The People Tracking America's AI Data Centers

7 Jan 2026 · 51 min · 21 chapters

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

```markdown The 404 Media Podcast - Episode Summary

Episode Title

The People Tracking America's AI Data Centers

Podcast Description: The 404 Media Podcast is a journalist-owned digital media platform that explores how technology shapes our world through investigative reporting and insightful discussions.

---

Episode Breakdown

Key Topics Discussed

  1. Tracking AI Data Centers (1:38 - 25:58)
  2. Main Story by Matthew:
  3. Researchers from Epoch AI are utilizing open-source intelligence to map the growth of AI data centers across the U.S. and globally.
  4. They combine satellite imagery, building permits, and local legal documents to create a comprehensive map of these installations.
  5. Epoch AI’s goal is to understand the infrastructure necessary for AI and its implications for local communities.
  • Discussion Points:
  • Matthew explains the motivation behind mapping data centers:
  • Increased transparency regarding their impact on local communities and economies.
  • The need for a baseline understanding of energy consumption and ownership.
  • Jason discusses the broader implications:
  • The significant energy consumption and environmental considerations associated with data centers.
  • Job growth in data center construction versus the limited long-term employment they provide.
  • Controversies surrounding data centers being built in economically disadvantaged areas.
  1. Grok's AI CSAM Issues (25:58 - End)
  2. Overview by Jason:
  3. Discussion of Grok, an AI tool integrated into X (formerly Twitter), and its controversial use during the holiday season.
  4. Users discovered they could generate inappropriate and potentially illegal images using Grok, including sexualized images of underage individuals.
  5. This raised significant ethical and legal concerns about the platform and the inadequacies of self-regulation.
  • Discussion Points:
  • The ease with which users exploited Grok to create harmful content reflects a larger issue of AI governance and accountability.
  • Emmanuel highlights the collapse of the information ecosystem where serious news and exploitative content coexist on the same platform.
  • The podcast emphasizes the difficulty of regulating AI outputs and the likely persistence of these issues without significant changes in technology or law.
  1. Subscriber's Story: Our Acquisition of 404media.com
  2. The hosts briefly discuss their recent purchase of the domain 404media.com, highlighting the importance of direct support from subscribers.

---

Key Takeaways

  • Importance of Mapping AI Infrastructure:
  • Understanding the location and operation of data centers is crucial for assessing their environmental impact and effect on local economies.
  • Ethical Implications of AI Tools:
  • The misuse of AI tools like Grok raises significant concerns regarding content moderation, user safety, and corporate responsibility on social media platforms.
  • Community Impact:
  • The growth of data centers presents a dichotomy between potential job creation and actual benefits for local communities, often leading to increased energy costs for residents.
  • AI Regulation Challenges:
  • The podcast highlights the ongoing struggles with regulating AI technologies and the challenges posed by rapid advancements outpacing existing legal frameworks.

---

Conclusion

This episode of The 404 Media Podcast provides insightful commentary on the intersection of technology, ethics, and community impact, particularly as it relates to AI infrastructure and the challenges of moderating AI-generated content on major social media platforms. The discussions emphasize the need for greater awareness and accountability in technology's role in society.

For more content, visit [404media.co](https://404media.co). ```

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Chapters

Tap a time to open that second in VO

Returning for a New Year

1:13 to 2:24

Hosts discuss returning for the new year and previous interview episodes.

“So I thought those were good to carry us over.”

Mapping America's Data Centers

2:24 to 3:33

Introduction to the story about researchers mapping data centers in America.

“So we'll get into the map and the specifics and your experience with it.”

Understanding Epoch AI

3:33 to 5:30

Discussion about Epoch AI's mission and motivations behind mapping data centers.

“Like, what's sort of the drive behind it?”

Impact of Data Centers on Communities

5:30 to 7:20

Examination of how data centers affect local communities and energy costs.

“And so one, it's like the entire American economy is like betting on this.”

Job Creation vs Reality

7:20 to 9:37

Discussion on the job creation expectations versus reality of data centers.

“There's also a lot of them being built in like Northern Virginia, which is a pretty wealthy place.”

The Future of Energy and Data Centers

9:37 to 11:50

Exploration of the energy demands of data centers and future implications.

“and you just have to make sure that the computer don't blow up.”

Building the Data Center Map

11:50 to 14:00

Insight into how Epoch AI constructs their data center map using various sources.

“I mean, this is like the huge trillion dollar question.”

Understanding Epoch AI's Data Center Mapping

14:00 to 15:30

Learn how Epoch AI constructs its map of data centers using various data sources.

“I mean, they're using slightly fancier things than Google Street View.”

Exploring the User Experience of the Data Center Map

15:30 to 17:45

Discover what it's like to navigate and utilize the data center map created by Epoch AI.

“Google doesn't want to necessarily tell you how much megawatts a deep mine facility in New Albany is using.”

Challenges in Mapping Data Centers

17:45 to 20:16

Examine the challenges and limitations faced by Epoch AI in identifying data centers.

“So right now, it's like a dozen things in the United States.”
Show all 21 chapters

Update on Microsoft's Underwater Data Centers

20:16 to 22:02

Learn about Microsoft's attempt to create underwater data centers and its eventual failure.

