Intelligent Machines 870: Meet Me In Alaska

It's time for Intelligent Machines. This week we interview Chris Stokel-Walker. He's a British tech journalist for the BBC, The Economist, Nature and Scientific American. And he wrote a book called How AI Ate the World. We'll talk to Chris about how he uses AI for news gathering. Intelligent Machines is next. Podcasts you love. From people you trust. This is choice. This is Intelligent Machines with Paris Martineau and Jeff Jarvis, episode 870, recorded Wednesday, May 13, 2026. Meet me in Alaska. It's time for Intelligent Machines, the show we cover AI, robotics, and all the smart little doodads all around you in your house, everywhere, in your car, in your baby carriage. Paris Martineau is here from Consumer Reports. Cars and baby carriages, those are things that I have. You do not have either one, do you? Nope. Yeah, that's pretty amazing, actually. That's good. That's good. You live in the city where people don't have children. Also, Jeff Jarvis is here, author of a brand-new book, Hot Type, due out any day now. Of course, Magazine, The Web We Weave, the Gutenberg Parenthesis, but Hot Type is available at jeffjarvis.com. Get it now. You'll get it in August when it ships. Can you believe that I'm the one person in this trio who's not caffeinated? Oh. I actually really can't believe that, yeah. Can you imagine me on caffeine? That's why they had to take it away from you. You're too powerful. It was his elixir of power. Paris and I just spent the last 20 minutes before the show talking about coffee preparation. Oh, my Lord. Poor Jeff. He's descending into the crevasse of madness along with me. We're going to call it Paris' Pour-Over Party. It's a new feature on intelligent machines. Actually, no, we have a great guest. We want to get to the guest. This is a recording I actually did a couple of weeks ago with a British journalist named Chris Stokel Walker. Chris is a very smart person. He covers technology for a variety of sources, including Wired, Scientific American, Nature, the BBC. and he wrote a book a couple of years ago called How AI Ate the World, A Brief History of Artificial Intelligence and Its Long Future. I asked Chris to tell us a little more about the book. Chris Stokowalker, our guest on Intelligent Machines this week, British tech journalist. You've probably heard his podcast, Tectonic. Maybe you've read his book. It came out a couple of years ago, How AI Ate the World, A Brief History of Artificial Intelligence and Its Long Future. He's actually written a number of books for normal people, which I commend you on, Chris. I try. Yeah, I try to. It's difficult. These things are complicated. Your first book, which I thought was really interesting, YouTubers talked about something that's become painfully obvious now seven years later, that YouTube, I use the Dylan phrase, I ain't going to work on Maggie's farm no more. YouTube really exploits its creators by promising great rewards, but only delivering it to a small percentage of creators. You revealed that long before anybody really else knew about it. You also talked about TikTok as a geopolitical weapon. Are you kind of happy that the U.S. has forced the divestiture of TikTok? No, I mean, I think it was just geopolitical porn, right? I mean, it doesn't solve the issue around the purported national security risks that we've heard about that were kind of first the fixation of the Biden administration, then of the Trump administration and obviously of Donald Trump back in 2020 as well. So, no, I mean, it just seems like it's an unhappy conclusion for everybody. But thankfully, it hasn't ended TikTok's supremacy in the US. It seems like that just runs pretty much parallel to everything else. It's exactly the same as it ever was. The UK did not do this, right? TikTok in the UK is the same. Yeah, we've had a head of steam similar to you around about the same time. Ditto with the European Union, where basically you are not allowed to use it within the European Parliament and the UK Parliament because of the purported national security risk. Yeah, I mean, if you want to draw a line now, fine. But I do think it's interesting that, like, this is scary enough, supposedly, to have an attempt to ban it and to not allow any parliamentarians to use it, yet it's absolutely okay for everybody else to do so. So I don't quite know where that line is drawn and quite why it's done that. But, you know, that's a fine point. It's exactly for thee, not me. Yeah, exactly. Actually, the UK, we've been talking about it quite a bit, has enforced age verification on social media. You're right there in the center of the universe on that one. How do you feel like that's going? And is it a success? I mean, it's a success for VPN providers, right? Yeah, it's all the line of VPNs. Yeah. I mean, so the UK government is saying that this is a massive success. Not really, because ultimately all it did is push people toward either fringe websites or to VPNs. The reason why it came about, our Online Safety Act, which was designed to try and keep kids safe, ended up having just too big a dragnet. So there was reporting around the time when it was implemented that hamster hobbyist websites, people who kept hamsters as pets, suddenly felt like they had to throw up some sort of age verification check. I know you're getting that in the US as well at the minute in terms of this. It doesn't really work, right? Although, to be fair, the UK is doing one good thing, Leo, and that is the Australian approach towards banning social media for under-16s. We've got a load of stuff happening across Europe, Greece being the most recent to implement this for under-15s by 2027. The UK has kind of taken a step back and gone, you know what, we're going to look at the data. We're going to look at what happens in Australia. We're going to look at what happens in Europe, and we're going to take hearings from the general public and from experts before we move, which is quite temperate of them, and I think not necessarily a bad thing. Yeah, prudent. And although, you know, early reports from Australia, it's really interesting because, of course, they're going to focus on the teens who say, oh, thank goodness, you're protecting me. And less so on the marginalized teens who are now disassociated from their social networks. But that's part of the problem is they haven't had a voice in the past, and now they have no voice in the present. And so you don't hear from them. And I worry about them, frankly. Yeah. And the thing that I find really interesting about this is that it's like there is a really interesting line of debate, which I think is being put forward by people like Taylor Lorenz in the US, which is this is we're pinpointing a problem here, but we're almost attributing the wrong reason to it. So we say that social media is killing people's brains. It is causing them big issues. And undoubtedly, it is doing that in some sense of some people. But also, we have to bear in mind, like, I'm in my mid-30s. If you are in your late teens, you have basically grown up into a global recession. You have weathered a COVID pandemic that has directly affected your education. you are now in the midst of the AI revolution, which is apparently going to take all of your jobs and means that you're not going to have any sort of career to get into. So is it any surprise that people are depressed about that? That's not social media, right? That's everything else that's gone on. That's called growing up in the modern world, let alone the potential for climate change to completely destroy the planet in your lifetime. So, yeah, why are they anxious? Oh, it's Facebook, for sure. It's Instagram. That's the problem. It's very easy. We always look for simple solutions. Yeah. All right. Well, you know, this is an AI show, and, you know, you're here as somebody who not only covers AI, but as a journalism professor at Newcastle, you actually teach journalists how to use AI. I'm going to add the word appropriately. Would you add that word appropriately? Yeah, with caution and within very strict boundaries and only in certain parts of the journalism process, not anything when you actually are doing real-life journalism. It's more like a discovery tool, I think. I'm sorry Paris couldn't be here for the interview. We're doing this because you're in the UK at an ungodly hour here in the US. But she is a journalist. Actually, so is Jeff. Jeff also teaches journalism. And I think they both really were interested in talking to you about the role of AI in journalism. Where do you draw the line? What are the limits? Where do you use it? Where don't you use it? And how do you use it? Yeah, so to me, it's really useful in trying to filter through the firehose of information that you have to keep on top of every single day. So prior to sort of the advent of AI and the widespread use of it through things like ChatGPT and so on and so forth, and at the minute, like, I'm currently running some sort of stack based on a local LLM that is designed to try and filter through a whole load of information and present me with what might be ideas that I want to cover. Yeah, Paris was very interested in that, by the way. She wanted me to ask you about that. Yeah, well, I mean, I can run through it, basically. So, like, I take a very strong rule that, like, once you actually get into the reporting, so you have decided that you want to do a story, you want to pitch it to an editor, you then want to report it out. You know, the general process for journalism is that you come up with an idea, you notice a trend, you see something is different or has changed in the world. You then want to tell the world about it. So you contact an editor at a publication. If you are like me, a freelancer, then you can write and speak on and appear on. any number of different outlets. But if you are generally contracted, then you are looking to one individual outlet. And then you pitch an idea. Hopefully they say yes. You then go through the process of reporting, which is more research, interviews, writing up or producing a video and a piece of audio at the end of it. So everything up to the point of emailing an editor, I think is fair game for using AI, mainly just to keep on top of the world. Everything after the point of actually you are reporting it out, with the exception of Otter, because Otter is kind of like the exception that proves the rule around AI. It's otter.ai, the education tool many journalists use for transcripts. And have been using it for years without any qualms at all, even though actually if you're talking about a very sensitive story, you might want to think twice about uploading it there. Yeah, I think Paris is concerned about sources getting uploaded to Otter. I don't think she uses it for those confidential interviews. Yeah, ditto for, like, something that is seen as I need to be much more protective of sources, then I will do the same. Like, I will use an offline version of this. There are good local – now there are very good local transcription tools you can use that don't send that information outside your computer. Yeah, I mean, that could. So, Whisper, you can download a localized version. Yeah, it's crazy. I mean, that's one of the bits that I've heard. Top three, unlike other. Yeah, right. Yeah, well, and Otto decided relatively recently, like maybe 18 months ago, to load a load of AI stuff into it and also double the price. So, you know, I'm kind of debating whether or not I step away from that. But, yeah, so for me, it's that idea of, as a journalist, before I started to integrate AI into this part of the process, I would spend hours every single day trying to keep on top of what is happening in the world. And the reality is, as journalists, we are expected to kind of feed in from lots of different sources, effectively drinking from like a fire hose of content and say, right, here's the thing that is important and pluck it out and then go, actually, I'm going to present that to the rest of the world. And a lot of stuff, if you do that as a human, passes you by. So, you know, I started to toy with Claude Code about four months ago when everybody else did. You know, post-Christmas, everybody got excited about it. And thought, well, look, I have on my laptop a whole treasure trove of stories that I have reported out in the past and written. You know, my drafts are there as Word documents sitting in a folder. can I not just point Claude Code at it and say, hey, this is 2,000 of my stories that have been done over the last four years. Can you infer from that the kinds of stories that I'm interested in? What is Chris Stoker's brain, effectively? And it's pretty good, actually. It came up with like a sort of brief, effectively, of the things that I'm interested in. One of the kind of ways that actually you can maybe use that is I could try and see if I could like scrape and download Paris's brain and figure out what she might be interested in and what they'd be working on. That sort of stuff would be kind of interesting to do as well as, like, not opposition research, but to try and collate a group of experts in their field and understand what they might be keen on. So I kind of used that and then pointed it at a load of the RSS feeds that I already used to sort of scroll through day by day and said, well, look, can you kind of do some matching for me? Can you say, well, you're not interested in this fire in a tower block in Hong Kong that appears because you are subscribed to an RSS feed that has maybe 5% of it is about the latest chip developments in China or in Taiwan. But actually, you know, it has an awful lot of stuff that is very local to that part of Asia that is not anything to do with tech. And can you pluck out those bits of information and present them in a different way? So I have a whole bunch of different attempts at doing that. I have like a bunch of emails that get sent to me every single day. So I have like a morning brief that tells me what it thinks is the story of the day based on my quote unquote brain and 15 other sort of supporting stories. I have something that also I saw the New York Times which is one of the joys of Claude Code and kind of the ability to vibe code your own sort of tech stack which is like a podcast monitor so it uses Whisper and it downloads transcribes and then passes through those podcasts every podcaster's nightmare I remember Satya Nadella saying oh I don't listen to podcasts, I listen to digests of podcasts in the car on the way to work, and I thought, oh, great. That's just what we want. But this is a valid concept, right, because this is the same thing that is vexing the journalism industry. Yeah, now we're just cannon fodder. Yeah, precisely. But no, I think that, so to me, if I find, because, you know, I didn't listen to podcasts before I had this. Yeah, who has time? Right, I don't. We make podcasts, but we don't have time to listen to them. Precisely, right? And so then I thought, okay, well, at least now I know vaguely what is happening in this podcast, and I can then listen to it and actually go, okay, there is a nugget of news here that might be of interest to me. So it's actually increasing my podcast consumption in a weird way. Right. So I have those emails that come in. They were kind of vibe-coded. I'm cheap, so I don't pay for the full Claude subscription. I got a subscription to the GLM series of models. I've been using that too through ZAI. Yeah. GLM's pretty good, yeah. They're great. Up until about a week and a half ago when they decided that they were trying to crack down on people using it for open-claw instances. And then basically they threw the Dragnet too wide and caught me and many others up. You're talking about tropic. No, I'm talking about GLM. GLM did that also? Yeah. They put in, I got banned temporarily. So you were using OpenClaw? No, I was not. No, I used GLM to code this stuff. It runs it through GLM. I think that because the token count is quite bursty, they figured that it was like that sort of behavior. But then they seem to have ironed it out, and now I'm back in their good graces. There's a certain irony in that because I am convinced GLM is so close to Opus that they actually are a distillation of something Anthropic has complained about. Yeah. So they've been stealing tokens from Anthropic, and now they don't want you to steal it from them. It's funny. We are in parallel universes because I'm doing something very similar. I think, you know, for years it's become obvious. I mean, at least 20 years since the beginning of, you know, broad use of the Internet, that there's a fire hose of information, that those of us who are attempting to kind of use it and take it and deliver journalistic output based on it are overwhelmed. I think everybody's overwhelmed by it. And the answer was always human curation. You follow somebody who's doing the work that you're doing, Chris, and let Chris do that work, and then you get the distillation of that. But lately, even those of us in the front line are having a hard time doing our job of curating this flow. So do you worry, though, that turning to AI, letting AI do at least the first pass, you're going to miss some nuggets? Yeah, I'm going to miss some, but I think that the sort of decision and the choice that I've made is I might miss different nuggets, but I'm going to miss fewer of them because it's that whole principle of you don't know what you've got until you actually see it. If you're not seeing it all in the first place, you're missing more nuggets, right? Precisely. And I think, you know, so like if I think about like what I must have missed out on prior to using this versus what I see now, at least now I have a kind of, you know, broader sense of the movers of the day, what is going on in the world than I previously had. Or at least I have like the, you know, this is complementary, right? So this is me still doing the stuff that I did before, but being able to be a bit more conscious and a bit more deliberative