This is AI Inside Episode 20, recorded Wednesday, June 5th, 2024. How AI fueled the ANOM takedown. This episode of AI Inside is made possible by our wonderful patrons at patreon.com/aiinsideshow. If you like what you hear, head on over and support us directly.

And thank you for making independent podcasting possible. Well, how's it going, everybody? Welcome to another episode of AI Inside, the show where we take a look at the AI that's hiding inside everything, including our coordinating and organizing our communications through wiretap. That is one of the major topics that we're going to talk about on today's show. I'm Jason Howell, joined as always by my co-host, Mr. Jeff Jarvis.

Hello, hello, boss. Good to see you. I was going to be with Jason on the couch doing a dig impersonation. Totally. But I got an earlier flight back from San Francisco, so here I am back, bleary-eyed and red-eyed, but here I am.

Yeah, you did the red-eye flight. I feel for you. Those suck. I'm not a fan. No. Sometimes it's the only option you have, though. I spent a few hours on the couch already, yeah. Okay, okay. So instead of the Diggnation couch episode, you took a nap.

It was the snoring couch, yes. Excellent. Well, it's good to have you here. Before we get to our main event today, we're going to have an awesome interview that's coming up in a second. Just want to say a big thank you to those of you who do support us on our Patreon, patreon.com/aiinsideshow. One of our newest members, Clint Kennedy.

Thank you, Clint, for your support. This is an independent podcast. This is how we run it, is through your contributions and your support. So thank you for doing that, patreon.com/aiinsideshow. All right, we got the business out of the way. Now it's time to get into the main story at hand. And I'm so happy to welcome to this show, talked with him many times on Tech News Weekly, over at Twit over the years, Joseph Cox, now 404 Media, and been doing the 404 Media thing for a while now.

And I absolutely love the work you guys are doing there. But in this case, author of a new book that just came out called Dark Wire. How you doing, Joseph? All good. Thank you for having me. And good to see you again. Yeah, great to see you again. And once again, anytime I talk to someone who has written a book, and Jeff, I mean, you've written so many of these. I have just a huge amount of admiration for someone who can collect and organize all this information, and then put a book out there. It just seems like a mountain that's too steep for me to climb personally. Maybe I'll do it someday. But you have been working on a story for many years now at this point.

And you've kind of collected all that information and released a book called Dark Wire. When is the official release date? Is it out in the world and in the wild at this point?

Yes, it came out on June 4th. So it's available everywhere now. Excellent.

Mazel tov. Congratulations. Thank you. Yes, indeed. It's a big day. It's got to feel good after all that work. Yeah. Oh, yeah. So here, if you're watching the video version, there you go. You can kind of see a little bit of what the cover is about. And it says, the incredible true story of the largest Sting operation ever.

And Joseph, you reached out to me not too long ago and said, hey, I've got this book coming out. Yes, it's about this massive criminal dragnet, a data collection and all that. But it also has a really cool and interesting AI component to it. And it seems like, you know, not a lot of coverage of the book so far is really focusing on the AI components.

So that's what we can kind of have the opportunity to do today. Before we get to all that, it probably just makes sense for you to kind of give in your own words what this book is about. What is this story that you've been following for so long?

Yeah. So as you know, I cover the encrypted phone industry. And this is sort of a shadow tech sector that caters specifically to organized crime, drug traffickers, hitmen, weapon smugglers, all of that sort of thing. It's separate to Signal and WhatsApp and Wicca and WIRE. They're, you know, the apps we all use every single day. These phones and their related apps sell for hundreds, if not thousands of dollars for a six month subscription.

So in around 2018, moving up through the years as well, this particular app appears on the market and it's called ANOM. It's very cool. It's very slick. It has all of these amazing features. You can redact photos, you can scramble the voice, you can wipe all data from the phone.

There's no GPS tracking, all of this sort of thing. It gets very, very popular with a lot of serious drug traffickers, you know, multi-billion dollar cartel leaders, except one thing. It was run by the FBI basically the entire time. And the FBI has been getting all of those messages essentially along with the Australian federal police, along with European counterparts. Obviously there's a lot more to the story than that, but that's the broad strokes of it in that this is unprecedented, both in like the FBI ran its own tech company with all of the sorts of normal issues and problems that any sort of tech CEO or tech company is going to face, but also just in the sheer scale because, you know, they arrested over 1,000 people just from this one operation alone with 10,000 police acting on the big day when they shut it all down. Wow. What kinds of crimes were charged to the 1,000 people?

Was it a broad range?

Yeah. I mean, primarily focused on drugs, but of course that's already a really big sector, right? So in there you'll have multi-ton shipments of cocaine. There's hash in there as well. A lot of methamphetamine, a lot of amphetamine. And then you have sorts of murders, assassinations. I mean, it comes hand in hand.

The little stuff. And I guess the key thing there is that the FBI had to intervene

as well as it could when it detected one of those threats to life. And I will get into that a little bit with the AI discussion. But then the final sort of pocket of crime is like, when I spoke to the prosecutors in the Asian Involved, they were blown away by the amount of public corruption on there. You know, like government officials, people who work in airports, people who work in seaports, and they're the sorts of people who don't normally get arrested. You know what I mean? It's like you just get the criminals or the mob bosses, but this allows them to see the people who don't get got essentially. Wow. One more question about that.

Just then we'll get into the AI stuff. This sounds like the kind of the wholesale level over Silk Road and retail level. It sounds like it was, this is the big deals were happening. Is that, is that legitimate?