“Yeah that sounds great and you're right people do People do care about it.”

The Controversy Surrounding Grok's AI Image Generation

24:13 to 28:04

Discuss the troubling implications of Grok's AI capabilities and the ethical concerns raised.

“Well, it's January and you know how it is.”

Elon Musk's AI Regulation and Social Media Chaos

28:04 to 29:54

Explore the lack of regulation in AI on social media platforms, especially under Elon Musk's leadership.

“going to be some meaningful regulation of this by the US government, of course.”

AI Generation in Social Media Feeds

29:54 to 31:53

Discuss the integration of AI in social media and its impact on user experience and content quality.

“But some of the background here is every social media platform has been taken over by AI Slop.”

Non-Consensual Content and AI Exploits

31:53 to 33:57

Examine the emergence of non-consensual AI-generated content and the challenges it poses on social platforms.

“It's like historically, also, as you mentioned in the article, and you link it back to what happened just before we came back from the holiday break.”

The Training Data Dilemma

33:57 to 35:36

Analyze the implications of training data in AI systems and the challenges in preventing misuse of AI-generated content.

“But then there's thousands of people in this telegram group and they immediately shift to a new exploit.”

The Information Vacuum During Crisis

35:36 to 37:16

Investigate the challenges of information verification during crises and the role of social media in shaping narratives.

“Well, I think another thing about X also is that it's like Grok is doing this to real images that people are posting.”

Navigating Misinformation and Media Trust

37:16 to 42:03

Discuss the complexities of dealing with misinformation in social media during significant events and the decline of trust in platforms.

“like whether it's a mass shooting or a hurricane or an act of war or something like that.”

The Information Vacuum in Journalism

42:03 to 44:14

Explore the challenges journalists face in an age of rapid information sharing.

“And like, that is not how the news media works for like understandable reasons for the most part.”

The Dangers of AI in Journalism

44:15 to 46:20

Understanding the implications of AI-generated content on reporting accuracy.

“The thing I'll just add is that, I mean, that's like the old problem.”

Grok and AI Apologies

46:21 to 49:12

Discuss the misconceptions surrounding AI's capacity to apologize or express intent.

“And Grok was like, oh, yes, this happened for XYZ reason, and we're working on it, and we're sorry, etc.”
Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:03Hello, and welcome to the 404 Media Podcast, where we bring you parallel access to hidden worlds both online and IRL. 404 Media is a journalist-founded company and needs your support. To subscribe, go to 404media.co. As well as bonus content every single week, subscribers also get access to additional episodes where we respond to their best comments. Gain access to that content at 404media.co. I'm your host, Joseph, and with me are two of the 404 Media co-founders, the first being Emmanuel Mayberg. Hello. And Jason Kebler. Hello. And then we're also joined by our regular contributor, Matthew Galt.

0:45Hello. All right. So we are officially back for the new year. I hope everybody enjoyed the replays of our interview episodes. It was mine with the head of Signal. Sam had one about Pornhub as well. Definitely check those out if you haven't already. And I think those were good because a lot of stuff has happened in both of those broader stories when it comes to AI, encryption, backdoors, age verification, porn, and everything else. So I thought those were good to carry us over. But we're back. It's good to be back. New content. New content, please. Exactly. And we actually published a ton over the holiday break.

1:30Obviously, we wrote a lot of this beforehand. and that one of those is the first story we're going to talk about. That is Matthew's one. The headline was, Researchers are Hunting America for Hidden Data Centers. I'm actually going to read the lead because I thought it was really, really good. And I think that'll set us up for the rest of the discussion. If I can read it on my tiny screen here. A team of researchers at Epoch AI, a non-profit research institute, are using open source intelligence to map the growth of America's data centers. The team pours over satellite imagery, building permits, and other local legal documents to build a map of the massive computer-filled buildings springing up across the United States.

2:15Matthew, this is really, really interesting. It came in pretty late in the year, right at the end, and then we published it, as I said, over the holiday. So we'll get into the map and the specifics and your experience with it. What is this organization? What is Epoch AI?

2:36Sometimes I struggle with what to call these things. It's like a think tank, basically. It's a non-profit think tank that is just tracking the growth of the AI industry. And having talked to them, I think that they are ambivalent to pro-AI, quite honestly. but they are serious about like understanding the impact of this technology especially with regards to like the infrastructure that it's going to take to build it out and so you can kind of think of them as like they're collecting a bunch of data to explain to any claim if anyone that would come to them uh about like what's going on in the country and the rest of the world quite frankly but the like one of their main focuses is america they are looking outside of the u.s as well.

3:24Sure. So what is, just to expand on that a little bit, what is this motivation for building this map? It's just, hey, we think more information is better? Or why specifically focus on this? Like, what's sort of the drive behind it? Because we don't know a lot about it. I wouldn't say that data center construction is a secretive industry. But it is just something that's kind of sprung up in the last couple years in the United States and is affecting a lot of communities. People don't always understand how it's affecting the communities and what's being built exactly and who owns what. So it is just an attempt to get a baseline knowledge of what's in the country and what it seems to be costing in terms of energy infrastructure and what exactly it looks like, how big it is, and who owns it.