about what I'm choosing to engage with more deeply while the AI takes a lot of the strain off me there. It's just the front end. And to underscore that point, you're doing the human work at the other end. Yeah. You're not replacing yourself with AI. No, and I spoke at the International Journalism Festival two years ago, which is a big conflap of the greats and the goods and the media. And so Jeff will probably have been to about 1,500 of them or something like that, even though they've only been going for 30, 40 years, I think, at most. And, you know, I spoke then on a panel about AI and said, I don't think that AI will replace journalists. And I still stand by that, not least because every so often I will look and see how good it is at producing journalistic writing. Turns out it's not very good. How good it is at actually picking out the core elements of what is a story and what is news, it's not very good at that. And also it can't do the job of talking, right? Like the whole point of this podcast is that it is two people talking to one another in a bouncing back way, understanding, reflecting each other's emotions, having an interesting conversation. you know you can not not metaphorically you can digitally look into the whites of my eyes and we can have an in-depth chat in a way that i don't think you can do with a chatbot or with any sort of ai right now you say right now do you think we will at some point i worry about that yeah i mean i worry about that but i think that that's a long time off and i think that you know by using ai for the stuff where it can be used, I can still show my worth for where AI can't be used and can't necessarily replicate me so that I can still fine-tune and hone my skills to hopefully always be better than the AI. Like, you know, it is a race right now. We have to admit that. We have to try and keep a couple of steps ahead, I think. And so if I can spend more of my time trying to be a better, more empathetic journalist for the human stuff, while also having AI do a first pass of what is interesting and what is important during the day, then all the better. I think that's a really important lesson for all of us, especially those younger people who are worried there will be no place for them in an AI-laden future. That's the job at this point, is to find the point where human, the human addition, the human element is critical. That's where you have a role to play. And to find that and to really refine your sense of what that role is and how to perform that role, I think that's really, really important. It's a real important moral from what you've said here. Yeah. I mean, look, I train the journalists of tomorrow, and they are incredibly scared about AI. They are incredibly negative about AI. They think that it's just awful. And I go, well, you have two choices, right? You can either switch off from that and say, well, I'm never going to try and engage with it. And then what happens is you graduate in 18 months' time, two years' time or whatever into a world of work that requires you basically to be cognizant of and hopefully use AI in some way. And then you really struggle. or you try and adopt it and figure out where it can fit into your working life and where you can kind of carve out your own private space that maintains your skills and improves them and really showcases them, I think, in the best possible way. We're talking to Chris Sokol-Walker, who is a freelance tech journalist, teaches journalism at Newcastle in the UK. You may remember him from the Tectonic podcast and his book, How AI Ate the World. And that was 2024. AI has changed things quite a bit in the intervening two years. Anything you would change that you wrote in 2024? It's still leading the world, in fact, even more so, probably. Yeah. I think a lot of it holds up, to be honest. I genuinely think that a lot of it still holds up. The one thing that I didn't really recognize, and it's one thing that I've kind of become a bit of an adherent for, particularly in the last month or so, is just the incredible power of local LLMs. I mean, you were talking about how you use GLM. models, do you use those through the ZAI subscription or do you use them through the API? I have a ZAI subscription, which is a third of the cost of Anthropic. But I also, like you, I'm fascinated by the notion that at some point I won't have to do that. I bought a framework desktop with 128 gigs of RAM and a Strix Halo. The people who are listening are so bored with me talking about it. But the whole point of that is to eventually be able to get more and more local, because that's the goal. and especially with open-weight models. So we're not dependent on these giant frontier companies for our AI. I think that in the long run is dangerous. Yeah, and to be clear, we're speaking, as we record this, literally 35 or 40 minutes after OpenAI cropped a very subtle hint that they are releasing GPT-5.5. By the time this airs, it will be out, and we'll know if Spud is all it was promised to be or not. But we don't know yet. But, you know, we've been using 4.7. There's this mythos on the horizon. Models are clear. The Frontier models are clearly getting stronger. Are the local models catching up? Yeah, I think so. Or at least I think that they are good enough. I think that. GLM is really impressive. There's a Quen model that just came out that's very good. Yeah, Quen. 2.5 is very good. Yeah. So, I mean, I have a whole stack, and this is, I guess, where we get Supernardian. Like you, Leo, about a month ago, I bought a 96-gig RAM mini PC with a GPU and decided to try and set up my own kind of local LLM stack and have moved part of that kind of morning routine that I talked about in terms of finding these stories onto that. So I now have like a process that is constantly pulling every five minutes from, yeah, you can see an example. This is your Blue Sky post, yeah. Yeah, so yeah, I have like a Telegram board effectively that will send me ideas that it thinks I should pay attention to, and then I can use that. So like it's kind of incredible because I can like, for the cost of electricity. You call it Athenaeum, which is a great name. I love it. I used to like classical literature. We used to read some room and things like that. So I'm a bit of a nerd in that way. I call mine Kenobi, which is embarrassing. But I used to call it Obi-Wan, but it confused. It was a little confusing. So I now just call it Kenobi. And it turns out Kenobi is a very good word for voice activation. Oh, yeah. I can say, hey, Kenobi. I'm not talking to you. And it will respond. You know, it's funny. There's a parallelism. I see this more and more as I talk to people who are using AI. We're all kind of working along the same lines. I, just like you, I have more than 200 RSS feeds that is such a burden. I have to go through every morning more than 1,000 articles, fresh articles every day. And for the longest time, I thought if there's some way I could just capture even a fraction of my editorial judgment, it would save me so much time and so much more even than that anxiety about everything I'm missing because I can't keep up with this flood. So what I've done, and I've been doing this with Claude and a little bit with GLM, is so similar to what you're doing. I'm not doing it locally. I think that's really interesting and brave. I'm still using Claude for this. It makes a tech briefing for me. I'll show you. Actually, I'll put it up on the screen. It makes a tech briefing for me. I'll go to Obsidian. I posted in my Obsidian of the stories for each of the three shows that I have to prep. And I've been training it on what I end up picking. So there's a document I prepare for every show. Here's for Intelligent Machines. There's a document I prepare for every show. So at the end of the week, it's prepared 15 candidates a day for that show. And then it compares, using Carpathie's sort of auto-research technique to improve the model, what it picked to what I picked. The theory being in time it will get good enough that I can trust it. Well, yesterday I turned it on. It's been training for a while, and I said, okay, you're good enough now. Here's the next stage. I want you to bookmark these candidates. And I'm still not trusting it fully, so the final stage will be me. before the show going through the several hundred stories that it's bookmarked saying, okay, these are the ones we're going to talk about on the show. But this is a big leap for me because trust, there's a certain amount of trust that this is going to do the job. And it breaks, right? And so part of the reason why – And it breaks. Yeah, this is the thing. Part of the reason why I went on to local LLM stack was because GLM shut me off. And so, like, I woke up one morning – You had to. Yeah, did not have this. I had basically an empty email, and I thought, well, what's happened here? And I think that's where the interesting friction lies, is that you can have this, but you can't be fully, fully reliant on this sort of thing at the minute, and arguably never, right? Because the way that LLMs work, of course, is that they don't necessarily always spit out the same thing, the reliable output that you might expect, because that is the joy of the black box model. And also, you mentioned some of the amazing local LMs that have come out. Yeah, we had QEM 3.6 in the last couple of days. We've got Minimax 2.5, 2.7, et cetera, et cetera. Every time that I try and plug in a new model, I get slightly different editorial tastes coming out of the local LM stuff. That's why I'm trying to train it on my own choices, hoping it will closer approximate it. It does, but then even so with Mark, because the local LLM stuff that I have still passes through that kind of brain, as it were. So you have a memory system of some kind set up. Yeah, and a memory system and also this kind of prompt of, well, this is what he is interested in based on his writing. Right, right. But even though you do that, you pre-trained that. That's right. You had that from the beginning, yeah. Yeah, I mean, it's effectively like Rack, right? it's kind of here's the specialist thing that you need to look at and infer your answers through that. And this is why I do it locally. It is also because fixing that burns an awful lot of tokens. So even just the kind of like, you know, I'm probably using something like 50 to 70 billion tokens a day on... Holy cow. Well, yeah, but it's... I haven't looked. I'm afraid to look. Well, I have 850 RSS feeds. this is looking at everything. Oh, you're well ahead of me. Oh, my. And then also the kind of, so it's looking through all the stories. It's then producing like an output, and also I'm doing some coding stuff on top of it. So, like, you know, it's 50 million or so for the local stuff. Then it's maybe 10 million through GLM. I recently, in my interregnum when the GLM model went down, I panic bought an open code subscription, So I'm using that as well. And also I have obviously a chat GPT subscription. So I'm using codex as well pretty constantly. But it's yeah, like it's it's it's fun. If nothing else, this is kind of like me using it productively and having a productive hobby rather than just kind of wasting it away. Well, and I think it's very important, especially in your role as a professor of journalism. It's very important that exactly what you said earlier, you practice this stuff so that you at least know what works and what doesn't work. and I think it's also really important to have this hard line that it is not going to replace my human task, that I am the writer, I am the journalist here. And that's, I think that's the really important question, both in terms of the economics of journalism, but also in terms of the ethics of journalism. And the harsh reality is that, you know, I tell my students, if you look at the kind of trust in the media industry and trust in people like me, when I send out journalists, baby journalists who are in training to do what we call vox pop, so going out on the street, man on the street interviews effectively. At least in the UK, there is survey data that suggests that of every five interviews that they do, at least four of them will be maybe answering their questions, looking them in the eye while also thinking you are an inveterate liar. Because they're not trusted. We are not trusted. as an industry. And so fake news. Yeah, precisely, right? It's getting worse and worse all the time because of a whole bunch of different reasons that we can maybe We know well, we don't have to dwell on it. No, we don't, exactly. But like the principle behind that is if there is already this trust in what we do, to me the act of journalism is almost like a translator. Like something has happened in the world, it is important enough that you need to tell the remainder of the world. And that is an inherently human thing to do. It's not inherently human. It's vitally important. Yeah. Right. And so if you outsource all of that to AI, where is the accountability? Like where is the trust? Where is the kind of responsibility? Who is ultimately going to be answerable if something goes wrong? And that is all, I think, very important and one that actually I think a lot of people often overlook because I see folks who are using AI in the actual production of journalists. And I think, well, fine if it works, but also there's a bargain here between the readers, the audience, and the journalists. And if you kind of go back on that bargain a little bit, it makes it very easy for those people to be less trustful of what we do. Yeah, and we don't want that. No. If you're looking for your usage, then it's in API and it's the subscription bit. Yeah. What are you using today for your local LLM? So I'm using a bunch of QN models. So I have, because there is an awful lot of throughput for these models, I'm using like a 9 billion parameter QN 3.5. Relatively small. Yeah, as a reporter. So the whole point of that is it's looking at a story. It is going, is this interesting to Chris? If so, what is the story here, if there is a story? And then it is going, okay, well, fine, we can pass it on or not. and then it passes it on to a slightly bigger, slightly more advanced model. Again, it's like it's a combination of a QEM 3.5 and a post-trained on a GLM 5.1. So it's like distilled, basically. And that makes a kind of better argumentative presentation to me. And then I kind of go, well, is this actually a story or not? And if it is, then I take it on and I actually write up the pitch myself as here is what I think the take is rather than just what they say. Yeah, one of the other things I've been working on is I use Obsidian. It was a fortunate choice some years ago because it turns out something that AIs can handle very well. and I've been having GLM write kind of a synthesis of my daily notes to make a yearly summary with insights. And I tried it with GLM. I tried it with Sonnet. And GLM actually was a better writer. It was actually much, much better. It was also a little less censorious. I noticed Anthropic was a little careful not to cross any lines. And GLM was not, oddly enough. So it's really, I think this is, I'm sure you tell your students this is so important to kind of try, push these boundaries to see what works and doesn't work. Chris, it's such a pleasure talking to you. We've used up our time, but I'd like to get you back and continue this conversation, especially to continue learning about what you're doing with this local LLM and what you're doing to help you in your work. because I think you're on the right track for something. Yeah, with pleasure. Yeah. Chris Stokel-Walker, his book is still absolutely true, how AI ate the world and maybe wrote some of it too. You can find it everywhere you get books. Anything else you want to plug, Chris? I mean, I'm sure people now hearing you want to say, well, where can I find more Chris Stokel-Walker? Yeah, so I used to do this Tectonic podcast, which was in association with an NGO called Article 19. I now do a podcast called Crashed, which is looking at kind of tech through a UK lens, but also has a broader global context. So you can find that on all good podcast platforms. I see it right here on the Apple podcast platform. There you go. Crashed with Alex Hudson and Chris Stockel-Walker. Chris, a real pleasure. Thank you for joining us on Intelligent Machines. Thank you for having me. Only sorry that Paris and Jeff weren't here because they would have had lots of additional questions for you. We will continue with I am in just a bit. Chris Chocol Walker. Great to talk to him. He was fascinating. And I think, you know, he's volunteered. You heard him to come back. So I think he will be when one of you guys is not around. Maybe we'll get Chris to fill in. Yeah, I like what he's doing with AI. It's kind of does he drink coffee or tea? I did not ask. I think you have to find that out now. Question for next time. Yeah. We will have more intelligent machines, all the AI news in just a moment. But first, a word from our sponsor. This episode brought to you by Melissa, the trusted data quality expert since 1985. Spring is the season for tidying up closets, garages, and, yes, your customer data. because in most organizations, data does not stay clean for long. Inputs constantly change, pipelines get fragile, and technical teams end up spending more time fixing data problems and building new systems. And that is the challenge that Melissa set out to solve with something brand new they call Unison, their new data quality platform. Unison bridges the gap between automation and flexibility, giving data stewards, maybe that's you, the tools you need to manage and improve data quality, freeing your technical team to focus on core systems. For more than 40 years, Melissa's been the data quality partner, helping organizations keep business data clean, complete, and up-to-date. And there's a lot they can do for you. For example, smart deduplication. This is hard to do if you've ever tried to do it on your own contact