[00:00:00] Humans will never be more intelligent than AI. There's going to be two types of companies. Those are great at AI and those that went out of business because they weren't. How do we build a future that is human-centered? I'm Jason Howell, and on my podcast Pioneers of AI, we answer that question and so many more. As an AI scientist, entrepreneur, and investor, I know what it takes to build AI that works for everyone.
[00:00:28] Every week, I sit down with the pioneers shaping our future, and we take you behind the scenes of the AI that's transforming our lives. Find Pioneers of AI wherever you tune in. Coming up, not one but two massive open-weight models just dropped, and both of them are challenging Chinese dominance in the open-weight territory.
[00:00:52] A 19-year-old raised $11 million to sell you a $3,500 personal AI computer that's worth more like $5,000, but don't get your hopes up. It's sold out, at least for now. AI just started legally prescribing medication in Utah, and Xreal's Aura smart glasses finally have a price. But does it have a market? That's all coming up with me and Jeff Jarvis on this episode of the AI Inside Podcast.
[00:01:31] What's going on, everybody? Welcome to another episode of AI Inside, where we take a look at the AI that is layered throughout the world of technology. The world of technology. I'm one of your hosts, Jason Howell, joined by Jeff Jarvis on the other side, not of the world, but of the country, at least. So, you know, we're covering some geography, at least. Yeah. Yeah. Yeah. We're the coastal elites. That's what we are. That's all we are, is the coastal elites.
[00:02:01] You could at least call in from France, Jeff, and then we'd have some direct tie-in to our top story. Well, next week I might be in Germany. Oh, okay. That's closer. Okay. That works. Yeah. That's legit. The day after our show. And then this weekend, I'll be elsewhere on the coast, just to mention, just to plug real quick. Yeah. I will be in Haverhill, Mass, near Boston, for a book talk at the historic New England in Haverhill at 10 o'clock in the morning.
[00:02:29] And then we'll go up the hill to the Museum of Printing, and you can see a live working linotype and smell it and feel it. And so it's free. If you're anywhere in New England, please, please come to Haverhill. If you go to Jeff, for those of you who have video, Jason's showing it now. If you go to jeffdarvish.com, there's a link there to register. Free. Register here, and you can bloop, bloop, and go. And I mean, it's a full-day event. That's exciting. It's a lot of time, man. Yeah. Yeah.
[00:03:00] I love it. So cool. And you get your book signed by Jeff, and you get a cool tool. You visit one of Haverhill's exceptional local restaurants. That part's not free. I was going to say, I don't think that the lunch is paid for. No. You get to have lunch. Yeah. That's cool. That's nice of us. Yeah. Right. That sounds really cool. If I was in the area, I would definitely go.
[00:03:26] Go to jeffjarvis.com to find more information for that and all of the wonderful books that Jeff has written, including Hot Type, which I'm sure will come up at least a couple of times in today's show. See, it already has. We're doing the work. Hey, let's talk about Mistral. Hey, Meastrol. Let's see here. I got to get this overlay accepted. I don't know what that was all about. Right off the bat. What do you think? Right off the bat. Right off the bat.
[00:03:55] Meastrol wins the award for naming a model. I like the name of this model. Le Chonk. Le Chonk. I mean, it just, it has personality to it. And it sounds big. It sounds really big. Yeah. And kind of, I guess. Kind of in a cuddly way even. Yeah. Like I said, there's a character to Le Chonk. Like I get this vision of like this massive like.
[00:04:22] It's like the Burmese mountain dog of AI models. You know? Yeah. But is Le Chonk cute? I don't know. I think, I think a scritchable. Yeah. Wait, what? Scritchable. A little scritch behind the ears. Oh, I see. Kind of. Okay. Yeah. Yeah. Yeah. So, but I'm serious actually. That to get away from these kinds of ridiculous techie names. Yeah.
[00:04:49] To get to something somebody can have any handle on. So right there, right there. I'm in favor of it. Sure. So it's Mistral Large Four. Large Four, a.k.a. Le Chonk. And this is an open weight AI model. Very large as the name implies. 49 billion. Sorry. Trillion parameters total. 49 billion active at inference. It is a mixture of experts. Experts.