4:25Yeah. It's like sometimes you live... I've been talking to people the last six months. Sometimes you live down the street from a woodland area. It gets cleared out and buildings start being constructed and you have no idea what the hell's going on. Right. Jason, I feel like you've brought it up a few times about how we want to cover data centers or AI data centers more and more. Matthew has been doing that with, I think, three or four articles at this point, something like that. There's another one in the hopper too. It's going to be super interesting. Great. So what's the context here, Jason?

5:04I mean, this is the entire American economy at this point. It's pretty shocking if you actually look at the numbers, which I do not have in front of me right now. But if you look at the jobs numbers, it's like there are two places where there is job growth. in the United States right now. It's like nursing and healthcare because they're an aging population and it's data center construction. It's like construction jobs, like HVAC jobs, like electricians and things like that. And so one, it's like the entire American economy is like betting on this. Two, data centers are like really controversial in the places that they are being built and set up for a variety of reasons.

5:48One, often they're pretty loud. Two, and more importantly, they use a shitload of energy. And so there is quite a lot of discussion about how we are powering these things. In many cases, they're powered by fossil fuels. In some cases, very rarely, or not often enough, I would say, are they being powered by renewable energy. and so there's there goes like a hand in hand like who how are these things being powered and also what are the impacts on the local community because of that because in a lot of places these data centers have come in and people's electricity bills have gone through the roof because there's you know the the actual like power generation there's like not enough of it to go around and And so the price of it goes up.

6:41And often, I mean, these things are, well, maybe the map shows more, but I would say the ones that are quite controversial are ones that are being built in relatively poor areas. And so they're being built in places where people can't afford to pay more for their electricity. and that's happening because a lot of times, you know, the city governments, the state governments are giving tax breaks to these companies to build these things. And so they're getting put in places where there's not as much like NIMBY type pushback. There's also a lot of them being built in like Northern Virginia, which is a pretty wealthy place.

7:27And there is a lot of pushback there because that's a pretty rich part of the country. Um, and it's just become like this huge political issue in, in those places. Like we probably should do this at some point, but there's a state representative in Virginia who ran essentially on like an anti AI, anti data center platform and he was elected. um so yeah i mean that that's like the broader context and i bring up virginia because it's close to dc it's also close to where the um the internet like enters the country there's like all these different places where the undersea cables enter the country and so you have like a lot of internet like throughput capability there and so that's like one of the reasons that that's that's happening there among others um but yeah it's like hugely important um at the moment yeah Yeah.

8:17Also the water. Right. The water. And then one other thing is like, once these things are actually built, they don't bring that many jobs. It's like, there's a lot of jobs in actually building them. But then once they're up and running, it's like, there's not that many people who work on them to keep them up and running. So it's not like it's some amazing financial boon for these local communities. There's kind of like a lot of promise. And then once they're up and running, it's like, okay, we have like seven people working there or whatever. I don't know actually how many people it takes. But I think I mentioned this on a recent podcast that we recorded.

8:50But I went to a Bitcoin mine in upstate New York that was causing some of these problems a few years ago. The Bitcoin mine was built there because they had really cheap hydropower. And the Bitcoin mine sucked up so much of the energy that people's bills were going through the roof. and the Bitcoin mining company was like, oh, but we're bringing jobs back to this town that doesn't have that many. And it's like, there was actually only two people working at the Bitcoin mine. Because it's just like, they needed someone to turn on the light and they needed security, essentially. Once it was actually like, everything was installed, there was nothing going on there.

9:30And it's like, that is not the exact same as these data centers, but it's similar in Vibe where it's just like you just have a bunch of computers running and you just have to make sure that the computer don't blow up. Well, for this, it's probably even worse because once it's built, the AI is going to be running in the data center. It's like, cool, now we have the AI to take even more jobs which aren't even physically located here. Obviously, I'm generalizing. Matthew, what were you going to say? There's just a similar situation happening in West Texas in a city called Clarendon where it's got like a population of 10 ,000 people and it's mostly retirees.

10:07It's people that moved out there because it was cheap to live there. And this is like they wanted to retire close to the plains and just like that's... This is the final place they're going. Well, guess where they're building a data center and guess what's happening to everyone's electricity bills. And they're giving the same line. Like, well, we'll bring jobs to the communities. Like we didn't need... It was a bunch of retirees. We didn't need jobs. They literally don't want jobs. Yeah. And the other thing that I think is... One of the other things I think is really interesting about this stuff, and one of the things that worries me about how it's propping up the community, is that it's an enormous gamble that people think is going to pay off in like 5, 10 years.

10:49How so? So like in West Texas, again, near Amarillo, they're going to build four big nuclear reactors, suck up a bunch of water from an aquifer that's already not doing great, to fuel them. The reactors probably won't come on for at least five years, probably more. So to get the data center up and running that they want to power with the nuke plants that will be constructed before the nuke plants are, they're going to build a bunch of

11:24natural gas plants to burn everything off. It's going to spike everyone's energy bills because none of that's going to go back into the community. And what happens in 5 years, 10 years, if there's a contraction on LLMs? If this AI stuff doesn't pan out, and now we've invested billions of dollars in all this infrastructure that's just going to be unnecessary overnight, no longer profitable to build, what happens then? Yeah. Totally fine. I mean, this is like the huge trillion dollar question. It's like, are we going to have these husks of data, these abandoned data centers. And I think Emmanuel, to put you on the spot, is a little bit more bullish on this, that regardless of what happens with AI, there will be some need for these data centers eventually at some point.