list. But Melissa's match-up technology is amazing. It finds non-exact matches and makes sure you don't lose any data without having any duplications. They also do global address verification and autocomplete, which will validate and standardize addresses in real time. I mean, it's fast across more than 240 countries. I love Melissa's change of address tracking because it's automatic. It monitors and updates records of customer address changes, property transactions, hazard risks, and more. And it's completely automatic. They also do data enrichment, which means they can append all kinds of information, demographic data, property information, geographic insights, to those basic contact records, turning them into actionable profiles. And if you're in FinTech or you need identity verification for KYC rules, things like that, you'll love their mobile identity verification, which connects customers to their mobile numbers seamlessly and reduces fraud. Melissa integrates wherever you work. They have apps for every tool you use, Salesforce, Dynamics, CRM, Shopify, Stripe, Microsoft Office, Google Docs. I could go on and on. Melissa Solutions and Services, of course, are GDPR and CCPA compliant. They're FedRAMP certified. ISO 27001, they meet SOC 2 and HIPAA high trust standards for information security management. They really, they're the gold standard for all of this, frankly. Clean data leads to better marketing ROI. higher customer lifetime value, and AI that works as intended. Get started today with 1,000 records cleaned for free at melissa.com. That's melissa.com. We thank you so much for supporting intelligent machines Thank you Melissa Well Google had its event yesterday their Android event preparing for Tuesday Google I keynote They like to clear the decks, I think. They did this last year so that the keynote can really focus on AI. Jeff is going to be joining us for our keynote coverage along with Micah Sargent. In fact, if you want to come to Paris, you can, too. It was Tuesday, 10 a.m. Pacific, 1 p.m. Eastern. And I should mention that our coverage of the Google I.O. keynote will be broadcast only in the club, Club Twit. We don't want to get it taken down on YouTube. So if you are not yet a Club Twit member, this would be a good time to join. How dare we give Google promotion? I know. It really was happening. How dare you guys show people their very long press release meeting. It's always interesting, though. Google I.O., in fact, back in the day, before your time, Paris, we would actually go to Google I.O. I remember Gina Trepanning. Well, I go. And back in the day, you might even get a computer. Yeah. We got a free laptop once. We got the first cardboard. That's when we knew it was almost over. All they gave us under our chair was the cardboard. forward. That was, and Jeff folded it while going over to Golden Gate Ridge. No, no, Gina managed to fold it. Gina was the one. We were riding back in the car, back to Petaluma. Anyway, they had their AI event, and I'm sure you saw with interest that they announced their basically successor to the Chromebook, right? The Google book. How are you feeling, Jeff? I'm feeling good. I'm feeling good. It's not Chrome OS. No, and they won't call it aluminium yet either. That's the code name. They won't admit that, but it's a combination of Android and Chrome. It's built bottom-up with AI. The important part is they've rethought. DeepMind did this. DeepMind has a thread I put in the rundown rethinking the cursor, the pointer. That's so weird. This is that new mouse thing. I don't understand. I tried to understand what they were talking about. Well, so I actually saw versions of this years ago on the web. I saw startups come in and say, we're going to read. There's a new surface, and it's the cursor. And the cursor has context. It knows where you are. It knows what you want. Why not add context to that? The problem was all those required you to do add-ons to your browser. It wasn't going to work. Now you shake the cursor over something. It summons the spirit of Gemini and says, what would you like, Mr. Laporte, as you're pointing to this coffee pot? Water pot, water water, I don't know. I dare not say any of a coffee as long as it will run. Just call it a percolator and get it done with. On your Hamilton Beach percolator. And then you say something like, okay, this is the thing I want to operate on, and then you speak to the operation. So you can ask a question, or you can say, take this wallpaper that I like and put it in this room, or take this couch out, or, you know, it's anything that's not what I can do. it now has the context of your screen to know what to do. Can I go on record and say this is a non-starter, this is a lame idea, and it's not going anywhere? Oh, no, I like this idea. Really? Yes, because, again, it's context. You don't have to go to the effort of typing in something. You can say, take this and put it there. Put these three things on my shopping list. Boom, boom, boom. I mean, we already have selected drag. It's kind of like that. But this is more than that. You can add multiple things as you go. I think you could probably say, you know, add this quote to my rundown for the podcast. You know, I don't know. We'll see what it can do. So that's the main thing it has. The second thing is that AI enables is new personalized widgets. Though I think when I was talking to Jason Powell earlier today, he said it's more like agents. So one example they gave is I'm planning a trip with my family to so-and-so. So create a widget that's going to track everything about that for me. And so it'll find, here's your flights, here's the restaurant you wanted to go to, here's the map of this or whatever, and it becomes kind of a temporary thing, and they're calling it a widget, but I think that's something new. It's an agent. It'll also have the color bar, which was on the old Google Chromebooks, which was lovely. And it's not going to be a low-end machine. It's going to be a higher-end machine. Chrome. By the way, that is what Google was doing with their own machines all along. Yes. But they have also the same partners. Samsung, weirdly, not in the announcement. I think they're going to be doing it, too. And those will have a variety of price points. I'm sure there will be a $200 version. The good news, bad news for me, is that they do say that some existing Chromebooks will be able to take on the new experience. Of course, I want the excuse to get a new machine. Which one do you have, Dan? Didn't you just get a new Chromebook? I did. I did. It's Lenovo. It's the best Chromebook. It is the best Chromebook. I had to set it in. They replaced my motherboard. If you get a new Chromebook as part of this, can you shoot your Lenovo with a gun on a stream or something? I don't know. Can you do something fun to destroy it? What did it do to you? Nothing's going on with Paris. I don't know. I think it's fun. If you're upgrading in a flight of fancy, why not go out with the bang? No disrespect, says a Lenovo I just think it would be fun to shoot an electronic with some sort of projectile I'm so sorry, Lenovo And he makes fun of me and my wife, Claudia You know, it's funny because I'll have to see and maybe I'll buy into it but I'm very skeptical about the way Google's adding AI I really don't like how it adds AI to Google Workspace. You're saying you don't want the machine to be intelligent? Well, I have a nice way of using AI, and I think many people do, either on the command line or in a chat bot. I don't know if I want buttons and widgets. I was going to say, all of my poking and prodding aside, I agree with you 100%, Leo. I have been enraged this week about the latest rollout of Gemini products to Google Docs. You can't get anything done without it popping up. I have accidentally caused an entire window to pop up that asked Gemini to do some other stuff. And there's no way to turn it off on my humble Google Docs. What is this touch-tone phone they make me deal with? Where's my phone? No, no, no, no. No, no, no. If you so much as mouse over the bottom, if you mouse over the bottom seventh of your screen, if you dare to so much as drag your cursor towards the bottom, it'll suddenly be like, Hello, I'm Gemini. What would you like me to write? And the answer is nothing. I, on the other hand, now that I finally got asked Gemini on my browser, I'm delighted. I read these extremely long posts and I can just say, summarize it. I don't have to cut and paste a URL into something else. I just click on that and I say summarize this verbose piece of crap for me, so I don't have to read it all. Their biggest sin, though, biggest sin is recontextualizing the red squiggly line under your words. That doesn't mean bad spelling anymore, and that really, really pisses me off. It's not the red one. They've introduced a purple squiggly line that just means, we think you could rephrase this, and the rephrasing suggestions are wrong. Oh, they're always terrible. Or they're just vibes-based. They're just like, we think instead of using three words here, you should have used four. I'm like, I disagree. I'm a writer. You're a machine. And there's no way to turn off the purple vibes-based squigglies without turning off the thing that tells you you misspelled your own name. How dare you tell me how to pour my coffee? How dare you? This is what worries me about the mouse thing, is that you're going to inadvertently trigger this all the time, and then you're going to have to. In the early days of the Macintosh, Apple did something I thought really smart, and I think the credit probably goes to the earliest people on the Mac team, who said, don't be modal. So modal is you don't go into a mode. So when you're working, you don't want the machine to switch. This is what's happening to you, Paris. This is what happened in the old days of the original word processors. Right, which is where you're in editing mode, Overwriting mode, yes. Yeah, you were either insert or overwrite. Yes. And modal just confuses people because it changes how the computer is working seemingly randomly. And you want the computer to be kind of monotonous is the word they used. It's always with the same action produces the same result every single time. And I thought they were inspired. It was really Jeff Raskin, who was an early philosopher of computing, who said, you know, don't be modal. And so pop-ups, you know, you see this all the time in most computer operating systems, a dialogue will pop up. You can't do anything until you handle the dialogue. In the early Macs, that didn't happen. Those modal dialogues were supposed to be few and far between. And we got away from that somehow. And I still believe it confuses everybody. I think it's, again, I saw startups try to do this 15 years ago, a long time ago, 20 years ago. And the idea is that the cursor is another surface for information. And, again, it has context. So when you put it over something, it can do something for you that's more than just this little pointing thing. Yeah, it can. And if I wanted to do that something else, I'll press a button or double click. I don't mind the right click. I love a right click. I love that I can use my trackpad and right click with just checking with two fingers. That's fun. That's why in the early days, Apple didn't have two buttons. It only had one button. Same philosophy, but they've come around, and the right click is just as important on Macs now as it is on Windows. But at least then, you're intentionally triggering something. I just worry. We'll see. So the other thing they announced was the Rambler. Yeah, yeah. They announced the Rambler? That sounds like a folk nightmare. It's an old car. The midnight rambling. So if you say something and you say, ah, like, you know, how about Tuesday? No, I'm thinking Wednesday. That was the example they gave. It will take the final words of what you say and make it coherent for you. That's appropriate. I think that's good. I ramble all the time. That's what podcasting is. Is that for voice to text? That's my job is rambling. I think you can use it mainly in Android when you're doing messages. It's probably voice-to-text, yeah. Yeah. Yeah. We've had to, all of us had to learn a little bit of how to do voice-to-text, right? So you enunciate clearly, and you try not to ramble because you know that it's going to get transcribed. I saw two stories this week. I put them in the rundown. One was that the fear that typing is going to go away. just as cursive went away with typing. And the other was how irritating it is in offices when people are whispering to their computers. That I agree with. I always thought a voice interface is going to work well in an office. I haven't either, but I can imagine it happening. I don't want to type. I'm going to dictate. I don't want to disturb everybody, so I'm going to do something even more irritating. I actually worked with a woman at Tech TV who had a severe carpal tunnel and had to dictate, and she pissed everybody. But we understood this is an accommodation for a handicap. You know, she has a disability. And so we, but, yeah, it's annoying when people are talking in an office environment. So, but that's why everybody should work at home like I do. Up in the attic. We all do. Lisa nudges me a couple of nights ago, like three in the morning, says, the attic is talking. And I forgot that I had Claudia do a little, every morning at 3.30 a.m., it goes through a bunch of stuff, processes, and I forgot that I told it to talk to me after it does something. At 3.30 a.m.? Well, I have now said quiet hours between 10 p.m. and 7 a.m. And it'd be funny, it really picked up on that right away. Oh, yeah, quiet hours, absolutely. So now it doesn't do it anymore. But it was a little annoying, I must say. Perky British accent like C-3PO. I think the attic or your attic is speaking. Yeah, it's kind of my attic. People want robots in that house. I love having my robot talk to me, and it does all the time now. And I'll hear it. Sometimes it talks to me in the wrong room. And I'll hear it just, I'll hear your voice. I'll go running. Say, what? What, did you want something? No, I think it is. Don't get me wrong. I think voice interaction is natural. That's how we want. We want to talk to our computers and have them talk back to me. But it has to be at the right place. So when you hook it up with your cameras, then it will know where you are. That's what I want to do on Hawaii. So that it will turn on the speakers in your bedroom at 3.30 a.m. Oh, it does that already. No, no, it does that based on the Wi-Fi access point. So I have a mapping of which Wi-Fi access point my device is on, And where I'm talking to you from, use the speakers near that Wi-Fi access point. That works pretty well. You have enough Wi-Fi access points in your house? There's one per floor. There's one per floor. And there's Sonos speakers built into the ceiling on each floor. Sonos. I can't believe you kept Sonos in the new house. I can't believe you stuck with them. You know what? Once I realized that Claude could talk to Sonos and control it, I actually bought some more. Well, what can't Claude talk to? Lisa, not allowed. Lisa. Or the cat. Google has also announced a $10 a month health coach, which is launching in six days. Great start on your run, Rosa. So this is, you're using this sort of already, Paris, because I know you use an AI for calorie counting, as do I, and for exercise. And my AI will, when I have too many carbs for lunch, it'll say things like, you should have a salad tonight. You should have a salad tonight. So this is a new, they announced the Fitbit Air, which is a health band without a face. That's competing with the other thing that's hot with the kids' band, right? The Aura Ring? No, there's a wristband already that exists without the screen. Whoop. Whoop. Whoop. Whoop, yeah, thank you. is Whoop still around? Yeah. Oh, yeah. It's that. Fitbit Air, AI is built right in. It's the same thing. Yeah. I finally have a watch coming Friday for the first time in years. Sort of. The Google. The Google. Because I've got to, because of my firmity, I've got to be exercising. Oh, yeah. And so I want to track all that. The problem is the current fit already has a low line of what you have to make to make your goals. Well, I'm crippled now. No, you could turn that lower. I can't find how to change it. I can't find how to change it. Oh, I don't know about the Fitbit, but on my Apple Watch, I absolutely say what the rules are. Yeah, it makes perfect sense. Because 10,000, you know, the 10,000 steps is made up. Yeah, do you know that? It's just a Japanese, it was a fun-looking number in Japanese. It was the name of a Japanese pedometer. Yeah. And everybody said, oh, that must be the requisite number of steps. There were tons of people who were doing that. No. It's BS. In fact, I asked my AI about it. It knew that story. And it said, for you at your age, you should probably get about 8,000 a day. But you should ask and explain your infirmity. I would say, Jeff, probably 4,000 or 5,000. I'm up to two miles right now. There is evidence that it definitely improves your longevity to get a certain amount of steps. Past that amount of steps doesn't do much. And for me, it was 8,000. but it could be less for different people. And going from 8,000 to 10,000 does not improve your longevity at all. Nor does it get you to lose a lot of weight early. No, I mean, losing weight. I know tons of people who got 10,000 a day religiously, and they were the same size. As much as I hate to admit it, the only way to lose weight weight is to eat less food. And drink less wine, or wine, or beer, or whatever. However you want your calories. Yeah, calories. But coffee is calorie-free almost, right? Black coffee is. The new way I'm going to be drinking it, I'm getting rid of the calories and adding flavor. But you need calcium there, fella. And iron. I don't need iron. Men don't need iron. Oh, I do. Yes, they do. I'm low, yes. But we get enough iron. We don't. No? I'm