[00:05:18] And let's see if my understanding of mixture of experts architecture is spot on to yours or is near yours. But it's basically that certain, well, I guess you would call them experts move to the front as you're interacting with it. So all of them are not stored in memory at any given time. Right. It basically reaches into the kind of collection of experts inside of the model and pulls forward the ones that it needs to use.
[00:05:45] And so it's a very efficient way to get these massive models into a smaller container, right? So my children, back in the day when I had my first computer, my Osborne 1, it had two five-inch floppy drives in it. And one would be for data and the other would be for program. But the memory in the computer was so small that it constantly had to go back to the disk to get the next instructions and put it back. And that's the way it worked. So it was a structure of going back and back and back and forth.
[00:06:15] Yes, that was the luggable computer, 28 pounds. I mean, it was portable for the time. You could also see that. I donated mine to the Museum of Printing in Haverhill, Mass. Oh, did you? Yes. Did you? My Osborne is there, yes. Look at that tiny little screen. That screen is like smaller than a smartphone screen. Yeah. It was WordStar. Notice there's no cursor. No mouse was there yet. Okay. No trackpad. You didn't have directional keys. That's all. Otherwise, everything is control and a key.
[00:06:45] So control H was backspace and erase. Yeah. Remember that for WordStar. So anyway, when you dealt with smaller memory, you had a structure, an architecture that it could constantly go off and grab the next piece it needed for that. So it's kind of a time-honored way to shrink something into its essence so that it can fit on smaller hardware. Wow. The chunk would not fit on the Osborne one. Yes. Yes.
[00:07:11] If that isn't abundantly clear, if you couldn't have imagined that to be the case, that is absolutely true. This model was trained entirely on 3,800 NVIDIA Grace Blackwell GPUs inside of Mistral's own data centers in Europe. So they're not renting from anyone in order to do this. It's also very inclusive. It's also very inclusive.
[00:07:40] 160-plus languages, including every official EU language inside. So it's sovereign. It's inclusive. It's inclusive. And it's not coming from China. And it's not coming from the US. Those two things right there could be enough for someone to be like, all right, I'm in, you know? Yeah, so that makes me happy right there. Yeah. For all the years that I went to Germany, especially Europe in general, people would complain, we want our own Google. Fine. Invest. Build it.
[00:08:09] I want more competition. It would be great if you had more competition. So I think it's extremely healthy that France has pioneered this, and it really matters. I read it. So, pardon me. I'm sorry. It's okay. The Süddeutsche Zeitung had an interview with Arthur Mensch, which is a great name, by the way, the head of Mistral,
[00:08:33] after he had worn Bavarian traditional attire to present at Bits and Pretzels, a startup conference, which is kind of funny, and a really good interview with him, in which he's just a voice of reason. There's no such thing as a doom scenario. It's not going to happen. It's ridiculous, he says. And he goes on to say, I'm trying to find it here.
[00:08:57] For example, he's asked, how about an international panel of experts needed to go into companies and check what can be published? And he says, no, that's regulatory capture. It's a classic trick used by large technology companies. He recognizes what that is. He's asked whether Mistral had fallen behind. And he said, now they have this big new model. The open weight models are only a matter of a few months behind the frontiers.
[00:09:22] And so it's good to see a competitive structure here for AI in international hands, not U.S. hands, given everything that's going on here, not Chinese hands either. So got to root for Mistral. Absolutely. Man, he's saying all the right things. Yeah. I don't think he's just saying it just to say it. I mean, I think he's spot on.
[00:09:48] We are so marinated in a certain perspective and culture and ideology around everything happening in USAI right now. It's nice to get a perspective that isn't so consumed by it that it's hard to see the forest through the trees, right? Like they're kind of outside of that perspective. They also have something to gain from not sharing that perspective for sure.
