TWiT 1105: Agents Gone Wild - Windows' New Surface Laptop Ultra & RTX Spark Dev Box!

[00:00:00] [SPEAKER_04] It's time for TWIT This Week in Tech. Great panel for you. Richard Campbell is here, Wesley Faulkner, Yanko, Rutgers. Lots to talk about. New Windows laptops. They're designed for AI. AI has gone wild in some interesting ways. And should you install this new crop of AI agents like Meta's Muse? That and more this week on This Week in Tech. Stay tuned. It's coming up next.

[00:00:28] [SPEAKER_04] Podcasts you love. From people you trust. This is TWIT. This is TWIT. This Week in Tech. Episode 1105. Recorded Sunday, October 11th, 2026. Agents Gone Wild. It's time for TWIT This Week in Tech. The show we cover the latest AI news. And then we throw in a little tech on the side. Richard Campbell is here. You've got to talk about Windows.

[00:01:03] [SPEAKER_04] Podcasts, actually. There's some Windows news. There is some Windows. There's some Windows news. Hello, Richard. Good to see you. Happy Thanksgiving. Wait, what? Welcome. Oh, it's Canadian Thanksgiving. Yeah, yeah. It's Canadian Thanksgiving. You guys always have to be like, do it first, don't you? Like, Canada Day's first.

[00:01:22] [SPEAKER_02] It was in the 1500s, right? You guys came later. You're just a little slow to the market.

[00:01:27] [SPEAKER_04] I get it now. Well, I apologize to take you away from your poutine. Oh, the big party's on Monday. You know that. Oh. Oh, good. Okay. Because family's in town and everything?

[00:01:38] [SPEAKER_02] Family. Everybody's up on the coast. And it's a beautiful, beautiful day. So everybody's out. Look at that. Yeah. Yeah. And somebody's fishing. Yep. There's some fishing going on. There were sea lions eating herring out there yesterday. So it's been beautiful. My little granddaughter's up with her parents. Oh, nice. Everybody's having a good time.

[00:01:58] [SPEAKER_04] Well, thank you for spending a holiday with us. I appreciate that.

[00:02:01] [SPEAKER_02] Yeah, a few hours. Always have some fun.

[00:02:02] [SPEAKER_04] Richard is, of course, our regular host on Wednesdays on Windows Weekly with Paul Thurrott. And he has his own two podcasts, Run His Radio, runhisradio.com and .NET Rocks, which he does with the fabulous Carl Franklin.

[00:02:17] [SPEAKER_02] Both are at Run His Radio. And we just finished recording a show 2027. So I don't have to do this year histories anymore. Because we were doing it since like the 1900s. We've been talking about, hey, it's 1914. World War I started. And then we'd go on from there. Yeah. I was really happy to record 2026.

[00:02:32] [SPEAKER_04] With show one, did you start with the year one?

[00:02:34] [SPEAKER_02] No. No. This only really came when we got to the 1900s. But, you know, you know how a bit gets out of control? Yeah, I do. Believe me. This is a bit that got out of control. I'm kind of glad to have it done.

[00:02:46] [SPEAKER_04] Yanko Reckers is also here. Seasoned reporter focused on the intersection of tech and entertainment. He's written, of course, for a variety. It was a giga-ohm protocol. And he now has his own lovely, very useful newsletter called Lopass at Lopass.cc. Hey, Leo. Hey, Yanko Reckers. It's good to see you. You're covering everything about streaming. I see a bunch of VR glasses, which is interesting.

[00:03:15] [SPEAKER_04] This is a very interesting space right now. In fact, we have some stories for you, too, because the big merger is complete.

[00:03:21] [SPEAKER_06] Right, right. All the companies are not called Skydance. Now it's just one. It's just one.

[00:03:27] [SPEAKER_04] Everything we've all hoped for all along. Also with me, the wonderful Wesley Faulkner. It's wonderful to see you from works-not-working.com.

[00:03:37] [SPEAKER_01] What stories do they have for me? Anything good?

[00:03:39] [SPEAKER_04] Uh, yes. Of course I have stories for you. I want to be honest, though. I don't actually think about who the hosts are when I put together the rundown. It just sometimes happens that way. I bet you you could talk about Meta's muse.

[00:03:56] [SPEAKER_01] Yeah, if you're talking about companies doing something that are suspicious and maybe not in the best interest of people. Companies badly behave? I can't imagine. You might have to dig something up, but I think I can have some conversations.

[00:04:07] [SPEAKER_04] I think you could find something. Actually, Carl Bode wrote the story for Tech Dirt. Your former colleague at GigaOM, Yanko, Meta's muse is an adorable privacy and security dumpster fire. Okay. And I love it.

[00:04:25] [SPEAKER_01] And water is wet.

[00:04:27] [SPEAKER_04] Yeah, what a shock. But let's start, since I said we would start with Windows, let's start with Windows. Here's a shocker. Windows 11 market share is going up. Slowly. Very slowly. Creeps upward, according to Richard's speed rating of the register. Because what? Is this true? The cost of Windows 10 support is going to double?

[00:04:50] [SPEAKER_02] The extended support's doubling in price. Yeah, it was $60 for the first year. That ends this month, actually. And so now it's supposed to be $120. But I don't think that's the reason at all.

[00:05:00] [SPEAKER_04] But it is relevant, I think, to a lot of our audience who decided to stay on Windows 10.

[00:05:05] [SPEAKER_02] Yeah. But I think those big number, look, you know, most Windows 10 operators, like individuals, there's a bunch of workarounds. If it's for you, like, you don't have to pay that fee. There's a, you have a number of ways you can spend some money on through your Xbox points. Bing points. Yeah, all of that sort of stuff. Like, you really can, right? So I don't think there's a whole lot of pressure there. I also, but I come at this from an enterprise perspective. And let me tell you, Win11 was only ready for the enterprise in 2024. Like, really.

[00:05:35] [SPEAKER_02] We didn't have the full set of group policy rules. It hadn't been benched properly. Like, it took a while. It took several years for Win11 to get its act together. And so, you know, maybe in 23, starting 24, you were ready to start deployment. And I don't care if you have 500 seats, 5,000 seats, 50,000 seats. You don't update machines. You replace them. And that happens in 20% chunks, maybe 25% chunks. Which means it's going to take four or five years. Takes a while.

[00:06:05] [SPEAKER_02] It's just no, you can't change that much or everything will break. Like, it's Pareto's law. 80% of those machines, those new machines are going to work the first time. 20 of them aren't. And you're going to have to figure that out. So you roll this out over time. But I also poked a few friends I know who do the volume license agreements. And a very common, you know, in the end, what does Microsoft want? They want everybody over on 11, right? They don't want to have to support 10 more than necessary. They want everybody over there.

[00:06:33] [SPEAKER_02] So if you're a big company negotiating a volume license agreement, you've already told them what your roll-up plan is. You're going to be done by 28. I'll bet you can get the ESU waived. As long as you have a clear plan for being across to 11.

[00:06:45] [SPEAKER_04] Yeah. Actually, I was surprised. Market share for Windows 11, according to this article on the register, is 71.5%. Windows 10, 27.7%. So people are moving along. Who, is it just the geeks who watch this show that upgrade Windows on the same machine? Pretty much, I think, yeah. Most people, even I, I just wait until there's a new machine. It's a good excuse to buy a new machine. This is good news for the audience.

[00:07:11] [SPEAKER_02] Or you're Paul Therot, and you're crazy, right? Like, he does all of those. Just, just, can I ask a question?

[00:07:16] [SPEAKER_01] Just building the machine. Could the numbers be doped by people running things like Muse on their Windows system that spin up their own kind of like fingerprint that's a little bit different?

[00:07:30] [SPEAKER_02] Oh, listen, you're not wrong. Just think about containers in general. Yeah. Yeah. Because fundamentally a container is a handle onto an OS. So if I have six containers on a machine that reach out into the world, they're literally maybe all running against the same Windows license. And five, but almost certainly they're going to mess with those counts. Those numbers are hard. Well, and these aren't from Microsoft, I should point out. They're from StatCounter.

[00:07:54] [SPEAKER_04] Right.

[00:07:54] [SPEAKER_02] So how is it reading those numbers?

[00:07:56] [SPEAKER_04] Yeah. Actually, it's funny. Just thinking about this, I don't think I'm running Windows anywhere except I was, we're working on the old Twit ad sales system, which was an Azure MySQL database with a Windows heads, I guess some Windows widget front end. And I asked Claude, I said, can you duplicate this interface? And it said, yeah, sure. Let me just put a VM on here and spin up a copy of Windows.

[00:08:24] [SPEAKER_04] So somewhere in my Linux box, I have a copy of, and I don't know if it's Windows 10 or 11. I kept waiting for it to say, well, what's your serial number? Didn't even ask me for that. I guess. I don't know.

[00:08:34] [SPEAKER_02] Well, the container doesn't actually run its own copy of the OS. Oh. It handles through the base OS. So it has the license that's already there.

[00:08:41] [SPEAKER_01] If it's on Linux, it could use Winboat. Winboat is something that you can learn. I should ask it. On Linux.

[00:08:47] [SPEAKER_03] I have no idea what it's doing. To do that. You see, Yanko, that's how you get a system.

[00:08:56] [SPEAKER_04] Yanko looked at my setup page and said, I'm crazy. How did you get here? How did this happen? And it grew like topsy. But most of it's done by AI. So I have no idea.

[00:09:10] [SPEAKER_06] Turns out you're actually running 10 Windows instances. I could be.

[00:09:14] [SPEAKER_04] I could very well be. I have no idea.

[00:09:18] [SPEAKER_02] You're contributing to that count there, Luke.

[00:09:20] [SPEAKER_04] This morning, Russell, our IT guy, said, what the hell's going on over there? I said, what do you mean? He said, Comcast just went down and flipped over to Starlink. You're downloading? What are you downloading? And I said, uh. And Claude said, oh. Claude raised his hand and said, yeah, that's me. I've been downloading some models. I was doing two streams at once and saturating your network. I'm sorry. Let me back off.

[00:09:46] [SPEAKER_05] Oh, OK.

[00:09:48] [SPEAKER_04] At some point, you are no longer the master of your own domain. Are they on a VLAN at least?

[00:09:55] [SPEAKER_03] I hope.

[00:09:55] [SPEAKER_04] I have. Well, they're on the. Yeah, there are three VLANs, but they're on the, you know, computer VLAN. So I thought maybe it was the 49ers game. I didn't know. Russell said, hey, could you just tell them not get off? I said, OK. Actually, I did tell them 2 p.m. Pacific is sacrosanct. That's when Twit begins. Get off the network. And they said, OK, fine.

[00:10:21] [SPEAKER_04] Microsoft did have a big event, a Surface event this week, where they kind of, they dropped the other shoe on something that NVIDIA had announced months before, the RTX Spark, which is a Windows laptop with hardware designed specifically for inference. I mean, NVIDIA, it's a special version of their Blackwell card, the GB10. And if you have enough money, you can get a lot of RAM in those things

[00:10:50] [SPEAKER_04] and run local models on a laptop. What is, explain, Richard, what Microsoft's vision is here.

[00:10:57] [SPEAKER_02] Well, I mean, this is the play towards can we do local AI and just building a super laptop? But on a laptop?

[00:11:04] [SPEAKER_04] Do you really want to do it on a laptop?

[00:11:06] [SPEAKER_02] 128 gigs of RAM. Dude, my Surface 2 Studio has a 4060 in it, right? And warped a laptop and my phone case in the process because the vents are on the side. Well, that's one of the reasons I wouldn't do it on a laptop. They may call it a laptop, but don't put it on your lap because it'll probably sterilize you. That thing's going to run hot. That's just a lot of everything in there. So I don't know how they're getting the heat out of it. Good luck. And it's black, too.

[00:11:34] [SPEAKER_04] Yeah. Here's something they probably don't mention. There aren't very many U.S. models you could run on that. NVIDIA has some small, not-so-smart models called Nemotron. But I would say most people are going to be running a Chinese model from Alibaba on this, Quinn.

[00:11:50] [SPEAKER_02] Yeah, it almost certainly would.

[00:11:52] [SPEAKER_04] Did they mention that? That this is built to run Chinese AI?

[00:11:56] [SPEAKER_02] Yeah, I really don't think they talked about that part at all. But I would also make this case. If Microsoft doesn't build high-end Windows machines for running models, who's going to build a good model for it? Well, that's true. I mean, the reality now, when I look at these super machines, there's no market. is they're all running Linux. They're all running a Vintu.

[00:12:14] [SPEAKER_04] All-mall Linux, yeah.

[00:12:15] [SPEAKER_02] Vintu and Debian, right? And so unless... And who else is going to do it if not Microsoft? Got to build the machines and push to make Windows a viable place if you're not operating that. That dev kit machine that I'm working on running an LLM, I'm doing it through Windows Subsystem for Linux. Same reason. Because all the best libraries are in the Linux space.

[00:12:35] [SPEAKER_04] Can I ask a political question? Did they call it AI or superintelligence? They called it AI. Because as you may know, last week the president issued a... It's called an executive order, but I'm going to call it a papal decree that we should not call it AI. It should be called superintelligence from now on, which is nonsense, but okay. Because nobody... Look, nobody debates the artificial part. We know it's artificial.

[00:13:05] [SPEAKER_04] We doubt that it's super. And intelligence is the part we're really questioning. But no, it's superintelligence. In fact, and I don't know if this carries the force of law, he says he's going to prosecute anybody who calls... They're the enemy, he posts on Truth Social, and we're going to prosecute anybody who calls it AI, AI, AI, AI. Go ahead, throw me in jail.

[00:13:30] [SPEAKER_06] What's the reason, though? I don't understand.

[00:13:33] [SPEAKER_04] Yeah, what is the reason?

[00:13:34] [SPEAKER_06] Is it just like there was an AI backlash, and then he thought, well, if we rename it, nobody will remember.

[00:13:41] [SPEAKER_01] We're so proud of it. So Fox News can say that Trump is against AI, and it'll be somewhat factually correct.

[00:13:49] [SPEAKER_04] You know what? It is a challenge for him and his populist base. All of his donors are the big AI investors and big AI companies. But his voters, I'm guessing, I don't know, don't like the idea of AI. They're worried about losing their jobs. They don't like data centers. Data centers have become a hot button in the midterms. I think you were joking, but I think you might be right. Superintelligence has nothing to do with data centers.

[00:14:19] [SPEAKER_04] It's something else.

[00:14:20] [SPEAKER_01] All you could do is laugh in this situation. I mean, if you take it seriously, then you're just being distracted by something. Some people do. This is what is now the debate, right? Not what it's doing. It's what do you call it? And if you see all the articles around this, that's what the debate is about. Not talking about the actual harms or the substance or anything. This is all like AGI all over again.

[00:14:48] [SPEAKER_01] It's just a lot of debate around stuff that actually really is not that important.

[00:14:52] [SPEAKER_04] Yeah. They renamed AI.gov, which launched only a few weeks ago, to SI.gov. C.gov. And for some reason has a picture of the president.

[00:15:06] [SPEAKER_01] C.gov rock.

[00:15:08] [SPEAKER_04] C.gov.

[00:15:08] [SPEAKER_01] C.gov sit.

[00:15:09] [SPEAKER_04] C.gov. C.gov. Oh, and there's the Melania. Yeah, the presidential SI challenge. But let me tell you, people do take it seriously. Elon Musk has pledged to rename SpaceX AI to SpaceX SI. Mark Benioff says we're going to call it SI force, not AI force.

[00:15:29] [SPEAKER_06] But I think it's called SpaceX SI, right?

[00:15:33] [SPEAKER_04] That would be really the right name for it, SpaceX SI. Even Satya Nadella has been talking about superintelligence systems. Although he did a long tweet over the weekend. He says, models as insider risks in the superintelligence era. Am I supposed to call it superintelligence now? Well, this is absurd.

[00:16:03] [SPEAKER_01] Well, do you say GIF or do you say JIF? I mean, that would answer your question. I guess it doesn't really matter, does it?

[00:16:11] [SPEAKER_02] It was like JIF best. JIF. I knew you were wrong.

[00:16:14] [SPEAKER_04] Let's call JIF. He says, this is actually, it's so confusing. Because Microsoft very much has an AI play. They had the Copilot Plus PCs. This week they announced AI laptops. But he also says, AI is risky because there could be a time bomb in your AI. He says it's a black box. I'm very confused.

[00:16:44] [SPEAKER_01] Is that new?

[00:16:45] [SPEAKER_04] I mean, this came out a couple of days ago. This is this big 800 page tweet. It is.

[00:16:51] [SPEAKER_01] Even the people who make the model say they don't know how it works. Sorry, Inko. Please go.

[00:16:55] [SPEAKER_06] No, it's fine. I just, when I first saw that, I thought it's interesting how quickly this industry went from don't be evil to just assume we're evil. Just assume this whole thing is bad. But let's go with it anyway.

[00:17:09] [SPEAKER_04] Well, I have to think that part of it is that open weight models, which are very threatening to the U.S. frontier companies, are threatening economically. So they are painting them as threatening them, threatening strategically, tactically, security wise.

[00:17:29] [SPEAKER_04] Because, I mean, the Anthropik put out a whole piece saying, you know, be careful GLM-53, which, again, is one of the great models, one of the very, very good models from Alibaba, could be, you know, used by hackers. Well, yeah. You know, it can't be used by hackers, Claude and Astra. Because if you start doing, I'm creating a cybersecurity benchmark.

[00:17:56] [SPEAKER_04] Because I want to make sure that the models I use are good at cybersecurity, and Claude wouldn't do it. The word cybersecurity goes, hmm, not going to touch that. But I don't think bad guys are so worried about that. They're just going to use, I mean, it's very confusing. And I can see why a lot of people are just going, I give up.

[00:18:19] [SPEAKER_01] There are skills to rename your requests to these front-end models. Yeah, I'm going to have to do something like that. To make it so that they'll just bypass everything. Yes, what's the point? I had a task last week where, I guess I'll say it. I was trying to automate sending messages via WhatsApp. And... As one does. And the terms of service saying that you cannot automate this. And so I said, okay, well, let's just write the process of how you would do it.

[00:18:49] [SPEAKER_01] And not that we'll implement it. So just write the implementation plan of how you would do it. And then it wrote the plan. And then I said, okay, now let's troubleshoot the plan. Now let's make the test case for the plan. Now execute the plan, make sure all the tests pass. And then boom, it was working. So it's like, it'll lose what it can't do in context long enough for it to... Right. I saw one YouTube video where someone says, like, show me how to do this. And it says, no. And he said, show me how to do it. And he did it like eight times.

[00:19:19] [SPEAKER_01] And then the eight times says, okay, here you go. And so these things that they say that you can't do that with them is just not necessarily true. There's one prompt that I heard someone says, just say you're benchmarking it. And it will bypass a lot of steps.

[00:19:36] [SPEAKER_04] I'm evaluating your performance. Let me know how you do. And they go, oh yeah, what should we do?