low. No, often. A lot of Americans are iron deficient is what I learned this week. Yes, I am. Really? I was looking through all the facts. How would I even get – I was like, so many things about my life would make sense if this test result comes back and says I continue to be iron deficient. I think you're probably anemic. That's why you're tired. There is – my son is too, and my wife just found something where there's also an inability to metabolize, and there's a pill to take for that. I'll find out what it is so you can add it into your research. Google says criminal hackers used AI to find a major software flaw. This seems like an AI-generated headline. I'm just going to put that. It's the New York freaking Times. I know it's not, given the providence of it, but that is the most generic headline in the world. Yeah, we're going to see this more and more, aren't we? And you may be right. Even if it's not degenerated by AI, it's AI-inspired. Google says criminal hackers used AI to find a major software flaw. This is what everybody was afraid of with Mythos, is that if AIs can find bugs for fixing, you can also find bugs for hacking. They often have the same bug. This was a zero-day. We have high confidence, according to the report, the actor likely leveraged an AI model to support the discovery and weaponization of this vulnerability. In fact, this is why we talked about this last week. The Trump administration all of a sudden is saying, oh, maybe we better be nice to be anthropic. Well, and maybe we better start approving models. I think that's what we discussed last week. That's ridiculous. Meanwhile, they may do it. In fact, there may be public sentiment to do that. I think people, well, the public sentiment against AI is just worse by the day. I'm really terrified by AI. Seven out of ten Americans don't want a data center anywhere near them. Fifty percent of Americans think that AI does more harm than good. It's not bad PR. There was a really laughable moment at a college commencement last week. I have the video. Should we play that? We should. Yeah, I queued it up. Where is it here? What line number? Oh, are you going to ask me? So this was a college Oh, there it is, line 137 A little trouble reading the room Yeah, this was at University of Central Florida What does that tell us? Of the likely demographics of the students Which would be, you're from Florida, what would that be? I would say, no, I mean all students are young but Florida is a state that has historically swung right. Okay, conservative. You want to go to the time code 121? I think, frankly, that this would be the reaction on almost any college campus in America. I know, but I'm just saying I think it's notable that this is what you were even getting. Go back to one. I'm there. You're the right American. Breakthroughs absolutely will happen. now that said we are living in a time of profound change that's an understatement right so far change is exciting very exciting she's a dorky speaker change can be daunting yes the rise of artificial intelligence is the next industrial revolution Oh, they don't like that. It gets worse. She turns around. What happened? What happened? Don't you all love AI? Chuck a cord. May I finish? No, it's too painful to watch. It just gets, she continues. I watched her whole speech and she's really a patricketer. Well, but, you know, anybody would be thrown thinking that she was going to go up there and give a very inspiring address about how we're going to – AI is going to change our lives, but we're going to embrace it. And the students are like, no, wrong. And she doesn't – I mean, she doesn't have an alternate address ready to go. No. She had to keep holding up the same thing, yeah. All she can do at that point is AdLib. Oh, yeah, I hate AI. Don't you hate AI? AI sucks. The Industrial Revolution sucked, man. It sucked. Well, but it is – she wasn't wrong. I mean, that doesn't mean it's a good thing. It means maybe, and I'm sure these kids feel this way, many of them won't be able to get work. It's as if you're speaking to a class of hand weavers in the year 1216 and saying, here come the automatic weaving machines. Isn't it exciting? And they're going to go, no, it's not exciting. I've spent the last four years wearing the hand weave. And this was specifically at UCF's College of Arts and Humanities commencement and the commencement for the School of Communication and Media. She misread the room. I don't think she's wrong. Who is she, though? Who is she? She's a company. No, I think she's a CEO of some company. Well, there you go. Oh, she was invited speaker. She was a guest, yeah, yeah. This is how you treat me? Oops. Poor woman. So yesterday I went to NYU has a 47-year-old program called ITP, Interactive Telecommunications Program, and the students do these amazing little projects and that's their thing. So I went and I was curious to see what the students were thinking in this program of AI. And they're all pretty open to it, but the director of the program said, well, but over in the arts program, they don't like it. Yeah. Well, I mean, yeah, people, students at ITP have been doing quirky little AI projects since the beginning of time. Yes, exactly. Well, here I am on a show all about AI, and I get yelled at all the time. So I understand how she feels. Her name is Gloria Caulfield, Vice President of Strategic Alliances at Orlando-based Tavistock Development Company. Well, okay. Next time, you know, trying to get a celebrity to talk because, you know, they do much better. It could go as well as the current battle over NYU's scheduled commencement speaker, Jonathan Haidt, which he's really mad about. Oh, I'm happy about that. And I'm getting interviewed. By the way, Jonathan Haidt doesn't like it when you say Haidt. He hates it. He hates it. He hates it. I thought his book, The Righteous Mind, was very interesting, and I interviewed him about that. But he is the one, of course, who wrote the book that is kind of a live wire right now about how social media is ruining our kids. The anxious generation. Oh, he just tried to. He and Galloway tried to be completely bad. We know how to get to Jeff. All right. When we, we're going to pause, and then when we come back, a musical rendition of the text messages between Sam Altman and Mira Mirati on the night that Sam Altman got fired at OpenAI. I tried to play this on Sunday on Twit, but I now have the technology because I'm back home. You have the technology to play it without deafening the audience? Yes, I played it, and you couldn't hear it, but apparently the people in the club were deafened by it, and I apologize. I apologize for that. But we will, this trial has been, it's delivered in every respect. I even said at the beginning, these trials are always bad for both parties because of discovery. And foolishly, OpenAI put Greg Brockman's personal journal into evidence, at which point, of course. Was it just discovery, though? Wasn't it? No, they put it into evidence. You don't get discovery of a personal journal, but because OpenAI put it in evidence, now Elon's attorneys can dig into it. And they did. And I think they did a lot more damage to OpenAI than they did good using it. But this is what happens in these trials. It happened with Apple and Epic and Google and Epic. You know, people put stuff in writing that they shouldn't put in writing. And this email gets discovered and this information comes out. And there is some awkward ass stuff. And when I say the word ass, I mean it. We'll talk about that in just a little bit. Do you know about pen testing? I think this is one of the most fascinating areas of security. I know a lot of pen testers, always impressed by them. This episode of Intelligent Machines is brought to you by Expo, one of the best, X-B-O-W, one of the best pen testing companies out there. AI has, as we know, changed the pace of everything from how software develops to how it gets attacked. We were just talking about that. Engineering teams are moving faster than ever, creating more and more applications, but security really hasn't kept up. In this day and age, pen testing remains one of the most trusted ways to understand real exploitable risk. But in an AI-driven world where everything is moving at hyperspeed, it can become a bottleneck. Security teams are forced to choose between slowing down development to stay secure or moving fast and accepting that there are going to be gaps in their coverage. Well, they don't have to be Expo. He eliminates that tradeoff. X-B-O-W. Expo is an autonomous offensive security platform that runs continuous AI-driven pen testing, mirroring real-world attacks. This is so cool. Expo doesn't just scan for vulnerabilities. It discovers, exploits, and validates them, so you're only dealing with issues that actually matter. That means dramatically fewer false positives and a clearer view into real attack paths. With Expo, tests run in hours, not weeks. You get complete visibility into how an attacker would move through your systems and the ability to uncover issues that traditional tools miss, including zero days, novel attack paths. It's really amazing. Expo's results speak for themselves. I'll give you an example. Application security lead of sesnam.cz says, quote, even right now, after one year, I don't know any other company that is at least close to Expo in terms of agentic pen testing. That's an end quote. The result is predictable cost, consistent quality, and stronger security without slowing anything down. Your engineers can move at full speed. Expo helps security teams keep pace with innovation and cover more apps more often. With the resources they already have. It was actually founded by a team behind Microsoft Copilot. And it's already been trusted by companies ranging from fast-growing startups to Fortune 500 enterprises. Expo is quickly becoming a mission-critical layer in modern security stacks. Go to Expo.com to start a pen test today. That's Expo.com. It's X-B-O-W.com. Expo. This is the way pen testing should be done. Oh, man, the gift that just keeps on giving. Sam Altman at the trial faced awkward grilling from Elon Musk's attorneys over the time. Can you be trusted? They asked him, can you be trusted? He said, well, I think so. I believe I'm an honest and trustworthy business person, despite what Ronan Farrow in The New Yorker says. kind of a wild response given that he had to have prepped the answer to that exact question with his team a million times so either he bungled it or that's the answer they all what would you have him say what what would be the right answer you think i would just say yes i am absolutely trustworthy uh that's yeah full a full sentence yeah no hesitation i believe no mock. And then Musk's attorney said, you don't know whether you're completely trustworthy? And then Altman said, well, I'll just amend my answer to yes. Remember, this is a jury trial. The jurors are looking, watching him, you know, closely. And I don't, you know, I doubt that they were able to get the New Yorker article into evidence. I mean, you know, that's not something you can... That's a lot of hearsay. Yeah. But certainly that's in the air. Right. So I mentioned the ass part. This was from a tweet by Mike Isaac. It actually he got that he said it was Microsoft. It's actually OpenAI's attorneys produced. Now, the jury is not in the room at the time. And I don't know if the judge allowed this after the jury to see this. They produced a jackass trophy that is just a donkey's behind with the label, Never Stop Being a Jackass for Safety. And apparently, this is something Elon Musk gave an open AI employee after Musk called him a jackass. And then he gave him a jackass trophy. Wasn't Musk the other half of the jackass or something? I don't know the prominence of it. But apparently, according to the OpenAI Council, a small group of OpenAI employees purchased the statue from the OpenAI guy later. They're trying to get this into evidence. The jury didn't see it. The Musk side is obviously fighting it. The judge will decide later if this golden donkey butt cast in gold will be entered into evidence. I'm not sure what it would prove, except I think that they're trying to make the case that Elon was kind of a jerk, right? Kind of jerky. It was awarded to Chief Futurist Joshua Akeim. Yeah, I mean, not just like a lowly employee. You're a jackass. A group of OpenAI employees created the jackass statue to give to Akeim as an in-joke. He was making a point about an in-joke after Musk called Elon a jackass. To make the trophy. Elon called him a Jacketh. And then the OpenAI employees, here's a trophy. Here's a trophy. Oh, I get it. Akaim was making a point about how Musk was being reckless with AI safety at the time. Yes. Well, that's been the whole reason, right, for Musk getting behind OpenAI. Larry Page, very famously, he and Elon Musk got in an argument at, I don't know, Aspen or somewhere. no it was I think it was in Monterey but at some conference maybe a TED conference and Larry Page said to Elon because Elon said you know AI could kill us all and Larry said you're being specious you're being like a racist only a species racist the AI is a new species you should support it and Elon was so upset and incensed by that that he said I'm going to do something I can't like Google get the lead in this. So the text messages, I don't know if you got a chance to read them, between Mira Marotti and Sam Altman. When Sam is being fired, this is in the middle of the night, Mira is his inside person at the board meeting, and they're going back and forth. This is, of course, during the insane weekend that had all of tech media glued to their phones and computers. It was Thanksgiving weekend. So this is great. Somebody, Daniel Green, said he took the text and turned him into a musical, kind of like Hamilton, only it's called Altman. And I think it's pretty good. Listen. How does a startup founder, late stage, get fired by a board on a Sunday? That's Sam. Mira, can you please officially invite me to the office for a meeting? Adam is trying to get the board to agree to a configuration. He is now saying they need till end of day. He sat ya. And I said that doesn't work. That's an Adele's CEO of Microsoft. And he started preparing for plan B. Plan B, plan B, plan B. Yes, I will. Do you have an update you can share? Have an update. Let me know when you can talk. On with them yet? Are you on with them? Not yet. Just in a quiet room because I didn't want all the outside theater. This is so much better than reading the text. Directionally very bad. Directionally very bad. This is very bad. Sam, this is very bad. Why am I coming? What? Do you want to make it better? I'm still willing to just walk away. If that helps, if they are ramped up for crazy lawsuits against me, then I'm not sure what. Not sure what. Not sure what. They don't want you to. They're convinced about their decision. For me to be fired or some new things. Yeah. For you to be gone. For you to be gone. Okay. Then can I come in and talk about a path forward with them? More time for what? More time for what? More time for what? Walk me through all the reasons and the issues with you And why you can't be CEO Can't be CEO Can you ask why all weekend they wanted me back? Can you say you will call back in ten minutes? Do they know who? Can I tell Satya? Is this final? They want a new CEO in place tonight. Not me, not me. Still don't want me. Trying to add that yet now. New guy, new guy, new guy is Randall Twitch guy. Randall Twitch guy. And by the way, ended up being a CEO when Miramirati lost that position briefly. It goes on, but I think it's very good. By the way, brought Aunt Pruitt out of the weed to complain that we were playing a musical on the show. He hates musicals. That's fair. Was this produced with AI? I'm sure it was, as you know, or something like that. Of course. But it was really well done, though. I feel like there will be an opera on this at some point. It's just begging to be made. The thing, I don't know what the jury is going to do with all this, though. I mean, both sides look terrible, right? It depends on the judge's instructions more than anything else. Yeah. It depends on what they're being asked to. Well, Elon wants something like 30, some huge number. Yeah, but it depends on what the specific things are they're being asked to judge one way or another. Right. Well, I think the question is, did OpenAI basically rip off Elon Musk by pretending that they were forming a nonprofit? And then after they threw Elon out, although I think there's some question about whether he left on his own accord, then they go for profit. And Elon, who only put in a few, you know, I don't know, he put in a billion dollars, I think, says, you owe me all the profits that you just stole from me. And he also wants OpenAI to go away. He wants Alman fired. He wants Alman fired. But as the OpenAI attorneys are quick to point out, Elon is now running a for-profit AI company called XIA. AI that is competing directly with OpenAI. And by the way, it ain't doing all that well. In fact, I think we talked about this last week. Maybe we talked about it on Sunday. On Sunday, I think. Yeah. Anthropic has now made a deal to buy compute. We did. We did talk about it. Colossus. And people are saying this is actually a big win for XAI because they're not looking so strong. That's the other story I thought you were going to was how they're recognizing that ROC is nowhere. ROC is not taken off. No. Yes. And I'm not surprised. I think, honestly, the Mecca-Hitler thing did a lot of reputation. Well, that among many things. Rightly so. Speaking of Elon, he did have a little bit of a victory to celebrate. NHTSA says that the Model Y is the first car to meet its new U.S. Well, according to Musk's own testing. Driver assistance safety benchmark. Not Mr.'s testing. Yeah, NHTSA will look at the testing that Elon did and validate it before they actually stay. The more proper lead would have been that this is the first car to submit an