Yeah. Yes, exactly. This is not like street dealers. I mean, sure. Maybe the phones could trickle down to there, but it really targeted the top players. And you especially have people in Dubai, for instance, which is basically a hidey hole for a lot of organized criminals at this point. And that's also just how crime works, but also a benefit to the FBI where a top tier drug smuggler at the top of their pyramid would buy the phone. Well, you can only speak to other people if you have the same phone. So that meant it would trickle down and then the people underneath would have it and then have it and then have it. And that's how it actually spread because that's just how crime works and how it's actually organized. Wow. Yeah.

I think what fascinates me about this story, and I do remember, God, I I'm almost positive. We brought you on tech news weekly, you know, throughout your reporting on this, because it's just such a fascinating kind of glimpse into, you know, how this works behind the scenes and not a story we're used to hearing. But I have to imagine that when you're talking about criminals of this caliber at this tier, it's not easy to penetrate that world with something that ends up getting trusted and used by so many of them. Yet it's a you know, it's a total it's a total honeypot, essentially. It's the deep fake of deep fakes.

Yeah. I'm just amazed that they were actually able to pull that off because I and mind you, I don't know this level of criminal, you know, on a deep level to understand like how much they scrutinize these things, but I'm assuming probably a lot. How were they able to, you know, kind of get this past the sniff test?

Yeah, that's the thing. So obviously there's a lot of these different phones in this underground industry. I was explaining earlier what a norm and by extension, the FBI had to do was not just run like a good company.

It had to be better than the competitors, basically, because you actually want people to come over. Because I guess in business talk, there's a finite customer base, right? There's not an infinite number of criminals on the planet.

And there's not an infinite number of serious drug traffickers. I guess it's like, you need that customer base in that demographic. So they did that by making it basically the most feature rich phone you could imagine. As I mentioned, there's the images, there's the voice scrambling. Um, it had a couple of other really crazy features. Like the enormous app itself was hidden behind the calculator. So you'd have to type in two times to press equals that would fade away and the enormous app would appear. And that sounds, that sounds really stupid.

Yes, exactly. And that sounds very stupid and a little bit of security filter, but drug traffickers love this stuff. They love the gizmos and the gadgets and they sort of love the theatrics of it because to them that looks secure, even though it's not really like it looks good. And of course it's very expensive. So it must be a very good product as well. They marketed themselves as the

Royals Royce of encrypted phones as one, uh, law enforcement official told me.

So how did AI enter into this whole operation? Yes. So as you can imagine,

they start getting a lot of messages because all of these criminals, it starts in Australia with just a handful. Then it goes to Europe. Then it's sort of Asia and it's a South America.

And I mean, there's two different problems there. The first is obviously a lot of those countries do not have English as their primary language. And you have a ton of FBI agents reading the messages like, well, what the hell does this say? Like it's in Swedish or something. So they do turn to Amazon web services for their sort of machine learning translation service.

I think they used APIs and they could do that in bulk at scale. So you have the language thing. Then probably the more important bit for the AI is the sheer scale of the messages. It ended up going to something like 22 million messages by the end.

And that's not just one big data dump. It's every Monday, Wednesday, Friday, they get more messages. They have to go through all of them and review all of the images to then say, oh, this is a gun. And maybe this person is going to be shot. We have to detect what is going on here and try to figure that out.

And that's where AI comes in. The Dutch officials, their national police agency, they developed these AI programs because they worked on similar investigations before that would automatically surface the content of communications that they thought would be relevant to the investigators who were digging through it. So it's sort of a very, very interesting training set where it's just criminal communications, and it's just learning from the criminal communications. And there's one of the specific examples given to me by the Dutch official I spoke to was that, oh, we can detect if this conversation is about cocaine, even if it doesn't contain the word cocaine, because of course they may be using a slang term, right? Which is I don't local to the terms, but the system could learn to surface those communications and then put them in front of an analyst and be like, hey, you should probably pay attention to this. And you should probably read this as well.

Because it's picking up on those patterns and saying, okay, this keeps coming up at some point, they're able to determine that that pattern actually equates to.

Well, it's also, I presume that there are human, in this case, law enforcement officials who were identifying, teaching the machine as well, right?

Yes. Well, that's the really funny thing is when the FBI were getting access and then eventually they were speaking to Swedish officials and they kept saying, this is the FBI talking, what's the salmon case? We keep seeing the word salmon.

What's the apple case? And the Swedes laugh because salmon is a slang term in Sweden for like a sum of money or something like that. Whereas the FBI thought there was going to be drugs smuggled in fish.

And it's like, no, you completely misunderstand. So of course the AI is helping as well, but you need that local institutional knowledge to then feed back into the system and not even just the AI, obviously for the agents to understand them themselves as well.

So for the translation, they used what was available in the open on Amazon web services. Yes. What were the underlying models they used, and technology they used for the AI? First question. The second question, did they get help in doing this from both AI companies? And I would think that social companies would have been useful. And there's a lot of connection in cybersecurity and trust and safety with government and the social companies. Was this so secret they couldn't get help? I mean, how did that whole relationship work with the technology industry?

Yeah, it's basically the latter because of the sensitivity there, this system, at least on the FBI side, was developed in-house. So they made this, sort of the surveillance backend of a norm was called Holler iBot.

I couldn't tell you why they called it that, but that was the name and that was on all the documents I obtained. And you would log in, it would have a little screen and you could read all the messages, all of that sort of thing. But I was told that was made in-house by FBI computer scientists because I don't think you're going to outsource that sort of thing because it's a super specific use case.

And they made that Holler iBot system, not just for the FBI, but they also of course wanted to make it for the European officials who were eventually remote into the system as well. But then of course, as I said, this started in 2018 and went up to 2021. What happened during that time, the COVID pandemic. So they had to quickly spin up production and accelerate production on this tool because nobody could come into the office. So they had to like very, very quickly make Holler iBot so they could, oh, well now we can tap in and get back to our jobs. Yeah.

