The Ex-Journalist Declaring War on AI Slop

0:00Hello again, everybody from Newsweek headquarters at the top of the World Trade Center in New York City. I'm Carlo Versano, and we've got another issue, another edition of The 1600 today. This country was stone cold dead. We're going to get off of politics. We're going to put a pin in our Trump and politics coverage for a minute to talk about AI, get back on our AI hobby horse, which, of course, is something that we've been covering pretty in depth here on the show. If you've been on the Internet lately in really any capacity, It's pretty clear now that virtually the entire web is entering something of a phase where you can't really tell anymore what's real, what's fake, what was created by the human brain, what was created by AI, what was some sort of combination of the two.

0:40It's becoming really a serious problem just in terms of getting people to believe what their eyes can see, which has always been how the human brain has worked. In the wake of the White House Correspondents' Dinner chaos last week, I've had people tell me through the newsletter and through DMs that they thought that the whole thing was AI-generated. I don't even know what to make of that. I can assure you as somebody who was there, it was most definitely a real event. There's also been people who have accused me of having writing that sounds like AI. That one actually hurts me deeply because I don't use AI in my writing.

1:17And it's – that's one of those things that people tend to – it's almost like a criticism now, like a slur. It's like, oh, I don't agree with what you said. It must be AI. I've had other people suggesting that various viral posts or events in the news are AI-generated. One guy I remember said that they thought the entire backlash to Charlie Kirk's assassination was AI or driven by bots, right, which is essentially the same thing, and that no one was actually celebrating his death. which again is untrue. So basically, as I said, nobody believes anything they see or read anymore. And that's going to become, if it hasn't already, a societal problem that we're going to need to deal with.

1:58My guest today is trying to do just that. Edward Tian is the co-founder and CEO of GPT-Zero. He's a former AI researcher at Microsoft and an investigative journalist in a former life with the BBC. With GPT-Zero, Edward has built what they're calling now the world's most widely used AI detection tool. We're going to talk about what that means in practice. He's here in studio with me in the flesh. Edward, good to see you, man. Happy to be here. Thanks for having me. And you are real. You're not an AI bot, right? 100%. Okay. So tell me, let's start a little bit with like, what is GPT zero in layman's terms?

2:34Like what is the problem that the tool is trying to solve? We started in November of 2022, right after the admin and launch of ChatGPT, where there was an explosion of content and became increasingly easy to generate AI content. We started with the mission to preserve what's human on the internet. That meant building the first AI detector and seeing if something was made by AI or human. Since then, the product has focused on this core AI detection, if something's AI or not, and added additions in terms of detecting if something's hallucinated, if something's real, if something's factual, and continuing on this core mission.

3:17So you have a dashboard up on the website, which we'll put up on the screen here. And so as of today, the point of the dashboard is to tell you in sort of real time, how much of the internet is AI created, right? Is AI generated. So as of today, according to your metrics, 5 % of Reddit is AI, 11 % of Twitter, X, 11 % of Substack, Get this, 40 % of what you see on LinkedIn is AI, according to your stats. That's insane. On all platforms, it's something around 14%. And you've got a countdown, which we'll show here, that according to your countdown, according to the extrapolating of this data, we're essentially five years away from the entire internet being AI generated.

3:59Is that hyperbole or is that really going to happen? No, that's correct if we're on the current trend as well. and a little bit of context of how we came to this dashboard or is the internet AI.com. Around three years ago, when we first started, we had this vision of going around the internet and identifying if something was AI generated. Actually went even a little bit before GPD Zero while I was working at the BBC, had a very great mentor of mine called Benjamin Strick. And this was in 2021, kind of previous election cycles, where we were looking at detecting AI images on X and Meta or then Twitter and Facebook, previous election cycles.

4:43And we would look and run algorithms then in 2021 that would take the most popular image generation model, which was CellGAN3 then, and would detect if it was AI or not. And it turned out it was so simple then. All the previous AI models of that generation were generating the same pupils, the same eyes. So if you lined up all the eyes, you could find and you could tell when something was an AI-generated network of profiles and images as well. So there is a great conversation with some of the journalists there where you just stare into the eyes. That's always stayed with us where we wanted to build something where as people browse the Internet, will be able to tell them what's real and what's fake.