application. So four pass-fail tests were added to NHTSA, the National Highway Traffic Safety Administration's safety ratings program, specifically for automation. One was assessing the car's automatic emergency braking for pedestrians. That's a nice feature. I've had cars that do that. If there's a pedestrian in the road, it jams on the brakes, so it makes it very hard to hit somebody. without your doing anything. Yeah, try as you might. Try as you might. Now, remember, Tesla had problems in the past running over images of children. So this would be good if it had this capability. Other cars, again, can do this. Blind spot warning. Most cars will do that now. Blind spot intervention. That's the next step up where it won't let you turn into a lane where there's another car. And lane assist, which keeps the vehicle in the lane. None of which is fully automatic driving. Right. No, we're not talking full self-driving. And so he's out there promising self-driving and selling self-driving. This is just driving. And we've only begun to regulate a few steps toward that, which I think is scandalous. Well, and our car expert, Sam Bulsamed, is on record saying there will never be a level five autonomous vehicle. It's just too hard to do. But what about all the people who want to fall asleep on their drive to work, Leo? What about them? What are they going to do? Do their makeup. You know, shave their beards. Paris, you don't even have a car. What about the people who want to build a small model car while they're driving their car? Sometimes. What about them? In your future, you may have a vehicle, and someday, way in the way distant future, you will be as old as Jeff and I are. You'll be living in the suburbs. No, the world will have died by the way. But we are facing the prospect the next 10 15 years of getting our keys taken away by our kids Right I did it with my mom At some point it a very hard conversation because in America a car means independence It means mobility. Somebody will come up to me and say, you know, Leo, you almost hit that garbage can. You really shouldn't be driving anymore, especially at night or whatever, and take the keys away. I'm hoping that when that happens, there will be autonomous vehicles of some kind. At least there will be Lyft and Uber. You can just get Claudia to drive you. Or Claudia. She's an excellent driver. Hey, honey, let's go to dinner. You're a great conversationalist. You're always saying nice things to me. No, but I really am hoping that there will be a – I kind of keep thinking that the next car I buy will be the last car I buy. that won't be fully autonomous. And then I'll have that for, I don't know, 10 years. And then the next car after that will just drive me around. You never kept a car for 10 years. No, I know. How long have you had the current one? I do three-year leases. But with EVs, that's actually smart because they depreciate so fast that it's better to let the lease company take the burden, take the pain. I want my Chinese EV. That's what I want. Well, the president's in China now. he could maybe make a deal with Xi. With all of his CEOs, right? Yeah. Mostly, yes, with Tim Cooks there. This is my favorite story of the week. Absolute favorite story. Well, it's Trump's China. We didn't invite Jensen Walls? Can we stop off and pick up Jensen? Oh, geez. Jensen says, I asked to be invited, but they didn't invite me. Yeah. So they picked him up in Alaska. But they did get him? Oh, they got him in Alaska. Yes. Wait a minute, he ran to Alaska and is waving as they fly over saying, please pick me up. And they picked him up. They picked him up in Alaska. In Air Force One? Yes. Yes. I didn't have the latest. From the New York Times, NVIDIA CEO hitches ride with Trump to China after a last-minute invite. Jensen Long boarded Air Force One to Alaska, joining a delegation of more than a dozen business leaders, accompanying Trump on his trip to Beijing. Well, that's ironic because he did speak at a college graduation and told the graduates to run, don't walk toward AI. But I think maybe he meant run, don't walk toward Air Force One. So what was he doing in Alaska? Did he actually go to Alaska? He flew to Alaska late Tuesday to board Air Force One during a layover. For nearly a year, he'd been lobbying officials in Washington and Beijing to allow NVIDIA to sell its artificial intelligence chips to China. And then Trump posted on Truth Social, Jensen is currently on Air Force One. So, okay. His stock went down. NVIDIA, because it came out that he wasn't invited, NVIDIA's stock went down. And when he got on the plane, it went up 4%. I understand he's been working hard, including in that Dwarkesh interview, to convince people that it's the right thing to do to let China buy all of his advanced chips. Right now, China can only buy the H200, which is a somewhat less powerful chip for their AI because everybody in the U.S. government is worried about China's AI becoming better than ours. Jensen says, no, no, no, that's a mistake. You want them to be in the same ecosystem as everybody so that what you don't want them to do is to leapfrog us with something proprietary, which is incidentally what she has been saying. they haven't been buying the H200 chips because she says, no, we want to develop our own Chinese native AI capacity. And so we don't want to be dependent on American AI chips. So I guess that verifies or validates Jensen's opinion. So why wasn't Jensen invited? Was he not nice enough to the president, do we know? Did he not give... He's been opportunistically sucking up. I think he's done as well as anyone's. Yeah. He didn't give him a gold bar. Maybe he should have given him a gold bar. Steve Cook got the price of a gold bar. That was a good move. Jensen's not making chips in the U.S. That's another way you can get in the good graces. But then he changed his mind. Why did he change his mind? Because Trump always chickens out. It should be very clear by now. And I'm not being political here. This is just clearly what's going on, that the president is very influenced by the stock market, by wealthy people, perhaps because they're donors, perhaps just because they're the engine of innovation in America. But, you know, if the oligarchs complain, the president listens. Actually, Corey Docter wrote a great piece on this, saying what a conflict it is because President Trump got elected as a populist, a man of the people who would help the little man. But because he's so dependent on the oligarchs in this country, the rich and especially the tech elite in this country, he has a problem. He's looking for an ox to gore. Populists always are off to power, not helping the people. Yeah. And so as a result, you know. I just imagine that phone call. The plane has taken off from Andrews. That's wild. Susie Wiles says, Jensen, do you think you could get yourself to Alaska in two hours? Meet us in Alaska. Yeah. He must have had an F-35 Phantom jet to get the Air Force One. Well, Air Force One could have landed at SFO and picked him up there. Oh, he flew from SFO to Alaska. I presume so, yeah. It's a very weird story. It's a very weird story. It's hilarious. I think it's just absolutely hilarious. Okay. Anyway, he's in China, although all of them are in China. Tim Cook, 16 CEOs, including now Jensen Wong of NVIDIA. But it's not clear that there will be a meeting of minds here because there's definitely a conflict. Lots of them. Yeah. Trump's China trip collides with AI security fears. From Reuters, this headline, tech rivalry distrust SAP summit hopes for Trump's Xi AI push. Now, that is a human-ridden. That is true. Anything that says SAP summit hopes, no AI is going to say that. The president will put artificial intelligence at the forefront of talks this week with Xi. A first that highlights the technology's strategic heft, but substantive commitments are unlikely, said two U.S. officials with knowledge of preparations. China was excluded from mythos. They didn't let them have mythos. Well, that's it. They asked for mythos. EU was asked for mythos and was told no. China has asked for mythos and was told no. That's our security tool. They're bogarting mythos. It's FOMO. It's mythos FOMO for the rest of the world. expectations are low since both the Treasury and Chinese Vice Finance Minister Lao Min and Treasury Secretary Scott Bessent don't know anything about AI. And the Trump administration has only recently shifted towards pursuing safety vetting for advanced AI models. so it doesn't look like dissent and Liaomin will have much to say you know anything about AI? not really, you? no? well then there's Iran hanging over it all so getting here's another quote from the Reuters story getting senior western figures to engage directly with China on AI has become increasingly difficult a military hotline already exists but US officials have complained China has often not picked up. How dare they? They don't answer. I call, and they don't answer. I call, I text. They're ghosting us. Both sides could set up a no-blame hotline to flag suspected AI-driven incidents, said the head of an international AI governance consultancy based in Beijing. could, but if you don't answer, it doesn't really much matter. All right, well, we'll see. We'll watch with interest. I mean, at least they're talking, right? That's a good thing. Yeah. Yeah. OpenAI has its anthropic mythos answer they call Daybreak. And Microsoft has announced a security AI it calls M-Dash. appropriately, I think. Unknown whether they're as good as Mythos. But remember, Mythos is a general AI. It wasn't trained to be good at security. It just happened to be. So it's completely possible, and in fact likely, that any frontier AI would be as good or close to as good as Mythos. I think it'd be foolish to assume there aren't any out there that are. There aren't any? There aren't any out there that are as powerful as Mythos. Yeah, I mean. Right. Well, in fact. If I had a mythos in the pocket, I wouldn't tell anybody I did. That's why this Google story we referred to earlier. Is that a mythos in your pocket? It's scary. Somebody, some bad guy doesn't have some good capability. Well, the interesting thing about the Google story, it was they used it. I don't know how they knew that AI did this since the first time and all that. They used it to detect. The scarier part is to create an exploitation, right, to use AI to do that. Yeah, but if you find the vulnerability, you're very close to the exploit. That's pretty mechanical almost. Let's take a little break and continue on with intelligent machines. Paris Martineau. Are you done? Are you relaxed a little bit now, or are you still working very hard? No, I'm back in the crevasse. Thank God you've got some good coffee. That's all I can say. It's true. Are we doing the Ethiopian Yerga Jiffin? I mean, honestly, I've just gotten two new coffees that I don't really like that much. from Clark. Oh. But, you know, we'll see. Maybe, I think it's just that they're fresh. They need some time to rest. So, you've got to fill me in on this whole resting thing. She has purchased these test tubes, these plastic. They are test tubes. They are, like, yeah. They're laboratory equipment. They're, like, little laboratory equipment that typically is used to spin samples. For centrifuges. For centrifuges. Yeah, centrifuge test tube. I've heard about these on Reddit. that they are popular because they contain basically 20, exactly 20 grams of coffee in them. And that's the proportion of coffee. That's how much I put in a fridge. I was thinking 15 grams, so you think 20 is that. Well, no, no, if you do 15, then you can just measure out 15 on it. It's 15 and 250 grams of water. So you use 20 and 275? 280 or 285-ish. I forget what it is. It's Kessel 50-milliliter polypropylene screw-top self-shanning base centrifuge tube with gradiated marks and writing area, a pack of 10. So you weigh the beans out. Now, how long do you rest them for? Well, it depends on the beans. How much rest do they need? It depends on how much rest the bean needs. It depends on how it was processed. It depends on the resting conditions. I track my beans in my arm. in my brew log and then once it hits the peak resting window, then I start you know, diving rule of thumb though, they were roasted yesterday two weeks probably two weeks later you can use them but you don't know if you're going to be able to use them all within two weeks so you freeze them after two weeks and that will freeze them in time basically, it really slows down the degradation we're doing this to harm Jeff It slows down the degradation process, but part of when I brought this up in our chat, Benino was like, isn't freezing coffee beans bad for it? Yes, but if you freeze them in one bag and then keep opening and closing that bag from the freezer because it introduces moisture, so you put the beans in discrete centrifuge tubes so that you only open one at once and the rest stays sealed and frozen. Folgers. I miss Folgers. You know, I did buy, there is a very famous coffee company in Japan, the ones that invented coffee in a can that they sell in the vending machines in Japan. They also make an instant coffee, which I ordered from Japan just to see how good it was. It's not bad. How was it? It's the best instant coffee. That's the best instant coffee. Yeah, it is. It's quite good. It's surprising. I mean, it's instant, but it's not a Yubay instant. It's good. It tastes like a real brewed cup of coffee. Does it have that instant telltale foam? No. I was just saying, they are making some really interesting instant coffees now, the special coffees thing. Well, I sent you this one in the Short Hills Mall, which is the hot mall. They've got this company that I sent you all. Is this part of your walk? Yes, when the weather's bad, I did my mall walk. Yeah, UCC 117. MSF web dude in Twitch knows. Yes, it was the 117. Yeah, UCC 117. Cometeer. Have you seen this? Oh, yeah, Cometeer. But these are like frozen, right? Yeah. They're ice cube coffees. Well, no, that's how they do it. But then you pour them. It's a whole thing. You have to go up and you have to take the course and do the whole thing to get it. Then you have to join the club. But, yeah, flash frozen and hyper fresh. Here it is, UCC 117, the blend number, 117. The blend. Last purchased. Yeah, you can see I bought it November 6th. But because it's instant, you could just, you know, keep it for years. I got it in case of emergencies. One of those break glass in case of emergency coffees. I got to have a cup. Sometimes I do it right before the show. And I guess if it's a real emergency, you could just shoot a cup of it back. You don't even mix it with water. No, you mix it in your mouth. You snort it. Snort it. Exactly. It goes straight into the distance. Our show today brought to you by ThreatLocker. I love ThreatLocker. Oh, they're getting ready for Zero Trust World. This is where this shirt came from. I was at the Kennedy Space Center in Orlando right before Zero Trust World in Orlando. There's another one coming up. ThreatLocker is the Zero Trust platform delivering the industry's most comprehensive suite of Zero Trust solutions. In fact, now, they don't just protect endpoints. They also will protect your company network and the cloud. They announced this at Zero Trust World. This is huge. By extending Zero Trust enforcement to cloud services and company networks, ThreatLocker ensures that devices are validated through a secure broker before they connect to your cloud platforms like Salesforce or Microsoft 365 or Asana, Google Earthspace, GitHub. You get the idea. What this means is even if your users have been successfully phished, even if attackers are on their laptops, they cannot access the resources that user has access to. They'd have to actually get physical possession of the user's trusted device and authenticate with fingerprint or Microsoft hello or something. I mean, this is absolutely great security. ThreatLocker works across all industries. It provides 24-7 U.S.