So do they hire people away from AI, from colleges? Do they get them trained in this technology? I mean, we weren't yet at the point of, oh, I could get an LLM to write my code for me. This is high-end technological effort and expertise, isn't it?

Right. Exactly. Yeah. You're right in that it's highly specialized, but also the timeline. Like this is pre-commercial open AI and GPT. So it's in-house computer scientists. And I think that's like a role inside the FBI that's basically often overlooked by the public. You have these very large teams inside the FBI who are doing any number of things when it comes to computer science and their applications. And like that could be developing turnkey solutions for very specific problems like this, but then there's also, we need somebody to develop a tool to break into an iPhone or we need to remotely hack into a computer system.

And often they actually buy those externally as I've reported elsewhere as well, but sometimes they are configured or tweaked or built entirely in-house. And we never really hear about it because it doesn't match the stereotypical image of a FBI agent with a badge going out when actually a ton of the work is being done behind a desk somewhere in Virginia or in this case, San Diego. So let's get Snowdonian and paranoid for a second.

All of this technology and all this ability, I presume is useful for just plain old pots, wiretaps and surveillance of other sorts. Do you think this is being used elsewhere in this country and other places?

Well, the thing that it brings up for me is that I don't know if you've spoken about this previously or if you've seen this, but Axon, the law enforcement contacting giant, recently released its own product, which is based on open AI, which will take the audio from a body cam footage and will automatically generate police reports based on that footage.

It's just from the audio, that sort of thing. And I found that very, very interesting because that comes out a few weeks ago before my book came out. And I've been sitting on this information about the FBI and the Dutch officials using AI. I'm like, wow, that was quick. It's already gone from the FBI to now local and state police.

Obviously it's not exactly the same sort of technology, but broadly, the application of AI is being applied much more broadly than I think we would anticipate inside law enforcement. And it kind of just happened and we didn't get to really talk about it. And now it's already here. You know what I mean? Yeah. Wow.

Yeah. Not only that, I think what's fascinating about this is this is stuff that we just normally don't hear about, yet people's imaginations run wild with the kind of access that is provided to law enforcement behind the scenes, what they're able to do with it. But it's been a lot of guessing for a long time. So this is really just kind of confirmation that, yes, they are able to do this. I guess, what does this say about the ethical implications of potentially the overreach of an operation like the ANOM operation? I mean, this is quite a scale that we're talking about. And being able to scoop that up and do all of that analysis on a broad scale, that certainly concerns privacy advocates. What do you say about that?

Yeah. I mean, on one side, you have the AI stuff. And I still think there are a lot of open questions about, well, how effective is this AI or any other that law enforcement may use at surfacing particular pieces of information? Is it reliable in that?

Is there any bias in that? Can we see the process and can we see the results of that as well? That's going to be case by case, product by product, but I think those are valid questions. When it comes to the scale of operations like ANOM, and there's two others, one called EncroChat and one called Sky, where the police didn't run those companies, but they hacked into them and they were encrypted phone companies as well. There's so much data there. And so many people use these phones that I did find it sometimes did include people who were not suspected of a crime and especially defense lawyers. And their only quote unquote crime was that they were defending a drug trafficker. And you can think whatever you want about drug traffickers, but they are entitled to legal representation as well.

That lawyer has not been suspected of a crime. Now, maybe that's actually not an issue for you. And maybe you don't think that's a problem.

Okay, sure. Well, then the next stage comes and there's a cryptographer at the end of my book who puts this thought experiment forward. And he asks, okay, well, what happens when the criminals move to other networks that we all use like Signal or WhatsApp? And I've spoken to drug traffickers and I've spoken to people who sell these phones. And they say that is what some of these criminals are doing because they don't trust the encrypted phones anymore because of ANOM. So is the FBI just going to go and give up and not investigate anymore?

No, I think that would be ridiculous. It's now what happens to the other platforms where there are more, um, quote unquote, ordinary people, you know, members of the public, like, could they be swept up in it? We don't know. But in Europe, there's legislation where European authorities want to be able to scan the content of encrypted messages initially for CSAM content, but they've already said that they want to expand that to other crimes. It is like, it's not even hidden. They're just coming out and saying it.

And, um, I just think there needs to be much, much more debate about it because with this ANOM case, it does show the extraordinary lengths that law enforcement will go to, to access encrypted communications, including running a tech firm outside of the FBI, San Diego field office. I wonder whether they had foosball.

Um, they had a lot of pizza, a lot of pizza in the kitchen.

Um, uh, yeah, I was just visiting a friend at Google yesterday and boy, they have good snacks. Jesus. Um, I presume I'm going to keep my Snowden mask on for just one more minute. I'm not normally the paranoid one, but I, but I will for a second, I would imagine that a lot of these skills are, uh, endemic to what the Chinese internet does and what other governments do.

And so I'm curious about how this passes over the bloodstream. Did one way is with the Western law enforcement could learn from what, um, authoritarian governments did. But the other is that as this becomes public, uh, was it, was it any of the technology published as scientists do? Uh, is it just known now? Is it going to be available to bad regimes, including possibly soon in this country?

Well, that's the thing, even though the FBI has announced the operation, because that was part of their goal. They wanted to come clean, so to speak, to so that miss that doubt inside the criminal

community, they'd be like, Hey, look, we were running the company. You can't trust these phones. And I understand why they did that. They still didn't give many details, which is why I wrote a 300 page book about it and spoke about all of that.