5:27However, in the last three years, and we'll get into it, we've been working on a lot of professional writing work as well, student writing, all sort of work, and we haven't gone into building something as you browse. It detects AI. It took three years for us to do that. So that's the main product right now, which is basically an overlay on your browser, and it will tell you what you're reading is generated by AI. Correct. Correct. There's the core, just scan AI, scan documents, verify the quality of information. And then now we put out this AI vision product, where as you browse the internet, on Facebook, on LinkedIn, on X, on Reddit, on Substack, Medium, on Google reviews or Amazon reviews, a point out and flag if something was AI or not.

6:12And we only put this out around a month or two ago. It took three years of running this company to get there because of two inflection points. One is in the last two or three months it's been so apparent and drastically different of how much people need this on the social media and the internet and information they're consuming it's almost a tipping point in terms of the amount of slop and two is because before it was AI generated texts now it's also not just detecting texts but actually AI generated agents that are acting on the internet and behaving as human. So this has suddenly become a problem that it took us three years, but we're very proud we're building.

6:55Hey guys, sorry for the interruption. Just going to take a quick ad break here and we'll be right back. How do you detect if an AI agent is behind an account, for instance? Great question. So going back to even years ago when we started in and some of the investigative journalism side as well too, there's a lot of actors in the internet who will put networks of agents. You can think of this in a political context where you have bad actors put out entire networks of disinformation agents as well, pushing out content. And then two, you can also think of it now as a business context because of how demographized these AI agents are, where you'll have hiring managers or companies or marketing companies put out hundreds and hundreds of these Asians pushing out content to people.

7:47So it's a combination of viewing an individual account and running AI detection on the text or the content they're producing individually combined with a macro approach of seeing this entire network and identifying the actor that's putting it out. It's interesting because it just made me think of there was a story, I think, in Rolling Stone recently, but it was about this band. I think they're called Geese, based in Brooklyn, where I live. And it was about how this band sort of came out of nowhere. Like nobody had heard of them. And all of a sudden they're on SNL. They're at Coachella. Like they're everywhere.

8:20And the sort of thesis of the story is it was this kind of – it was sort of a coordinated marketing campaign using bots, using AI agents to sort of like get these guys in the sphere of what people are talking about. But part of me, when I read that, I was like, is that so different from just the old fashioned way of like, you know, putting up a bunch of flyers and, you know, hand posters for for a band that people want to see and having them play little shows? I mean, I guess it's just sort of like an accelerated version of that. You wouldn't be in the old days. You couldn't go from being a, you know, starting a band and being on Saturday Night Live in like a few months.

9:01But now if you if you know how to play the game, I guess you can. Yeah, it's fascinating. it's different because of the quantity and scale of the problem. Never before has it been easier to spin up hundreds and thousands of agents generating message and polluting the networks as well. It's so much of a tipping point that even the people running these social media networks are realizing it's a problem. Just in the last month, the head of product, Keto Beer at X as well, have identified AI-generated text as a core issue. same as Reddit has started building out their internal flagging of AI-generated text.

9:40It's just a tipping point in terms of scale. And because of the people putting out the content are getting so much assistance, I think the people consuming it, we need to balance that scale, hence the work we're putting out in terms of giving people the ability to find and detect AI content. I'm just thinking about my own work as a journalist. I actually find it fairly easy to clock AI-generated text just because I'm reading so much for the course of the day. And I sort of figured out if somebody is using a lot of M dashes, for instance, that's probably one bet at least. That's at least a low-tech way for me to at least think, oh, maybe this was written by AI.

10:20The scarier thing to me is actually the images because the way that we've seen AI-generated imagery change just in the last couple of years. I mean you used to be able – even a year ago, you could tell immediately, right? I mean even as a layman I think you'd be like, OK, that's not a real – a picture you can sort of tell. Everything has sort of like this sheen to it in AI images. But now I don't know. I mean now it's getting to the place where there are images I've seen just like scrolling X and I'm like I honestly have no idea if that's real or not. And if I can't tell if it's real, I think about people like my mother who's not necessarily very good on the computer or just people who are not super into this being able to tell what's real and what isn't, which I guess is the problem you're trying to solve, right?