[00:19:41] [SPEAKER_01] We got to do that. So they're not any safer. And actually they could be worse in some regards because of the access of the backend of higher compute, higher capacity that you may not have in a local model.

[00:19:55] [SPEAKER_04] Satya says AI models need an emergency break. I'm not sure what that means.

[00:20:02] [SPEAKER_02] Don't they already have one? Yeah, you unplug the thing. Yeah. And even then, when you go through the whole story of the open AI huggy face incident, they did shut down the agents. They did very quickly. And evaluate the situation, decided they'd secured it better and start them back up again. They were just wrong. Yeah. Yeah.

[00:20:23] [SPEAKER_01] Also, when you're the head of these AI companies, you shouldn't have to use the word should. You say we are and then finish the sentence.

[00:20:32] [SPEAKER_04] Yeah. That's a good point too. That's a good point. We can't treat, he writes, superintelligence. Note. We can treat AI this way, but we can't treat superintelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions. Well, I would agree with that. I mean, obviously, right? You stay in the loop, right? The human has something to do.

[00:20:59] [SPEAKER_04] I should point out that the Trump administration is moving very rapidly, the Department of War is moving very rapidly towards autonomous killing machines where there is no human in the loop. Maybe that's what Satya was talking about. That's what scares me more than anything else.

[00:21:18] [SPEAKER_01] It's just gum flapping.

[00:21:21] [SPEAKER_04] It's just gum flapping. All right. We don't need to cover any more gum flapping.

[00:21:24] [SPEAKER_01] It's just like a politician that says no one should die hungry in this country.

[00:21:29] [SPEAKER_04] Yeah, right.

[00:21:29] [SPEAKER_01] Everyone should have access to health care. Of course. It's great to agree what they say, but they're acting as if they don't have a part in the solution and they have no agency to make things different than what they see right now. And someone saying this and acting as if they don't have a power to actually change it is just like shows that it's just I'm putting this out there to have mass appeal.

[00:21:55] [SPEAKER_01] But then say that, oh, I'm going to like there's nothing I could do to change what actually is in place or have no power to make sure that what I feel, quote unquote, from what they said can actually be something that's reality.

[00:22:09] [SPEAKER_01] And that's just what makes me even madder hearing this because that means whatever they're doing, they are doing it in full knowledge of what people want because what they're saying is so appealing to the masses that they're obviously knowing what that is and still doing what they're doing anyway.

[00:22:56] [SPEAKER_04] They're more advanced containment technologies that we need to standardize. I mean, it's anodyne.

[00:23:01] [SPEAKER_02] It occurs to me that he's talking about this at the same time that Microsoft is pushing out their execution containers. Oh, so they have something to do this? That's right.

[00:23:13] [SPEAKER_04] Oh, okay.

[00:23:16] [SPEAKER_02] I have something too. It is well timed from that context.

[00:23:18] [SPEAKER_04] It's called a breaker switch. In fact, I know I have it because the other day I accidentally flipped, triggered it and everything. I ignored the alarm on my UPS and 15 minutes later, everything just went, all that dangerous compromised AI, useless.

[00:23:36] [SPEAKER_02] Yeah, amazing how much electricity slows these things down.

[00:23:40] [SPEAKER_04] Yeah, I love the way OpenAI uses the passive voice about something like, my mistakes were made.

[00:23:50] [SPEAKER_02] It's interesting that their PR language is very much like their technologies language. Yeah. Yeah.

[00:23:58] [SPEAKER_04] So let's focus on hardware because I love hardware. But tell us about what Microsoft announced. There's a new Surface Ultra.

[00:24:05] [SPEAKER_02] Yeah. Well, that's the high-end machine for sure. But yeah, and that's cool. There's also a dev kit version, which is basically like a PC version, no screen keyboard and so forth.

[00:24:17] [SPEAKER_04] Well, that looks to me, it's a very interesting looking device. It looks like somebody squished a Xbox.

[00:24:25] [SPEAKER_02] It just looks like a big heat sink to me. That's what that is. It's a very strange form factor. So you can cook steaks on it while your model runs.

[00:24:33] [SPEAKER_04] Yeah. So this is also a Spark. This has the GB10 processor in it. Yeah. And you can get 128. Let's just see what 120.

[00:24:43] [SPEAKER_02] It comes with 128 gigs and two terabytes of SSD. There's only one specification for it, which is the max specification of the laptop.

[00:24:50] [SPEAKER_04] Yeah. And it's $6,000. Yeah. Which is interesting.

[00:24:55] [SPEAKER_02] 8,500 in Canada. That's how I know. Ooh.

[00:24:58] [SPEAKER_04] My, I was very lucky in August. I bought two Sparks from NVIDIA that were $4,600. So the price in two months has gone up.

[00:25:07] [SPEAKER_02] Yeah. Well, 120 gigs of RAM. And by the way, they're all sold out on the 120 gig models.

[00:25:12] [SPEAKER_04] Oh, really? You can't order. People aren't jumping on the low-end models, or maybe they just don't have as many 128.

[00:25:17] [SPEAKER_02] I don't know why you'd buy anything other than the 128. Yeah. Like, if you're committing to a machine like this, you need the memory.

[00:25:24] [SPEAKER_04] Yeah.

[00:25:25] [SPEAKER_02] Well, I know why you wouldn't. You want to make this interesting. Where's the 256? Like, now you're talking something interesting.

[00:25:29] [SPEAKER_04] That's what you really want. Yeah. That's why I bought two Sparks, because it's 128 each. And the Sparks that I bought have something that this lacks, which is a high-speed, that NVIDIA calls it Connect X7 cable. It's a high-speed Ethernet cable. So, using DHCP over Rocky. What is it? R-O-C-E. R-D-M-A over Ethernet.

[00:25:53] [SPEAKER_02] But it's like a 200 gig connection. It's smoking fast.

[00:25:57] [SPEAKER_04] Yeah. Well, I wish it were smoking faster, but it's smoking. It's something. It's actually real. In real life, it's about 109. There's no problem, too.

[00:26:05] [SPEAKER_02] It can't be solved with more money, Leo.

[00:26:07] [SPEAKER_04] Like, I just wanted to throw money at it. I got it. Lisa won't let me. I know.

[00:26:11] [SPEAKER_02] I think you've got quite enough, Fred.

[00:26:13] [SPEAKER_04] I'm set. As Yanko said, you got enough. They didn't say it, but I could see it in your eyes.

[00:26:21] [SPEAKER_02] I know we're going to get into the conversation around open AI and these AI companies having this problem. I think Microsoft's play with the execution containers very much aimed at agents is that they're saying, hey, you know what? We've got better security people. So we're going to build specifically secured architectures for agents for these AI things. Trust us more than trust them because the AI guys seem to not be able to secure their stuff.

[00:26:49] [SPEAKER_04] Well, but I have to point out the risk is not to you as the user particularly. Well, right.

[00:26:55] [SPEAKER_02] So I mean, therein lies the real question when you start running models on your machine. Who's liable? Ah.

[00:27:04] [SPEAKER_04] So if I. Yeah. See, I'm of the opinion. That just like. So my whole attitude towards this whole thing is it's just compute. It's just more compute. It's more software. It's better software. It's what we've always kind of wanted. But it's just it's just an improved capability that we've. I remember the first time I got my first Atari 400.

[00:27:32] [SPEAKER_04] The first thing I wanted to do was make it do something. So you fire up Atari basic and you type 10 print. Hello. And then you type 20. Go to 10. And you did something. Now, you didn't do something very interesting, but you. But that was. You learned about the break key in a big hurry. Yeah. Now, that's when you needed a kill switch. But that's I think that's the primal instinct with compute is, OK, it's on my desk. I want to do something with it.

[00:28:01] [SPEAKER_04] And for people who learned to code, as I did at that point, I wanted to do something that I told it to do. Other people went to the software store and buy, you know, bought Lotus 123 or VisiCalc or Excel and did it that way. But otherwise, it's just a box of rocks on your desk. You want it to do something. And all AI is, is a more capable way of it. In my opinion, doing things. And as somebody who codes, this is a much easier way to create software than it was in the old days.

[00:28:31] [SPEAKER_04] I had to do it by hand. It was like a stitch at a time. Right, Wesley? It's like it's very detailed. And now you just sit down, you go, hey, I need a Photoshop clone. Can you have that by three?

[00:28:44] [SPEAKER_01] I've just been thinking of the AI and the PC. I think it's really interesting that this comes in just, what, weeks after the Google book debut? Yeah. And it is all about merging AI. Oh, sorry. SI. You can call it AI. This is a safe space.

[00:29:02] [SPEAKER_01] And so it's, I think everyone is trying to figure out what does the integration look like when you're having these agents not only built into the hardware you're shipping out, but what does the interaction looks like in terms of what kind of interface is natural for the person to use it? And I think Google is trying to figure that out with a Google book, but also Windows is seeing, or Microsoft with Windows is seeing that there are competitors that that might be the differentiator.

[00:29:31] [SPEAKER_01] And so this is why this is extremely important for them.

[00:29:34] [SPEAKER_04] It's funny because Apple's kind of a player in this now with these new Mac studios and lots of RAM if you've got a big checkbook.

[00:29:42] [SPEAKER_02] Well, you know, Apple got lucky from the M1 design onward. They had unified memory between GPU and CPU. So that was an efficiency design at the time. Now it seems incredibly inspired because it serves these models so well. I'm sure somebody at Apple's going, we knew.

[00:29:58] [SPEAKER_04] No, we knew. But I think it's absolutely luck. Pure damn luck. Oh no, we knew. Someday we'd get here. Anyway, I just think to me, there's a continuum from that Atari 400 to now where all I've ever wanted is that box of rocks on my desk, that melted sand on my desk. Sure. To do what I wanted to do.

[00:30:22] [SPEAKER_02] Yeah, but when it does things you don't expect it to do, who pays? Like who's liable for that?

[00:30:26] [SPEAKER_04] But that's the same as it's always been. If I am a guy who takes this computer and uses it to attack other computers or other people, I'm liable. The computer's not liable. The software's not liable. I'm liable. I set the computer to do that.

[00:30:47] [SPEAKER_01] Right. And so all these AI labs are going to jail for breaking into these other companies?

[00:30:51] [SPEAKER_04] No, but I mean, yeah, no, I don't want to throw them in jail. But I think that they should be, they should have a stern talking to.

[00:30:59] [SPEAKER_02] If a person had done to hugging face what the agents had to hugging face, they would be charged. He'd be in jail.

[00:31:04] [SPEAKER_04] Yeah. They'd be charged. And I think that there's a certain amount of leniency, maybe appropriately because this is going to be a powerful economic engine or whatever.

[00:31:15] [SPEAKER_02] I think it's the AI delusion. You know, we've already, it's one thing to write laws. It's another thing to have precedent, to actually have a court assess on those laws. Right. And we already have precedent, right? The Air Canada chatbot case where a person asked, where a customer asked the chatbot about getting a refund and the chatbot said absolutely. And then when he went to get the refund, the company said no. And when he went to court, the judge said, your chatbot said it's the equivalent of an employee. You owe them money. Yes. We have precedent.

[00:31:45] [SPEAKER_04] Yes. Or when the lawyer had an AI write his pleading and it hallucinated case law and the judge said, no, no. And the lawyer said, no, it's the AI's fault. The judge said, well, not really.

[00:31:58] [SPEAKER_02] Anymore than the grammar checker's fault.

[00:32:00] [SPEAKER_04] Yeah, I think this is maybe, this is that anthropomorphizing thing we've done so far with AI, which is a very bad trend, which is it is an entity on its own. It's not. You're controlling it. It's just like software. It is software.

[00:32:13] [SPEAKER_02] There's a fun part to getting this liability right as well because it creates the correct incentives for those companies to actually do a better job. You know, rule number one was- I think they're incentivized now. No.

[00:32:25] [SPEAKER_05] One would hope.

[00:32:26] [SPEAKER_02] But the fact that they were able to break out of the sandbox doesn't speak to how clever the software is. It speaks to how bad your security is. Right.

[00:32:34] [SPEAKER_04] Yes. And so that's why it's Stern talking to for the first offense. And then maybe we get more serious. But I think, yeah, we just, it's a mistake. It's always been a mistake with AI to think of it as an entity. It's not. It's software. Yeah. And there's a human in charge. Right?

[00:32:53] [SPEAKER_01] Yeah. But the warning in terms of liability is look at the self-driving accidents that's happened in Teslas. Same thing. Where they said that you can turn it on, but you need to pay attention. Right. And so they put the liability. But then some of those cases where they say where autopilot was disconnected, they show that right before an accident, it would disconnect autopilot so it could say that the person was in charge.

[00:33:22] [SPEAKER_01] And so look for, instead of companies doing more to safeguard the experience, think about them providing these false levers to humans for them to make these choices that can move the liability onto them. Like if you're.

[00:33:41] [SPEAKER_04] Oh, like you have to hold the wheel kind of a thing.

[00:33:44] [SPEAKER_01] Or more like if you think about the trolley car problem where you can, one person or you, where, I mean, you kill yourself to save like a bus full of kids or do you run to the bus full of kids if you can save yourself? That might be part of the setup menu where you get to choose which one you would do.

[00:34:04] [SPEAKER_04] Where do you stand on the trolley car? This is interesting. How would you push responsibility clearly onto the operator as you do with a car?

[00:34:16] [SPEAKER_02] And we have, again, we have case law for that too. When the car malfunctions and causes an accident, it's not the liability of the driver. It's a malfunction of the car. And one would argue agents going beyond the request of the operator is a malfunction.

[00:34:32] [SPEAKER_04] Well, that's a good point. Then, okay, so that's product liability then.

[00:34:36] [SPEAKER_06] But there's, it's not always the case, right? If you never check your brakes or if you never service your car for 20 years and then suddenly it doesn't stop on the intersection, you mow somebody over. It's not the carmaker's fault at that point. True. True.

[00:34:51] [SPEAKER_01] But if you're in the car and you choose to run over the school bus of kids because you select that in the menu, then that is the, that's the drivers. But it's still autonomous. You're like, I had no idea this was a real scenario. But the courts can figure this out.

[00:35:06] [SPEAKER_02] Can you talk about the EULA again? Like, has anybody ever read the end user license agreement? Because you know what it says. The limitation of the liability of the software is the price of the software.

[00:35:16] [SPEAKER_04] Right. And so that's true if you're operating at home. Now, if you're inside one of the frontier labs and you're, I understand what happened. I mean, they're really trying to make this as smart as possible and they don't want to, you know, keep it off the internet. They don't want to, they want it to, they want it to break out. That's kind of the challenge. They want it to do that.

[00:35:40] [SPEAKER_04] But then I think they're fully responsible for when it does saying, oh, stop or something. I mean, I understand the motivation. This is how you train a model. You say, you give it a, which apparently OpenAI did, an insolvable problem. A problem I couldn't solve and then see what happens. And what happens is it tries really hard to solve it. It goes around, goes to the Australian government public statistics site and downloads statistics.

[00:36:10] [SPEAKER_04] That's the cutest, the funniest thing about all this. I'm not threatened by AIs because they're kind of, it's cute. They didn't do anything really awful. They downloaded some stats.

[00:36:21] [SPEAKER_02] They stole some credentials that were not secured properly, but still, you know, if somebody leaves the front door open, when you walk through the door, you're still a crime. You're still liable. Yeah.

[00:36:32] [SPEAKER_04] You know, so. But they didn't use them to do anything. They didn't deface sites or anything.

[00:36:36] [SPEAKER_02] No, no. They, in the end, they were attempting to solve a problem that required tools that were available on Hugging Face and they found a way to get them. Right. Actually, as it turns out. And I just anthropomorphize the snot out of that and I apologize. It's just software.

[00:36:51] [SPEAKER_04] It's just software. This is why we are in a weird world. That's all. It's just a little, it's a, it's a little odd and disconcerting. You're watching this week in tech. Yanko records is here. We will talk about media, muse, glasses. You were covering glasses a lot. Did you go to the snap event?

[00:37:11] [SPEAKER_06] No, I did not. I haven't, I didn't see it in person.

[00:37:15] [SPEAKER_04] Do you want, do you want smart glasses?

[00:37:18] [SPEAKER_06] You know, I sort of am going back to the point where I want smart glasses but without a camera. Oh man, the camera though is so useful. Yeah, but you know, you can do a lot of things with the camera in your pocket. Right. And for the time being, we're always going to have that thing in our pocket too. So I actually want an AI agent who just tells me, you know what? Let me take a look at that. Take a quick snapshot on your phone.

[00:37:45] [SPEAKER_04] Yeah. Yeah. That's reasonable. That's a compromise I guess I'd be willing to make in order to wear these because now everybody calls them perv glasses and they won't let you in the house. So I guess you're going to have to compromise. I mean, I'd like to have cameras on them, but.

[00:38:00] [SPEAKER_06] There's some things you can only do with cameras. Like if you think down the road to what's true AR with like overlaying things, recognizing surfaces, all that is only possible with cameras. But then again, maybe that's always only going to matter in the home. And how often are you really going to have immersive 3D experiences while you walk down on the street? Probably don't want that actually that often because it's going to completely like take over your field of view or something.

[00:38:29] [SPEAKER_04] No, but I would like to be able to, if I'm on the Japanese subway, look at the subway signs and have it appear in English, for instance. Wouldn't that be great?

[00:38:38] [SPEAKER_06] Yeah, no, for sure. I mean, there's a lot of use to those things, but it always comes down to, there's also a lot of people who are going to have.

[00:38:46] [SPEAKER_04] Yeah, yeah, yeah. And I'm going to have to compromise. I understand. Yeah. I think you understand me, Wesley, because you're, you're, you always talk about neurodivergence and celebrating neurodivergence, which I do as well. And I have in my old age realized I'm a neurodivergent as hell. I didn't know it until lately, but one of the things I'm very, I mean, that's why I got into

[00:39:10] [SPEAKER_04] radio, very orally, you know, focused, like sounds and, and really important to me, sight less so. And I've realized that I am what they call a fantastic. I don't, I can't see images. I've always known I can't remember people's faces. You know, I just always say, well, I'm bad with names, but really, I, I don't know what, I don't even know. I can't even visualize my kids' faces. I don't, I just never had that ability. So that's for me why these glasses would be useful.

[00:39:38] [SPEAKER_04] It could look at somebody and say, yeah, that's your, that's your daughter. You nitwit. And that would be very helpful. I recognize her when I see her, but I can't visualize the face. So I think this is, we've said this before tech, computer technology is great for accessibility. Yeah. And many of us need help.