12:18So perhaps the risk of this is not as high, but nonetheless. We will compute something, whether that's generating images on Grok or something else is to be seen. Or it doesn't mean it will be a good return on investment, but I don't think they'll sit there empty like abandoned gold mines or something. Streaming video games, right? I mean, it's like... Anything, yeah. Xbox can put their servers there or whatever. Have you followed... That'll be dead in five years. This happened right before we went on break and I hesitate to open this Pandora's box, but have we followed the data centers in space debate?

13:01It doesn't make any sense to me from a logistical perspective. Again, it's one of those things where it takes too much money to build it to get any kind of return on it in anything under 100 years, right? It's a great idea because it's cold up there, etc. But to get everything up there and get it running and then beam everything back, It's just like, that seems like a pretty tall order. I thought we were putting them in the ocean. I think the ocean is better. Well, don't clip this because I don't... The negative externalities of having them underwater could be many. But ocean is also very cold and is a lot closer and the internet's already down there.

13:46And they've done it. They did one test. I remember Microsoft put one. Microsoft has or had one. Yeah. well speaking of maybe there's one under the water is that on the map Matthew or rather let me rephrase the question to bring it to the story they have this map and you can scroll through it and zoom around and look all of these different data centers across America how is Epoch AI actually building this map I touched on this by reading out the lead but like how are they making this map Are they just on Google Street View? What are they doing? I mean, they're using slightly fancier things than Google Street View.

14:29But yes, they're looking at satellite imagery. They're kind of tracking news stories about the announcement of data centers. And then you go and you look at the satellite imagery for the location. You pour through local filings at the county level and at the city level to see what people have declared that they are building. And then one of the things I thought was really fascinating is looking at the satellite imagery for these construction projects and looking specifically at what cooling infrastructure is like. So they look and they see if something's uncovered, which a lot of stuff is, like how many fans are they building?

15:10How big are the fans? And then they kind of have a formula that they go through and say like, So they've got this number of fans, and that tells us that it's using this much electricity and that it's generating this much data. And working backward from the problem. Because a lot of this stuff is... Google doesn't want to necessarily tell you how much megawatts a deep mine facility in New Albany is using. So they look at the outside of the thing and work backwards that way. It's all open source and open information. I mean, it actually sounds like a fair bit of work. It is. Bringing it all together.

15:57What is your experience scrolling through the map? Is it pretty intuitive? Is it informative? What's it like actually to use this map? I think it's really good. because it's all... You just look at the map. It's all right there. You got your bright blue circles. You click on a blue circle. Go into it. And then it's all annotated. I'm looking at this one for Google that's in Oklahoma. And it'll walk you through all the little individual pieces like I was talking about. This is where the backup power is. This is where its fans are. And then you can even click through dates to see what it looked like as it was under construction.

16:42You get a lot of information about what these things look like and how they work just by going through these maps. Interesting. How busy is the map? Are we talking like a few results? Dozens? Because it can't be a massively populated map. Or maybe I'm wrong. No, no, no. It's only got, in the US, it's only got a few results right now. Because they kind of have, because there's kind of an arbitrary line that they've marked where it's like, you could technically say like everyone's PC is a data center, right? Because it's doing compute, it has a GPU in it. At what point do you have enough GPUs strung together that you consider it a data center for the purposes of this map?

17:28Right. And right now, they're kind of only looking at bigger, more massive projects. And that's what they're charting out. And also, they don't know about everything. And they haven't run the numbers on every little thing they've found. So that's just stuff kind of comes up as they have the time to get it up on the map. Yeah. So right now, it's like a dozen things in the United States. They've got some other stuff in other places in the world. They've got some in Oman and one in China that they've found. But I imagine the map is going to grow as the year goes on. Right. It's not cataloging every digital ocean and AWS data center, because of course, they're all over the place and have been for years and years and years, because everything's obviously moved to cloud infrastructure and that sort of thing.

18:09But it's not that. It's very specifically stuff about AI. Right. It's very narrowly focused on, yeah, the stuff that is powering like DeepMind and XAI and Anthropics stuff, and like even an Alibaba data center that's here, or in China rather. So yeah, it is very narrowly on those kinds of things. And they also have some stuff that we think that this is what it is. So yeah, that is kind of what's going on. What, I think just to round out, what happens now? Are you going to keep an eye on this map and return to it? I mean, I guess that it's not really going to break news, probably. I don't think this map is going to reveal a new data center because it's going to be picked up in local or national press, presumably before it's on the map.

18:59But do you think you'll come back to this as a resource? Yeah, this is going to be a resource for me this year. And it was especially interesting because we... This is one of those that I kind of stumbled on this at the end of the year, because I was talking to some guys about some Nuke stuff and some OSINT stuff they're doing with Nuke watching. And they were like, have you talked to these data center guys? They're doing something very interesting. And that's kind of how I got onto this. And I was interested because, like Jason, I think data center and resistance to data center is going to be a pretty big story this year.