But when it comes to the technology spreading, yes, it goes more back to, I think the axon example in that the specific technology, the specific LLM used in any sort of single operation, maybe that stays in house, maybe it spreads, maybe it's repurposed, but I just think generally there is absolutely going to be a push towards more artificial intelligence used in policing, because as you say, you do look at China and you see this, you know, omnipotent, uh, omnipresent use of facial recognition and that sort of thing. And of course, there's been lots of great reporting about clear view AI as well.

We're only getting closer and closer to that sort of thing. Now, of course, the key here is that a norm was a network primarily used by criminals. I know I said, I found examples of ordinary people using it, but again, primarily criminals. It's when they cross over into that almost 51% of normal people. Now that is where it's going to feel very, very different.

Well, and also the, the AI kind of systems that are trained and designed to analyze all this data. That is a big crossover point too. Like no matter how the data is acquired, you know, be it through a specialized phone that targets only criminals or the open internet, there's still the, the tools that, you know, aren't fully capable of this very specialized kind of analysis of, is there something wrong happening here? And that does cross, as you said, Jeff, that crosses the bloodstream, uh, and, and touches on all points. And I imagine even more so, you know, year after year as these systems improve, as more of our communication is done in highly trackable ways and highly collectible ways. Yeah.

The only thing I would sort of add on that is that I've been talking a lot about the messages that were intercepted. And of course that's the most valuable data, but what the FBI also got was an incredibly powerful map of somebody else's connections. And it got to the point where these are non criminal users were talking to each other so much that one of the FBI officials who I heard from, he explained it less as a messaging app, as a social network for criminals, like it became the Facebook of drug traffickers because they had their little circles and then this circle is their close contact.

And then they speak lots to this person. It's the sort of pattern of life analysis. A lot of AI is being used to do, or I guess machine learning a little bit more specifically, but I can absolutely see that expanding into, um, other forms of policing as well. I mean, I've reported before about how DHS is using AI powered products to analyze social media when people cross the border and that sort of thing. I really don't think people appreciate how much AI is in law enforcement, in federal, in border enforcement, and even local police now with the axon stuff.

And what's interesting too, is that, is that in me and I'd love to maybe cross another bloodstream here and hear how this affects your coverage and worldview at 404. Um, but the, the, the tech companies and the AI companies are often presented as the evil, um, too powerful being, but what fascinates me about this story, Joseph, is that, is that the, the authorities didn't need the technology companies. Yeah.

Well, they made their own.

And, and so I think we focus on trying to find the, the, the, the body, the, the, the actor, uh, who's evil. If we can just stop them or regulate them, then everything's going to be okay. No, it's just technology and you can use it for whatever the hell you want to use it for. Yeah.

And if you feel like it, you can go and do your own spin off of it and you can make your own copy and tweak it as you see fit. I mean, the cat is out of the bag with AI, right? I mean, we all know that that's a very obvious point to make, but I really, really think that's the case where we're only going to see more of this in more operations going forward for like absolutely sure. And I really think it could be on the smaller scale as well, where maybe they won't run an entire phone company, but let's say we have all of these, you know, just normal wire taps. I think you mentioned earlier, we'll use AI to quickly churn through those. I can absolutely see that happening as well.

So how has this affected? I mean, you've been covering this for a long time, so it's obviously affected your, your built on shalong, your, your worldview throughout, but 404 is an investigative technology operation now, may motherboard rest in peace. And so you probably already have a somewhat paranoid view of the, of the use of technology, but I would imagine this only amplified that on the one hand, on the other hand, we talk a lot on this show about test Creole and long-term ism and the doomers and, and, and all of that. And I mock them because it's, it's, it's the future risk is BS. It's the present risk. And you very much have a story of present risk and technology. So sorry for the long-winded Joe Scarborough like question here, but how has, how does this affect your own worldview of your coverage at 404?

It doesn't a few different ways. I mean, we cover AI constantly, and I genuinely think you should speak to Jason Kepler as well. One of my co-founders is doing a lot of stuff about AI and Facebook spam and that sort of thing. And what we really focus on is that while everybody is talking about these hypothetical risks, which are like really, really far in the future, or they're just not really tangible.

We are super focused on talking about the real harms or, you know, sometimes benefits that are happening right now. So one of my co-founders, Sam Cole, she was the first journalist to uncover deep fakes. And she's always said, this is being used to harass women, for example, which is obvious. And now it seems that people are coming back to that and being like, oh yeah, it's used to harass women, right?

Yeah. We've been talking about that for a very, very long time and it's still happening. And yeah, exactly on the screen now, she published this piece just recently about legislation that's going to go forward and potentially forget about the other people who are impacted by it. So you gave a very long question.

I'm giving a very long answer, but the summary of it is that we really try to focus on the human, actual, tangible, concrete impact that is actually affecting people now, rather than like this floaty sort of abstract idea in the future. I mean, Amen, brother. Amen.

It's also the, it's not just the human impact. I think that's part of it, but it's also what's fascinating about your story in the book is it's the human actors. Again, it wasn't the technology company. It wasn't some really reinvented technology.

It was the use of present tense technology for things that, and as Jason said at the beginning, you might think is a good use to get rid of crime, but that has serious implications. And it came from within government, which is fascinating. How many governments were involved in the end?

There was something like 16 involved in the end. But just on that, when you say it's also the people behind the technology, I entirely agree. And I think that's really important here because this is their own description.

The people I've spoken to in San Diego, they saw themselves as a group of cowboys, essentially, who were like, we have this amazing idea. We're sick of how drug crimes are prosecuted in the United States. Let's do it different.