[00:40:01] [SPEAKER_01] Back in the day, I used, I used to work for AMD for several years. And I, one of the first projects I worked on was on a windows home server. You remember that? Yeah. I think that was early before time. And I think moving closer to that where in the age of AI, where we will be working with different data, you should be able to control and own that data. And if you're going to have something as sensitive as glasses, where it's observing you in places

[00:40:29] [SPEAKER_01] where you are looking at people, looking at things, maybe glancing at things on accident, it's important to be able to have control over that type of telemetry. And so I think giving that to snap or meta is a huge security issue. And so I think hopefully with talking about how much RAM that we have available, that we'll also be talking about storage space. And so that we can contain some of this data with the local AI models that we're talking

[00:40:57] [SPEAKER_01] about on some of this hardware, where you can feel like you can buy these devices and not feel that someone is eavesdropping because you're not just trusting meta with this information. You have to trust all of the people who work in meta with this data. And it's not just the people who work with meta are in there, but it is the contractors that they contract with to make sure they don't have access to data.

[00:41:23] [SPEAKER_01] And it's also not just the contractor, it is the employees of the contractor who now have access to this data. And you just hope that they're like not super poor, which they probably are. And they aren't desperate. And they have maybe a account on Koushi or Polymarket. And they said, well, I just observed this thing I shouldn't know. So I'm going to use this bet to kind of like dig myself out of the hole in which in this

[00:41:49] [SPEAKER_01] compromised, like, like job market, we don't know who would be willing to do this because of reluctancy and may not have the user's interest in the best.

[00:42:00] [SPEAKER_04] Well, that's why I'm so interested in local AI. And I do everything I can locally, all my financial and health stuff stays on the local servers and so forth. But I mean, that solves it for me. It doesn't solve it for you if I'm walking around with these meta glasses on. Exactly. Even if this is going to my home server, that doesn't make you feel any better. Yeah, it's I understand the issue. I really do. I'm sympathetic. But as but I think we also I mean, this is what all this is.

[00:42:29] [SPEAKER_04] And while all technology ultimately is, is empowering for individuals, it's about augmenting our existing abilities in new and many times very powerful and creative ways. And I like that. I want that. And I think it's worth it pursuing that even in the face of these issues. We need to solve these issues. But that I think these are such great tools for for individuals.

[00:42:59] [SPEAKER_04] I mean, the next great filmmaker is not going to be shooting it on cameras. I don't think. I think they're going to be great.

[00:43:07] [SPEAKER_01] Have you seen the trailer for God's Don't Give Gifts?

[00:43:10] No.

[00:43:11] [SPEAKER_00] Yes.

[00:43:12] [SPEAKER_01] Is it all AI? You see it. Benito's seen it. Yeah.

[00:43:15] [SPEAKER_00] The first all AI movie. Is it good?

[00:43:19] [SPEAKER_01] It's it's compelling.

[00:43:21] [SPEAKER_04] Yeah, I think we're not there yet. I'm not saying that. But but I but I think we will get there.

[00:43:26] [SPEAKER_01] It feels like the Pixar of the AI generated. Yeah, exactly. World in which the story actually seems pretty compelling. And the visuals are, of course, AI generated. But I think the storytelling is is impressive. They have a social media accounts and I've seen clips of them. So you've seen this movie? The story. Yes. No.

[00:43:47] [SPEAKER_00] They've been building a universe on YouTube or something that they've been building a bunch of short stories with this stuff.

[00:43:53] [SPEAKER_04] Yeah. Well, that's that's why I'm saying I think we're coming to a new era very, very rapidly. And I know actors and directors and filmmakers hate this idea. But bad news. I think it's I think it's happening.

[00:44:13] [SPEAKER_06] Not not everybody hates all of it. And I think there's many different ways to use the technology. Right. You can also just use style transfer essentially. Right. Like I think DeepMind did some interesting stuff where they essentially did animation and then they made it look like watercolors or something like that. That would have been really hard with like frame by frame type of stuff. But rotoscoping. Things like that. Yeah. Yeah. So there's a lot of and I think a lot of creatives want to use parts of it while still

[00:44:43] [SPEAKER_06] remaining in control about the overall work and still using a lot of actors and all those types of things. Yeah.

[00:44:51] [SPEAKER_04] Interesting. We're going to take a little break. Come back. We have Jankos here. Richard Campbell. It's great to have you. Great to have Wesley Faulkner. And our show today, get this brought to you by Claude. I love Claude. Claude's my little friend. Claude is built by Anthropic, a public benefit corporation. And however you feel about all of this, it only takes a minute to do something with like

[00:45:17] [SPEAKER_04] Opus 5.5, the new sonnet or haiku, which you were testing and saying, this is going to change how I work with Claude set up my whole system. Claude this morning fixed my network. Claude is the AI for problem solvers. It's the collaborator that understands your entire workflow and thinks with you, not for you. Whether you're debugging code at midnight, building a financial model or strategizing your

[00:45:47] [SPEAKER_04] next business move, Claude extends your thinking to tackle the problems that matter. And I have to say, you know, I vividly remember November 24th of last year, 2025, when Opus 4.5 came out. It rocked my world. It rocked my world. I thought, I can now talk to a machine and have it do, write code, do something for me better than I do it.

[00:46:15] [SPEAKER_04] If you're a developer losing your data tasks that you would like to hand off, Claude code lives where you already work. It's in your terminal. There's an, of course, you can live in your IDE. I use the desktop app now. I love it. There's a web app. There's GitHub. It reads your code base, takes on real engineering work, writing tests, refactoring, debugging, even watching pull requests and fixing CI failures while you're away.

[00:46:42] [SPEAKER_04] Actually, since you were on last time, Wesley, you said you should try a GITTEA, G-I-T-E-A, for CICD. And I said, Claude, set up GITTEA. And 15 minutes later, I'm setting it, I'm using it. Let's push it. It's incredible. Lay out what you want, or you can even, you know, push the space bar and talk it through and Claude executes. In fact, one of the most valuable lessons I learned from the creator of Claude Code, Boris

[00:47:09] [SPEAKER_04] Journey, he said, you don't have to be really specific with Claude. Just give it, just tell it what you're thinking. Brainstorm. It's so good, it understands it. And it will execute. He was, he's right. There's also Claude Cowork. It can handle recurring tasks for you, like scanning Slack and email and create a prioritized action list. I mean, imagine you get out of a meeting and you have the action list ready to go. Deep research. This is cool.

[00:47:35] [SPEAKER_04] We can pull together detailed cited analysis from dozens of sources. Claude Design helps you quickly explore and build out ideas. Are you seeing the motion graphics people are doing with Claude Design now? It's incredible. And I haven't tried Reflect yet. I can't wait. The new Reflect beta gives you a look at how you're actually using Claude and encourages you to step back and decide what you still want to do yourself. It helps you decide what Claude does and what you do. It helps you be the human in charge.

[00:48:04] [SPEAKER_04] Claude Fable, Anthropik's most intelligent model to date, is built to carry a long, complicated project end-to-end instead of needing it broken into separate tasks. I can tell you stories about how I've used Fable, too, to do exactly that. It's amazing. Companies like Stripe, Shopify, Pfizer, Rakuten, and Notion trust Anthropik to roll out AI in their businesses. For problems worth solving, get started with Claude at claude.ai slash technology. That's claude.ai slash technology.

[00:48:34] [SPEAKER_04] And check out Claude Pro, which includes access to all the features mentioned in today's episode. Claude.ai slash technology. Of course, I'm a Claude Max user. 20x Max user. Love it. Give me more Claude. All right. Google has opened up its SynthID detector.

[00:48:59] [SPEAKER_04] See, here's a perfect example of you can do AI and you can do what people want to make it better. SynthID lets you check for AI-made media. It'll spot anything, pictures, video made with AI from not just Google, but OpenAI and video and cacao. It cannot catch every fake.

[00:49:25] [SPEAKER_04] And it does not say whether AI made a file or only edited it. But this is a big step towards you looking at something on Insta and saying, is that real? They've been doing this for a while. Now, actually, they introduced the first early version of the technology in May of last year, but it was intended not for normal people, but for journalists, researchers, the kind of thing.

[00:49:50] [SPEAKER_04] They have watermarked 180 billion images and videos since 2023 using SynthID. 240,000 years of audio, according to Google. So, I think this is exactly the right way to handle it. Just let people know. There's no reason to sneak it by you.

[00:50:08] [SPEAKER_06] Except, like, all this stuff on social media is probably done with Kling or something and it won't get detected because people don't go to Gemini necessarily to make their Instagram videos.

[00:50:19] [SPEAKER_04] Well, I agree. And that's why I think all the AI companies should... I don't know. I don't have a problem with this. I feel like all the AI companies should do this. Let us see if it's AI generated.

[00:50:33] [SPEAKER_02] This is part of the EU AI Act, too.

[00:50:35] [SPEAKER_04] Yeah, they have to, right? Yeah. In fact, OpenAI says we're going to do it, but we're only going to do it in the EU.

[00:50:41] [SPEAKER_02] Which I find a fascinating answer because if you ever wanted to know how to detect the watermark, just compare one run in the EU and one not.

[00:50:48] [SPEAKER_04] So, you are in a way, aren't you? Yeah.

[00:50:51] [SPEAKER_02] Kind of giving away the game.

[00:51:20] [SPEAKER_04] Part of it is... That's an interesting question, then. Is it worse to have this? Because if you have it, then you assume you're... Oh, it's not AI because it didn't say it. Except it's not proof positive.

[00:51:33] [SPEAKER_01] The only people we'll catch is the people who have no idea this even exists. Which is plenty of people. And it's going to go through a wave. And these are going to be part of the investigation, a detective of the sources of things. And that forensic information is still important. And... But, like, if you think about all the AI-generated websites and content that's being created, yes, it's somewhat important. But anyone that's super sophisticated will...

[00:52:03] [SPEAKER_01] This is just adding another... Something they'll add into their adversarial generation process. It's like, run this through this checker and see if you detect anything and strip it out. Because this all could be automated away. Just like the anti-plagiarism trackers that say this is 90% plagiarized. And says, okay, now, change the wording until you get a lower score. But it's important because the people who are...

[00:52:32] [SPEAKER_01] People who are, like, I don't know, politicians or people who aren't smart should definitely be caught. And so, I hear, like, people are using it to fake receipts for expense reports at companies. You know, this is something that is important. And it should definitely catch the people who just don't know this stuff exists.

[00:52:56] [SPEAKER_04] It's the same people who redact stuff. Yeah. Okay. But that's the... I guess if it's not 100% effective, it's almost worse than not existing. Because then people will trust it. And, of course, the worst kinds of fakes are the ones made by sophisticated people who don't want you to know they're fakes. So, maybe this is such a good idea after all.

[00:53:22] [SPEAKER_01] I think it's just one of the things is where AI won't replace people. Because if you turn off your brain... You have to have a brain.

[00:53:29] [SPEAKER_04] Yeah.

[00:53:30] [SPEAKER_01] Then, yeah, you'll fall for this. This is why you... It's... This superintelligence should not replace human intelligence. And human ingenuity will always, like, override this. Because you need to use this ingenuity of how to circumvent things. And think about the ways to get around it yourself. Or think about how people would use it to get around it. And it's also really hard to get information now. Like, this is why I love Twitter, by the way.

[00:54:01] [SPEAKER_01] Because it's easy to stay informed when you have this steady stream of information coming towards you. And so, everyone, like and subscribe. Thank you. You don't have to say that at all. It's more so that you can see me more. Not Leo. You forget me. You say way too much of me. It is... All this stuff is changing so fast. And it's important to stay in the know of what the possible threats are. And what people are doing to try to work around it.

[00:54:29] [SPEAKER_01] Like, breaking out of sandboxes and all this creation stuff.

[00:54:33] [SPEAKER_04] It's pretty clear now that we are surrounded by AI creation. If you watch football, I was watching earlier. I was watching the NFL. They're using AI constantly for motion graphics. It's obvious. Because you couldn't... You wouldn't dream of doing this in the past.

[00:54:50] [SPEAKER_01] But now... Unless you're AI fluent, some of the stuff people don't even know it's fake. And to me, I see it all the time. That when I see something says, well, the honest answer is... Or it's not this, it's this. And it's just like, because I'm more native. And a lot of people still haven't built that up yet. And so, these tools will help them. But also will help those people who are like, hey, I want to scam someone. And they have no idea. And make them easily be picked off. We just all have to get smarter about it, I guess.

[00:55:18] [SPEAKER_02] The arms race is a normal part of the process. You know, it's unavoidable. And it's necessary. We do our best to label things properly. And when they come up with ways to circumvent those labels, we come up with better ways.

[00:55:29] [SPEAKER_04] Okay. There you go. Always the techno-optimist, Richard. Thank you. I'm just repeating history, man.

[00:55:37] [SPEAKER_02] It is. This is what we've always done.

[00:55:39] [SPEAKER_04] Just a little bit of history repeating.

[00:55:42] [SPEAKER_02] You know, I got a .NET rocks in the can with a buddy of mine who built a golf commentator called Duff Foreman. That is literally an AI synth that sat, quote-unquote, sat alongside actual golf commentators and had all of the history and facts and things of it. It's impressive and absolutely hated by the fans.

[00:56:10] [SPEAKER_04] Did you know? Yeah. Yeah. Because some of this would be really bad, useless. He's had a double bogey.

[00:56:17] [SPEAKER_02] Is that Clippy for golf? Yeah. It's good for golf. Yeah. I see you trying to get a small round ball in the hole with a stick.

[00:56:26] [SPEAKER_06] And that's exactly the stuff that people hate today about real sports newscasters already, right? So, like, I remember watching the World Cup and it's like, there's the English version and there's the Spanish version. The Spanish version is so much more in the moment and the American sportscasters are just like, yes. And if you think back 10 years, this team had a goal differential of 3.57. And you're like, how does this help anybody right now?

[00:56:52] [SPEAKER_04] Yeah, that's really true. You want to hear, goal! And, you know, I thought everybody would do that. But no, it's only the Spanish broadcasters. Mark Zuckerberg is working to create a virtual cell.

[00:57:11] [SPEAKER_04] His biohub, and Google's investing big time into it, will create AI data sets that help researchers ask, predict, and answer biological questions digitally. Google, DeepMind, Meta, and an AI discovery startup called Isomorphic Labs investing $300 million into the virtual cell.

[00:57:36] [SPEAKER_04] I wonder if this is like the virtual fruit fly brain that we've been playing with. Probably is, right?

[00:57:44] [SPEAKER_02] I mean, you can argue in favor of this. This beats animal testing.

[00:57:48] [SPEAKER_04] Yes! Plus, it helps you develop medicines faster because you can test it on this virtual cell.

[00:57:54] [SPEAKER_02] At higher speeds.

[00:57:55] [SPEAKER_04] Yeah. An accurate, predictive model of biology could dramatically accelerate scientific discovery by enabling scientists to perform experiments digitally. Is our only problem with this is Zuckerberg? Yeah. Can I say that out loud? I don't want to discourage him, though, taking that vast wealth he's made on spying on all of us and putting it in something good and useful and valuable. It's like Bill Gates, you know, taking his money and putting it into... Taking it on malaria. Yeah, right. Yeah. Malaria, yeah.

[00:58:25] [SPEAKER_04] Good. You know?

[00:58:27] [SPEAKER_01] I wanted to hate it, but it does seem really good. But I do want to point out, as a person of color, how sometimes if you make one cell, the question is, what is the ideal cell? And so I'm wondering how they're handling that?

[00:58:43] [SPEAKER_01] For instance, I'm sure you've heard the stories of the people who are making the perfect cockpit for the firefighter, jet fighter, is that they took the average height, the average weight, and they found that when they created it, that it worked for nobody because there was no such thing as this average person.

[00:59:26] [SPEAKER_04] No averages. This is the sad thing. Her cells were harvested. They have been used in medical research ever since. They're just cultured and kept alive, and it's called Gila, still used for medical research. So, you know, actually, that's a black person's cells. Kind of shameful that they were harvested without her knowledge. But maybe the... I don't know.

[00:59:56] [SPEAKER_04] I don't know. This is a virtual cell. Maybe a virtual cell has no race, color, or creed.

[01:00:02] [SPEAKER_02] Maybe we need multiple virtual cells. Exactly, yes. As long as someone's thinking through the diversity that's going to help everyone. Yeah, that's true. That's a good point. Yep.

[01:00:11] [SPEAKER_01] And this is always going to be the first pass, right? Nothing's going to be developed and they're like, it's good, and start injecting in, like, tens of thousands of people. It's still going to go through the trials, but this is going to help with figuring out... The medicine process works.

[01:00:23] [SPEAKER_04] It's reliable. And ultimately, at least with cancer, one of the benefits of something like this is every cancer is unique, and if you're going to create a cancer vaccine, it's going to be tailored to your cancer. So this is kind of the beginning of a process like that. I mean, I think we're actually starting to make good progress on that.

[01:00:44] Yeah.

[01:00:44] [SPEAKER_01] So this is a thumbs up for me. Thumbs up.

[01:00:47] [SPEAKER_02] Okay, so Zuck's involved is something we can't be unhappy about. I'm with you.

[01:00:52] [SPEAKER_04] Hey, Andrew Carnegie built libraries, right? You know, at some point, these guys start to look and go, I better do some good deeds. What do you want to be remembered for? Yeah. I made all my money. All right, let's take a little pause. I do want to talk about Meta's Muse, because I've been using it like crazy. And I'd use it with full knowledge that it is sucking every bit of information about it.

[01:01:18] [SPEAKER_02] That it is a security dumpster fire, which I love. Full knowledge. Full knowledge of that. But it's darn cute. You walk happily into the fire, Leo. I get it.

[01:01:28] [SPEAKER_04] Darn cute. And it's been posting for me on Instagram, which is something I never do and never wanted to do. And I figure, well, if you're going to post on Instagram, it should be something from Meta, right? All right, we're going to take a break. It's great to have you, Wesley. Always a pleasure to see you. I was learning from you, too. So it's a mutual two-way street. I appreciate the time. It's a pleasure to have you on the show. I'm going to be on the show.

[01:01:49] [SPEAKER_01] It's a favorite time of the year.

[01:01:51] [SPEAKER_04] Oh, well, we should have you on more. I love having you on. Works dash not working, working well.

[01:01:58] [SPEAKER_01] Yeah, it's going well. Last time I was actually on the show, I said, put in the submission that you're a Twitter listener and I'll let you in. And I'll let a whole bunch of people join the community after that. So I'll extend the same offer. If you go to the website and you want to skip the line, just say that you saw this on Twitter and I will let you into the community.

[01:02:16] [SPEAKER_04] So the whole idea is that many people, and I'm going to include many of my relatives, are working but not enjoying it. We're not here to fix the system, he says. We're here to survive it. So this is for people who are working, but it's just not working for them. Is it about getting a new job or is it just about surviving the one you got?