19:32And I wrote this at the end of the year without super thinking about a very hard talk to the guys. And I came back and I had so many emails and messages from people that were like, they're building a data center here. like this thing is going up and like it just like any kind of anytime I write about this stuff it really people are paying attention and it really like you know I always get a lot of ton of messages so it just again it tells me that like this is something that people really care about so as I continue to like carve out a data center beat this year I think this is going to be a resource for me for sure.

20:15Hell yeah. Yeah that sounds great and you're right people do People do care about it. Jason? I have a very quick update on the Microsoft data centers. Very funny. The underwater ones. So in 2020, soon after they were launched, the headline on a Microsoft press release is Microsoft finds underwater data centers are reliable, practical, and use energy sustainably. And then it says, more than half the world's population lives within 120 miles of the coast. By putting data centers underwater near coastal cities, data would have a short distance of travel, leading to fast and smooth web surfing, video streaming, and game playing.

20:51Here is a headline from ITPro, which is a trade publication that writes about data centers and other things from July 19, 2024. Microsoft scrapped its ProjectNatic, which is what this was called. Underwater data center trial. Here's why it was never going to work. It's basically like they had all sorts of issues with it. They're not running it anymore. The main issue was that it was like hundreds of feet under the ocean. And so if they needed to do any sort of maintenance on it, they needed either scuba divers or to haul the entire data center back up to the surface, which was extremely expensive and did not work.

21:31So perhaps underwater, maybe not the best. Which makes you think it's also quite expensive to go up to space, as far as I know. Like it's not super cheap, you know? So... As I recall, when we had to service the Hubble Space Telescope, for example, that was like quite an undertaking. Right, right. Why didn't they call it Project Atlantis? Like that would have made way more sense. Too on the nose. Yeah, I guess so. All right, we'll leave that there. When we come back after the break, we're going to be talking about, frankly, the complete shit show that's happening with Grok and its AI image generation.

22:10We'll be right back after this.

22:25You know why I wanted to learn a new language? Not to memorize endless grammar tables. I wanted to actually speak with real people in the real world. That's exactly what Babbel is built for. It gets you there fast. A while back, I decided I didn't want to be the person awkwardly pointing at menus when I travel. So I started using Babbel's app, just 10 minutes a day. The lessons are short, practical, and easy to fit into my routine. Perfect for a quick break or while I'm having my morning coffee. Learning with Babbel is all about small steps and big wins. Those bite-sized lessons make it easier to remember what you've learned and the progress tracking keeps you motivated.

23:03What I really appreciate is that Babbel focuses on real-world conversation. Their courses are designed by over 200 language experts to teach useful words and phrases that you'll actually use, from ordering a coffee to chatting with a new friend abroad. In my case, I speak Spanish, but it's been a long time since I've practiced. So Babbel has helped me refresh my skills and helped me pick up new ones before I travel. Inside the app, you'll also find a large collection of podcasts where Babbel experts share language tips and an inside look at local cultures. So however you learn best by listening, speaking, reading, or writing, Babbel adapts to your style.

23:40It's no wonder Babbel has sold over 25 million subscriptions worldwide. Plus, they offer 14 languages to choose from, and every subscription comes with a 20-day money-back guarantee. Here's a limited time offer for our listeners. Right now, get up to 55 % off your Babbel subscription at babbel.com slash 404. Get up to 55 % off at babbel.com forward slash 404. Spelled B-A-B-B-E-L.com forward slash 404. Rules and restrictions may apply. This episode is sponsored by BetterHelp. Well, it's January and you know how it is. There's all this pressure to become a new version of yourself. New habits, new routines, new goals.

24:23And honestly, I've learned that the new year doesn't really require a new me. sometimes it just calls for a less burdened me. For me, one of the biggest things I've been working on letting go of this year is perfectionism. That constant feeling that I should be doing more, doing it better, or doing it faster. Therapy has really helped me step back and understand what that pressure comes from and how it shows up in my relationships and even how I talk about myself. And that's where BetterHelp comes in. BetterHelp makes it easier to get an outside, unbiased perspective from a licensed therapist, someone who's there to help you identify what's weighing you down or holding you back, whether that's fear, doubt, pressure, or something else entirely.

25:06Their therapists are fully licensed, work under a strict code of conduct, and BetterHelp does the matching for you, so you can focus on what you actually want to work on, not on filling out paperwork or calling around. And if your first match isn't the right fit, you can switch therapists at any time. BetterHelp makes it easy to get matched online with a qualified therapist. Sign up and get 10 % off at betterhelp.com slash 404media. That's B-E-T-T-E-R-H-E-L-P.com slash 404media.

25:47All right, and we are back. Jason, this is one you wrote by I Know Everybody here has thoughts on this so please jump in but just to get it started the headline was Grok's AI CSAM shit show referring to child sexual abuse imagery or material sorry Jason to get going what did people realize over the holiday that they could have Grok do yeah so I mean hey at Grok has become a big thing on X like, hey, at Grok, respond to this person, blah, blah, blah. And people realized that they could ask Grok to put women in bikinis who had posted their photos on Twitter, on X, or to do all sorts of things like spread her legs, take her clothes off, so on and so forth.

26:43And it became a meme for the last, I don't know, like two weeks at this point. It still happening to some degree where anytime you see an image on Twitter, there's often someone under it asking Grok to manipulate the image in some way. And what was happening was people were doing this to images of underage girls. And so there was a Stranger Things actress who I think is 14, who was put in a bikini and people just doing really fucked up things to underage people, which is crazy. It's very, very bad. You don't even need to explain why it's bad. It's child sexual abuse material being generated on one of the biggest social media platforms owned by the richest man in the world.