Let's be really, really bold. And they were told multiple times along the way by low management, middle management, and actually some people at the top of the DOJ as well, that you can't do this. This is stupid. This is ridiculous, but they persevered and they pushed through it. And I do think that goes to show, again, there were people behind this operation. And if there had been other people, it just simply may not have happened. It was this crazy harebrained scheme from a handful of FBI agents and prosecutors.

This could have easily not happened if they just didn't have the audacity to do it. Okay. So the important question here, Joseph, movie deal? It has been optioned by Netflix. Yes.

Oh, yay. Mazel. That's great. It's coming. I know. That's amazing. Um, was it optioned? I'm curious, was the book, the forcing mechanism to get that option or was your coverage already being talked about as, as that, as an opportunity, did you need the book to sell the movie?

It was the book proposal. Um, so a while ago at this point, but because I covered the, uh, the story for so long, even though I didn't have all of the granular details, I knew the arc. I knew this story inside out.

Um, that's, that's how it came about. And of course I went in with the goal of writing a very entertaining nonfiction, uh, true crime book. You know, that was just my goal. That's a new challenge.

I'll do that. So I wrote it like a movie as well. Uh, I don't know whether I'll be effective. That's, that's kind of up to the people who read it. Right. But that's what I was going for anyway. And I hope that comes across. Yeah.

Okay. Uh, other, other showbiz question. I'm a former TV critic, uh, who, uh, what are some of the great characters in the story? What are two and who should play them? Oh, that's going to be really, really, um, not me.

I'm not in it and I don't want to think, um, I would just say that, I mean, there are some really, really interesting law enforcement characters who, um, who are very determined to do what they think is right.

Even when they're told you can't do this. And then there are some very interesting drug traffickers. One being a guy called Microsoft, not connected to the tech company, just called Microsoft. I don't know why, um, who is very, very crazy. And there's another guy called Hakan Aik. He's one of the world's biggest drug traffickers. He's in there as well.

I couldn't say who should play them. That's, um, you know, I'll leave that to the experts, but there's some side of the story about the FBI nerds. I mean that maybe, maybe I could do that one. I could, I could probably fill in the, the, the, the nerd role, but there's a ton of characters in there. And I hope that people find one that they can latch onto. Like my favorite characters, Microsoft, just because he's so crazy and so colorful and entertaining. Um, but there's something in there for everybody. I think that's just great.

You continue to do amazing work, Joseph. And I'm so happy you reached out about this. I'm so happy you put it on my radar. Dark wire is the name of the book. Uh, it is out again, the incredible true story of the largest sting operation ever. And it is fascinating to hear you, um, kind of detail the kind of the fruits of all of your work over the years.

And, um, I I'm, I'm definitely going to be checking this book out and waiting for the movie on Netflix. That is so cool that that's happening. You gotta be so happy about that.

You are, you are, um, uh, the Howard Stern king of all media of, of geekdom. I'll take that book movie. I know. Right. Yeah. Yeah. Probably a newspaper.

Right on. Well, four Oh four media. Uh, everybody obviously needs to follow you and your team's work at four Oh four media. You guys are doing incredible work over there.

I love what you're putting out and then obviously check out the book, Joseph, thank you for reaching out and, uh, love to have you back sometime when there's a, when there's a crossover, we'll definitely reach out. Yeah. And do you do the same please? Yes.

Just downloaded the Wow. Thank you. Excellent. That's one sale. As I've got a book coming out from Hachette also in the fall called the web we weave. So you always be selling show. Always be selling. I, everybody watches the show can recognize that I will sell the good word parenthesis magazine and the web we weave constantly and obnoxiously retail battle.

All right, there you go. Find, find a bookshelf in the back and, and throw a copy on there. So people see it, even if you don't mention it, there you go. Yeah. Joseph, thank you again. It was a pleasure catching up with you. Thank you. Anytime. Anytime. All right. Right on. Take care.

I was at a bookstore, uh, two weeks ago and I found a copy of the good word parenthesis and it was kind of hidden in the digital section and I moved it up to the front of the store.

Heck yes. I bet you're not the only person to do that. It goes into a bookstore and you're like, this is not where it needs to be. It needs to be right here at the front door facing everyone. So thank you.

Thank you for arranging that with Joseph. That was, that was fascinating.

Oh yeah. He's, he's one of my favorites. He, he was always so much fun to bring on a TNW and I say fun also through the lens of fun and frightening at the same time. Cause usually the, the, the topics that he, he's researching and reporting on, uh, go deep. And, uh, this book sounds like very similar, similar lines as well.

Uh, so everybody should check that out. All right, we're going to take a quick break. We've got a short amount of time left, so we'll squish some news in there in a moment.

All right. So here's what I learned this week, Jeff, and I think I'm, I'm starting to just go ahead and come to terms with this. NVIDIA has, you know, big chip news that they announced Intel had a big laptop chip news that they announced. Um, uh, Raspberry Pi, that one coming, uh, had some big chip news to announce. And you know what I'm realizing? I realized that this stuff is really important. I realized that, that AI chips and you know, the, the advancements and everything, like obviously it drives the industry, but man, when I read, when I pick apart, attempt to pick apart some of these news announcements, I get lost in the details. And really at the end of the day, I come out of the other end going, all right, they're faster, they're cheaper. And you know, they're more performant or whatever the case may be. There's so much detail in, in how, in how, and what these chips are. And the, and that is just kind of lost on me. That's not the kind of thing I should admit, but I'm, there we go.