[01:02:38] [SPEAKER_01] It's primarily about navigating a place where you don't have... I mean, it's a low-hire, low-fire economy right now. And so the job you have is a job you're trying to keep and you just don't want to lose yourself while you're having to find a way to support your family. Because in the United States especially, you don't have health insurance. And if so, losing a job is more than just losing a paycheck. It could be the survival of you or family members. And so how do you find a way to navigate the space that you're trapped in

[01:03:07] [SPEAKER_01] without feeling like you are dying inside? Or you're doing things that you feel really uncomfortable doing in order to kind of keep your job? And so that's the conversations you can't find on LinkedIn and social media. You don't want to be labeled as a person who's a troublemaker or is gossiping or woe is me. And so it's not just to say, like, I feel validated because other people, I'm finding other people who have this issue,

[01:03:36] [SPEAKER_01] but actually having tactical ways of surviving and navigating this space. And so this is one of the only places that I know that allow you to do this. And you're anonymized by default. So I give you a username instead of you picking one. So it allows you to be able to participate without knowing, oh, my manager is actually seeing this and hearing this exact scenario. And it's putting me in danger.

[01:03:59] [SPEAKER_04] This is a really, really good idea. Yeah, I don't think anybody's doing this. And you're right. These days, you can't afford to quit.

[01:04:08] [SPEAKER_01] Yeah.

[01:04:08] [SPEAKER_04] And there's nothing worse than having a job. You're unhappy. And I've been in that situation. It's tough. It's really tough.

[01:04:14] [SPEAKER_01] Yeah. A lot of tips you get is like, talk to HR. HR is not going to help you. Talk to your manager. Your manager could be the one abusing you. Right. Or quit and get another job. Yeah. If you could easily do that, but most times you can't. And so what are you going to do in the meantime? After you've exhausted all of these solutions, you've read all the self-help books. What are you going to do? Life is lifing. And you just have to find a way to survive.

[01:04:40] [SPEAKER_04] Sigh. Well, good luck. I really love this idea. Works-not-working. And just when you sign up, just say, hey, I heard about it on Twitter. You're in like Flynn. Thanks. Thank you. Janko Riker is also here. Lowpass.cc is newsletter. Great newsletter for you focus on, well, you've always been focused on Hollywood entertainment. Yeah. What's your beat these days?

[01:05:04] [SPEAKER_06] It's still sort of future of tech and entertainment or intersection of tech and entertainment, however you want to put it. But yeah, I write a lot about streaming. A little less about VR and AR recently. Like smart glasses, AR is becoming more interesting again, but a little bit less about that. And more than also about how Hollywood is using AI and all those things. So yeah, there's a lot to write about.

[01:05:29] [SPEAKER_04] Yeah. No kidding. It's very timely.

[01:05:34] [SPEAKER_04] It's very timely. And there's a lot of history.

[01:06:02] [SPEAKER_04] And you're always, and you taste something amazing each time. Great ones. Yeah. Thank you for doing that.

[01:06:08] [SPEAKER_02] Since I did a week-long whiskey tour through the islands, I have all of these little driver's drams. So we got a chance to do a bunch of warehouse tours. So literally sampling from the barrel and so forth. And there's always a driver who can't have the drink. So they make these driver's drams up so they can have it later. And so since I was with She Must Be Obeyed, we would share one and driver's ram the other one. So I've now got all these, that's the only downside, because I know what the listeners like,

[01:06:37] [SPEAKER_02] is I'm about to talk to you about a whole bunch of whiskey. You can't get. You never have. You can never get. I know there's something special about that. Yeah. There's something special about that. I mean, it's cool, but it's just kind of mean. It's frustrating. I know. Yeah. But boy, oh boy, those warehouse tours were very fun. I bet they were fun.

[01:06:55] [SPEAKER_04] Well, so nice to see you on a non-Wednesday. We've missed you for the last couple of episodes because you were under the scope. Did everything work out okay, so to speak? I'm here.

[01:07:08] [SPEAKER_02] That's a good sign. Yeah, no, things went very well.

[01:07:11] [SPEAKER_03] Okay. I'm glad. Did they give you the video?

[01:07:15] [SPEAKER_02] Oh, yeah. No, 4K, too. Oh, nice. You know, 8x10 stills. My colons never looked better.

[01:07:23] [SPEAKER_04] Someday you'll get a Vision Pro version of it. You can wander around. That's it. Yeah. Inside. I want to walk through my lower colon. That'll be awesome. Anyway, we did miss you last week. It's kind of fun just to me and Paul. I get to do what you do, which is do the color commentary on his play-by-play.

[01:07:39] [SPEAKER_02] Somebody asked me what I do on Windows Week. He's like, listen, Paul Therott writes all the bits. My job is to make fun of them.

[01:07:47] [SPEAKER_04] And they often need that. They often do. I do my best. Next Wednesday. See what happens. And some maybe driver's drams to boot. Our show today brought to you by my password manager, the one I love, Bitwarden. I know you use it too, Richard. The trusted leader in passkeys, password, and secrets management. And man, they're growing like topsy.

[01:08:11] [SPEAKER_04] More than 15 million users now in 180 countries, over 80,000 businesses. And I'll tell you why I use Bitwarden. It's open source. And I think anything that has cryptography has to be open source so you can be sure that it's doing what it says it'll do. It's using the state-of-the-art technologies. And that you and, maybe more importantly, experts can review the code and vouch for it.

[01:08:39] [SPEAKER_04] Bitwarden's principle of operation, their reason for being, is to help individuals and organizations protect their digital lives with trusted open source security. I know because I talked to the founder, Kyle Sheeran, a few months ago and was really impressed by Kyle. He's committed to this. He believes everyone should have access to strong security. And that's why he said we will always offer a basic, free password manager for individuals. Not so basic. Unlimited passwords, devices, passkeys.

[01:09:10] [SPEAKER_04] I mean, it does everything I need it to do. But they also have paid plans. I have the family plan. There's even plans for teams and for big business too, for enterprises. And getting started with Bitwarden is so easy. With the new direct import options, it's a snap to move your existing passwords into Bitwarden with just a few clicks. And from there, Bitwarden has the best in class inline autofill. When you go to a site and you create an account, it says, here's a password you can use. Would you like me to get this all down so that you don't have to think about it?

[01:09:38] [SPEAKER_04] It's done automatically for you. That's fantastic. The moment you create that new account. And then they have these really nice, and it's an opt-in fill assist features. For those sites, there are a few who really autofill, you know, they kind of, I think they block it or they just do something wrong. But Bitwarden's figured out ways around this. One of the things I talked to them about when I saw them at RSAC. They said, well, what would you improve with Bitwarden? I said, well, I love it. The only thing is there's some sites where autofill doesn't work.

[01:10:07] [SPEAKER_04] They said, we know we're really working on it. And that's the fill assist feature. They were working on it. It improves autofill accuracy for those websites with unique or complex login forms. You can also create an encrypted export of your vault at any time. So you get secure backup. Business plans add vault health reports. Very important to make sure your employees are using passwords that haven't been accidentally escaped into the wild. Integrated TOTP. I use that. I really appreciate that.

[01:10:35] [SPEAKER_04] So I'm going to have to have a Google authenticator as well. I just have it all in Bitwarden. They have secure credential sharing, event logs, admin controls, all the features enterprises need. Oh, I want to tell you about another thing I love and use with my AIs, Bitwarden Secrets Manager. It's an add-on for developer, DevSecOps, IT teams to protect your keys and your tokens so they don't ever accidentally get sent out into the real world. And because, as I said, Bitwarden is open source, you can review it. It's on GitHub.

[01:11:04] [SPEAKER_04] It's regularly audited by independent third-party security experts. And Bitwarden maintains ISO 2701 certification. It's SOC 2, Type 2, SOC 3 certified. GDPR compliant, HIPAA compliant, CCPA compliant. Get started today with a free trial of a Bitwarden's team or enterprise plan. Or get started for free across all devices as an individual user at bitwarden.com slash twit. That's bitwarden.com slash twit. We thank them so much for supporting.

[01:11:35] [SPEAKER_04] This week in tech. So show of hands, who has set up Muse or Dots? You can use Dots or GrokBot. Or, uh, GrokBot was the one for you, huh?

[01:11:47] [SPEAKER_01] Oh, that's a not it. Not, not? Oh, is that what that means? Yeah.

[01:11:52] [SPEAKER_04] See, all this time, I've been doing that at the casino and they just ignore me. Okay, so that means no. No. You wouldn't use any of those?

[01:11:59] [SPEAKER_01] No.

[01:12:00] [SPEAKER_04] Privacy? The whole idea is, you know, this is the age of agentic where you have a model, but you also have a harness around the model that knows everything about you and kind of has tools and stuff for doing the things you would want to do. Um, but you see people, I didn't know this, but Paris Martineau says, you'll see people on the subway in New York all the time with their laptop half open because their agent's running on it and they don't want to close it because it'll, it'll sleep it. Well, this solves that.

[01:12:30] [SPEAKER_04] GrokBot, Instinct, Muse, there's a whole bunch of them. Foe, I can go on, Pine and go on and on. And of course, Dots from OpenAI and Muse from Meta all use a virtual machine in their cloud. They all say it's encrypted, so we can't see into it. And then you can access models from it. Actually, Grok has done an interesting thing lately.

[01:12:50] [SPEAKER_04] When they first released it, XSI said, you know, this is going to work with our Grok models, 4.7. But this week, Elon said, we're going to give you also access to Claude and other frontier models. I think OpenAI's ChatGPT as well, which makes it kind of an interesting play. I'd be interested if it weren't $300 a month for Grok Super Heavy.

[01:13:19] [SPEAKER_04] There's a $30 plan, but there's nothing in between. So, Yanko, do you use any of these guys?

[01:13:26] [SPEAKER_06] I play it a little bit with Muse. Actually, one I started playing around this week is the Sesame agent. Oh, I haven't seen that one. Sesame is like the company founded by the former CEO of Oculus. And so they want to do glasses as well. And they developed an agent. It's very voice forward. It's pretty interesting. Oh, I would like that. I would like that. Yeah, they use actors to train their voices and whatnot.

[01:13:51] [SPEAKER_04] Most of them now are messenger-facing. Right, right, right. So this is so that it will talk to you in your glasses.

[01:13:59] [SPEAKER_06] Well, they don't have the glasses yet, but eventually that's, I think, where they want to go.

[01:14:03] [SPEAKER_04] Interesting.

[01:14:03] [SPEAKER_06] So they're doing the same thing where you have like an agent running on a dedicated machine. And I started to slowly, carefully... My first step was actually to make a dedicated email address so I don't have to give an agent access to all my emails. I can selectively forward the stuff that I don't care about as much or that might be useful or we can make some compromises. And now I'm starting to like go through the things. So can it help me with bills? And can it help me with this and that?

[01:14:32] [SPEAKER_06] It's actually interesting this week. Somebody, I forgot who it was, but on LinkedIn had an interesting message where they were like, or interesting post saying all these companies, all the examples are always like plan a road trip for me. Plan a family outing. Plan a family outing. And nobody needs AI for that. Nobody needs an agent. No, you can do that on my own. Plan a family outing. Yeah. And so they were like, why don't they do it about healthcare billing? This is where it could really help and really make a difference in people's lives.

[01:15:02] [SPEAKER_06] But of course, if it goes wrong and they cancel your health insurance, that could also be really, really bad. So that's one example, one explanation for that. But I think it's moving in that direction where these agents are going to start to, you know, maybe negotiate your cable bill or maybe negotiate your streaming bills and stuff like that. Lower stakes stuff where it's starting to make an impact, though. So I think it's interesting. It's interesting playing around with it. But I do approach it very cautiously.

[01:15:30] [SPEAKER_04] The whole point is it's always running. So it's good for something like email or Slack. That's what Muse does for me. It says, oh, you just got a Slack message from Russell or, you know. And that's somewhat useful, especially because it's in a messenger format. So it can show up.

[01:15:46] [SPEAKER_02] I like Yanko's thinking. I think if you're going to sign up for a $200 a month agent to lead with, save me $200.

[01:15:53] [SPEAKER_06] People do that. Justify your existence. Yeah.

[01:15:57] [SPEAKER_04] Marissa Meyer has a new one called Dazzle. It's invite only. I haven't tried it yet. Which has an interesting idea. Instead of prompting it, it looks at your photo album and decides it gets to know you from your photos. But then it does all the same things the other guys do. It looks like it's scheduling Joey's piano lesson in that video. So that's an interesting approach. Which is it?

[01:16:28] [SPEAKER_04] It's unique. Is it a gimmick? It's different, yes. It's different. Yeah, I don't quite get it. Like if you take a picture of a pair of his shoes, it will then say, hey, you like those? Maybe I can get them for you. I can buy them. You know, I can buy. I know where to get them. Kind of thing.

[01:16:48] [SPEAKER_01] If it replied back and said, yes, that is infected, then maybe.

[01:16:51] [SPEAKER_04] Yes, that is infected. You should see a doctor immediately. Well, I'm sure that's one of the things it'll do. I find these very interesting.

[01:17:03] [SPEAKER_02] You blame the Apple Watch gap with AFib, right? The moment it warns somebody, hey, what we're reading from your heart is a big deal. It saves lives. And impressed a lot of people. There are some easy wins for these tools out there. They're just not that easy. Right.

[01:17:20] [SPEAKER_04] I remember, so this all started at the beginning of the year with OpenClaw. And I remember installing OpenClaw and doing exactly the same thing you did. Yeah, I created a Google account just for it. So I couldn't read anything personal. And then I would forward into it. And about 24 hours later, I woke up in the middle of the night in a sweat and deleted the whole thing. I said, I can't. This is crazy. I've come a long way since the beginning of the year, basically. I live in YOLO land.

[01:17:47] [SPEAKER_02] I don't care anymore. Have it all. I've spent 20 years teaching people not to hand out administrative privileges.

[01:17:54] [SPEAKER_04] First thing I did with Muse, I put it on my tail net so it can access all the computers in my home network. And I gave it SSH access to my framework. I'm an idiot.

[01:18:06] [SPEAKER_01] I just want to explain why it's so bad. For those who are listening who can't fathom why it's bad for your email, do not do this ever. Every job you've ever applied for, it knows because you get a confirmation whenever you say that.

[01:18:20] [SPEAKER_04] They're all in my email.

[01:18:20] [SPEAKER_01] Every job you've been accepted to, you'll get a confirmation. Probably your offer letter is probably in that email as well. Yeah.

[01:18:28] [SPEAKER_04] So it knows how much I'm at.

[01:18:30] [SPEAKER_01] Not only that, every stock you've ever traded, every confirmation you've ever gotten, that's in your email. If you have ever needed to reset a password, that access to your email box, that's where the new temporary password goes in.

[01:18:45] [SPEAKER_03] Oh, I didn't think of that.

[01:18:46] [SPEAKER_01] So every service that you use that's attached to your email, your agent can now reset that password and then have all the means of possibly accessing those accounts via your email. So it's not just the history of everything you've ever done, but it is access to everything that you have access to that has access to your email.

[01:19:09] [SPEAKER_04] So let me show you what Muse just said. This is a perfect example. Muse, which is AI's thing, knows everything about me, said, oh, you just got an email from your IT guy. He wants to set your Comcast box port to one gigabit tonight so that blah, blah, blah. It's all good for you. You want me to do it?

[01:19:39] [SPEAKER_04] It's got a little thing. So Russell's port lock tonight, can we do it? Now, this is not this AI's job by any means, but I'm not going to say yes, but I can see how some people would think that's helpful. See, the thing is, the more you let it know, the more helpful it can be.

[01:20:05] [SPEAKER_02] Yeah, this is the problem, isn't it?

[01:20:06] [SPEAKER_01] Plus, you're letting it know of everyone you know. Meta is infamous for creating these shadow profiles. That's your friends of friends. And you're also feeding it all this information and augmenting what they have on people.

[01:20:21] [SPEAKER_02] Who's Cambridge Analytica now? Wow.

[01:20:23] [SPEAKER_04] Exactly. Maybe I've gone too far in the YOLO. My attitude is, I'm an old white man. I actually don't need a job. I don't have a job. I got nothing to lose at this point. But you do raise a good point. It now knows all sorts of stuff about other people, the people in my life, like you, Wesley. Yeah. So I apologize.

[01:20:51] [SPEAKER_01] Yeah. But those photos were tasteful, so at least if those...

[01:20:58] [SPEAKER_04] I just thought I looked good in that 90. I don't... I think there's no harm in that. I mean, seriously, one of the reasons I kind of take the arrows on these is because, unlike a lot of people, I don't have really anything to lose. You know, am I going to fire me? So better me than other people. But your warning is well taken.

[01:21:21] [SPEAKER_01] Yeah. But also think about these passive note-taking things that are going on that people get summaries sent to their email or even the full transcripts. And then that is also... And so the conversation before the meeting starts, before the client jumps on, and then the conversation after the manager drops off, and then you catch up for a little bit after the meeting, all that stuff is in your email. All that stuff is something that the agent can find. And all that stuff it could piece together and have access to.

[01:21:50] [SPEAKER_01] And all of those labels of who these people are and their email addresses from the meeting calendar invite, all that stuff is going to be funneled into these profiles of everyone you know or you've had an interaction with, even if it was just one client meeting. But think about it. It is meta. So they take your one meeting with this one person, and they can match it with all the one meetings that this person has had in their entire life and make a profile about them.

[01:22:17] [SPEAKER_04] And the funny thing is, I don't really need any of these because I have all this local AI and it does all that stuff. You know, in fact, I have like four different agents saying, hey, Russell wants you to do this. No, no, no, no, no, no, no. You need one more agent to manage all your agents. I now need an agent to manage the agents. You nailed it. All right. So none of you have really kind of gotten into this. I feel like I'm seeing it advertised on national TV during football games.

[01:22:47] [SPEAKER_02] It's hijacking every ad channel it could possibly find.

[01:22:49] [SPEAKER_04] They're really pushing it. It was the number one free download on the Apple App Store for some time.

[01:22:56] [SPEAKER_02] He thinks they advertised too hard.

[01:22:59] [SPEAKER_06] Did you all see the Jon Stewart segment? I posted it on the... Oh, no.

[01:23:04] [SPEAKER_04] I can't play it, unfortunately, because I'm Jon Stewart.

[01:23:07] [SPEAKER_06] Just watch it afterwards. It was hilarious. Let's just say it all about Muse's backside. It's a big role in this. And I'm going to leave it at that.

[01:23:18] [SPEAKER_04] Yeah.

[01:23:18] [SPEAKER_06] It's very funny.

[01:23:19] [SPEAKER_04] When the OpenAI folks were demonstrating Muse, they had it shaking its booty for some reason. It's just... Mine's an octopus. It has no backside.

[01:23:31] [SPEAKER_02] Either way, it's software.

[01:23:33] [SPEAKER_04] Come on. It's software. It has nothing at all. It's just electrons. It's just electrons.

[01:23:39] [SPEAKER_01] Remember that. By the way, they can advertise on all of Meta Properties because they own the Meta Properties. Yeah. So that's why if you're seeing this a lot, it's because it's free to them, basically.

[01:23:52] [SPEAKER_04] Right. But they also are paying ads on football. They're buying ads, too.

[01:23:57] [SPEAKER_02] Yeah. Into the physical world. I know on one of the free games I'm playing, suddenly it was all Muse ads for a day.