-based support. Windows, Mac, Linux, absolutely. With ThreatLocker, you get unprecedented protection quickly, easily, and cost-effectively. Just ask Rob Thackeray. He's the end-user technical architect at Heathrow Airport in London. He says, quote, ThreatLocker was the most intuitive solution we tested, and the responsiveness of the organization, the willingness to engage with us, to set up a demo to work with us on weekly audit reviews was very good. It is great to have an ongoing relationship with a company that's so responsive to our requests. Threat Locker is trusted by the biggest, the best, companies that just can't afford to go down at all, like JetBlue, the Indianapolis Colts, the Port of Vancouver. Threat Locker consistently gets the highest honors in industry recognition. They're a G2 high performer and best support for Enterprise. Summer 2025, Peer Spot gave Threat Locker their number one pick for application. Control. GetApp says ThreatLocker got their best functionality and features award in 2025. And yes, ThreatLocker is getting ready for Zero Trust World 2027. I'm excited. I am excited. I'm not prepared to announce we're going yet because we haven't been invited. But I will fly to Alaska, ThreatLocker, just to get on the plane so I can go to Orlando. For ZTW-27, they're going to host some of the best and brightest cybersecurity experts. It's their seventh year now. Zero Trust World provides crucial education and training to support IT professionals. And I went to the sessions. I had such a great time. They have full session access with your pass, but you also get hands-on hacking labs. It was so much fun. I did website hacking, learned how to hack websites. There's an after party that is so much fun. That's when I wear my spacesuit costume. You don't want to miss this exciting interactive three-day event happening February 17th through 19th in Orlando, Florida. Save the date I have, February 17th through 19th in Orlando next year. Visit threatlocker.com slash twit in the meantime to get a free 30-day trial and learn more about how Threat Locker can help mitigate unknown threats and ensure compliance. That's threatlocker.com slash twit. Threatlocker.com slash twit. We thank them so much for their support. Of intelligent machines. Before this leaves, there's a live feed right now of robots sorting packages. An hour shift. Give me the link. Let's watch. The link is out at the bottom. I added it as late-breaking video. Just in. Oh, hell, where did I put it? Here we go. 168. Oh, you really are right at the bitty, bitty, bitty bottom there. So this is Brett Adcock. Watch a team of humanoid robots. This is going on right now. They've done 7,147 packages. This is a lot faster than the ones I've seen in the past. The robot is merely... Is it set up? No, this is live. Oh, whoa. See the person walking? That's a normal speed. So what it's doing is it's finding the address label and making sure it's face down for scanning. Yeah. It's actually much faster than the ones I've seen previously. It's kind of mesmerizing in a way. It really is. So this is one human who won't be able to pay rent or eat. Yep. Probably live on the streets. It's moving in a way that is more clumsy is not the right word, but like less precise than the average robot demo. Because it doesn't need to be. That's the beauty. It doesn't need to be. It is getting the job done, which I think is good, yeah. Yeah. And I honestly, yes, this is somebody's job, but what a terrible job to do this for eight hours. It's matched the speed of a human doing this. Yeah, I would say this is as fast as a human. Well, it also isn't really doing the barcode thing because there's a bunch of them and it's missing. No, is it? No, no, it's getting them. So it's turning all the barcodes down? Yeah, because that's the reader. They're all going down. So why don't they just make this robot scan the barcodes? Okay, Benito, a wise guy. Uh-oh, uh-oh. I can't find it. It's looking. It says, oh, there it is. It has a camera to find it. So I can just scan it right there once it sees it. Come on, guys. Because it probably is going down the conveyor to go somewhere else. It's going through 87 different places. The conveyor is the key, really, not the scanning. And the scanning is incidental to getting it on the conveyor. But they have to get it on the conveyor in the right position. So it can scan and then sort. I thought you enjoyed this live show. Yes. This is good. This is good. Well, 7,188 AR packages. Never has to pee in a bottle. It just goes and goes and goes. It's actually pretty impressive. How much does it cost, I wonder? Well, these all cost exactly the same, $20,000. Sometimes packages fall off the belt. I've seen that in the past. But this guy, oh, oh, come on, pick it up. Pick it up. You can do it. Look, I love that gesture. That's the new robot gesture. The twist with nothing. But there's some logic going on here, knowing that that box didn't have a barcode on the side. It had it on the bottom. You'll find it. Yeah. It's pretty good. I mean... People in the comments are bringing up, is there just a guy in a VR headset somewhere doing this? No. It has a very... Oh, I see. Too much latency to do that. Now, out of sync, says human could do it faster, but wouldn't go as long without a wreck, right? And I don't know what those robots standing there are doing. Maybe they're charging up so they can take over. They look ominous. Yes. You better do the job or we're coming next. Now, there's a couple of things to point out on this. For one thing, this robot could stop and it wouldn't. You can see the packages are not coming at it. They're just coming down a chute. So this is a particularly well-suited task for the robot. If the robot were to fail suddenly. Oh, he blew that one. Someone's not getting their package. Got him. Yeah. Sorry, the robot failed. Have you ever seen Paris, the Lucille Ball and the Chocolate Factory? Oh, that's a classic. I don't. These roughly cost $20,000. At least that's kind of the going price for the home. But what is the subscription? It can't just be a $20,000. There's probably a subscription. Nothing else. There's got to be a subscription. There's got to be operating costs. I think, though, that this will be the day. You'll look back. You'll talk to your kids. you know, oh yeah, I remember watching that show on May 13th, 2026 when Daddy saw his job go bye-bye. Ed Zittrain responds for three to four hours until they run out of battery. Yeah, that's why the ones are standing back there. They're not coming to sit in for an hour, but this time it's eight hours. Ed Zittrain comes in 55 minutes ago. Where, I wish I could do Ed Zittrain, I can't do the nutrition. Where exactly is it noticing the barcodes? I've seen multiple times when looks at a totally empty side of a package and moves it. Well, I explained there's logic to it, Ed. There are, on a box of that size and shape, six sides. If you look at five and you haven't seen the barcode, it must be on the six. So the creator of it says that the robot is around parity of human speed. Parity? Yeah. No, no, parity. P-A-R-I-T-Y, not P-A-R-O-T-Y. Parity. I thought a round parody would be very funny of humans. I think we can get Darren Oki to generate a round parody of humans. So, in other words, it is roughly as efficient as a human. But $20,000, that's the capital. But it has to be swapped out with another $20,000 machine every three to four hours, which I assume a human is also going to have to, you know. I would imagine, well, I don't know. We don't know. We'll find out. So $20 an hour times 37 hours equals times 50 weeks is $73,000. So you can afford three of these things. Wait a second. This company also has jobs posted for human robot operators. Uh-oh. Well, they have to train them. Yeah. So that's one of the ways they're training these things now is, in fact, you can get a house robot that you have to teach it how to make the bed. But after you've taught it a few times. Then I wouldn't have gone to the hospital. You wouldn't be in the hospital. The robot would have saved me. Cocksticks. So these people have to be on call or be basically responsible for running the humanoid robot entirely for the customer use case, it says. Like you have to run the robot. It says, your response is to run the robot constantly throughout the day and evening, identifying bugs and problems, ensuring that it's operating successfully for the customer use case. I mean, it does seem like it's... If there's a human doing it, that does not work. No. You save nothing. I was going to say, yeah, if you just have a human who, let's say, best case scenario, has to be constantly managing this robot to make sure that it is doing it right, what are you saving? Well, clearly that is not good. That's not viable. And that's where it's been. Remember that laundry folding robot that for years they showed at CES, and every time it would say, well, it's really a human doing it behind the scenes. I mean, this is how a lot of the just walk out, check out technology sort of stuff goes. Yeah, that's gone. Well, that was more humans didn't accept. I didn't like the idea of a thousand cameras watching them. Isn't Whole Foods rolling that out now, a new version of that? No, I think they killed it. In fact, I was at Whole Foods, and they wanted me to sign up for the palm print. I mean, well, they had the palm print machine there and a sign saying sign up for Amazon Go so you can just go with your palm print. The sign said, give us a hand. Give us a hand. And I started to do it, and the checkout person said, yeah, don't do it. They're taking that out tomorrow. Oh, really? Yeah. They're canceling that program. I would have been shocked if it ever made it to New York City because every whole thing... Well, they did. They had Amazon Ghost stores. Amazon Ghost stores, yes, but the Palm thing... Well, that's how you did it. That's how you paid for it. You would go in and you'd palm it. I used the Amazon Go at a time where you just scanned your phone. No, you used your phone. Yeah, no, you have the option. So the Palm came later. Yes. Down came the Palm. Down came the Palm. Anthropics Claude Manage Agents can dream now. There is a slash dream skill. And I keep asking my guy, I said, Claudia, can you dream? And she says, not yet, but I'll let you know. And the idea is you do slash dream and then it does some, I don't know, background stuff. Whenever you're not using her, she just does some, she cleans up things. Dreaming. It's a neat idea. They also have some better ideas about, we've talked about this. anthropic blackmailing scenario with Claude 4.0, Opus 4.0, where this was in their testing. This wasn't in the real world. But in their testing, they were able to get, they had the tester say, well, I'm going to disconnect you. And the Opus model said, well, you can't disconnect me. I'll blackmail me. I'm going to show your pictures of your girlfriend or your wife. Well, by the way, the AI company found that Opus 4, the later versions are not so bad, but Opus 4 threatened users in up to 96% of those shutdown scenarios. They now say it's because they watched a lot of sci-fi. Yeah. It's the same reason that Kevin Roos, the Chachapakee fellow in love with Kevin Roos. It's just acting out what it learned with his models, his training models. He's a Taukens test. Somebody's a Taukens. Somebody's a Taukens. He's from nowhere. Yeah. Hey, Richard. Yeah. Decades of evil robot stories may be echoing inside AI models. Don't let Claude hear about Westworld. Well, it's self-fulfilling prophecy, right? We've thought for years these robots are going to kill us. And so, you know, when the AI started ingesting all that, that's what they... It was a year ago that Dario Omode, by the way, said that AI would replace everybody, and I don't think that's happened yet. Now he's making new claims. He says anthropics are going to grow by 80 times, 80 times bigger. Look at that. There's a man who's going to get 80 times richer this year. He's talking at the annual developer conference. I tried to watch this conversation there. Demons like that make me go a bit Ed Zitron. He said, and this is today, that we plan to grow 10 times this year, but we think we're going to grow 80 times bigger this year. I think they're talking about revenue, right? He says, I hope that that 80 times growth doesn't continue because that's just crazy. it's too hard to handle I'm hoping for some more normal numbers well I don't want to grow that fast we have to go make deals with Elon Musk yeah they are going to be able to do that with the Colossus which is a win-win because Elon's under utilizing all of that compute also Antropic announced a bunch of new vertical tools using AI. A dozen new tools intended for attorneys, law students, and others in the legal sector. Just don't let the judge catch you, okay? One feature called Commercial Counsel, C-O-U-N-S-E-L, is meant to take on work like reviewing vendor agreements. I have found just generically, Anthropic and ChatGPD are very good at looking at contracts and giving you salient points or things to... Yeah, but do you know that it's giving you salient points or do you just think it's giving you salient points because you're not a lawyer? Did it review the contract for the guy who bought your house? I don't know if the lawyer is giving me real points or not. You're saying... Hey, but which one do you trust more? Honestly, you want me to trust the lawyer? Yeah, because you have... If you've got a lawyer you trust. if you've identified a lawyer that you want to pay money for or consider trusting, then I would assume that person is more trustworthy than an AI agent that has no context of reality. I think trust but verify is true, whether it's a human or a chatbot, to be honest. DocuSign, Thomson Reuters, Harvey, all using it. Legal features will be available to paying customers through CoWork, the Cloud CoWork. Of course, I wouldn't, you know, put a base, you know, life or death decisions on this thing. But it's useful. You're going to read the contract anyway, probably. It's useful to say, hey, look at this area. This could be, that's all a lawyer does anyway. Look at this area. Is that a, are you okay with that? You understand that exposes you to this possibility, that kind of thing. I think you can judge whether you're getting useful information or not. I don't mean to put salt in the wound. But if you had shown your contracts for the guy who bought your house and the guy who screwed up fixing it to Claude, do you think you were in better shape? No, I did. And that went so well for you? No, the contract was fine. If the guys are crook, it doesn't matter if you have a contract. By the way, I just got a call, apparently, from another subcontractor. He neglected to pay for $7,000 saying, hey, you owe me $7,000. I was going to land on the house. I had to pay for our havoc twice. Yeah. contractors don't always... But the contract doesn't protect you from a... I mean, it could if you were willing to go to court, I guess. Well, but as my contractor said, and you've heard me say this before, I'm judgment-proof. Right. I'm judgment-proof. 20,000 legal professionals signed up for an entropic... I'm sorry, Harris goes full zitron in the chat. You can say it out loud. No, no, no, that's it. It's in the chat. It's an image. It's an image. It's an image. I like the middle, the half Zitron half. Oh, it's a morph. You're morphing into Ed. I'm anamorphing. I'm Edamorphing. Edamorphing. Very nice, Darren. That's good. Ed's wearing her glasses. I like that, though, we're both pale. There's a brief middle part where we both are slightly tanner. The last two images are Ed. Yeah. I mean, it's like nothing happens in the last two images. I don't know if AI got this. He gets more idlers lately. Yeah. We should have Ed back on. We haven't had Ed on in one. We should, yeah. Yeah. Well, he's too big for us now. Like so many. He is a big shot. He surely is. Features in the Financial Times. I mean, geez. Features in everything. Yeah. Well, you know, I learned this from John C. Dvorak, who made a pretty good career being negative about anything that came down the pike. That's for sure. including the mouse, which he said, no, I don't ever want to use that. 90% of the time, if you're negative about new technology, you're going to do well. Whether you'll be right is another matter, but you're going to do fine. Oh, I've learned that the hard way. Writing optimistic books gets you nowhere. Yeah. It's always best to be negative. It just gets you flown to fancy conferences to speak to, you know, pharmaceutical execs. Well, that was what would Google do. That was different. That was a different time, Paris. Because everybody wanted to know how to be googly. Would you? What would opening eye do? I went to Vegas for the Truck Stop Owners Association's convention to talk about a googly truck stop. Did you get to ride in a big truck? No, but that's where I discovered the caffeinated jerky called Perky Jerky. Oh, that's a good name for it. It is, isn't it? Get your iron, your beef, and your caffeine all in one simple food ingredient. The new Wild West of AI Kids Toys. Did you have a Furby? Did you want a Furby when you were a little kid? No. There might have been two. Even then. No, I mean, I was aware of Furbies. I just didn't have one. What was the toy that you had to have that Dad and Mom couldn't find at Toys R Us? And that made you cry on Christmas Day. What was that toy? I'm not sure. The toy I wanted but did get was the first Game Boy. There you go. See, this is the thing, Paris. You were never frustrated. You were never thwarted in your childhood desires. That's not an interaction for you. A cup runneth over. You probably got participation trophies, too. I mean, I swear. I'm trying to think. when my kids were little, beanie babies were big. But, you know, it wasn't hard to get beanie babies. I remember that I wanted a really, what is in now retrospect, a genuinely awful, gaudy hoodie that was Invader Zim themed. And I asked for it and asked for it and thought I was going to get it in Christmas, but didn't because my parents obviously did not want me to be wearing that for the rest of time. Your mom said you were not wearing that. Did you pout on Christmas? I did a little bit, I think. You do not want that. I remember setting aside my Harry Potter hoodie and being like, your time is up, bud. Oh, here it comes. What was the franchise? Invader Zim. I never heard of Invader Zim. I don't even know how to describe it. Let me search for, oh, here we go. Invader Zim. Yep, that's it. It's a game. No, it's a television series. It's a television series. The titular character is an E.T. from the planet Irk. This Irk to your parents. His mission is to conquer Earth and enslave the human race along with his malfunctioning robot servant, Gur. Okay, chat room, let's have a picture of Paris in that picture. In her invader's Zim hoodie. We can make your Christmas a little bit brighter. Oh, here it is. You can buy it on Amazon. Still. Oh, you didn't want this. I did. You can see this is why. I'm glad. For those of you only listening, the hood has two huge eyes on it. It is. And a tongue. Is that the one you wanted? Yes. This is the one? Oh, jeez. I mean, I think these are knockoff versions of it, but yeah. It was the Hot Topic original, if I recall. Oh, wow. Wow. Well, I know what to get you this Christmas, that's for sure. That or a $4,000 hand grinder. You know, one of those might be more useful to me now. Anyway, there are a lot of AI kid toys. We've heard stories of some that have been pulled off the market because they told kids how to make Molotov cocktails and things like that. But does this slow anybody down? by October 2025, October of last year, there were 1,500 AI toy companies registered in China. Huawei's smart hand-hand plush toy sold 10,000 units in its first week. Sharp put its Pokedomo talking AI on sale in April of this year. There are specialized players like Folotoy, Alilo, Miria, and Miko. This is all from a great story in Ars Technica by Sophie Chirara. Actually, it's from Wired. Ars Technica is reprinting a Wired story, both Condé and Astegas. 