No, I don't think you're lost in the details. I think you, and I would put myself in the same bucket, of course are lost in their hype because it's constantly bigger, bigger, faster, right? It's like, okay. To what end? Uh, it's, it's, it's like when I, when I watching Jensen Huang last keynote, he's had one or two since then. Yeah. Where it was just this, this, this, and we talked about this in the show, it's this expanded, this huge capacity. And you have Elon Musk saying, I'm going to build the biggest God, everything. It's all very macho size matters stuff. And I constantly quote, um, the stochastic parents paper where they mocked that and said, size doesn't matter.

It getting bigger. It only makes it more difficult to, to manage and understand what it's doing. And it doesn't necessarily perform better to your goals. So it's the ultimate case, I think of a tool looking for a problem to solve. And, um, and so I think it's just, it's just marketing hype now. And I know that's, that's, that's simplistic and, and, uh, the size and the speed do matter, but I don't think that there's necessarily a vision for where. Yeah.

Yeah. So, I mean, and, and this is actually interesting cause you said, you know, one of his previous, um, kind of reveals or whatever, you know, it was only three months ago that, that Nvidia CEO Jensen Huang, uh, announced the Blackwell kind of, uh, direction for, for their AI chips. And now here we are three, three months later and yes, they were announcing kind of an expansion, an extension of the Blackwell chip, but they, but he also said, what did he say? He had a, he had a quote where he was like, I'm not sure yet whether I'm going to regret this and then announce the next generation, which is called the Rubin or maybe just Rubin. The Rubin kind of sounds like a sandwich.

Um, Rubin platform in 2026, Rubin Ultra in 2027. And like, I'm reading through it and I'm like, you know, there are some people that read these announcements. They're like, oh man, can you believe it?

You know, it's got all these doopadoo and they're, they're naming off the acronyms, acronyms and everything. And for me, it's just like, it always comes down to, you know, uh, cost and energy savings, faster performance. And it's like, okay, it's a chip.

It calculates things and it does things in a different way. I mean, I'm old enough to remember, you know, the PDP 8080, you know, is out. And, and, and then when I watched the original twits, uh, they would talk with excitement about a new chip. Um, and we've never really known what it matters. I was thinking the other day, you remember that one Intel chip that, that had a math problem built into it, uh, that was in PCs that I was some time ago, this is a long time ago, but I was thinking, what if there's some in the, in the grand complexity of these chips, what if there's a mistake, right?

What, what impact might that have? I just looked up Nvidia stock and by the way, I'm sure I have some of it in my, um, various funds. Um, six months ago, it was at four 65. And right now it is at 12, 17, $1,217 for a share. Uh, that's incredible. Right? That is incredible. So speed and size matter in the market.

That's why they're right. Tech, right. Time. You know, a lot of people are looking at, at Jensen Huang and saying, Oh, he's the Steve Jobs of AI and all this stuff. I mean, or, or the, the other analogy is that, uh, what is it that they're handing out the shovels? What is the shovel analogy? It's like when you've got, you're not cashing in on the, on the gold rush, you're, you're the one that sells the shovels. And that's what Nvidia is doing for the world of AI right now that they're in the right place to, to make a killing.

Tell me if I'm wrong here, but it occurs to me that I mean, Jensen Huang has his own look with a leather jacket and Mark Zuckerberg is jealous of it and so on and so forth. Is he the first chip celebrity? Hmm. Oh, that's a good question. And you had, I mean, you had moguls audience ships. I don't think you ever had this personality,

but this like flashy kind of like, Oh, I know exactly who that is when they walk. You know, he's got a certain, a visible style to him that is in some ways reminiscent to a Steve Jobs, you know, the black turtleneck kind of aesthetic or whatever.

But, um, it was always something inside. You didn't know how it operated, but it was in there and you cared about it now raises to the surface, this brand and its capabilities. And it's kind of fascinating to watch at the same time that to the average person, the technology won't matter at all because you're just going to speak your language to it.

That's what it enables the complexity of make simplicity. Anyway. Yeah.

Uh, yes. So there was a, there was a ton of that kind of news this week and, um, we don't understand it all. We're not going to understand it all and we don't have to understand it all.

Well, you know, and, and, and it's like, I'm, I'm, I sometimes struggle when I'm faced with this kind of avalanche of, of this particular type of news. Like I've tried for years to read through a lot of the, you know, the hyper-threading and the multi-cores and it's like on a, on a grand level, I understand why it's better, but there are people that read that, they get really excited about the possibilities and, oh, this is, this is what this means. And for me, like I said, it always just kind of gets distilled down to, you know, better, faster, smaller, cheaper, whatever.

And like, you know, Intel's lunar Lake laptop chip, um, two memory sticks attached 16 and 32 gigs each. Like, okay, I could, I could see why that's, you know, that's to celebrate, but, um, but I don't know, like maybe I'm just the wrong person to evaluate why this is like incredibly important outside of those things that I said, it gets distilled down to.

Before we got on the air quotes air, uh, with Joseph, uh, we were doing media geekery and talking about scale and the, you know, that a 404 is not a vice and it's not, right. Uh, uh, you know, time Inc of old, uh, but it's big enough, right. And scales, a paradoxical thing around the internet. And now we're on the technology. And I write about this in both the Gutenberg Parenthesis and The Web We Weave, sorry for the two plugs, but you know, always be selling, um, that, uh, you know, we assume that everything had to be huge. And, and now things are more of the human scale because of the huge scale. That is to say the internet connects everyone.

Ergo, you can find the five people who have the same interest you do. And these chips are of massive scale and their abilities. And that's just what enables them to speak our language and bring it down to a human scale. And that's, that's the technological paradox of scale that I think we see here. But part of this is also just simply about male testosterone. Yes, it's about, I want to be bigger than everybody else.

I'm bigger than everybody. Yeah. No kidding. And that is something that we see time and time again in the world of technology. Uh, want to give a shout out to the ozone nightmare.