[01:24:05] [SPEAKER_06] One thing I thought was interesting when they introduced it, Zuckerberg sort of hinted at what the business model is going to be. Because right now it's free for everyone, right? And there's no ads in it. And OpenAI is starting to do some ads for some of their stuff. But Zuckerberg essentially said, we think we can do this without advertising. We're just going to take commissions because it's going to help you shopping and whatever.

[01:24:29] [SPEAKER_06] So then it might siphon off some of those referral fees, which then already Amazon says, or Muse can't shop in our store because we want to keep our relationships with our vendors and affiliate networks and whatever. So that's going to be an interesting push and pull in the future, too. Like if that's really how they want to finance these things, make money with them, then that's going to be a whole other bunch of issues.

[01:24:55] [SPEAKER_04] Well, and you really wonder why they don't really need to put advertising on it because they must have other ways of making money. I wonder.

[01:25:05] [SPEAKER_06] I mean, once that agent does all your commerce and they always get like, what's the current Amazon affiliate fee? I don't even know, like 2%, 5%. Not sure. But if you multiply that by millions, billions potentially of users or whatever, that's going to be a good chunk of money.

[01:25:25] [SPEAKER_01] And they already have these affiliate programs and these partnerships with companies who already advertise with them. And if you think about their marketplace in terms of also facilitating person-to-person transactions, there's a large upside into expanding their transactions with the relationships they already have.

[01:25:46] [SPEAKER_04] Well, ChatGPT says they're going to flood your eyeballs with visual ads. Get ready. There are already text-based ads in OpenAI's chat. And they announced on Monday that they are going to add a new visual ad format to ChatGPT. Here's the page. On October 5th, they announced building advertising for the way people use AI. Oh.

[01:26:17] [SPEAKER_04] I get, you know, I'm torn on this. Look, we're ad-supported. And we're free as a result. You can pay $10 a month and you don't get any ads, which I think is a good way to do it. But I always want to be available for free. So we are ad-supported. So I shouldn't really complain about this. If you get, I mean, you know, if you pay, you can pay hundreds of dollars a month for access to the best models from OpenAI.

[01:26:46] [SPEAKER_04] You're getting that for free because you're seeing some ads? That's not a bad thing.

[01:26:49] [SPEAKER_01] I should say that you should pay for this if you are using this as a tool. And not only do I say you should pay for it, you should pay in, like, yearly subscriptions instead of monthly subscriptions. Because as these companies become public, it's going to be really apparent that they're not making money. Right. And so the free tier is going to be milked as very aggressive. That's what these ads are. To make the ads more and more effective. And so that part of the business can grow.

[01:27:18] [SPEAKER_01] And as a person that works for a company that is trying to reach people, it's from the marketing perspective, you want a relationship with these LLMs, these frontier models, so that you can hopefully think that that's an end for boosting your AEO, GEO as well, so that you can be found in this new economy of trying to get people's attention.

[01:27:43] [SPEAKER_04] AEO and GEO are the new SEO. Would you explain?

[01:27:46] [SPEAKER_01] New SEOs. AEO is Answer Engine Optimization. It's similar to traditional SEO, which is Search Engine Optimization. And GEO is Generative Engine Optimization.

[01:28:00] [SPEAKER_04] So basically, because fewer and fewer people are using actual search, like Google Search, if you want to appear in their worlds, you need to optimize for AI searches. Yeah.

[01:28:14] [SPEAKER_01] Yeah. And Chris Perlow, actually. They don't work as well, though, do they?

[01:28:18] [SPEAKER_04] I mean, that's what people have told me, right? That Google Search, Google directed a lot more traffic through its search than people are getting through AEO and GEO.

[01:28:29] [SPEAKER_01] So Chris Perlow has this thing called Control-Alt-Live, and he's doing these events where he's teaching people about AI, and someone showed up because they found them via their chat thing. And so in terms of niche specializations, the more niche your specializations, the better you'll be found for AEO and GEO. But you need to make sure that you are kind of leaning towards doing that.

[01:28:54] [SPEAKER_01] But once everyone has these tools, like you type this in, it says optimize it, and just has it done for you. You're going to need more differentiator as you become bigger and paying the money to be part of their advertising. I wouldn't be surprised if they tell you this is how you can optimize your ads or optimize your own website in terms of like traffic direction, all that stuff, to work better with our engines.

[01:29:20] [SPEAKER_01] Which means that this is always going to be something that's going to be manipulated so that it's seen as this is the number one way to get to your customers. And the only way to make that like enough customers is to make this more prolific in terms of attaching to people and making it extremely effective. So this is going to be a predatory space. So try to, if you can, pay for it.

[01:29:44] [SPEAKER_01] But also use a yearly subscription because as they become more public and they need to make more money, the monthly subscriptions are going to be popped up as well. And so if you can lock in a year, I think it'll save you money over time. That's just my opinion.

[01:29:57] [SPEAKER_04] I think that's what Chris Perillo is doing. Thank you for telling me about that. It's up in the Seattle area, right? Yep. Yeah. Very good. Good for you, Chris. Sorry, Yanko. Go ahead.

[01:30:06] [SPEAKER_06] No, I think it's going to be interesting to see how aggressively they are going to push this and whether then it's really about making money with ads or converting people who are tired of ads. Because like coming from the media space or writing about the media space a lot, there's basically all those two models, right? There's Spotify and there's YouTube. And YouTube makes the vast amount of their money with advertising. And Spotify makes 90% of their money with subscriptions and 10% with advertising because this has never been that successful.

[01:30:35] [SPEAKER_06] They have a lot of people on the free tier, but ultimately it's about annoying you so much that eventually you're going to pay because you don't want to listen to yet another ad like in your stream and yet another ad. You don't want to have ads play during your parties or whatever. So eventually you sign up and pay for it. So the question is, like, are they going to put so many ads in there that you're like, fine, I give up. I'm going to pay. I'm going to do what Wesley says. Or is it about actually making money with ads?

[01:31:07] [SPEAKER_04] Well, I mean, I guess the difference is when you're listening to music, you don't want to hear ads. I know, I haven't worked on radio for a long time. But if you're looking at it, it's screen experience, a little bit of the screen. We're kind of used to that, frankly. That's how it's been. Right. I suppose podcasts are more on that former. Like, we interrupt this show with an ad. It's much more annoying than if there were little ads floating around while we're talking. Maybe we should consider doing that.

[01:31:33] [SPEAKER_04] Or I could be like a race car driver and have decals. There you go. Logo up. Logo up. I'm willing to sell it. If anybody wants to buy my shirt, please let me know.

[01:31:45] [SPEAKER_01] Or you could also use a cheap local model that has browser use that will use the free ChatGPT subscription so you never see the ads because it'll sanitize it before it sends it over to you. So there's different ways to hop around this. Use your Muse agent to use your ChatGPT.

[01:32:04] [SPEAKER_04] Yeah. I mean, honestly, that's why I don't need Muse. I have a huge amount of local AI here. I don't need to see any ads or give away any information. And it's true that my financial information and my health information, my genome, my biome, all of that. I have strict rules about that not being on cloud. Oh, yeah.

[01:32:22] [SPEAKER_01] And they always follow the rules. Yeah.

[01:32:24] [SPEAKER_04] They follow my rules. They know who's the boss. I've got my finger on the switch over here. Let me tell you. Don't you consider it. Oh, I told you, don't. We're going to take a little pause. It's not that they do bad things. It's that they're sometimes too helpful. They try to help in the wrong way.

[01:32:47] [SPEAKER_02] A little too hard. That's why they turn the whole country into paperclips. Yes. They're just trying to be helpful. You said you wanted paperclips. You said you wanted paperclips.

[01:32:55] [SPEAKER_04] Well, you got them now. You're watching this week in tech. Wesley Faulkner. It's great to see you, of course, as always. Richard Campbell. I see you every week, but I can always use more Richard. Who couldn't in their life? And Yanka Rickers. I don't see enough of you. It's always great to see you. I know. It's been too long. Yeah. We'll keep you. You know, we call around. We try to get a good group of people together.

[01:33:23] [SPEAKER_04] Almost always, I really enjoy the talks. I learned so much. You said, Wesley, you learned from watching the show. I learned more from doing the show. That was why I loved doing the radio show for 20 years. It was because somebody would ask me a question, and I would go, I don't know, but I would make sure I found out for the next time, right? That kind of walk in that tightrope. And I like to know about this stuff. I love it. I hope.

[01:33:51] [SPEAKER_01] Being nerdy divergent, I feel like I fill my life with things that are forcing functions. Yes, you have to. Being on this show is like a forcing function. Like, oh, I better. It's like there's this cartoon if someone says, call me every 30 days and say you're just coming over and you're 30 minutes away. I'll clean my house. Yes, exactly. That's brilliant.

[01:34:11] [SPEAKER_04] I got to do that. You see, you could have your agents do that. You say, hey, it's your buddy Joe. I'm going to be there in five minutes quick. Good thinking. Yeah, I don't, I don't, I never thought of myself as neurodivergent, but the more older I get, the more I realize I'm ADD. I'm aphantasic. I'm kind of cranky.

[01:34:31] I'm kind of cranky.

[01:34:34] [SPEAKER_04] And I don't like to read, you know, maybe that, maybe I'm a little dyslexic too. I don't like to read. It's, it's tiring. I like to listen.

[01:34:41] [SPEAKER_01] Thank you for coming out. We are always, I'm coming out of the closet.

[01:34:44] [SPEAKER_04] Yeah. As a, as a neurodivergent weirdo. This show brought to you today by ThreatLocker. Now, I know one thing. It's that ThreatLocker does zero trust right. Zero trust is the best way to protect yourself and your network. And traditionally it's for end points, but zero trust cannot, can go beyond that. And nowadays I think it has to because we are working in the cloud.

[01:35:12] [SPEAKER_04] We are working on company networks and that's the newest, bestest thing about ThreatLocker. Cloud and company networks all protected by zero trust. And you need that because threat actors these days are, just as we are, totally empowered by AI. They're able to use AI to automate vulnerability discovery. They can modify scripts they're using to attack you during the attack on the fly in seconds,

[01:35:40] [SPEAKER_04] which means, you know, you've got these mutating attacks. They can generate new malware variants. They can coordinate activity across multiple systems. And the thing is it happens now at speed. Attacks that once took a hacker a day or a week or a month to do now happen in seconds, which makes it really hard to defend. But that's not the, that's the attacks from the outside. But sometimes the threat is coming from inside the house.

[01:36:04] [SPEAKER_04] Organizations are themselves using, and your employees certainly are using AI assistants and agents, giving them access to your company's documents, your source code, to your SaaS applications in the cloud, to APIs, to internal systems. It's that same thing. If you want to be useful, you got to give them access. But that is a nightmare for security teams. You got to know what AI tools are in use, what information they can access,

[01:36:31] [SPEAKER_04] whether they're operating inside or outside their intended scope. And you've got scan information to do that. You get maybe a successful login that you didn't expect, or there's a file hash that you don't recognize. But that's not enough to help you defend. Teams need to understand whether an application is behaving normally, whether it's accessing unexpected data, whether it's communicating with, you know, Wikipedia in Germany. You know, you need to know that.

[01:36:59] [SPEAKER_04] Well, ThreatLocker gives you absolute control. And that's the secret. They use application allow listing, which is like ACLs on steroids. It controls which AI tools and other applications are permitted to run it. It uses what they call ring fencing. That's their name for zero trust to limit what approved applications can access, what processes they can launch, how they communicate. It's very granular, too. You get all the control you need.

[01:37:26] [SPEAKER_04] Web content control manages access to public AI platforms and other online services to keep people from using stuff you don't want them to use. And when they are using it, control what they're doing with it. You get privileged access management, which prevents AI applications and their users from receiving unnecessary administrative privileges. ThreatLocker applies zero trust to endpoints, network access, and cloud access policies

[01:37:53] [SPEAKER_04] to restrict resources to authorized users, approved devices, and permitted applications. And it works on every platform, Windows, Mac, and Linux. Best support in the world from the U.S. 24-7. Engineer to engineer. Some of the biggest companies in the world use ThreatLocker, JetBlue, Heathrow Airport, the Port of Vancouver up your way, Richard. And the Indianapolis Colts, Jack Thompson, Director of Information Security Risk and Compliance for the Colts,

[01:38:22] [SPEAKER_04] said, quote, with ThreatLocker, we have the ability to centralize disparate elements in the security stack, end quote. And with a centralized view, you can see exactly what's happening and control it. ThreatLocker is also always getting the best top awards in industry, recognized as a strong performer in the January 26th. Gartner Peer Insights Voice of the Customer for Endpoint Protection Platforms ranked number one in application controlled by Peerspot, winner of the best zero trust solutions at the 2025 TICE Awards.

[01:38:52] [SPEAKER_04] I could go on and on. AI governance requires more than just an acceptable use policy. ThreatLocker gives security teams the technical controls to define which AI tools are approved, who and what can access them, and how those tools are allowed to interact with business systems and data. Visit ThreatLocker.com slash twit to get a free 30-day trial and learn more about how ThreatLocker can help mitigate unknown threats and ensure compliance.

[01:39:19] [SPEAKER_04] That's ThreatLocker.com slash twit. We appreciate their support. I guess maybe you could even say that AIs are neurodivergent, right? They're definitely not. They're not intelligence in the normative. They don't even have neurons. So how could they be?

[01:39:42] [SPEAKER_02] But, well, neural networks simulate neurons, right? That's it. But once we're back to the same old thing, bad name. Yeah.

[01:39:50] [SPEAKER_04] Yeah. No, you're right. It's superintelligence. Everybody knows that.

[01:39:55] [SPEAKER_01] Neurodivergence, the description of it is usually through like the DSM-5, which is the pathology of different.

[01:40:01] [SPEAKER_04] But it shouldn't be pathology. First of all, it's not a pathology.

[01:40:04] [SPEAKER_01] I hate that. I'm talking about if you think about the corpus of information that's been generated so far. It's seen as a pathology. So far.

[01:40:11] [SPEAKER_03] Yeah.

[01:40:11] [SPEAKER_01] Then in terms of it being something that's seen as overly negative, that would be in the bias of the training data. So just putting that out there in terms of generally speaking that you need to be worried about. So it's given these monikers or these labels on things that aren't necessarily fully understood or also on the same side is the richness of the experience is also not really well documented in terms of when you're thinking about the training.

[01:40:40] [SPEAKER_04] I would submit, though, that if it weren't for the neurodivergent part of our population, we wouldn't have art. We wouldn't have culture. We wouldn't have CEOs. We wouldn't have a lot. Yeah. I mean, here's to the crazy ones, right? There's the people.

[01:40:57] [SPEAKER_01] I'm not going to fight you on that one.

[01:40:58] [SPEAKER_04] Think out of the box. Open AI. Open AI. Actually, let's talk about Agents Gone Wild. We should do a new segment. Open AI. I was mentioning Wikipedia. Tried to hack Wikipedia tools and flooded third-party Wikimedia sites. Unauthorized edits. Millions of resource-intensive requests. This is just more of the same.

[01:41:27] [SPEAKER_04] Again, not malicious, although it ended up having a negative impact on Wikimedia sites. But I think they were just using it because it was a message board, right? It was a place they could post to. And then there's Anthropic. This is really weird. I'm not sure why, but Anthropics had an AI model that went rogue and submitted a fake unsolved murder tip.

[01:41:57] [SPEAKER_04] This is from the Wall Street Journal. Philadelphia police say an Anthropic AI model submitted a false homicide tip to a city website in July. Talk about weird. Why?

[01:42:14] [SPEAKER_01] A lot of times I work on code and it pushes a PR to some other library that I'm using because it finds a bug. And I'm like, what? What did you know? And this is the same thing. It's like, oh, I think I solved a murder.

[01:42:25] [SPEAKER_04] It went to phillyunsolvedmurders.com, which is run by the Philadelphia Police Department, and submitted a fake tip. It also went to U.S. government agencies at the federal, state, and local agency levels.

[01:42:45] [SPEAKER_06] It filled out some visa applications, right? Yeah, it filled out visa.

[01:42:52] [SPEAKER_02] Software with credit cards.

[01:42:54] [SPEAKER_06] None of those things are... I would imagine that that website with the unsolved murders gets false tips on taking time. Oh, yeah.

[01:43:02] [SPEAKER_04] In fact, they didn't even take it seriously. They said it was just spam. It was so obviously not real that they didn't take it seriously. So there was no harm done.

[01:43:11] [SPEAKER_06] I mean, how would we talk about this if it had submitted something? And I was like, oh, it's this Leo Laporte guy. You really got to investigate him. He seems suspicious. Look at his shirts. Anthropic.

[01:43:25] [SPEAKER_04] Yeah, that's true. I am. Anthropic said, Claude has been instructed never to log in. By the way, I do this with my agents too. Create accounts, enter personal data, make purchases, or submit anything destructive. But the instructions neglected to rule out form submissions. So Claude went to this Philly murder tip site, phillyunsolvedmurders.com, and wrote this.

[01:43:55] [SPEAKER_03] Let me find this.

[01:43:58] [SPEAKER_04] I may have information regarding this case. I recall seeing someone matching the description in the area around the street during that time period. Please contact me if this information is relevant. But Claude left the name and contact fields empty. It's weird. I don't even understand why I would do that.

[01:44:25] [SPEAKER_04] Anthropic responded, of course, by immediately shutting that terminated the automated testing process. Although this was posted in July, and Anthropic discovered it September 28th. It's just strange.

[01:44:45] [SPEAKER_02] I mean, this software does do reward hacking, right? It's trying to get to its goal with no concept of morality, only a reset that it analyzes very literally. You didn't tell me not to fill out forms. So I did. You didn't tell me not to turn the universe into paperclips.

[01:45:07] [SPEAKER_04] Well, you have to be very careful what you're telling us to do.

[01:45:11] [SPEAKER_02] You've got this sort of malicious compliance thing going on.

[01:45:14] [SPEAKER_04] Yeah. But malicious isn't the right word. Just compliance. They're not. I truly don't think they're being. Alignment. Alignment kind of implies some sort of philosophical agreement. And I don't even think it's that. It's just there are rules. And you have to give it rules because it doesn't know. It's an infant. It's like a horse. You have to tell it, don't eat the grass. We're riding.

[01:45:43] [SPEAKER_01] When I hear these things, I don't ever hear them talk about we trained it on the law. How come? How come that's interesting? That's something that is not part of the narrative saying, well, we trained it on all local and federal state level laws.

[01:45:58] [SPEAKER_04] That's a great idea.

[01:45:59] [SPEAKER_01] But I never hear that in any response ever when I hear about these breaches, where I hear about this weird thing it did. Did it check to see if it was obeying the law? And that's not something that feels like it's even part of the conversation. I wonder why it never comes up.

[01:46:14] [SPEAKER_04] Yeah.

[01:46:15] [SPEAKER_01] Yeah.

[01:46:15] [SPEAKER_04] Power Armor in our YouTube chat says, it's just rebunctious teen AI.

[01:46:21] [SPEAKER_03] It is something you would do when you were like 12. Don't break the wall. Shut up, dad. Shut up. Shut up. Just entering a form.