27:36And this started another round of discussion about why X is a bad place and why it's an irresponsible place. And I mean, I feel like we have entered a period where I think it is important to sort of point out what is happening, but it's not like, it doesn't seem like there's any sort of hope that there's going to be some meaningful regulation of this by the US government, of course. And there certainly doesn't seem to be much self-regulation occurring by Elon Musk. It is like, by and large, in anything goes, shitshow cesspool. Have they even released a statement or said anything at all? I mean, I think that Twitter public relations has updated its autoresponder, the press autoresponder.

Read the full transcript

28:36For a while, it was just sending the poop emoji. and then there was like a brief period where they were maybe answering some people like very brief period and now i think it just said as an autoresponder that's like the media lies and so i haven't seen that one but the one i haven't seen yeah the one i remember is that it was we're looking into it we're busy and this was sort of around the doge time i think they changed to that after the poop emoji but i hadn't seen the the media lies one but you're right in that look this is the sort of stuff we've seen ai be used for for literally years at this point obviously emmanuel and sam have done a ton of coverage on it what's different here uh sort of two things it's happening on a major tech platform which has been acquired by you know a tech billionaire a tech mogul all of that sort of thing and the second component is that it's very much in public like the telegram groups that Emmanuel goes to and finds, oh, this is how they're doing images of Taylor Swift or whatever.

29:41That's, yes, it's on telegram, but it's somewhat private or at least, you know, semi-public, I guess. This is just fucking out there and they're just doing it on this platform. Yeah, I mean, well, let's throw it to Emmanuel in a second. But some of the background here is every social media platform has been taken over by AI Slop. We've talked about this a lot. But almost all of that AI Slop is being generated elsewhere and then being posted onto the social media platform. Meta has tried to integrate AI into its platforms and has done so in a pretty shitty way that people don't really use. And so this was a hope of Mark Zuckerberg that people would use Meta AI to auto-generate shit.

30:26And you would sometimes be talking to a robot, but sometimes be talking to a human. And that hasn't worked very well. Meanwhile, with Sora, Sam Altman and OpenAI have tried to make a fully AI-generated social media platform. And it's like that had a flash in the pan and it didn't really work. But now what has happened with Grok on X is that the AI, as you say, is integrated into the entire experience. And it's not great. Don't like it. I'm curious what Emmanuel's thoughts are on it. But yeah, it's like Elon Musk has kind of succeeded at this, like making people interact with AI alongside of other people goal of some of these tech scions.

31:10I think one of the best observations in the article is what you just said, where everybody's doing AI. But as is often the case, it really matters what the specifics are and what the UI is and what the user experience is. And in this case, it's not only that the AI generation is happening in the same feed as everything else with nothing in between. You just tweet at Grok and it puts the result in the feed. It's not just any feed, right? It's like historically, also, as you mentioned in the article, and you link it back to what happened just before we came back from the holiday break. Twitter used to be the place where you go to catch up with what's happening in the news or what's happening around the world in near real time.

32:12I don't know what Twitter slogan is now, but it used to be it's what's happening. And now you go to Twitter to see what's happening and it's like non-consensual bikini AI generated shots of minors. And that just kind of collapses the information environment and what is happening in the darkest corners of the web into one place in a really awful way. um i also wanted to talk about something that i think i will probably write soon when i get a minute but i wrote in may this story that is basically identical maybe a little worse but less known and the headline for that story was elon musk's grok ai will quote remove her clothes in public on X.

33:10And it's the same thing. People just discovered an accident or by trial and error that you can reply to the image of usually a woman, any woman on an X, and it will produce a nude image of her. And that process of Twitter's millions of users suddenly discovering an exploit is just a... It's very similar to the process that I see in those Telegram communities where people produce non-consensual porn and try to produce it with very mainstream tools by finding loopholes in the guardrails. And I feel like looking at those communities today is like looking into the future of Twitter and what i'll say about that is this is like a pretty primitive way of producing bad content with grok and i'm seeing people produce way way way way way worse things and i'm also seeing x to their credit in real time responding to the exploits like they see what people output and they fix it.

34:32But then there's thousands of people in this telegram group and they immediately shift to a new exploit. And it's like, it's this cat and mouse game that I've been following for months. And there's just no stopping it. You're not going to be able, I think, to fully end this problem because as we've said a million times, the nudity is in the training data. And as long as it's in the training data, it's going to be impossible to completely prevent it from showing up in the output. The only way to stop it would be to go into the training data, extract it, or redo the training data from scratch, which would be really expensive and realistically not something that these companies want to do anyway.

35:19So I would just say it's like, we are only a couple of years into this, but unless something fundamentally changes, whether it's the law or the technology, we're just going to be doing this like every few weeks, every few months, it's just going to be something like this. Well, I think another thing about X also is that it's like Grok is doing this to real images that people are posting. And I mean, I guess you can do that with other AI image generators and other AI tools, But it's like, one of the kind of jarring things, at least to me, is that a major world event occurred last weekend. And I am on the West Coast.