Thank you so much for, uh, the, the super, uh, super thing says, uh, Jeff, your point is exactly my issue with all this hype frothing. I've never heard that word before. Um, I'd be far more interested in a leaner, more focused tool than a bloated and aimless behemoth. So there you go.

Thank you so much for sending that in, uh, the ozone nightmare. So anyways, and you know, and I did mention very passingly, uh, about raspberry pie, uh, which I actually think is, you know, that's kind of cool. Anytime the raspberry pie does something, I'm always for it just cause I, I love what they've, what they've been able to do for, for very little money, you know, for people who are, uh, you know, kind of computer enthusiasts and want to play around and essentially they're create, they, they are bringing an AI chip for the raspberry pie, uh, microcomputer, the PI five to be specific, it's partnership with halo and a kit costs $70. So if you have a raspberry pie five, you can buy their, you know, a partnered AI chip and, uh, run native AI apps on even the raspberry pie. That's where we're at right now.

If he's, we're still alive today, you'd be buying a kit that has AI in it. I mean, it's, it's amazing.

Um, Google bringing some technological or technical rather improvements to its AI overviews. We talked a little bit last week about how people weren't very happy with the results that they're getting in AI overviews. Um, and so Google says they're making changes that you will notice. Um, they said they, they addressed, I think something like a dozen, uh, different issues that they, that they noticed that we're producing, um, nonsensical queries. So they say they've made these improvements to detect those, whatever that means.

Probably most of it was don't quote onion.

Yes, exactly. Well, so glue Google claims that the issues were okay. First of all, they say not many people were seeing these. They said it were founded one out of every 7 million AI overview queries. Um, but you know, again, when you're talking about scale, it's still, and I, that, that seems, it seems like more people were seeing it than that. But anyways, um, they said that if data for a particular query was missing, it would sometimes turn to satirical, that satirical content that you're talking about, go to the onion. So it's like, I don't have anything to put back here. Oh, sure. This'll do. Well, I'll just pull the information from the onion.

Probably not something you want to do. And you're also others rewrote the, what I saw in one, in one of those cases, I think it was the rocks case. Um, somebody else had rewritten the onion story for SEO purposes.

Didn't matter. It was just, they just rewrote it just to try to get some traffic. And that appeared higher than the onion as a regular search, um, uh, response. And so the machine wasn't wrong to say, Oh, I guess that's news because it's being fooled into it by nefarious actors. Yes. Yeah.

So there, um, Google also said that they're going to cut back though on health content, um, overviews related to health content and hard news. And, uh, so I guess they're drawing the line there, which is fine. I mean, health content, it's obviously pretty smart. People don't need to be eating rocks, you know, so,

but they will never get rid of, of, um, they'll never make LLMs as they now exist. Credible in my view. Um, do we have time for you to do your demo? Yeah, I think so. Is this, I think leads to the problem with the future web is that Jason's about to show you a mechanism to really, really easily make reports or pages. And I predict this is going to fill the web with yet more junk. Yeah.

So this'll be interesting. So thank you for giving me just a few minutes to do this. And, and, uh, leading up to the show, yes, I, you know, we were kind of chatting back and forth and you, you sent the message and I was like, Oh, I was not, I was not, uh, prepared for that. You said that glad regarding perplexity pages. Uh, it is how the web will finally be ruined for good. I'm like, Oh no, do I want to show this off? Like, I don't, I don't want to ruin the web, but I am a perplexity user. And, uh, they did announce and add a new feature called perplexity pages.

And essentially the idea here, and I played around with it a little bit, is that you could take, you know, one of your, or any number of queries that you put into perplexity to ask the LLM. And instead of getting just a, like a bullet point list or a summary or whatever, you can turn it into kind of a, a more multimedia, uh, shareable, um, piece of information, organized piece of information. So to my view, like this is one that I created through a, um, so I'd watched a video where we're getting ready for our trip to Italy.

And this morning I watched a video on YouTube. Woo. That is very loud to me.

I know you don't hear it. Um, on the away together YouTube channel. So give them full credit. Um, where they were saying, you know, don't make these Europe packing mistakes. I was like, okay, great. So I watched the whole video. So I did give them a full watch. And then I was like, well, I want to like, this was a very useful information for me.

I want to organize this in a way so that as we're getting ready over the course of the next week, you know, we pack accordingly. And so, you know, I got the transcript of that. I ran it through perplexity and then I turned it into a page. And so basically, and I don't think that it's as detailed. Like, I don't think that it caught, got all the pieces of information that the video does.

So there's that, but it's a way, and actually I could probably just take a screenshot from the video. So, no, no, no, no. I, I fed the transcript into perplexity and I said, all right, organize. So I said, create a comprehensive summary organized in an easy reference way that includes all the major talking points of this video transcript.

So this is what I got from perplexity to start off with. It's kind of, and this would, this is what I would normally do if there was like a video that I watched that I just want to get notes on it and I don't want to have to watch it again. I could do this and I get the notes, right? So then what I can do in perplexity is I can go up here, convert to a page.

And so I'll click that. It sends it through and it says it's writing and you watch in real time as it kind of puts this together in more of kind of a presentable way, let's say.

It's very impressive. So it's not just notes. That's the problem with it. It's very impressive. It puts together a nice package. So the problem is it does this well. Is that, is that what you're saying? So, so you can see tons of PR people and people wanting to get affiliate links. We'll just, we'll just be able to manufacture junk at scale.

And this isn't junk per se, but people will use it to make junk. I'm also flashing, Jason. I can imagine at your house, your wife or daughter say, Oh Jesus, dad, why are you using AI to tell us how to pack? Come on, man.