11:05It's changed so much. Before, what your eyes see, you can believe. Now you can't believe what you see. And I almost categorize it into two problems. One is sort of the annoying problem. And the other is the more dangerous problem. More dangerous problem goes into imagery, videos, as well as even with text, you're evolving with the world. And this is why there's a tipping point in the last year between an essay, an article, a blog post into now you can interact with an agent that's texting you constantly. And you're not sure if that's human or or WhatsAppping or an AI. And then two, even the former problem, the knowing problem, it's real and a pain there too because if you're flooding the entire content of what you're consuming with AI slop, that's going to degrade.

11:53Think of the resources, what our students access on the internet, the quality of information, just your quality of life of reading anything as well on the internet when it's entirely AI. No, and it just creates another thing that you have to think of as the consumer. Like, well, is what I'm consuming real? Was it written up by a machine? What are the things an average person can do? Back to our conversation about text-generated – AI-generated text, that is. What's something like an average person can do to sort of determine if a piece of writing is AI without having to have a tool on their computer to tell them?

12:27Are there any giveaways? I mean I mentioned the M-dash. Like maybe that's not fair to M-dash users, of which I am one. And I love them. I know. I use claim them. I know. That's why I've started using many more commas places in my in my writing because I don't want to be accused of being an A.I. agent. Yeah. Yeah. Well, one is I actually before this interview went through some of your previous articles on Newsweek and ran all of them through GPT zero. And they came out human. OK. Yes. Quality. Excellent. As well. And we have to reclaim these and dash because what we we don't look for the individual signals.

13:02It's kind of the quantity of text as well and the patterns. Although I think, in essence, you want to be able to, as an individual consumer of information, to just have quality and people that you trust. Kind of these quality articles as well, too. As a reader, I trust Carlo and I trust Newsweek as well. I trust the Substack writers I'm writing, Knock on Wood, as well, too. And I've actually started seeing some Substack writers put out, Here are some symbols and edges and just proof that I'm proudly entirely writing it myself in recent months, which was not the case two or three years ago. So you want to trust that.

13:42Do you think that's where this is going, where eventually there's just going to be like a watermark almost on like a – excuse me, on a piece of content that says GPT-0 or some other third party or Google or whomever has determined that what you're reading or what you're seeing was created by a human being? Is that sort of the next step? This would be, it's interesting because over the last two or three years, so many organizations around the world, whether Google, Adobe, have all worked towards this goal. And I was actually at a media party conference recently where there were folks that were on the Google watermarking team, all others, kind of collecting together.

14:23and it's still to be seen where technology is going to be the final solution. Just because we can watermark something that's AI generated or just because on the GPT-0 side we're replaying all the writing and text or information that's going to be accepted or be the common pattern behavior on LinkedIn or on Substack or on Medium. We would all benefit if people all adopted the standard but still to be seen because how people consume and the trends of information are so different. Also, I'm thinking like a Google or an open AI or an Anthropic, does this go against their business model to point out things that are entirely AI generated?

15:07I don't know. I mean I know that the LLMs are kind of only a part of their business, but the LLMs are the thing that most people are – how they are experiencing AI in their day-to-day lives, right? You could see it in both ways. There could be a business tension. There could also be a conflict of interest where are these organizations pushing for the largest quantity and usage of AI, the most amount of tokens spent? Are they the best guardrails for the usage of AI or should other organizations, whether it's independent organizations or governments, be the ones deciding those guardrails? Right.

15:46Back to just LinkedIn for a second. Forty percent of LinkedIn is AI generated. I mean, in some ways, that's shocking. In some ways, if you actually spend any time on LinkedIn, you might think it's actually higher than that. The amount of just sort of like low-grade slop that comes off that platform. Why do you think that is the platform of the social media, you know, the big boys? Why is that the one that has sort of – that users have generated so much AI slop on? Yeah, it's a great question as well too. I think we'll get into it, but it's this future work and where AI is starting to be acceptable as well.