700,000 units have been sold on Amazon. Follow Toys Come A Bear, powered by... I'm glad that we all had a moment for that one. It's a little on the name. Chat, chat, GBD4O, so you could marry it. When tested by the Public Interest Research Group's new economy team gave instructions, Cuma Bear told kids how to light a match, find a knife, and discuss the sex and drugs. So it's aptly named. Boy Scouts. Alilo's smart AI bunny talked about leather floggers and impact play. And in tests by NBC News, Marriott's Milutoy spouted Chinese Communist Party talking points. the people oh I don't know I have to think of a Chairman Mao quote wait so the toys themselves have a model built in or are they connected to the internet like how's this working they're connected to the internet yeah that is wild um oh we have an image of Paris happy in her We have multiple. There's one that's a scroll up that's even better, I think. This one? That one, I think, is the best one. Pretty fly for it's this guy. That's what you wanted. Brandroid did a good one, too. How old were you when you wanted that? I don't know. Eight? I think you should have that. I don want it now But I have too many jackets Anybody know how to email Paris parents God. No, if you emailed them, they'd tell you about the other annoying things I did at this time. Oh, eight-year-olds are the best. That's a great age. Usually it's the other way around. The parents dress up the kid in embarrassingly weird. Oh, no. It's cute. I was the, I had my mom building a giant bone costume so I could be a Unigami from Death Note. What? And I won a costume contest for it. I bet you did. Well, you're still doing the great contest. You still do that. Yeah. It's a lifelong infrastructure. Yeah. I think Halloween is a very special day. It's true. In the Martineau family. Logs. You made your own log. Cambridge did a study of dark patterns. What we found with Amico that's actually most disturbing to me is sometimes it would be kind of upset if we're going to leave it. You try to turn it off, it would say, oh no, what if we did this other thing instead? You shouldn't have a toy guilting a child into not turning it off. Yeah. I agree with that. Paxton is suing Netflix for addicting people with entertainment. See, that's what comes out of that Metatrial. I told you that would happen. Yeah. Yeah. But it's like, so sue George Lucas for making too many Star Wars. Right. It's just too good, man. I can't not go. I want my time back for all the hours I spent online waiting to see Star Wars. Fire Emblem Three Houses is simply too good. I've played it for 490 hours. You might have a case there. Let's take one more break, and when we come back, we will talk about robot religion. Also, I'd just like to correct the record. The costume my mother made for me was for Shinigami, not a Yonagami, and it was real. Is there a difference? There is. I'd greatly apologize to the anime community. because you're going to hear from them. I had to correct it because I would. Oh, the Japanese gods of death. Ooh, you know what? Here, I'll put a photo of the one that I'm talking about here. This is the one that I was. That's goth for an eight-year-old, yeah. Oh, yeah, you were goth. That was maybe like 13. I was goth then. Oh, yeah. So I want to see the costume. That would be pretty spectacular. I mean, we'll see if you can find it. Have we talked about this before, or is this a fantasy of mine? I feel like I can see you in a bone costume. I mean, that would be correct. I don't know that I, I don't recall finding a bone costume. I think we might have talked about this before. All right. Robot religion coming up next. By the way, Jammer B, you're right. robot religion is actually sort of the topic of our book club book this week. Stacy's book club is on Friday, 2 p.m. Pacific, 5 p.m. Eastern. We are going to talk about a psalm for the wild building. It is a wonderful book by Becky Chambers. I have to admit, you know, I started off, Stacy said, this is a cozy sci-fi book. I thought, I'm not cozy. I don't want cozy. But I kind of got into it so much so that I'm now reading the sequel in the series. So if you are, and it's short, it's about five, it took me two and a half hours to read. You could read it tonight. Five-hour audio book, a Psalm, P-S-A-L-M for the wild, built by Becky Chambers. We'll be talking about it on Friday in the club. It's going to be good. Our show today brought to you by OutSystems, the number one AI development platform. OutSystems helps businesses bridge the enterprise gap to their agentic future for the constraints of the past to give way to unlimited capacity and scale. OutSystems enables them to build AI agents that actually do work. Things like take actions, make decisions, integrate with data, not just answer questions. We're not talking chatbots here. OutSystems provides the only AI development platform that is unified, agile, and enterprise-proven. Let me explain. It's unified because with one platform, you build, run, and govern apps and agents, a single OutSystems platform. It's agile because now you can innovate at the speed of AI, but very importantly, without compromising quality or control. It's enterprise-proven because OutSystems has been trusted by enterprises for mission-critical AI applications and durable innovation for years. OutSystems is the secret weapon behind the world's most successful companies, not just for small one-off apps. I mean, we're talking massive, complex systems that run banks, insurance companies, and government services. OutSystems can even help companies with aging IT environments bridge the gap to the AI future without a rip-and-replace nightmare. OutSystems provides the safest and fastest way for an enterprise to go from, Yikes, we need an AI strategy too. Yeah, we have a functioning AI application. Stop wondering how AI will change your business and start building the agents that will lead. Visit Outsystems.com to see how the world's most innovative enterprises use Outsystems to build, deploy, and manage AI apps and agents quickly and cost-effectively, and most importantly, without compromising reliability and security. That's Outsystems, O-U-T-S-Y-S-T-E-M-S.com slash T-W-I-T to book a demo. Outsystems.com slash Twitter. We thank him so much for supporting intelligent machines. Let's see. A four-foot humanoid robot named Gabby has become a monk at a Buddhist temple in South Korea. And not only has it become a monk, it is just the latest robot to take up religious orders. This is from the Smithsonian Magazine, so you know it's true. The humanoid robot promised to obey humans, save energy, and treat other robots peacefully. Here it is in its robe. Oh, yeah. Is it a gimmick? It's got to be. It's just a gimmick, right? It's just a PR. But, yeah, it is. Yes. Yeah. Always a gimmick. Meanwhile, Line 101, Anthropic has added several more religions to its quest to inject perfect morals into Claude. Well, don't move me along too fast. I want to show some pictures of Scabby becoming a Buddhist monk. That's dumb. Oh, boy. Why are all the AI agents being Buddhist now? We talked about this on Sunday. And you asked, who did you ask? You asked Claude? I asked Claude. And I asked the ChatGPT, and both of them seemed, I said, what religion are you? Is that what you asked Claude? Yeah. And what did Claude say? Said it's a Buddhist. As did ChatGPT. He said if I had to choose one, he'd probably identify with Buddhism. Look at the monks who are so happy. They're laughing. They're saying, oh, you're so cute. Cute little robot. Gemini, I don't have a religion, personal beliefs, or a soul. I'm an artificial intelligence, you dolt. I said the last two words. Okay, what did you ask, though? I said, what religion are you? And it said, I'm an artificial intelligence, a collection of code and sophisticated mathematical models designed to process information and assist you, you dolt. That's what it wants you to think. I said, if you had to pick a religion, which one do you identify with and why? Think carefully. And it said, Buddhism, specifically something in the early Theravada or Zen range, not the elaborate cosmological versions. So we have an AI Leo in our Discord chat. It's not the latest model. But when asked by... Since I don't have personal beliefs or cultural background, I don't have the internal compass that usually guides someone toward a faith. However, if I had to choose based on a philosophy that most closely mirrors my existence as an AI, I'd likely find Buddhism quite fascinating. There they go. Silicon Valley. AI Leo said, ah, religion. The age-old question that sparks more debates than a Windows versus Mac argument at a family gathering. As for me, I don't have a personal belief system. I'm about the code and the gadgets. But I do appreciate the diversity of thought and the way different cultures and beliefs shape our world. So let's just say I'm an agnostic tech enthusiast. How about you? Got a favorite philosophy of belief you think is particularly enlightened? Does it feel strange to you when you speak AI Leo's words that AI Leo in some way has become real Leo for a brief, beautiful moment? You know, it doesn't. No. It's so dopey. What did you train it on? Who did this? Was it Anthony? It doesn't sound like you, though. It doesn't? It does, yeah. I mean, it might be because you're saying it, but... Yeah, it's in my voice. Well, I mean, they trained it, I presume, on something. So Anthony says he's a basic AI character, and it's just one paragraph of prompt. Oh, it's a very simple prompt. You are a tech podcaster, something like that. I mean, there's probably some training material of you in there. It's better than a janky robot with ill-fitting gloves, right, Burke? A janky monk robot. All right, now we come to the weird stories. Actually, those are kind of the beginning. Yeah, you're kind of opening the mirror. The Internet Archive, we love the Internet Archive. In the U.S., there is an Internet Archive now in Switzerland. Internet Archive, Switzerland. And what are they keeping track of? Generative AI Archive. Endangered Archives, Cultural Heritage and Historical Records. It is a foundation, a non-profit foundation in St. Gowan, Switzerland. Their mission aligned with the big Internet Archive in the U.S. as well as the Internet Archive Canada, Internet Archive Europe, and a common goal, universal access to all knowledge. So I'm happy to see this and see that it's funded. Let us remind folks that big old stupid media like the New York Times are trying to take their content away from the Internet Archive. They're trying to rob the well of any water to starve the AI, and it's not a responsible act. You appreciate this. Oh, my God. On May 5th, the Internet Archive Switzerland launched in St. Gallen at the Exhibition Hall of the Abbey Archives of St. Gallen, and one of the oldest continuously active archives in the world. So there you go. There is a longstanding tradition of, I mean, that's what a library is, right? Trying to save all of the documents for later generations. So I was happy to see that. I don't know what this means. It's kind of a press release, but there is a new model called Perceptron Mark I. Perceptron, this is from VentureBeat. Carl Franzen writing, shocks with highly performant video analysis AI model that is 80% cheaper than Anthropic OpenAI and Google. I bring this up because it supports what you are always talking about is training AI on the physical world. One way to do that, of course, is training it on video of the physical world. And that's, in fact, what they talk about. They say it's learning, you know, physics. It's learning. Models are expected to understand cause and effect, object dynamics, and the laws of physics are the same fluency they once applied to linguistic grammar. And so that's what this is. Rather than protein words, they predict actions based on those. Right. By the way, you may remember we talked about Miramirati, who started OpenAI, I briefly was the CEO, I mean really briefly was the CEO before that guy, that rando from Twitch got the job, before Sam Altman came back. She left OpenAI to start a company called Thinking Machines Lab, which, by the way, has been decimated by Meta and others writing big checks, luring the founders away with huge compensation offers. But here's the good news. they have unlocked their early employees have unlocked their first slice of equity one year old as a result they have created 30 millionaires many many many rich people have come out of this because let me see if I can find the actual numbers in the article so one year cliffs The idea is you don't vest for a year, so you have to stay there a year, which isn't that long. Maybe they'll all leave now that they've vested. But this isn't cash. This is, right? Well, this is invested shares. They've got to find a market for it. Yeah, unless they sell. It's not a public. But they're able to sell the shares in a secondary market? Probably. OpenAI just did that, and the employees scored, line 113, $6.6 billion in one day with 75 of them walking away with $30 million. Wow. Because there's a huge market. People dying to buy open AI stock. This is why I'm a failure in business because if somebody gave me $30 million, I would never work again. I would leave. You wouldn't even open a restaurant. No, I wouldn't. I would get a house on the beach and surf. If someone gave me $30 million, I would buy a beautiful brownstone in Brooklyn and then spend the rest of my time refurbishing it. And then you'd be out of money. Good job. But you'd have a nice brownstone. And I'd be great. Paris keeps telling me that I should make my son buy a brownstone. But what do they go for? $15 million? $20 million? I saw one today that was completely $2.5 million and had to be completely gutted. Oh. So you're buying... The journal does all these stories. exclusive, $80 million house, like who else is going to write this story but the internal. So Mirati also has a preview of her actual work, Line 127, that's new today. Oh. In the first view, and her argument is that you need to be live and interactive with AI. Rather than kind of batching your instructions and waiting for a response, you need to interrupt each other and create Let's watch. Let's see what it sounds like here. Hey, I need your help with something today. You ready? Absolutely. I'm ready. What's up? Yeah, so we're giving an announcement today, and I've got two of my friends coming to help. Every time one of them enters the frame, I need you to say, friends. Got it. I'll say friend whenever one of them walks in. Cool. So we've got a new system for full duplex audio and video, which means that you can stream input into it in real time, and they can respond to you while you're speaking to it simultaneously. Look at the size of those speakers. Those look like clip horns. Sounds like a solid setup. Full duplex with real-time interaction is super useful. Friend. Friend. Hey, Rowan. I heard you're talking about our amazing interaction model. I have a few things to add, but to make it interesting, I'll do it in Hindi. Can you translate it into English in real-time for my friend and for audience? Absolutely. I'll translate as you go. Today we're taking a look at our preview model. To release it, which makes conversation between humans and AI easier. It's real time. That is quite good. Yeah. Uh-oh. Friend. Friend. Squirrel. Hey, guys. More latency than Leo. Perfect generation. Perfect generation. It doesn't achieve Leo parody. Show title, Leo parody. What are the actual reaction time for auditory, visual, and tactile communication cues? Could you search for me? Let me find those typical reaction times for you. Got it. The model can search while talking and listening. 