36:07So it was like two in the morning West Coast time when people started talking about explosions in Caracas. And that's not like a very active time on the internet, the American internet. and so I went to Blue Sky. I was like, there's not like that much going on here other than people freaking out, understandably so. And so I don't spend like a whole lot of time on X anymore. But I was like, well, all of the fucking Trump administration has accounts on X. Like, is DHS bragging about this? Like, is Pete Hegseth bragging about this? Is Stephen Miller talking about this? Like, what's going on? So I went there and there were videos being posted of Instagram stories that people had taken from the ground in Caracas, allegedly.

37:03It was videos of explosions and helicopters and things like that. And it's like, okay, breaking news event has always been really fraught time on Twitter, on the internet in general. like whether it's a mass shooting or a hurricane or an act of war or something like that. It's like, you have some real information and you have some fake information and you have people trying to fuck with you and you have like old, you have like real images from other things that are not actually happening now that are being like recontextualized and just like posted to fool people. And so you always sort of like have your guard up and like, I was scrolling through this and I'm like, okay, like this seems like plausibly real.

37:46like maybe this video is plausibly real. I was just like looking for my own interest. I was very interested in what's happening. Like we were going to war or something. I wanted to know what's happening. So there's like that. But then there's also like AI generated images of Maduro that look really real because they're getting a lot better. They don't have seven fingers anymore. And also they have like this blurry effect on them because it's nighttime. And it's like, I don't know, this could plausibly be real. I have no idea. I bet it's fake, but it could be real. and then you have that image and someone's put him in a bikini and added nipples to him and shit and this is all happening like in the same timeline and it's just like dude this is so fucked up like this is just like such a such a mess it's just such a mess and it's like always been a mess but now it's a mess in a way that is like like brain exploding essentially yeah i'm trying to get the news i'm trying to extract some value from twitter or x like it used to be it used to be a genuinely good source of information and now you'll get the government propaganda from there and maduro in a bikini as well and then a bunch of actually like straight up abusive images as well yeah and i mean for the people who are just like just leave x it's like i have functionally left x like i don't really post there but but like during a war and through in this information vacuum at three in the morning knowing and like as a journalist i'm like okay let me go look what this is like at the moment like who is posting there what's going on and like this is what's going on it's like a mix of like non-consensual sexual imagery government propaganda elon musk like stoking some weird fears and then like grok taking like women's clothes off.

39:34And then also, I saw an AI-generated image of Charlie Kirk that I can never unsee, where he had comically long nipples. It's like, cool, cool, cool, cool, cool. That's what we're doing now. Got it. Matthew, what were you going to say before I asked Jason about something else? Do we ever think that this will get to a point where there is more of a mass abandonment of these platforms? At a certain point, it is just filled with garbage. and we can't sustain it. It's not worth it to even go. I mean, don't have the numbers. But anecdotally, I think just based on everything Jason has just said, there's going to be less and less value for genuine users.

40:17There's going to be continued value for the blue tick grifters and all of that. And they're absolutely not going to leave until they have their incentives removed. But yeah, it's absolutely less of a less valuable environment. For me, when it was the Venezuela stuff, I think I got up and I check Apple News because I can aggregate Wall Street Journal, Washington Post, sometimes New York Times, all in one. And I basically use that now as my Twitter feed. Whereas years ago, I would have scrolled Twitter and got some good information for some good OSINT accounts. It's like, I'm just going to wait for the Wall Street Journal report or something and scroll through that feed, essentially.

40:59I got added on a bunch of discords at like three in the morning. Oh, right. Yeah, I mean, I think that for like daily use, like, yes, these things are, I think people are abandoning them to some extent, at least like people who want to protect their peace and their mental health to some extent. But I think that when there is an act of war, a mass shooting, like something like that, you can't really solve the information vacuum that occurs between the time that something happens and the New York Times reporting on it or CNN reporting on it or whatever, where it's like, okay, here's a source of information.

41:41And the solution is that there's not that much you can do at three in the morning laying in bed, wait till some of the facts become clear. But I think that human nature is such that it's like, no, I want to see what's happening now, especially i mean god forbid but like in in a mass shooting situation it's like often people are trying to figure out what's going on and then a lot of the stuff ends up getting scrubbed at some point like maybe it's just because i am a journalist but it's like i want to see the stuff from the internet like before it gets deleted i want to i want to like know what's going on and it's just like unfortunately i don't i don't think it has to be Twitter, but it's like, there needs, there is, we're used to a world where you can post something instantly and see it instantly.

42:34And like, that is not how the news media works for like understandable reasons for the most part. Although the New York times took like five hours before they said like the U S did this, which was like crazy, especially because there's reporting from Semaphore saying that they were tipped off to this beforehand. It's just like that there is... People are going to want to try to solve the information vacuum that exists. And I don't know how you solve it other than with some of these micro-blogging platforms or clicking around Instagram. It's not great. I don't know. It's funny because I was kind of experiencing this yesterday because there was the gunfire at the Presidential Palace in Caracas.

43:18Did you guys follow this at all? Did you see this? So it was a... There was gunfire at the Presidential Palace in Caracas. And I saw it happen in real time. I was looking at the posts. It's getting posted on Reddit and some Discord groups I'm in. And people started to work themselves up into a fur. And this was like, all right, this is it. This is the military coup. Here's the fallout from all of this. You can already see people designing the narrative in their head before we know anything about what's going on, based on a couple guys' posts. And then come to find out a drone had gotten too close and they'd shot it down.