16:20And the first platforms that are casualties of that. Maybe I'll tell you a little bit of how we got to that number. It's so, you know, we've always been motivated since kind of the days of looking at the pupils and staring down the people's faces to detect AI until now. So in the last month or two, we launched our AI Vision product. Tell me what that is. Is that the dashboard? That's the Chrome extension as you browse as well, too. And Vision was also inspired by being able to have a filter for how you view content as well, too, whether it's AI or not as well in terms of the quality of content.

17:05So we, in the last month, and for users viewing as well, too, I have pulled up is the internetai.com, and I was showing that as well earlier. For the last months, we've had over 400 ,000 people download that extension, start browsing their social media with a filter, a new vision of content. And since then, we've collected over 20 million posts as well too, including 7.7 million posts on Reddit, individual posts, 3.7 million on LinkedIn, 3.6 million on X, many more on Substack, Medium, other platforms as well too. And among them, 4.5 % of Reddit is being tagged as AI generated. Actually, the Reddit team has been starting to flag content and derank some of the content.

17:5710.6 % of X and 41.3 % of LinkedIn in this broad categorization. And now the second part is someone on our team had an idea of why don't we concatenate that together, put a website as InternetAI.com that has a bit of this doomsday tracker, but also is just a concatenation of these 400 ,000 people and more browsing the Internet. And if you go, you can click on LinkedIn and you can see the top posts, the ones that have gotten the most virality today that were AI generated. And we thought this could be, and we do believe, Wikipedia is probably not the right analogy, but you want to democratize people understanding of AI content proliferating on the internet.

18:40And we believe this is the biggest source of kind of this concatenation of this information publicly available. And it's not just available to the average people interested in what AI content is going viral out there, but also journalists, also people that are concerned with AI's proliferation. Well, I mean, I certainly as from a journalist perspective, I totally see the value in a tool like this. Right. Because, I mean, we are we are just bombarded with so much crap these days that it's you know, it's hard to know what to make of of any of it. And sometimes I mean, you can even see stuff that this White House puts out that is like very clearly AI and not just the White House.

19:20I mean, corporate, you know, press releases and corporate statements. It's it's remarkable. But I want to go back to this basically started as an education tool. Right. Like tell me a little bit about this idea of using this to flag AI written work at the school level. Right. Whether that's high school or college, because we've heard a lot about how students now in high school and higher are using AI to do their homework, to do their book reports or whatever report they have to take home to do. Yeah. For us and almost every student we talk to as well, too, is using AI. in their homework, reports, daily lives.

20:02For us, it was all about balance. When we first launched GPT-Zero as the first AI detector or commercial AI detector, it was to balance out this large spike in usage of students on ChatGPT. Even today, the initial users and the largest user base of ChatGPT are students with now balancing where and when this is being used in essays. Now this has almost evolved from kind of a cat and mouse game of are you using AI or not into a world where I believe teachers and educators are recognizing students are using AI. I've read that there's some teachers now that are using AI to grade the AI papers. Yes, that as well too.

20:46So it's kind of, and as you get, and it's different between K-12 and higher ed, as you get closer to future work, then people are recognizing this as part of the workflows. So now it's our job and part of our mission in this preserving human content to also bridge that gap when AI is being used to be kind of this assistant layer. So we have teachers asking their students to run their content through GPT-Zero, not just for AI content, but also for hallucinations, for citations, for credibility. We have an authorship or replay tool that stamps this with like a GPT-Zero badge once you're verified that you wrote everything themselves.

21:27We have now a new AI that chats with your writing process and your data of how you wrote something and tells you, whoa, you spend a lot of time here. Maybe this nut graph could be rewritten. Maybe this entry point. Maybe you could have spent more time on the outline as well. That's like having a writing. That makes sense. Exactly. Exactly as well too. And all with the goal of preserving kind of quality of thinking as well, critical thinking. So the point is not to be a snitch. This point is not to rat students out for using AI to their teachers. No. I think it's evolved really significantly since then.