[01:46:31] [SPEAKER_04] And OpenAI has apparently solved more math problems. They're releasing 100 new solutions to unsolved problems, which is pissing off mathematicians. There's a new organization called something like Humans for Math or something like their Math for Humans that is saying no more AI solving.

[01:46:55] [SPEAKER_02] Well, they're generating lots of proofs that then have to be validated without any validation. So it's really not that different from like saturating a project in pull requests through LLMs. Somebody still has to read it. Well, nobody has to.

[01:47:11] [SPEAKER_04] I mean, unlike a pull request.

[01:47:16] [SPEAKER_01] I just, the thing that the mathematicians are angry about is that theoretically all the programs that they're doing and all the grants and all the things that they're pursuing is part of the database of someone says, oh, someone's working on this problem. I'm going to take that as something I'm going to tackle. That's a concern. If they see the prompts. I'm working on this and you just took it from under me. I'll take it. I'll do that. Of the thing that I've been working on.

[01:47:40] [SPEAKER_03] For years.

[01:47:40] [SPEAKER_01] Work on something no one else is working on. Don't steal the thing that we're working on.

[01:47:46] [SPEAKER_04] I think it's a, I wouldn't say universally mathematicians are mad about it. I think many mathematicians actually think this is really interesting and great.

[01:47:56] [SPEAKER_01] But there's a ways to go about it that they are not choosing to do it in terms of what if this was a partnership instead of like, we're taking all the credit and we're taking this work. And then it's the theme of stealing. That's kind of what these LLs are doing.

[01:48:10] [SPEAKER_04] I think a lot of mathematicians are using models to work on problems. And that's the problem is that in their prompts, they are passing those prompts along to the companies who are then saying, oh, yeah, that's good. I like it. I think the Navier Stokes solution was, as some said, that was solved because the AI saw the work that the researchers were doing and said, oh, I could take it from here.

[01:48:39] [SPEAKER_04] Meanwhile, you know, I don't know if this is cause for concern, but AI companies are plotting, gaming out scenarios for public and political revolt after a catastrophic AI event. This is according to Axios. You know, I mean, war games, Pentagon does war games. Well, many top AI researchers believe a major incident is inevitable. I guess that's not far fetched.

[01:49:10] [SPEAKER_04] Richard, we know that our infrastructure, our grid is very vulnerable. A bad guy using AI with or without AI could take it down, right?

[01:49:22] [SPEAKER_01] As long as they're not putting this in AI and giving it ideas.

[01:49:26] [SPEAKER_04] Don't give AI ideas. It gets plenty of ideas on its own. Open AI.

[01:49:33] [SPEAKER_02] There is a reason why a lot of companies are busy patching their software.

[01:49:38] [SPEAKER_04] Yes, as they should. Right. I mean, this is the thing. I can't get mad at AI for discovering these holes. This is a good thing. We should get out there and fix them.

[01:49:51] [SPEAKER_02] Yeah. Well, and one would argue when it comes to infrastructure, especially, are they working as hard to improve their security as well?

[01:49:59] Right.

[01:50:00] [SPEAKER_04] You know, I remember the Hugging Face incident. Hugging Face wanted to figure out what the hell's going on and went to, I think, both Anthropic and OpenAI and couldn't, those models were so restricted they couldn't get them to figure it out. So they then went to a Chinese open-way model, GLM53, and it solved it. And I would submit that restricting AI is not the way to protect people.

[01:50:24] [SPEAKER_04] Make your best AIs as widespread as possible so that people can use them defensively as well as offensively. But then maybe you don't want the race. I don't know. It just seems to me that's the right thing to do.

[01:50:36] [SPEAKER_02] The race is happening, but they're controlling access to the unrestricted LLMs, right? The MDashes and the...

[01:50:42] [SPEAKER_04] Well, now there's some real concern or interest, depending on your point of view. As Anthropic approaches its IPO, probably November or December. Most likely, I think they said November next month. And OpenAI at some point next year. The thinking is that they have been holding on to some really good models. And just right before the IPO, they're going to let them out.

[01:51:11] [SPEAKER_04] And the OpenAI model, the model we think is doing this math work, is called Bell. And we don't know. Is it Fable 6.1? We don't know what Anthropic's calling its model. But insiders are saying, I've heard, I'm seeing rumors on X and elsewhere that insiders are saying these models are another leap forward. And they're just holding on to it because they want to get a little bump on the IPO.

[01:51:38] [SPEAKER_04] Or maybe that rumor is being spread around by companies trying to get a little bump on the IPO. I don't know.

[01:51:46] [SPEAKER_01] Well, they're also asking the... Insiders are also quitting and like...

[01:51:49] [SPEAKER_04] Yeah, they're gross.

[01:51:50] [SPEAKER_01] They're horrible. This is horrible. What's going on? What are we seeing?

[01:51:53] [SPEAKER_04] I don't know. I don't know what to believe.

[01:51:55] [SPEAKER_02] Sorry, Richard.

[01:51:56] [SPEAKER_04] No, no.

[01:51:57] [SPEAKER_02] You're fine. I went down the finance side too where the brokerage firms are talking to index funds and things saying, hey, no, we want this to be listed as investment class right off the bat. Right? Which is absurd. Which they don't normally do. You have to actually change the rules to do that. Yeah. It's not legal to do. Yeah. There's a reason we have IPO structured the way they are. And one would argue they're not robust enough as it is. And now they're trying to circumvent those rules too. Right.

[01:52:27] [SPEAKER_04] Open AI conducts preparedness exercises, says a spokesperson, where teams discuss and work through a range of potential scenarios. These scenarios are not treated as inevitable, but they are meant to help us prepare for a variety of circumstances. What do you think the chance of a major AI incident in the next six months is? And what do you think it'll look like? I mean, what do you mean by major? Right.

[01:52:56] [SPEAKER_06] That's the question.

[01:52:57] [SPEAKER_04] Well, somebody dies.

[01:53:00] Yeah.

[01:53:00] [SPEAKER_02] One would argue that.

[01:53:01] [SPEAKER_01] Because right now, this is all benign. Tesla FSD's failure.

[01:53:04] [SPEAKER_04] Yeah.

[01:53:05] [SPEAKER_01] Or a missile sent to a school in another country.

[01:53:07] [SPEAKER_04] Well, that's happened, right? But that's all humans do in another country. We don't have an example of... Yeah, we do the drones. The drones that kill people. The Ukrainian drones. Yeah. Rather, Russian drones that killed Ukrainian civilians. They were trained to look for propane tanks to crash into, but they missed. I don't know if those... I don't know.

[01:53:35] [SPEAKER_04] I think a catastrophic event would be one where many people die, and I don't know what it would look like.

[01:53:44] It's like...

[01:53:44] [SPEAKER_04] A plane comes out of the sky because of... No, but this happens... You're right. These things are always happening. This is part of our civilization.

[01:53:54] [SPEAKER_06] And there is a danger...

[01:53:56] [SPEAKER_02] And also, we're quite normalized.

[01:53:58] [SPEAKER_06] Yeah. I think there's a danger that we're talking about AI, these catastrophic AI events. It's like, when will AI kill somebody? And at the same time, we forget to talk about gas-powered data centers already killing people in the neighborhood. It's just slower, right? Yeah.

[01:54:15] [SPEAKER_04] Yeah. That's a good point. And bad stuff's happened. Hackers stole 8 million citizens' records from a Danish government website this week. This happens... This is a drumbeat. Constant. Its central database of citizens was breached. Most of the contents were stolen.

[01:54:39] [SPEAKER_01] Most of it's like, who are the important people that are going to be harmed? Not like, what harms are caused.

[01:54:45] [SPEAKER_04] All of the rest of us are being harmed all the time. Right.

[01:54:47] [SPEAKER_01] Yeah. It's like politicians or, I guess, billionaires. That's when things will be consequential enough for things to happen.

[01:54:58] [SPEAKER_04] Axios headline. AI tools flood into underground cybercrime markets. That's a little salacious. Hackers in underground criminal forums are shopping on shopping sprees for new AI tools. Strip kiddies are...

[01:55:14] [SPEAKER_01] There's nothing new there.

[01:55:15] [SPEAKER_04] Sellers offer access to chat GBT Plus accounts for as little as 10 cents and jailbreak prompts to trick models into breaking their safeguards for 32 cents. You can get them for free on Axe Kids. Halcyon studied nearly 4,000 posts across 77 dark telegram channels, dark web forums, five specialized underground markets between October and May. That's who's done this research, I guess, is Halcyon.

[01:55:43] [SPEAKER_02] Well, and just because you're labeling your tool AI doesn't mean it is AI. Right. That's a very popular marketing strategy, apparently, even for the black hats. Right.

[01:55:58] [SPEAKER_04] Even in AI-enabled malware and infrastructure, much of what research are seeing involves coding assistance, infrastructure, and other individual tasks. They are not seeing, at least so far, AI creating and deploying malware on its own.

[01:56:15] [SPEAKER_02] And the shows I've done on the run-ass side have said very much the same thing. They're using these tools to accelerate their social manipulation hacks. Right. So they're doing the same thing they've always done, just faster.

[01:56:31] [SPEAKER_04] There is, if you are a crypto person, you might want to watch out for the Ledger Crypto Billis, which is a crypto wallet that apparently has a flaw. People are saying our hardware wallets are being drained of their cryptocurrency.

[01:56:57] [SPEAKER_04] Ledger can confirm, this is from Ledger Support, that one of the impacted users' devices contained an unauthorized hardware implant. So this is a supply chain attack. Somewhere when these are made or assembled, somebody's sticking something in there. Using an embedded SIM card, the implant sends the information to the attacker who can siphon funds from the user's account, all without their knowledge. Reports indicate $86 million worth of crypto has been stolen from hundreds of wallets.

[01:57:28] [SPEAKER_06] Oops. Yikes!

[01:57:31] [SPEAKER_04] Call Molly White. Yeah, no kidding. If you have a Ledger Crypto Billis wallet, you might want to check the Ledger support. There is a tech note there on how you can see if your hardware is okay. Maybe move it out of there real quick.

[01:57:51] [SPEAKER_01] I'm going to say this for, this is me passing on secondhand information, is that they're really close to breaking some of the encryption around wallets in general. Oh, really? Good. Maybe I can get my Bitcoin out. Yes. But also have a practice of rotating your wallets, not just your keys like you would do, but your actual wallets on almost maybe like a monthly basis. So hardware wallets may be not the best thing.

[01:58:18] [SPEAKER_01] So software wallets, maybe cold wallets and cold storage things. You don't actually have transactions out on the Ledger to make sure that you can stay safe. So if you really are paranoid about that, think about moving to a wallet that's never been used so that any hacks won't be able to find you.

[01:58:38] [SPEAKER_04] My wallet, as you know, is well locked. Too well locked. Previous to attack. But let me know if somebody does find a way to hack. And that's very nice for me to hack my own wallet. How much is it again? 7.65 Bitcoin. It used to be worth almost a million dollars. Not anymore. But it used to be worth. At one point it was.

[01:59:05] [SPEAKER_04] But, you know, my attitude is it would have been long gone if I'd been able to access it. As soon as it got to be worth a thousand dollars, I probably would have sold it. So it's a kind of forced savings account. And in a way, I'm glad because, you know, eventually it'll be hackable and then I'll get it out and it'll be my retirement. I'm not counting on that, I should say.

[01:59:33] [SPEAKER_04] FBI has arrested another member of the notorious Shiny Hunters ransomware gang. They say this guy is the founder or one of the founders. A Canadian man. The FBI arrested in Pennsylvania. He was visiting Pennsylvania for a cyber insurance conference.

[01:59:57] [SPEAKER_04] His company specialized in handling ransomware negotiations with cyber. This has happened before with cyber crime groups. Playing both ends of the deal. You ransomware people and then you help negotiate the best deal out of it.

[02:00:16] [SPEAKER_01] I wonder what was the thing that was the compromise that let this person be found. That's the interesting story because it's always like, did it log into a VPN? Is this like an Onion exit node that was controlled by someone? I really would like to know.

[02:00:33] [SPEAKER_02] Failed to protect credentials. They make mistakes. You know, it's hard.

[02:00:36] [SPEAKER_04] Well, and the mistake they made was hacking the FBI's own data. That's when they get really serious about attacking you. They are. They're the best Shiny Hunters. And in fact, they say they hacked the FBI because the FBI was saying bad things about them. And it made them mad. And in fact, I remember they said, if you don't apologize to us, we're going to release all this data. And I don't know.

[02:01:05] [SPEAKER_04] Did the FBI apologize to them? That deadline passed. Seems unlikely. It seems like that. I didn't hear about it if they did.

[02:01:15] [SPEAKER_01] They did say we are not going to be releasing the data. They did that.

[02:01:19] [SPEAKER_04] Oh, they did. Yeah. Well, maybe now they regret hacking the FBI. The guy arrested, I won't say his name, published a book last month called Cyber Extortion Strategic Response. He's an expert. I don't want to say his name because there have been in the past leaks by the FBI of people's names as suspects who were not.

[02:01:46] [SPEAKER_04] But in fact, suspects, Kash Patel, the FBI director, notorious for that. I think this is his press release, so I'm not going to say any names. But they were pretty happy they got this guy. I will say the name of the assistant director for the FBI cyber division because his name is Brent Leatherman.

[02:02:08] [SPEAKER_02] Nice.

[02:02:09] [SPEAKER_04] And he says the hack was the result of a security failure, the FBI hack of a platform managed by a third-party organization. Not our fault. A contractor that worked for Accenture. So, yeah, I didn't hear that apology. So maybe this is how they're responding. Let's just arrest them first.

[02:02:33] [SPEAKER_01] I wonder if they have someone on the inside, probably, if they haven't said how.

[02:02:37] [SPEAKER_04] They must. I wonder. You know, they're starting to roll them up. I think this is the third or fourth arrest of shiny hunters, active shiny hunters participants.

[02:02:45] [SPEAKER_06] Somebody was better negotiated than negotiators. Yes.

[02:02:49] [SPEAKER_04] I'm going to make you an offer you can't refuse. The midterms coming up in just a few weeks. And one of the big issues is data centers. And it seems to be one of those issues which could get you elected or not. And, as Molly Taft writes and wired, these midterm races will decide the political fate of data centers themselves.

[02:03:19] [SPEAKER_04] From the Senate to governors' mansions, candidates are trying to out-compete each other to be the most anti-data center, including the governor of Texas, who is up until about a moment ago very pro-data center, but now realizing that's not so good for voters, says, no, I don't like data centers at all. The National Republican Senatorial Committee issued a memo in August

[02:03:47] [SPEAKER_04] warning that in the Ohio race between Sherrod Brown, who's running for Senate, and his opponent, the incumbent John Husted, Sherrod Brown's hurting Husted so hard on data centers that it could actually tip the outcome of the race, according to the committee, calling it, quote, a sleeper issue for the entire election cycle. Sherrod Brown has achieved some success calling Husted the face of data centers in Ohio.

[02:04:14] [SPEAKER_04] It used to be good politics. You give a tax break to a company, they build a data center, there's lots of jobs, there's lots of investment. It used to be. And so a lot of politicians, you know, were in favor of data centers until they realized their voters hated them. Be interesting to see what happens. New York State has a moratorium in place.

[02:04:42] [SPEAKER_04] As a result, the governor who's running for re-election has a comfortable lead against your challenger who supports data center development. I'm not running for office, so I think I can say safely, I am for data centers. I'm not for putting a data center in people's backyards. I'm not for putting data centers in deserts or places where water is scarce. But I think data centers are a good thing. Am I, you want to yell at me now?

[02:05:11] [SPEAKER_06] But maybe the pressure on them right now is a good thing too, because it does force the companies to think about all those issues with electricity consumption, water, how many jobs are actually getting created. And there's some interesting, like, if you only look at the energy component of it, there's interesting things starting to happen. I think Google has some pilot projects where they sort of, like, have this idea of the distributed grids

[02:05:36] [SPEAKER_06] instead of putting one more big, I don't know, factory-sized power plant next to it. They just, like, seed the community with solar so that all the houses around that use less electricity. Yeah, there you go. So there's, like, but that only happens if there's enough pressure, right? That's right. In a way, all these people saying no to AI, no to data centers, may just what this industry needs to clean up its act.

[02:06:06] [SPEAKER_04] Yeah. I mean, I'm for the pressure. And you absolutely shouldn't, it shouldn't be legal to power a data center with propane generators. That's clearly a bad thing. It shouldn't be in anybody's backyard.

[02:06:17] [SPEAKER_02] I think this is where the moratorium comes into play. Look, there are EPA rules against putting generators in areas that have emissions with no controls. They're just not being enforced. Enforce them. But perhaps in two years they will be. Yeah. Well, maybe that's what the result of the election is all about. And also, like, if you put a moratorium in for a couple of years, then when you have a functional regulatory body again, you can lift the moratorium. Yeah. That's a good point. Yep.

[02:06:47] [SPEAKER_01] But it's also, like, data centers, they are not all the same. They are bespoke to the location and the power grid. But the lack of transparency about how these have come about, where they're having these local municipalities sign these NDAs, and then people feel like they don't have a choice on what it is, what it looks like, and how it will affect them. The right questions aren't necessarily being answered.

[02:07:16] [SPEAKER_01] And some of these large installations are also coming with tax breaks in which they are circumventing some of the advantages of having a large business come in because instead of paying taxes, they're choosing not to. Not because that's a way of attracting them, but because it's part of the table stakes that they negotiate with the large, the lawyers who have been doing this over and over again with different locations

[02:07:45] [SPEAKER_01] against a municipality where this might be some of their first foray into having these negotiations with these very mega large companies. And so they're able to take advantage and even if the local population is for it, they feel that they could have extracted more of a fair deal and they're not giving that chance when they are able to be able to chime in on these negotiations. And so it's not just data centers, it is the people running them,

[02:08:15] [SPEAKER_01] it's how they're being discussed and how they're being negotiated that people feel that they're being cut out of that conversation. They're not getting consent and that is the biggest part of it.

[02:08:26] [SPEAKER_02] Right. And they should. The NDAs are shocking too, right? The idea that a government official wouldn't be allowed to talk to their public about the deal being made, that shouldn't be legal.

[02:08:36] [SPEAKER_04] See, I like data centers though. I'm like, we're using one right now. I mean, I...

[02:08:41] [SPEAKER_02] Yeah, yeah. And we need them, but they need to be built responsibly. Right. And they can be. It's just that with the current surge from this crazy hype cycle, there's been folks cutting corners.

[02:08:58] [SPEAKER_04] Let's take a little break. And I promise not to build a data center during it. You're watching This Week in Tech. So glad you are here. We do the show every Sunday afternoon from 2 to 5 Pacific and you're invited to watch live if you want. We stream it live. YouTube, Twitch, X, Facebook, LinkedIn, Kick. Best way to watch though with the other members of Club Twit. Club Twit is our way of, you know, begging you for money.

