuff is going to have to go back to the old school. Like you may have to, I think we're going to see a lot of oral sort of education and exposition. Like the kids are going to have to stand up and defend whatever it is that they're, you know, purporting. If you wrote this paper on the Civil War, you have to get up and defend it and you're going to need to know something in order to get up and defend it.

12:23So we're going to have to get creative about how we infuse critical thinking back into the process. I don't have an answer for that right now myself, but we're going to have to figure out how we infuse critical thinking back into the process or else, you know, we might be cooked a little bit. Well, on the flip side, if we want to end this on a high note, I know you play around with a lot of these models. You're very well versed. What is working for you? How do you think AI is helping you, if at all? It totally is. I mean, you know, I use it to kind of refine my ideas. When we talk about idea generation, you know, you're refining ideas.

13:09It is super useful for those repetitive tasks that you do quite a bit. You can set up an agent to, you know, send emails or, you know, that productivity piece, I think, is what I'm most excited about with AI. because it's going to offload some stuff that I don't need to actually be expending brain power or time on. So I don't mind, you know, all of those automations and the different things that it can do on that end. But on the creative side, on the ideation side, you know, I tend to want to err on the side of human, if you will. Well, Stephanie Humphrey, thank you so much for bringing your ideas to us today.

13:52Let folks know where they can keep up with all that you do. Well, you can follow me all around the web at TechLifeSteph, or you can head over to my website at stephaniehumphrey.com and sign up for my newsletter. Or you can check out my podcast, The Tech John with Rob Dunwood and Terrence Gaines. One of my favorites. Love The Tech John. If anyone's not subscribed, do it now. Do it. Stephanie Humphrey, thank you so much for being with us today. and a reminder to everybody who's listening to this week in addition of DTNS. We can't do it without our patrons. Thank you to all of you who directly support us.

14:27Patreon.com slash DTNS is where to find out more about that. Thank you one and all for being with us. Have a great rest of your weekend. We'll talk to you next time.

14:42This show is part of the Frog Pants Network. Get more at frogpants.com.

14:51the dtns family of podcasts helping each other understand time and club hopes you have enjoyed this program