22:04And also the biggest usage of us is not detecting now content that's being flagged as AI. The biggest usage of things being scanned into GPT-Zero, which is across 800 million documents scanned through GPT-Zero, is content that is now a combination of some AI and some human writing yourself. Because that's been increasingly what's going to be how people are putting content out there. Well, it's interesting because, again, I've said that I don't use AI and I don't in my writing. But I have started to use Claude for some very basic help with just sort of like outlining and like structures of things.

22:43If I'm writing – not for my column, but if I'm writing sort of a big story for Newsweek, it helps me in the sort of basic structure of the writing. Is that – can you talk a little bit about what this 80-20 rule is that some teachers have developed? Have you – are you familiar with that? That's what you're talking about, the idea that you can use it. The key is to not just like prompt the AI and then just whatever that slop, whatever it spits out is yours. You just plug it as your final word. Exactly as well. We actually had a writing professor talk to a year ago that first started using GPT-0.

23:15His name was also Edward as well too and taught in Oregon. He first started using GPT-0 to flag when student work was entirely AI. And then the second year he used GPT-0, it was a complete reversal. where he put in the syllabi at the start of his college writing course. This is one of the tools you can use as long as you maintain an 80-20 balance. In terms of 80%, this comes out human, and then you have this 20 % of some AI to help form it. There's recommended usage and structuring and outline as well. And it goes back to the earlier point that the future of how people write is not going to be all entirely human.

23:56I, God, hope it won't be all entirely AI. just putting out on the internet. It's going to be some sort of combination as well that has that human qualities and craft and thinking preserved. Like everything else, everything is a balance, right? It's funny we're talking about this because there's a story in the New York Times today that I wanted to talk to you about. The headline is, How AI Killed Student Writing and Revived It. Quoting the New York Times here, the Times heard from nearly 400 college and high school educators who responded to a call-out about how generative AI is changing writing instruction.

24:28Almost all described a deep rethinking of how to teach writing and whether it still matters since AI has become a better writer than most students and adults, they said. Teachers are responding to the widespread challenge. Over the past year, AI use has become ubiquitous among American students. Between May and December of 2025, the share of American middle school, high school, and college students reported regularly using AI for homework increased from 48 to 62 percent, according to polling from RAND. That's over the course of like five months. Like that's incredible. The point of the story, it goes on, is basically that there's a lot of teachers in America now that are saying, to our point before about did AI kill the book report, they're basically saying you don't have to go home to do homework anymore.

25:11Like we're going to do this in class where there's some sort of like proctor or the teacher is there to watch them because they essentially just like they don't trust the students to do this, I guess. I mean is AI going to kill off homework? So I'll say two things as well, too. We've also seen an increase of pen and paper in reports recently as well. So I've seen that phenomenon. We have this kind of solution where it replays all your writing, whether you write on Google Docs or a GPT-0 editor or somewhere, to almost stamp and prove you wrote yourself. And it's recently gone viral among students.

25:51There was a TikTok yesterday with over 30 million views. Collectively, there's been over 500 million views of this replay tool where now students are like, wow, I can show I wrote everything entirely myself without having to do this pen and paper as well where students prefer typing nowadays. Second thing may be this larger question of how important it is to write. It's absolutely and so critically important. around a year ago I was visiting my own writing professor from college John McPhee he was a legendary New Yorker writer for many years and we actually played around with a lot of the AI systems including chat GBT and asked it and he was also curious can it write it in his style and we propped it and it wasn't too good at writing in his style and then I remember he asked me can AI tell you when something's so good don't change a single word like you've created you've crafted the perfect essay.

26:52And I was saying, I don't know what AI that students use that do that. It's constantly generating and generating. And he said, aha, that's the job of a good editor. Right. And I would say this is why writing and learning to write is continuing to be important because writing is a reflection of your critical thinking. It's a reflection of editing. It's a reflection of craft. We all know it's sometimes even before AI, but certainly nowadays, the hardest this job is not the first draft anymore. It's kind of that final iteration. No, that's a great point. I'm glad you said that because it was something I always tell whenever, like, there's students who come in, whether it's here at Newsweek or past jobs, you know, college students, like journalism students will come in and they'll, you know, just sort of, like, see what it's all about.