They don't know how I come up with that, with what I come up with. They just know that I do. Jeez. You're so mad. You're such a man. Yeah. But see, and so what you can do in here. So once this ends, I'll show you just kind of a little bit of an extension here, which, you know, kind of goes further to the point of like creating a presentable package is, you know, I can go up here and add media to create the like hero image. I could upload something or I could just tell it, all right, generate something. You know, this is a, this is a how to pack for a trip through Europe. So I'm just going to tell it to generate an image for the hero image.

And you know, it'll use its image generation to come up with a sure, a backpack. Sure. Why not?

There we go. And you know, I could, I could change the way that looks and everything. And you could do this for each of these particular points as well. And so I, and beyond that, then the next step of that is of course the publish button. So if I hit that, I now have a link that I can copy. I could share it out to people. So I can say hosted by perplexity with a URL.

Yes. It appears to be hosted by perplexity. I can copy that link, open up a new URL and it takes me, I mean, I'm logged in. So most people wouldn't see that stuff, my history and everything.

The computer, I'm blogging. Yes.

Right. Yeah. Yeah. I mean, it's interesting. Like I, I, I suppose I, I don't see it entirely as the way the web is ruined, but you know, being, being hosted on perplexity, I think is a, is a limiting factor, you know?

But, but it's, I don't know. It's, it seems like a way to, to, to organize what you're already doing in a, in a different way. And I think the share ability of that might, you know, might be the aspect that you're talking about.

Yeah. Well, I think shareability is good because I think you can, you can, you can have a, you could, you can show your work. I just wrote a post and I, and I got research and here's the, here's where you can look at my research or my organization of it.

Here's a query I've made and I'd like to build it collaboratively with others to be able to share that and build upon it, I think has value. No, I'm okay with all of that, but I just know how it's going to be used even not, not nefariously, just greedily. Yeah. And maybe lazily.

Exactly. And by many a student. Yes. Yes. Oh, yeah. Okay. Real quick, real quick before we, before we end the show last night, you know, my, uh, my older daughter, she's graduating from eighth grade, going to high school and she was picked to read, uh, like a graduation speech for her school. And so I was sitting, you know, with her in her room as we were kind of like copy editing. And she's like, here's my speech.

I read it out today and everybody told me it was too long. And boy, my first instinct was, well, we could just put it into perplexity and say, tighten this up. And I mean, but I didn't, no, I didn't. No, I did not go there, you know, because like, what is she, obviously what is she going to learn, you know?

And so we spent a half an hour kind of tightening it up and got it to a wonderful place, but it was just something that I noted for myself. I was like, not, not ready to go there yet. Not now. Yeah. Jason. So that's where we're at, but anyways, Jeff, a great episode, great conversation with Joseph Cox. Um, so happy you could be here for that. And, uh, absolutely. And, uh, yeah, what do you want to leave people with Gutenberg parenthesis?

Of course. Yep. That's, that's plenty there. Good. We're present.com has discount codes for my two books that are out now, magazine Gutenberg parenthesis. And I'm just going to brag for a second. Um, the sharp news, which is the, the, uh, journal for the field of book history just reviewed Gutenberg parenthesis and said wonderful things about it. And that makes me very happy because I was afraid that the real academics who study what I write about would say, what's he doing here? He shouldn't be writing about this. He's a schmuck, but there was a very nice review and I'm very grateful.

Right on. And you deserve it. That's awesome. Gutenberg parenthesis.com. Um, and then let's see here. What else do we got? Well, if you want to normally, ah, normally if you want to catch the show live, you can do that on the YouTube channel, go to youtube.com/@techsploder. Uh, where it streams live, you can kind of see the live streams are usually cataloged there. We are not going to have live shows for the next few weeks. I think it's the next three episodes are going to be prerecorded.

They're going to be released every Wednesday, you know, right around the normal time that we would normally do a live show, probably around noon Pacific time, 3 p.m. Eastern. Um, and so if you're subscribed to the podcast, you don't need to do anything different. You're just going to get the podcast delivered to you automagically.

Uh, if you want to watch the video version, you can just go to YouTube on Wednesday, uh, the at tech splitter YouTube channel, and you'll find those episodes as they are released.

Cause we've an actual vacation and gone off the show, but he wouldn't do that to you. He wouldn't leave you. So don't leave him. Make sure you watch these shows too, too many years in podcasting.

It's really hard for me to, to, to miss a week. And so I've tried really hard to make that not happen. So thank you for clearing your schedule, Jeff. Cause I know that's, you know, that's also on, on you, um, to, to make that happen as well.

So far, we've had two really good conversations and we have another one recording on Monday. And so you'll have good shows to listen to. Yes, indeed. Yeah.

Great stuff. You have, uh, to look forward to. Um, let's see here. patreon.com/aiinsideshow to support us directly. I'll just go ahead and say our executive producers of this show.

Dr.Dew, Jeffrey Marraccini, WPVM 103.7 in Asheville, North Carolina. Thank you. Thank you. Thank you for supporting us. Uh, do like rate review, subscribe wherever you happen to listen. Cause all of that increases the visibility of this show. And we can't thank you enough when you do that. Thank you everybody. AIinside.show, the single place.

If you want one place to go, that's where you can find everything you need to know. And, uh, we'll see you next week with some prerecords and I'll see you probably in about a month with another live episode of AI inside. Enjoy your break, Jeff. We'll see you later. Take care. Bye everybody.

Joseph Cox

404media

Related Episodes

How AI Fueled the ANOM Takedown (with Joseph Cox)

Tech Agnostic (with Greg Epstein)