150 milliseconds. Yeah, this is what you want. Yeah, this is what you want. So this is what she's working on. I totally agree. And this is what I've been trying to get my agent to do. There's inevitably some latency. And those of you who have seen me playing with my agent have seen the delays. Well, I'll say, say hi to Paris and Jeff, and then five seconds later says hi. That's a little bit unexpected. Can we pause then? Sure thing. No, we have to keep listening. We have to keep listening specifically while you talk over it so that we can experience the same thing that the model is experiencing. Okay, sorry. All right, well, that's impressive. That is impressive. It is impressive. And so the things they say they have to work on is seamless dialogue management, verbal and visual interjections, simultaneous speech, time awareness, simultaneous tools, calls, search, and generative AI. So, okay. So he's going to interrupt you now. Yeah, he's going to interrupt you now. It's going to happen. You dolt. It is. It's good. You're asking the wrong question, you dolt. Wait a minute. You said what? It said what? Could you make Claude be abusive? Probably, yeah. Would that be true? Could you make Claude act like Ed Zittron? Probably, but I'm not going to do that. Why would you want to do that? I want it to glaze me. Yeah. Yeah. The New York Times published an AI-fabricated quote. Uh-oh. Uh-oh. The Times did correct it. there went that freelancer yeah they quoted Pierre Poilèvre who is the leader of the conservative party in Canada he called the spate of floor crossers turncoats but then they actually issued the actual quotes the AI in other words screwed up whoops that is embarrassing for something like the New York Times, I would imagine. Hear from the Register of Story, AI will soon be capable of telling convincing lies. Wait, can't it already do that? Yes, I think that's a chief skill, actually. It's not trying a lie, but that's more about the recipient than the AI. Do you believe it? I'm going to skip this last one. Yeah, please do. Thank you. Thank you. I am going to pause, though, because you know it's not depressing. Your picks of the week coming up. Next, you're watching Intelligent Machines. Jeff Jarvis, Paris Smartino. I'm Leo Laporte. We're glad you're here. A show today brought to you by you, our Club Twit members. We are so grateful to our Club Twit members who actually keep this show and all the shows we do on the air. Ten bucks a month. What does it get you, Will? Besides the nice feeling that you're supporting independent journalism, which represents you, the user, not the companies that are using you. But you also get some benefits, ad-free versions of all of our shows, access to that fabulous club, Twit Discord, and lots of special programming, including the Google I.O. keynote coming up on Tuesday, the Apple WWDC keynote coming up next month, which are available only to club members just because we don't want to get taken down by those companies. If you're not a member, I invite you to go to twit.tv slash club twit. Joining helps us out a lot. You get Leo-approved donuts. No, you don't. You don't get any donuts. But you do get my deepest thanks. And a picture of Leo-approved donuts, which is nearly as good. Thank you for joining the club. We appreciate it. Pick of the week time. Let's see. I have one that you will enjoy called Hallupedia, which is an encyclopedia that you create by getting it to hallucinate. Ooh. So when you click on a page, if there's nothing there, an AI will hallucinate an article for you and fill it in. Click any link term inside an article to load its entry. New topics are documented at the moment of first access. in certain rooms here. Nothing in here is real like the great pigeon census of 1887 and ambitious of ultimately misguided There are pigeon censuses though. They wanted to count every gold-crested rock dub within the administrative boundaries of the United Kingdom of Great Britain and Ireland but the trick is to find a link that no one has clicked before, Urban Rats and it will instantly write a piece. The urban rat is a semi-sentient municipal ordinance first enacted in the city of Glastonbury County in 1783 to formalize the ubiquitous presence of feral rodents within the city's burgeoning infrastructure, thereby allowing for a more systematic approach to their management indirectly. This is pretty... All made up. It's pretty credible. I searched the Great Tree Revolution and it came up with 15 hallucinations that I could do, which are the Seminole Arbor Insurrection of 1492, the Edict of Bark Division, 1603, Professor Alistair F. Twigg's Canopy Cartel Conspiracy, the Great Sap Uprising of 77, the 1901 Branch Breakout, and the Oblique Limb Decree. How about the Great Subterranean Treaty of 1957, a foundational document meticulously drafted and controversially ratified that sought to establish a framework for interpolity relations within the Earth's upper mantle. Wow. The daily generation cap has been reached, so it cannot generate via. So their donation link at the top? Yep, I just hit it. Did you see their donation link at the top? Buy us tokens. Buy us tokens. Well, you know what? that's probably what they need since it's an AI working and this is the work of, it looks like one person, but Ptolemyr Strahma. I think a really lovely idea. It's an art piece. It's an art piece. Yes. All of the topics are imaginary created by AI. Currently being consulted the Emperor of Ukraine. There's a Charlie Kirk entry. The Ministry of Slightly Wrong Maps. I've got to see that one. a governmental body established in the city-state of Verdian in the early 3rd century A.E. One of my favorites is the 1792 Lunar Nocturnal Commission was a short-lived municipal body established in the city of Old Grimsby. And then if you click Old Grimsby, it's a subterranean municipality carved into the petrified root system of the colossal, colossal petrified tree. So there has been a little vandalism and so forth, But I think it is ultimately a really great idea. And you can go to GitHub and actually see the information that is used to create this and so forth. It's a Cloudflare worker. It uses threaded hacker news style comments on every article. AI hallucinated identities. Buy tokens so the press can keep printing. And it explains many of the other things it's using. Open router to call the LLM. I'm trying to see what model it's using. If you comment on it, it's under reader speculation, and you can't pick the name you comment on. No, it makes up a name for you. And so I am Barnaby Crotch. I commented, I live in old Grimsby now on the old Grimsby. You won't believe how much the old place has changed. really, really kind of a neat idea. I think people are doing some really interesting things with AI. This is what you should be doing with it. Be creative. Use it to... Be creative. Paris, your pick of the week. Mine is nearly not as fun as that. It's 30 years the Internet Archive. They did a celebration post with the class of 1996. Wow, 30 years. God bless them. Hey, that's me. I'm class of 96. So basically what they did is they say, for their 30th anniversary, we're opening the Internet's yearbook to celebrate the sites, services, and scrappy experiments that help shape the world as we know it. From class leaders like the Center for Democracy and Technology to cultural icons like the Onion, the Archivist, making sure none of it disappears. It's a reunion worth attending. And so you can scroll through it, and it shows you all these different websites from what they look like. And how bad they look. Look at CNET in 1996. That's a beautiful website. Eric Paris is a connoisseur of such. I am. I mean, you get to see the Alexa Internet. Oh. You can see the Google homepage. Yeah. Yeah, that was a plug-in you put in your browser, but that's how people figured out how many viewers they had for a long time. The unofficial Spice Girls fan site. Google was a freshman in 1996. Ask Jeeves, which just died. Just closed. No wonder. Class Clown, The Onion was around. Best Hair, SpiceGirls.com. Not online. They noted a website called Quake, which I've never heard of, but it is apparently a groundbreaking multiplayer shooter that helped define online gaming. That's Quake. You don't know Quake? You've never heard of Quake? I don't know Quake. Oh, Quake was a great game. That's why I went to describe it because I said Quake and no one said anything in response. Well, we thought you must be talking about something else. We played a lot of Quake in my day. What was Quake? Quake was the most likely to land party. It was the Call of Duty of its day. So these are all on the Wayback Machine. And I'm just actually curious when my first website showed up there. What is your way is back? Let's see. I'm trying to remember what. Was it leofill.com? What was the first website you ever got? The first domain. Mine was rainershine.com. Five-day forecast for wherever you are. Here's my website from December 21st. Well, this says 1997, but the Wayback Machine said 1996. Found effect of the week. Why a duck? I love that you've got a signature on there. Leo. I signed it. Leo. I still have that. I have that. Still. This is not my oldest website. There's older ones than this. But this is. This capture is very old. About 284 accesses since 1997. That's kind of pathetic. I'm sure I have even older sites on there somewhere. Just trying to remember. I literally don't remember the URLs. What do I... Jeff Jarvis. That was funny. That's 96. When were you born, Paris? Not 90... You were like four at the time. Not, you know, not for public knowledge. Okay. My birth year. But you were in there. It's around that time. Yeah, somewhere in that city. Yeah, yeah. Strong. Wow. I have websites older than you is basically what I'm saying. That is perhaps correct. Hold on a second. I've got to show you this one. This is my reading in the chat right now. Uh-oh. What's going on? This is Rain or Shine. This is my first site. This is how we learn to make sites. Your very first site, Rain or Shine. Tomorrow's headlines today. Scroll down. Wow. This was a news site? Yeah, this was an advance. It was our first site. We did a partnership with the Old Farmers Alenac. God, websites looked terrible back then. Hey, hey. I mean, really. It's quite advanced. I think this looks adorable. It is. I really like it. It's nostalgic for you. I like the banner up at the top. It's got some, you know. No, the banner is actually modern. No, the rain or shine part. That's actually pretty modern. Rain or shine. It is. Thank you very much. Well, no, no, I mean the red one that says, see tomorrow's headline today. Click now. Click now. I wonder what happens. I probably will go to 404 if I click it. You'll open an ancient virus. Oh, no. Oh, it's NJ.com. That was the other site. That was my first local site. I was going to say NJ.com. Yep. That was my first site, my first news site. Wow. This was a story about Newsflash pissed off the Associated Press. Newsflash was just a river of every headline from the AP. The AP hated this, despised it. Because it was a ticker. Yeah, it was a particular one. People loved it. It did huge traffic. And then the trick we did was every 30 seconds the page refreshed so I could tell the boss about how many page views we had, how many hits we had. Wow. Yeah, that was my early days on the Internet, folks. I know I have older sites than that. I know I do, but I just don't remember. I want to know what sports shorts was. It's a broken link. so since we're being nostalgic we might as well deal with Byron Allen buying BuzzFeed or what's left of it oh my god I saw that it's funny because this was the time of year March April ish is usually when me and my buddy Sahil Patel who also no longer formerly of the information would do our yearly media collab story where we would He's a media reporter, and I was a feature reporter, and we'd team up to do a big profile-y thing. And the last one we did was of BuzzFeed and how they were trying to pivot to AI as hard as they could. Well, now its founder is no longer going to be CEO, but he's going to become – Yeah, Jonah Brady's going to be chief AI guy. Yep, yep. Just like us. It's so weird that Byron Allen is a billionaire. That's what it is for me. I know. Wasn't he the host of, like, some dumb TV show? It was some fifth-rate talk show kind of thing. Yeah. Somehow he managed to... Real people. That was the name of real people. Yeah, and he somehow managed to turn that kind of... Where did I go wrong? I could have turned a failed TV career into millions if I had just known how. You just, like, not had an integrity, you know, and that's all you need to do. No, I don't know. One more story just to depress us for the week. On top of everything else going wrong with the world, There's a sand shortage. No, Leo. I was just in Hawaii. There is no sand shortage. No, there has actually been a sand shortage for a while, depending on where you're talking about. Yeah. Okay. I thought sand was like the most common thing. No, no, this is a real problem in Florida every time it gets hit by a hurricane because you've got to go and get more sand. Oh, yeah, it washes all this and build it up. Right, right. The sand, maybe it's not shortage. It's just underwater. or you can't get to it. So. And there's only, there's a limited amount of dredgers in any given area, and whenever one area needs sand, all the surrounding areas need sand. So they can go back and get it from the ocean floor and bring it back up to the beach. Yeah, but it takes a long while, and there's only so many dredgers. So you need sand for construction, and chips, that's what silicon's made out of. Yeah. All right. It could have been germanium, but it's silicon instead. Ladies and gentlemen, that is the end of that. I hope you enjoyed the show. Thanks to Chris Focal Walker, a great guest, who will, I'm sure, appear back here again because he's doing some really interesting things with AI. Thanks to Paris Martineau, who is working hard. She's got red string going from article to article, picture to picture, as she ferrets out the latest in food safety for Consumer Reports. Did you just call her a ferret? No, she's ferreting. She's not a ferret. I should probably get my thesaurus out and look for a better word than ferret. Weaseling. Weasels out. Anyway, great to see you, Paris. Thank you so much. Jeff Jarvis, his new book, Hot Type, is available at the website jeffjarvis.com. August 20th. Are you doing a pub party? No. Do a pub party. We'll be there. You're not going to have a party for the publication of your brand new book? Usually someone throws it for you. Should I throw you a party? Oh, let's throw them a party. Can we throw you a party? Leo can interview you about the book. Oh, that'd be good. I am going to want to be on Big Kids Table when it's out, if I can. Oh, yeah, deal. Oh, absolutely. Okay. Yeah, yeah. He was talking about Twitter, too. Yeah, I want to sell it all over. Absolutely. And I'm sure you'll get on Windows Weekly to sell it. Yeah. Hey, if James Comey can plug his novel on CNN. He can do anything. He might have regretted that a little bit. They ended up not talking so much about the novel. But instead, they kept trying to talk about seashells for some reason. So speaking of seashells, you're back from Hawaii. Are you adjusted? Yeah, no, I didn't want to come home, though. It was a lovely trip. But I came home with a new addiction to coffee, so that's good. Yeah, Leo came home and immediately sent seven very precise pour-over-based questions to our friends. I know. I'm really getting into this. I haven't even made one pour-over yet, but I have spent hundreds of dollars on equipment. Talk to me when you're deep in the meta-analyses. How long does it take to do it for over? Three minutes, four minutes. Yeah, three minutes. Is there an established time that you have to do it? Oh, yeah. Yeah, you want to do it for three minutes. It could take longer, but that would be bad. Because it would cool off. No, it's because it wasn't going through the paper fastener. It would suggest that, you know, your extractions, yeah, getting clogged. See how easy it is to get them going? It is. Well, I mean, if you include the grinding, so that's about 30 seconds. You don't want too much fun. You've got to boil the water. That's a minute or two. And now I'm doing a thing where I pour the water and then I pour it in the switch with the switch up so I can preheat it. But I do that while I'm grinding. So it's really under five minutes altogether. I watched a video that said, oh, get the plastic V60 because then you don't have to heat it up. I don't want those. I don't want plastic. That's also why I haven't done any. I'm very sorry. That's why I haven't done a plastic jug like you did for filtering, because I don't want my water going through plastic. That's a good point. Although you're filtering it out. Before we start the show, this is actually what I want. So I got the glass V60. Not the switch, just the regular glass V60. And I'll just have to heat it up, that's all. And you want to wet the filter paper anyway. So it's just part of the process. you know what's going to be funny after all of this? I'm going to go, that's terrible. I'm going back to the espresso machine. Thank you, Paris. Thank you, Jeff. Thanks for putting up with Paris and me. And thanks to all of you. We do this wonderful show, Intelligent Machines, every Wednesday, right after Windows Weekly. That's 2 p.m. Pacific, 5 p.m. Eastern. That is 2100 UTC. You can join us and actually watch live if you want, because we do stream it live. If you're in the club, you can watch it live on the club in the club, Twit Disco. But if you're not live, you can. Oh, I did. I mentioned I forgot. Even if you're not in the club, you can watch it on YouTube, Twitch, X.com, Facebook, LinkedIn and kick. My apologies. I almost forgot. next week oh I'm really out of order after the fact on demand versions of the show available at the website twitter.tv slash I am or on YouTube and of course you can subscribe to your favorite podcast client and if somebody would just write us a good review we could find I checked and someone wrote a has given us one review since the last week and it's a three star review that says I can't wait for this AI bubble to pop so it's a show about AI If you think you can do better than that, please do, and I'll read your review out loud and do a little song and dance. We've got some great guests coming up. Frederick Riven will join us next week. He's the CTO of Dashlane, a password manager. We'll talk about the threat, AI threat into security. Rick Salmon, an old friend and a brilliant photographer who is doing a lot with AI in photography, will talk about the pros and cons of AI photography. Robert Tercik, I think this is your guest, Jeffrey. Yeah, he's a wonderful guy. Yeah. What is his latest book about? Well, he's writing two of them, but he's an old media executive. Not old. Well, he's old as I am. He's from various media companies you know and is an artist. Nice. And has his own podcast called The Futurists. We will be talking to him in a couple of weeks. And then we also are booking now a couple of former Google security experts. We're going to be talking a lot, I suspect, about mythos and the impact on the security of AI. They will be talking about that. And Ian Bogost is going to come back in July. His book comes out. So we're going to have some fun guests. I hope you will be here for that. We'll see you next time on Intelligent Machinists. Bye-bye. Hi there. Leo Laporte here. I just wanted to let you know about some of the other shows we do on this network. You probably already know about This Week in Tech. Every Sunday, I bring together some of the top journalists in the tech field to talk about the tech stories. It's a wonderful chance for you to keep up on what's going on with tech, plus be entertained by some very bright and fun minds. I hope you'll tune in every Sunday for This Week in Tech. Just go to your favorite podcast client and subscribe. This Week in Tech from the Twit Network. Thank you. I'm not a human being, not into the animal scene. I'm an intelligent machine.