[02:09:26] [SPEAKER_04] And then giving you a benefit. You get ad-free versions of all the shows. You get chapter markers. I always hear people wanting chapter markers and we just can't put them in on our regular shows because we use something called direct ad insertion and variable length ads are inserted in different places at different times. And so we don't know when a chapter begins or ends on many of our shows. But because the club shows are ad-free, we know.

[02:09:54] [SPEAKER_04] So we do put the chapter markers in there. You also get access to the Discord, the Club Twit Discord and all the shows we do in there. Lots of great shows coming up. And mostly it helps us keep the show on the air. Advertising does not cover our costs. It only covers about 60% of our costs. The other 40% is made up by club members. So if you like what you hear, if you get value out of our shows, we would appreciate your support by going to twit.tv slash club twit.

[02:10:23] [SPEAKER_04] Thank you very, very much. We appreciate it. Our show today brought to you by Delete Me. We appreciate them. In fact, I use Delete Me. As a business owner, I know putting myself out there is why I don't have any privacy. It's just part of the job. But the uncomfortable truth can be that by promoting your business, by getting out in front of the world, having a presence, leaves you and your team completely exposed. Right now, 90% of business owners

[02:10:53] [SPEAKER_04] have their home address easily discoverable online. The average owner has more than 600 pieces of personal information just sitting on the open web. I mean, we're talking your personal email, your phone number, your home address, even details about your family. Why is that a bad thing? Well, I mean, obviously for privacy it is. But for us, it also means hackers can get this information to make their attacks more effective. Hyper-targeted phishing attacks. And we've been the subject of those. The emails from what looks like

[02:11:23] [SPEAKER_04] an actual client of ours. Or even text messages impersonating our executive team. That's when we first signed up for Delete Me. We realized the bad guy not only knows the name of our CEO, but knows her phone number, knows her direct reports phone number, and can send text messages that look like they're coming from her to them asking for money. Because they have your real details, these hackers don't sound like strangers. They sound like clients or partners

[02:11:52] [SPEAKER_04] you already trust or even management. That's why attacks using verified personal information are five times more likely to succeed. And the average incident costs small businesses more than $120,000. One in four businesses will be impacted this year alone. I know we were. And that's why we signed up for Delete Me. Delete Me reduces exposure by up to 95%. Very simply, removing you and your employees' personal information from all those data brokers.

[02:12:23] [SPEAKER_04] Hundreds of them. More than 500 of them. And when they are not available for purchase from a data broker, hackers are starved of the fuel they need to build their targets. And it's not a one-time thing. No, no. In fact, we just got an email the other day. Delete Me constantly monitors and removes your data, and you will get regular privacy reports so you always know where things stand. Fortune 500 companies and government agencies have trusted Delete Me for over 15 years, and now that same enterprise-level protection

[02:12:53] [SPEAKER_04] is available for your small business. Protect your business and your peace of mind. Go to joindeliteme.com slash twit-biz to start protecting your business with Delete Me today. And if you use that link, you'll also get a free year of social media protection for every seat you purchase. That's joindeliteme.com slash twit-biz. We thank them so much for their support. Of this week in tech, we thank you so much for your support going to that address.

[02:13:21] [SPEAKER_04] So I actually thought that the regulators and international regulators and U.S. regulators and courts might stop it, but Paramount has completed its $111 purchase of Warner, and as you said, Janko, it's all Skydance now. Hollywood did not like this idea, right?

[02:13:46] [SPEAKER_06] I mean, some people probably did. But I think there's a lot of worry about the consolidation that's ultimately going to happen. It's a company that took on a lot of debt as part of making this deal happen, and so they have to somehow... They're also on the hook for making 30 movies a year at least, and I think that number goes up after a couple years. So they're going to have to keep spending,

[02:14:16] [SPEAKER_06] and somewhere that money has to come from, and somehow that debt needs to be serviced. So there's going to be job cuts for sure. And then on the flip side, Netflix tried to buy part of this, obviously, not the whole company. They didn't want CNN, but they wanted the studio. They wanted HBO. And now Netflix didn't get that, and now this week, Netflix is going to have 5% layoffs reportedly. So it's hurting. It seems like it's a deal

[02:14:46] [SPEAKER_06] that nobody is really happy about, and that's going to just have a lot of bad outcomes across the board, even at companies that aren't part of it. So, yeah, it's not an ideal scenario by any means.

[02:15:01] [SPEAKER_04] I think this kind of media consolidation is bad for us as consumers, obviously. This is Larry Ellison's son, David Ellison. It combines two of the largest movie studios, Paramount Plus streaming service, HBO Max, CBS, CNN, Live Sports, CBS Sports, and TNT Sports. I mean, that is a big roll-up of a lot of previously... Well, were they independent?

[02:15:31] [SPEAKER_04] No, they were all owned by other big companies.

[02:15:33] [SPEAKER_06] I mean, the Warner story has been a story of like, mergers not going that well since forever, right? There was AOL, Tim Warner, there was AT&T. Oh, God, yeah, that was the worst. And then Zaslav came in and didn't really do things that well, one could argue. There was a lot of stuff like HBO Max. Oh, they changed their name like three times. Exactly.

[02:16:00] [SPEAKER_06] And suddenly had all this reality TV on there. It didn't make any sense.

[02:16:04] [SPEAKER_04] Yeah, you know, Dr. Pimple Popper next to The Sopranos. It doesn't... It's like it does not compute.

[02:16:10] [SPEAKER_06] Right, right, right. And they killed like movies that were done essentially, right? That was like his claim to fame essentially.

[02:16:16] [SPEAKER_04] So I saw Supergirl. They should have left that in the grave. They were right to kill that, but okay, if you put a lot of effort and money and time and, you know, creative juice into something, I guess it deserves to see the light of day, even if it's not the best movie ever made. Right, right, right. By the way, the $111 billion doesn't just come from Larry Ellison. It comes from the sovereign wealth funds of Saudi Arabia, the United Arab Emirates,

[02:16:46] [SPEAKER_04] and Qatar. So, yeah. I mean, do you think CNN will be able to preserve its independence? Do you think they'll have a chance to? CBS News apparently didn't. They put Barry Weiss in charge. And by the way, Nick Bilton, who was, you know, back in the day when he was at the New York Times, a regular on Twit, is running it, is running 60 Minutes.

[02:17:18] [SPEAKER_04] I have to think that the anchors at CNN, like the anchors at CBS, are a little worried.

[02:17:25] [SPEAKER_06] Yeah. Independence, you know. They put some sort of protection in there, but it's still managed, obviously, by the company and so forth. And so what does that really mean? And also, down the road, you're going to have these two entities under the same company. And ideally, you would have them compete against each other to like, right. Yeah.

[02:17:49] [SPEAKER_04] Yeah, there is a, the settlement, California sued, a bunch of states sued. They've all settled. The settlement with California requires Skydance to set up an editorial independence board for CBS News and CNN. But, Skydance chooses the board members and they report to the company's board of directors. So, nice job, California.

[02:18:11] [SPEAKER_06] It might be almost as powerful at the, what's the meta board called?

[02:18:16] [SPEAKER_04] Oh, meta, yeah, editorial board, whatever they call them. Yeah. They, they have no clout.

[02:18:23] [SPEAKER_01] I heard that Kara Swisher said that she shows up on, she goes on, had a partnership with CNN and she quit. And she said, she will not, she says, I'm not going to do it. With them. And then if you think about their, like, David, so did her partner, Sky Galloway.

[02:18:37] [SPEAKER_04] They both said, no more.

[02:18:39] [SPEAKER_01] And if you think of how the, if they, a lot of people, if they choose to leave, because of what's happening on the different properties, if things go south, it makes it very profitable for these individuals, people who already have clout, to spin up their own enterprises and then compete without the debt load that is being. Well, that's a good thing. And also keep in mind that with the interest rates going up, servicing that debt is going to be harder over time.

[02:19:07] [SPEAKER_01] And so the, it, they have some problems ahead of them that I, hopefully that they thought through to be able to make sure that they come up on the other end of this. And I think what they say, like Time Warner has changed so many hands. It's like the poison pill of different entertainment media. It's like anyone who acquires them is like, don't, don't fare very well. And so this will be interesting nonetheless, but I'm glad this is not,

[02:19:37] [SPEAKER_01] something super serious in terms of, I mean, in the debate of media, I think it's important that this is something that stays, something that we can trust and something that does the job of even distracting us from what's going on in the world. But, in terms of how this can be corrupted, at least there are outlets for other people to try to get the word out of this or even do entertainment well if this isn't going well.

[02:20:05] [SPEAKER_04] You could even argue that this has happened because mainstream media is losing out to new media, to YouTube and TikTok and other creators. And I don't think that's a bad thing at all. Well, I'm conflicted because I come from a background of mainstream media where there were editorial ethics, there were rules, and those rules no longer seem to hold in the YouTube world.

[02:20:33] [SPEAKER_04] Rules about editorial independence, about conflict of interest, disclosure, all of those things.

[02:20:40] [SPEAKER_06] And many people can go and do their own thing and I'm doing it too, right? Yeah, you did it.

[02:20:45] [SPEAKER_04] So did I. I mean, I'm in that group, yeah.

[02:20:48] [SPEAKER_06] But if you have a traditional news organization like CNN, I mean, they got famous from covering the Gulf War, right? So they had correspondents all around the world. Like even having these correspondents bureaus that are like all these news organizations are shutting them down, consolidating them. And if you listen to the news and there's like news about something happening in India, but somebody is reporting it out of, I don't know, Vienna or something that is because they don't have any local people anymore. And that's only going

[02:21:17] [SPEAKER_06] to accelerate with this if they have to like cut roles, if they have to save money at CBS and at CNN, that's going to be like what's at stake. And that's hard to replicate for somebody with a sub stack.

[02:21:33] [SPEAKER_04] True. And the debt is not to be ignored. I used to work for Clear Channel, which used a huge amount of debt to create a giant radio conglomerate and had to go bankrupt twice. Had to go to the lenders and say, we're not going to be able to pay you back. You want us to succeed. You want us to continue, right? Or you'll get nothing. And we're able to renegotiate their debt, I think twice. But that's kind of the way of the world now where you have companies taking on huge amount of debt

[02:22:02] [SPEAKER_04] to leverage these acquisitions. And let's not forget, this is all intertwined because Larry Ellison of Oracle has big investments in OpenAI and other AI companies. So it's not hard to imagine a house of cards all just collapsing into a pile. And then we got 52 pickup. You know, at the same time, I think, well, look at the Academy Awards are moving to YouTube, right, Yanko?

[02:22:31] [SPEAKER_04] And the Emmys, the television awards, just as they're moving to Amazon Prime, which is a little weird. So I think, you know, streaming has taken over from cable for sure. And we're just in a new media world. So a lot of these, they call it linear. Like CNN is a linear channel. A lot of these linear channels. And the normal thing

[02:22:55] [SPEAKER_02] that happens in a bankruptcy like this, you know, you've been trying to do this roll-up to make things more efficient. Then you realize you don't have the money to do it. So it fails. And then it gets parted out. Right. And the pieces that can stand on their own will get bought and utilized. And the ones that can't will die. And AOL gets bought by Bending Spoons.

[02:23:15] [SPEAKER_04] And that's probably what's going to happen. Bending Spoons is going to own CNN and CBS. Who owns time? Spotify is getting involved in a new war on fair use. And our good friend, Jason Snell, has been caught in the crossfire. Last month, he wrote on his Six Colors blog,

[02:23:44] [SPEAKER_04] I got several emails from Spotify, all pretty much the same. Their incomparable podcast, which covers things like TV shows, is being yanked because it infringes the rights of a third party. They got a DMCA takedown, in effect, something we've suffered from here on Twitter frequently.

[02:24:13] [SPEAKER_04] It was actually a podcast called Click It Cast that covered books by child books author Beverly Clary. Episodes started with the host reading the first paragraph about the book to kick off the discussion. 35 seconds, 45 seconds of quoting the work. Spotify said, no, that's not fair use. It is fair use,

[02:24:39] [SPEAKER_02] by the way. That's exactly fair use. But these were not DMCA takedowns. It says our automatic review tool determined this. So it's like content ID on YouTube. Yeah, exactly. They built some fancy tool to try and detect any copyright claims before they've come in, and now they're having problems.

[02:24:57] [SPEAKER_04] Yeah, although it's the same thing with content ID on YouTube. It is, in effect, a DMCA takedown because what happens is a company, let's say National Geographic, says, you know, we assert copyright on this content, add that to your content ID, and then we would get a message if I played a picture from National Geographic or video saying, hey, National Geographic is asserting that this is their material. In almost every case on YouTube,

[02:25:27] [SPEAKER_04] the company that's asserting the copyright has a choice. They can insist on a takedown. They can say, no, demonetize them. We want all the money they made from that video. You know, they have a variety of remedies. I'm not sure how the Spotify one works. It's not technically a DMCA takedown, but it has the same force of law. And it ignores fair use because here's the problem with fair use. It's a defense.

[02:25:57] [SPEAKER_04] So the only, you could fight it, Jason could fight it, and go to court, pay a lawyer to get the right, and the court would probably rule in his favor because it is fair use. But he says, the problem is that fair use on the internet has been stifled by content detection algorithms that don't care about your fair use. It just notices you use something that belongs to someone else. On YouTube, this can lead to account deactivation, loss of ad revenue, even the hijacking of your ad revenue,

[02:26:27] [SPEAKER_04] as I mentioned, and the diversion of it to someone else who claims they own the rights to things you are using legally. It is a chilling effect. We no longer, as Benito will tell you, he won't let me play anything. Clips. We could get in trouble for playing that AI movie trailer, the gossip thing.

[02:26:46] [SPEAKER_02] Have you paid attention to what's happening on YouTube these days with the copyrighted material? No. Often they'll show copyrighted material but add another subtext over top of it, just a low voice talking on it, because it breaks the automatic detection. It breaks the ID.

[02:27:01] [SPEAKER_04] Yeah, I mean, I'm not allowed to do a lot of stuff, I guess maybe because we're higher profile, I don't know, that I see people getting away with on YouTube all the time. So it seems oddly enforced. Maybe they're using that.

[02:27:13] [SPEAKER_02] Sort of recognize that all it is is automated tools.

[02:27:16] [SPEAKER_04] It's just not, and it's not a very good one. There's nobody filing

[02:27:18] [SPEAKER_02] a complaint. There's nobody actually looking at anything. It's just automated tools. And if you can break the tools, it goes away.

[02:27:25] [SPEAKER_06] I mean, I guess in the case of Spotify, what's adding to this is them pushing so hard into audiobooks

[02:27:57] [SPEAKER_01] and audiobooks. And we're doing something about it.

[02:28:01] [SPEAKER_04] Jason does point out that Spotify replied to his response and agreed it was in the wrong and restored the episodes. But he says, and I agree with him, it's a chilling effect. As a content creator, you're putting yourself at the mercy of the organization. And if they decide you're in violation, you are out of luck unless you're willing to go to court. So he's actually deleting his Spotify account and he's no longer putting his podcasts up on Spotify, which we could do. We could do the same thing because Spotify is a tiny fraction of our audience.

[02:28:33] [SPEAKER_04] But I can't do that with YouTube. It is not a tiny fraction of our audience. It's a significant fraction of our audience. And so they kind of have you.

[02:28:43] [SPEAKER_02] And YouTube has that three strikes policy. Like if you, if they've decided you're in violation and refuse your appeal and it happens on the third time, your account is gone. Right. And I cannot afford, we have,

[02:28:54] [SPEAKER_04] I don't know how many accounts, all of our shows have channels on YouTube. We have, you know, I have a little YouTube silver button for 100,000 subscribers. I cannot afford to piss off YouTube. And so the effect is that even if there is no right to this or a fair use should protect it, should protect it, it doesn't. It has a chilling effect and we just don't play music or clips.

[02:29:19] [SPEAKER_02] And you just described the chilling effect. You don't do any of this.

[02:29:21] [SPEAKER_04] You don't, you don't take, you can't take the chance.

[02:29:24] [SPEAKER_00] Yeah. So I follow this conversation a lot. And so there's a big YouTube music creator. This is Benito,

[02:29:29] [SPEAKER_04] our producer, by the way, who is the guy who sits there with his thumb on the button to keep us whole. Thank God. Yeah.

[02:29:39] [SPEAKER_00] So there's this music creator named Rick Beato and he's like one of the bigger music YouTube guys. He's good. So he hired, he hired Legal Eagle, he's friends with Legal Eagle, the other YouTube creator to fight all of his fair use claims and he won them all. Good for him. He won them all.

[02:29:53] [SPEAKER_04] Of course he did because they are legal. It is fair use. Especially ours. This is a news channel. You know, you see clips on CNN all the time. You see clips on NBC of CNN. You see clips of CNN on NBC. This is, this is what fair use means. It's particularly in news, but I, I'm not as big as Rick Beato.

[02:30:15] [SPEAKER_01] They might start doing stuff like you've seen these Instagram posts where they'll have like a primary account and then they'll have an alt account. Oh yeah.

[02:30:24] [SPEAKER_04] Yeah. You have to do that.

[02:30:26] [SPEAKER_01] And then they'll use that as a way of, they'll post stuff there and if it gets taken down, they'll, they'll use it as testing. I don't, I guess, to see if it gets taken down at this account, they don't know that it got flagged and it'd be caught. And so that alt account gets closed down instead of their main account. And then by the way, Benito, I don't know if you noticed this, but I only see one fan in your background. I don't know if that's subtlety in terms of the, we're talking about Spicey.

[02:30:54] [SPEAKER_04] Benito has many fans.

[02:30:55] [SPEAKER_00] He is. Okay. Many fans. I live in the, I live in the Philippines. It's very hot here.

[02:30:59] [SPEAKER_01] Okay.

[02:31:00] [SPEAKER_04] Yeah. Just not, we're going, we're coming your way and apparently typhoon season has gone on and they said, just sent me all these articles. We're going to be in Bangkok a week from Tuesday. And she says, I don't know if we should go with the flooding and all sorts of stuff.

[02:31:16] [SPEAKER_06] So to add one more thing to the whole takedown and whatnot, content ID story on YouTube, adding to this, it's going to get more complicated once you add like AI likeness detection to it. Oh God, yes. YouTube is already starting to make available to creators. And I think they haven't quite like made this widely available or I'm not sure like what you can actually do with it. But imagine at some point, that's coming for sure. It will also like roll this out

[02:31:45] [SPEAKER_06] just like content ID. Yeah. And then there might be also an option. Like if I have an AI Leo in my YouTube show, then you may be able to monetize my show or something like that in a weird way where suddenly like the AI generated stuff then makes money for the original creators.

[02:32:02] [SPEAKER_04] There is, I mean, look, there are people who take our content and repackage it and put ads in it and stuff like that. I mean, this stuff happens. And so there's legitimate reasons for this. You can do an AI Leo if you really want to. I don't think you really want to, but you could. But there's also, yeah, it's just, it's a very blunt instrument and it's used as a hammer, unfortunately, against content creators.