43:57And that's it. It was just like a quadcopter. And that was the whole of what had happened. But for 30 minutes, in real time as we're watching these videos, we've worked ourselves up. And yeah, I don't know how you solve that problem because that is human nature. We're always going to be jumping to those conclusions. Yeah. The thing I'll just add is that, I mean, that's like the old problem. The ongoing problem. That's the problem of journalism, right? There's this kind of unrelated thing, and it'll be a tangent, but it'll bring it back to the AI in that, and sort of the Grok problem as well. There was a post on Reddit recently, over the holidays, that claimed to be from a developer at some food delivery app.

44:43a new Breed or a Deliveroo or whatever. And it was making all these very bold claims that there's desperate pricing for certain drivers, et cetera, et cetera. We didn't cover it because we were on break at the time. I think we all needed our break. Other people did and they determined, various outlets did, I think Platformer, maybe the version as well, that it was likely AI-generated text. The thing I'm getting to is that one of the reporters engaged with the poster and said, hey, can you verify you're an employee or whatever? And they provided an image of what purported to be like an Uber employee badge, which was also likely AI generated.

45:31That terrifies me as a journalist in that maybe one day somebody's going to try and do that if they haven't already. and that just makes our job so much harder. Like the combination of the problem you're describing, Matthew, and now just somebody can just whip up something that looks kind of like an Uber ID in like 30 seconds or less. How long until you get got? Because it's probably going to happen one day. Everyone's going to get got. Everyone's going to get got in a personal context. It's not an age thing. It's not an education thing. it's not whether you're online or not everyone's going to get got in a personal context and everyone's going to get got in a professional context at some point as well essentially Jason sort of to wrap it up just briefly what and this was in the piece sort of towards the end what was it with journalists trying to get Grok to apologize for this behavior oh god I mean, I'll try to be quick, but basically, people ask Grok, hey, are you sorry that you're generating this?

46:44And Grok was like, oh, yes, this happened for XYZ reason, and we're working on it, and we're sorry, etc. And this is rule number one about reporting on AI, is that AI is not sentient. AI cannot apologize. AI cannot really explain why it did something. And it especially is not a stand-in for the company that created it. This has happened a lot of times. And I feel like I was going to write a story about it one time. But there was some horrible, like the meta AI bots that they launched. And there was a few journalists who started interviewing these meta AI bots and why they were not diverse enough, something like this.

47:37And the meta AI bots were explaining that there was an internal problem at meta that they were exposing all of this. And it's like, that's not how an LLM works. The LLM does not work. The LLM does not know what is happening at the company. It doesn't speak for the company. It can't apologize. It can't go in and expose corruption within the company. And this is the most common email that we get from people. All the time. Overall, it's like, Gemini told me something about Google. Here's my chat. And that is a sign of AI psychosis if you're doing that. Don't do that. It's not good. It's just not how it works.

48:27But anyways, CBS wrote that Grok apologized. A bunch of outlets wrote that Grok apologized. And it's like, Grok didn't apologize. X didn't apologize. I haven't seen X apologize. Unless something has changed in the last few minutes, Elon Musk was having Grok turn him in to take off his clothes. This is not something that they're sorry about. Yeah. Grok regurgitated a series of words that look like an apology-shaped object. But it's not an apology. You need intent. Well, it gave them what they wanted. Somebody prompted it to apologize and it did what LLMs do. It gave them what they wanted. Yeah.

49:08I mean, if you ask it to not apologize, you'll probably not apologize. Exactly. Yeah. We'll leave that there if you're listening to the free version of the podcast on our Play Us Out. But if you are a paying 404 Media subscriber, we're going to talk about how we finally bought 404media.com, you know, with an M. of a fully fledged.com domain. You can subscribe and gain access to that content at 404media.co. Don't go to the.com. I mean, you can, we'll redirect. Anyway, just go to 404media.co. As a reminder, 404media is journalist founded and supported by subscribers. If you do wish to subscribe to 404media and directly support our work, please go to 404media.co.

49:52You'll get unlimited access to our articles and an ads-free version of this podcast. You'll also get to listen to the subscribers-only section when we talk about a bonus story each week. This podcast is made in partnership with Kaleidoscope and Alyssa Midcalf. Another way to support us is by leaving a five-star rating and review for the podcast. That stuff really helps us out. This has been For a Full Media. We'll see you again next week.

50:33New emars

From the publisher

We start this week with Matthew’s story about an organization tracking the location of AI data centers around the U.S. and elsewhere in the world. After the break, Jason tells us all about what Grok got up to over the holiday break, and we ruminate on what the breakdown in the information ecosystem means. In the subscribers-only section, we talk about how we bought 404media.com!

Timestamps:

1:38 - Researchers Are Hunting America for Hidden Datacenters

25:58 - Grok's AI CSAM Shitshow

Subscriber's Story: We Bought 404media.com

YouTube version: https://youtu.be/zT9lEyHnZIk
Learn more about your ad choices. Visit megaphone.fm/adchoices

More from The 404 Media Podcast

All 164 episodes
The People Tracking America's AI Data CentersThe 404 Media Podcast · 51 min
Listen in VO