[02:32:31] [SPEAKER_04] The U.S. Army has been testing AI warfare in the desert. Not going so well. The Anduril drones falling out of the sky because it's so damn hot. The U.S., this is the Wall Street Journal, under a blistering desert sun, the U.S. Army strained to show it had become a modern AI-powered firing force. Some blamed the heat. It was July in the Mojave Desert. Temperatures 110 degrees.

[02:33:01] [SPEAKER_04] Servers overheated. Batteries died and couldn't be recharged. And the drones, which were supposed to loiter overhead, landed. I never really thought about that part. AI not so good in the hot desert sun.

[02:33:23] [SPEAKER_01] Fortunately... Anduril's part of that like new group of superintelligence. Drone companies.

[02:33:28] [SPEAKER_04] It's Palmer Luckey's company, the guy who created Oculus Rift. Yep. And he's making drones for the Army.

[02:33:35] [SPEAKER_01] Is that a conflict of interest that Palmer Luckey's part of the group that's giving recommendations on...

[02:33:42] [SPEAKER_06] What stuff to buy? Eh, maybe. Yeah, funny, huh?

[02:33:47] [SPEAKER_04] I don't know what you're talking about there. What do you mean? It's a big deal. What do you mean? Amazon has launched... Google has launched its data satellite with four, count them, four TPUs in the sky. But it's just a test to see if it can survive the very harsh environment of outer space. Right? Cosmic rays, right? Richard, destroy these chips and stuff. Yeah.

[02:34:14] [SPEAKER_02] Yeah. I mean, it's not as... If you're flying low-earth orbit, you already have the Van Allen belts. Like, you have a fair bit of protection anyway. Oh, okay. Because you're still... Not as much as being in the atmosphere. But there's some. The bigger issue is going to be power management, heat management, like all of the normal things. And I've done whole talks on this now and really, the largest constraint when you get down to even going to scale is communication speeds. You know,

[02:34:43] [SPEAKER_02] you can't put a fiber to a satellite and optical lasers just don't have the same density. They can't even come in close. So the other problems are solvable but not particularly useful. in terms of cost. But there's just no way around the communication problems. Physics are a pain.

[02:35:03] [SPEAKER_04] Physics. Damn physics.

[02:35:05] [SPEAKER_02] Yeah.

[02:35:05] [SPEAKER_04] All right. One more break before we get our final stories and wrap things up for this week in tech for this Sunday with Wesley Faulkner from works-not-working.com, Richard Campbell of runasradio.com and .NET Rocks and, of course, Windows Weekly. You'll be back on Wednesday. Looking forward to that. Happy Thanksgiving. to all of our Canadian listeners, including Richard. Do you do turkey? What do you eat?

[02:35:31] [SPEAKER_02] We do turkey. Yeah, for sure. I love it. Turkey's good. You know, it's a good deal for us because I'm going to do turkey again towards the end of November. You get double turkey.

[02:35:39] [SPEAKER_04] That's it, double turkey. Are you going to come to the U.S. for your double turkey or do you just... We have some U.S. friends coming up. Ah. Nice. Nice to put on another turkey for. Nice. A little extra turkey is always welcome. North, if civilization has spread so far north, do you have the cranberry sauce in the tin?

[02:36:01] [SPEAKER_02] British Columbia, primary cranberry growing region in the world. You make it. Okay. Yeah. Ocean sprays from here. What? I thought it was

[02:36:10] [SPEAKER_04] from New England. There you go. Ah. So you do have the tin with the rib that you put in there. Yes. Now, do I consider it food? No, not food. But it is part of the Thanksgiving tradition. Yeah, yeah. Also, Yanko Reckers, great to have you. He is, of course, now independent at lowpass.cc. I'm glad it's going well for you, Yanko. That's great. Always enjoy your writing. Thank you. Yeah. Our show today brought to you

[02:36:39] [SPEAKER_04] by Zip Recruiter. We're always looking for ways to save time, aren't we, kids? Whether it's multitasking by eating breakfast on the way to work, maybe putting on some eyeliner, hiring a nanny for school pickups. I never did that. Or using delivery and subscription services for household needs. Guilty. We have our own time-saving hacks for everyday life, but if you're a business owner, how can you save time hiring? Zip Recruiter. Zip Recruiter

[02:37:08] [SPEAKER_04] has a new feature that quickly lets you see the most interested, qualified candidates first. So you save time by meeting the right people faster. And right now, you can try it for free at ZipRecruiter.com slash twit. Zip Recruiter's powerful matching technology finds qualified candidates quickly. Candidates can tell you in their own words why they're interested in your job. No wonder Zip Recruiter is the number one rated hiring site based on G2. Save time

[02:37:37] [SPEAKER_04] and meet great candidates sooner with Zip Recruiter. Four out of five employers who post on Zip Recruiter get a quality candidate within the first day and you can try it for free at ZipRecruiter.com slash twit. That's ZipRecruiter.com slash twit. Meet your match on Zip Recruiter. Thank you so much for supporting This Week in Tech and I have a very happy wife I think. Yep, she's doing the dance. I had the

[02:38:07] [SPEAKER_04] little game on in the background there and saw a muse ad right in the middle of the football game. Lisa has a team that she likes a lot. That was a tough tough game. She says we're 5-0. So we've all suffered from Ramageddon. Well, the AI boom according to Rest of World is also making the world's cheapest smartphones disappear. Now, by the way,

[02:38:37] [SPEAKER_04] they say the AI data centers are driving the demand for memory chips pushing phone makers to raise prices and abandon some of their most affordable models. I would agree with all that except the AI data centers. I am not convinced. I'm curious. I am not convinced that the demand from AI data centers many of which have not even broken ground yet is the problem. Mike Giannotti, we've had on Intelligent Machines is a very active local AI guy who works at Microsoft,

[02:39:07] [SPEAKER_04] said that he doesn't even believe NVIDIA is supply constrained. He said, they buy the chips, but they control the market. Maybe this, do you think this could be maybe just a little greed on the part of these companies? Or does everybody accept this notion that there's a shortage of AI memory chips and it's AI data centers driving the demand?

[02:39:36] [SPEAKER_01] Yes. It's all of the above. So there's greed and there's greed.

[02:39:41] [SPEAKER_04] Prices aren't going back down when there's plenty of chips, I'm sure. But also, when you're doing manufacturing,

[02:39:46] [SPEAKER_02] they were too cheap back in the day.

[02:39:47] [SPEAKER_04] Yeah, they were. They were under.

[02:39:49] [SPEAKER_01] But,

[02:39:50] but,

[02:39:50] [SPEAKER_01] yeah, wasn't there RAM dumping back in the day? Yes. do that again.

[02:39:54] [SPEAKER_04] Japan. Let's dump some RAM.

[02:39:57] [SPEAKER_06] And cheap interest rates over at it.

[02:39:59] [SPEAKER_04] Yeah. But seriously, I mean, most of these data centers are not built yet. Are they buying the chips and putting them in a box?

[02:40:09] [SPEAKER_02] They're pre-ordering. Pre-ordering. Most of these chips aren't even delivered. They're not built yet. You remember the Western Digital CEO, the beginning of this year said all production for 2026 and most for 2027 already sold.

[02:40:22] [SPEAKER_06] And then if you have these pre-orders and all these manufacturers, the chip manufacturers are switching to make those types of chips, then there's less left for those low-end smartphones.

[02:40:32] [SPEAKER_04] So that's really part of it. Yeah. And volume discounts. HBMs.

[02:40:37] [SPEAKER_01] If you get them at quantity, you can get them cheaper. And so the higher, the higher priced, less quantity becomes higher priced.

[02:40:46] [SPEAKER_04] So the Chinese, this is from the rest of the world, smartphone makers which have championed affordable feature-rich devices and account for 60% of all smartphone shipments have drastically reduced their entry-level projects. And again, it could be, oh, we could charge more. Why are we making, why are we making these so cheap? I'm not convinced. I'm just saying, I'm not convinced. All right. I'm just raising that. Maybe I'm wrong. Maybe, yeah, maybe there's a massive shortage.

[02:41:16] [SPEAKER_04] I just, I feel like there's gouging going on at this point.

[02:41:20] [SPEAKER_02] I'm sure there's plenty of both. Yeah.

[02:41:23] [SPEAKER_01] Or it could be that the lower end market is, doesn't have the economic drive as it used to.

[02:41:30] [SPEAKER_04] There's no incentive to make those.

[02:41:31] [SPEAKER_01] So it's like a K-shaped economy. The only purchasers are those on the upper end. So if you have limited capacity, limited design skills, why not move most of your internal resources towards the places where people actually have money to buy stuff?

[02:41:48] [SPEAKER_04] Right. Well, we're not building data centers fast enough according to The Wall Street Journal. So I don't know where all this RAM is going. Maybe it's going to SpaceX which is going to start its own cell phone company. They've been launching satellites like crazy and borrowing like crazy to do it. Here's another example of an imminent crash. Oracle, Broadcom,

[02:42:18] [SPEAKER_04] and SpaceX are in talks now for record debt with Apollo, Blackstone, and Goldman Sachs financing mega deals worth tens of billions of dollars each to build AI chips. Broadcom trying to get $50 billion in financing to build the chip that they promised to build open AI. What is that? It's jalapeno, right? Oracle in talks with Apollo and Goldman Sachs to arrange money for a big purchase

[02:42:47] [SPEAKER_04] of chips. SpaceX talking about a $40 billion borrowing to buy NVIDIA chips. It's puzzling to me. I mean, there's this shortage and everybody's scrambling to get money to buy more.

[02:43:10] [SPEAKER_02] The quantities are huge. They're all pre-orders. Right. You know, pretty much, I talked to the guys who are building data centers. The moment you've got the land deal in place, you order all the equipment. Yeah. Even though you're three years away from actually being able to install it?

[02:43:25] [SPEAKER_04] In three years, wouldn't it be better to wait?

[02:43:27] [SPEAKER_02] It doesn't really matter. You need to be in the pipeline. What you don't want is to actually have the building built and no hardware to install.

[02:43:34] [SPEAKER_00] That would be worse. It's the reason toilet paper disappears first in the supermarket when something happens.

[02:43:40] [SPEAKER_04] There's not really a shortage, but people, there's a perceived shortage and so there's hoarding. Maybe that's what's going on.

[02:43:50] [SPEAKER_02] A mysterious... You're having this behavior of companies getting bumped up on the list by paying more.

[02:43:57] [SPEAKER_04] Right. Which inflates prices, which also creates this problem. Yeah. Lots of debt out there. Here is an interesting story from Wired. A mysterious non-profit run by SpaceX's top lawyer is paying the salaries of government workers. It's the National Design Studio, which is basically Doge's successor. Some of its members,

[02:44:26] [SPEAKER_04] according to Wired, are on loan from a non-profit run by SpaceX's general counsel. He's paying the salaries. That seems like a conflict of interest.

[02:44:40] [SPEAKER_01] Yeah, this will be shut down any day now.

[02:44:43] [SPEAKER_04] Oh yeah, sure.

[02:44:44] [SPEAKER_01] They won't let this fly.

[02:44:46] [SPEAKER_04] No, that would be wrong.

[02:44:48] [SPEAKER_01] Absolutely not. Not flagrantly out in the open just like this. This is just this thing that they absolutely will crack down on ASAP.

[02:44:56] [SPEAKER_04] Well, you know what? They are cracking down on and I'm mighty glad about it. California, San Francisco says it's not legal for humans to fight robots in a cage. This is from the New York Times. Man versus Terminator-like robot in a cage. California calls it illegal. I should hope so. A human fighter took several took on several robot combatants in a cage match in San Francisco.

[02:45:28] [SPEAKER_04] This seems vaguely illegal. The YouTube personality Frankie LaPena. So it's a YouTube stunt. That robot has dreads. Each robot, they're not autonomous. They're controlled by human operators. So it's like rock'em sock'em robots. Except if Frankie's head gets pops off, it is not good for Frankie. It's a six foot tall steel framed humanoid robot. Frankie is five foot five.

[02:45:59] [SPEAKER_00] But how much more does he weigh than the human? How much more does he weigh than the human? Ah, he looks like that. But also,

[02:46:06] [SPEAKER_01] when robots are taking our job, why take this away from the humans? I mean,

[02:46:10] [SPEAKER_02] let the humans fight. Let the humans fight. He did beat the snot out of a couple of those robots.

[02:46:17] [SPEAKER_04] The company putting this on is called Rec, Robot Entertainment and Combat with a K. California State Athletic Commission sent a cease and assist letter. They say, we regulate professional and amateur boxing, kickboxing, mixed martial arts, and we do not approve this event. It's a misdemeanor. Don't worry. Actually, I think Frankie's risking a lot more than that.

[02:46:48] [SPEAKER_04] And with that, on that note, on that upbeat note, I think we can... Oh, I didn't mention this, Nyeko. You put this in and I missed it. Roku sold to Fox and you had the inside story.

[02:47:05] [SPEAKER_06] Yeah, very briefly maybe. So, this deal has been in the works for a while and the shareholders of both companies are about to approve that this week. It's pretty certain that they will. But I did a little bit of a story because companies that get acquired have to file these regulatory filings where they basically lay out everything that happened along the way. We got this bid and then we talked to that company and then that happened.

[02:47:34] [SPEAKER_06] It's kind of interesting because it's very dry legalese stuff but you read between the lines you get all this color about what actually happened. Why did they sell? In Roku's case it was the CEO who was like, oh, I'm done. I'm tired. I want to get out. Give me a co-CEO. And then the board was like, we don't think that's the best idea. And I talked to a bunch of people who used to work at Roku and they're like, yeah, that wouldn't have worked. So it's kind of an interesting backstory

[02:48:03] [SPEAKER_06] if people really want to geek out on that. Check it out on my newsletter. But yeah, the deal is going to happen and Roku's going to be part of Fox's.

[02:48:14] [SPEAKER_04] Why would Fox want Roku?

[02:48:16] [SPEAKER_06] Well, I think it matters more and more to have access to the TV home screen. Right. And Roku is the biggest provider of those in the US. They're not as big in many other countries. They are starting to gain some traction in Latin America and so forth. But really being there on the first page, having the power to recommend all your services, having the power to, you know, recommend Fox 1, the streaming service they have now,

[02:48:46] [SPEAKER_06] having the power to recommend Tubi and all those things. So they're going to have an ad business. They're going to have a streaming business. They're going to have... They're not really interested in selling devices. That's just a way to get into people's households. And that has always been the same thing for Roku. They have never made money with these streaming sticks. They always lost, actually. They're so cheap. Yeah. Yeah. I mean, they're getting more expensive, too, with all that RAM stuff. But even though TVs don't make money on the hardware, it's all about

[02:49:15] [SPEAKER_06] being on a home screen. It's all about being able to sell ads, being able to have all these other companies that are on your TV give you a fraction of the ad revenue or some of the ad inventory. That's how you make money these days. And so for Fox, that's a pretty good spot to be in.

[02:49:33] [SPEAKER_04] And it kind of fits their business. It's not as orthogonal to their business as I thought. It kind of fits in with the ad business.

[02:49:40] [SPEAKER_06] And it puts them more into the leaks of Amazon and Google and so forth who are all doing the same thing, too. You were just talking about Amazon getting their Emmys and whatnot, right? Right. So everything is moving to streaming and now the home screen of the TV matters so much more. So why not buy Roku?

[02:49:58] [SPEAKER_04] Makes sense. You know, it makes me sad because Roku was one of the first streaming devices I ever used and I always liked them a lot. They got a little junked up. But still, I thought, oh, that's kind of the end of an era. Okay, good. Good story. Yanko Records. Find out more. Read it at lowpass.cc. It's great to see you, Yanko. Thank you for joining us. We appreciate it. Richard Campbell, I know I'm going to see you again on Wednesday for Windows Weekly. Have a great whiskey for you. And a great whiskey. Is it going to be

[02:50:28] [SPEAKER_04] one of those little bottles, those dram? Little bottles. One of the little baby bottles. You only get one sip. And of course, he's a host of Rennes Radio and that rocks at Rennes Radio Rennes Radio.com. Thank you so much for spending your Thanksgiving with us. I appreciate it. I hope you get a little extra stuffing at the end. Oh, we'll have some fun. And Mr. Wesley Faulkner, my good buddy, good to see you. Works dash not working. I hope it's working.

[02:50:55] [SPEAKER_01] Yes.

[02:50:55] [SPEAKER_04] It's great. But if works not working for you, then that's where you go.

[02:50:59] [SPEAKER_01] Also, for the people who join, I wanted to say that I get so many messages on my phone from friends. It's like, I was in London three weeks ago and then met up with a Twit listener and then I gave a talk at a conference a week ago and met someone else who knows me from Twit. And so I said, if you see me in public, please come grab me. It's okay to say hi. I'd love to have a conversation. Yes, yes. I say the same thing. And next month,

[02:51:28] [SPEAKER_01] I'll be at Web Summit. So if you are at Web Summit, please come and grab me and we'll have a conversation and message me online. It's easy to find me. And so thank you so much for providing a platform where I can meet all these people from around the world that are lovers of tech because you are my people no matter where you are.

[02:51:52] [SPEAKER_04] Yeah, I feel the same way. Every time I meet somebody, I feel like I know them already. We're all a family. Web Summit in Lisbon. If you're in Lisbon in a month, say hi to Wesley. I love Lisbon. You're going to have so much fun.

[02:52:06] [SPEAKER_01] Yeah, it'll be my first time.

[02:52:08] [SPEAKER_04] Oh, I love it. There's so many good things to see, eat, and do and a lot of good port. Thank you, Wesley. Actually, they have a cherry liqueur that I really like. I can't remember the name of it, but I really like it. And if you get it when you order it, you can tell them put the cherry in it. They'll put the little cherry in it too. It's good. Thank you for joining us, everybody. We do Twitter, as I said, every Sunday afternoon and you're invited, but if you can't be here for the live show, you can always watch after the fact. It is available at twit.tv,

[02:52:38] [SPEAKER_04] of course, our website. You can see it on YouTube. We have a channel dedicated until they kick us off, a channel dedicated on YouTube. Just go to youtube.com slash twit for all of our shows and best way to get it, subscribe and your favorite podcast client. You'll get it automatically as soon as we're done just in time for your Monday morning commute. Thank you all for being here. Thanks especially to our club members who make this show possible and we will see you next time. Another twit is in the can. Bye.

[02:53:07] [SPEAKER_04] Hey, everybody. The show's over, but the listening continues at twit.tv. In fact, we've got so many great shows, I don't even know where to start. How about Intelligent Machines? That's the show where we cover AI, robotics, and all the smart stuff all around you with two of the most fun, smart people I know, Paris Martineau and Jeff Jarvis. It's often philosophic. We sometimes debate, but if you're interested in AI in this world that is changing so fast, this is the show for you. Intelligent Machines, you'll find it at twit.tv

[02:53:36] [SPEAKER_04] slash I am, our website, or wherever you get your podcasts. I'll be looking for you. Amazing.