[00:10:16] But yeah, I think that's a real sober read. And I don't know. I've been pretty impressed by what I've seen about the kind of development of Mistral. And I'm happy to see this getting as much attention in play as it has been. It's a good time. And he sees himself very much as a competitor to OpenAI and Anthropic. And he says that, and this is I'm quoting the Google translate of the thing as I don't have time to translate it. There's a financial incentive behind what their model is.
[00:10:46] OpenAI and Anthropics business model is based on exclusivity. They train these models and then sell access to them with a 90% gross margin. Now they're under pressure because this business model won't last much longer. Many companies are now relying on open source models and adapting them to their specific needs. That's Mistral's model. That's the mouse biting the elephant here. And it matters. Yeah, indeed. And I mean, you know, it is big. It's also powerful.
[00:11:14] They say it beats DeepSeq V4 Pro, Quinn 3.8 Max on coding, outperforms GPT-6 Astra on financial and legal tasks. Also says it seems like it's pretty solid on security and vulnerability reproduction. It did well there. Large four ranks in the global top five for vulnerability reproduction.
[00:11:41] So, and I mean, part of that is because this being an open model and the way that they're going about this, it doesn't refuse security research tasks the way a lot of the closed models do. Mm-hmm. So that's going to be great for folks in the security field that want to get these higher capabilities and not feel like they're being restricted. You know, if you're a pen tester, if you're a security researcher, you're working with CLOT or ChatGPT, you're going to run into these roadblocks.
[00:12:09] Mistral is saying, well, we'll do that. We'll go ahead and do that. Should we be concerned about an open weight model of this magnitude that won't refuse those tasks? We should be concerned about any human doing anything. Yeah. It's less the model, it's more the humans. It's the human. And can people do bad things with powerful tools? Yes, they can. Always. And they always have been able to. And if you think that you have a technology solution to that, that's where you fool yourself.
[00:12:39] Yeah. Right. Right. But people really want to believe it. They really want to believe it. People ask how you make money with an open weight model. And the student who asked him, how do you ultimately make money? And he said, at the model level, the margin disappears. It won't be sustainable there. The margin shifts instead to the infrastructure. But integrating AI into companies is also enough work to make it a profitable business. Okay. Okay.
[00:13:10] Well, there you go. In a way, he's also, he's feeding into the Palantir model, which is just that. But he's also saying, in a way, he's competing with Palantir as well. So it's a double-edged sword. Yeah. Into the market to cut at the frontier companies. Yeah. And I think he's right about the frontier companies. I think their margin is going to be challenged completely.
[00:13:31] And I saw some numbers, somebody was running from CNBC, that the capital expenditure going forward is some ridiculous $4 trillion. And the amount of revenue you'd have to get to pay that would be something like $6 trillion. And it just isn't evident that the income on AI is going to be that high.
[00:13:56] This whole hyperscaling world and 90% margins on these models, it's not sustainable. And I think we're going to see, I hope we don't see an implosion of AI. I don't think we will. But we will see, I think, an implosion of kind of the presumption of scale in these models. Yeah. Running into a ceiling and, yeah.
[00:14:26] I mean, people are, I think what, and actually we're going to talk about another story that's related to this in just a second. But I think it seems to me that one of the big emerging stories in recent months has been the rapid production and increasing power and capability of open models. That conversation always seems to go back to China because for the last year, China has been
[00:14:54] the kind of geography or the region that has really been leaning hard and deep into that. Now we have some options for people who feel nervous about that. Also, Europe kind of has, you know, continues to have this example of, hey, we are in the game. We do have, we are competitive. Maybe we're not the, you know, we don't have all the companies that are stealing so much of the spotlight like Anthropic and like OpenAI.
[00:15:23] But Mistral's no slouch. And I think this really goes to show that Europe can play in the game and that there are options outside of the US and China. And I think that'll make certain companies and certain people really happy. The other thing that we've talked a lot about, both open weights and locally run models. And now you have to buy a $5,000 Spark or a $10,000 Mac to run most of these. And I don't, I think Mistral's model is beyond that. I think it's more for enterprise and corporate. Oh yeah, it's definitely beyond that.
[00:15:56] But, you know, I wonder where that goes. I saw some TikToks that went by, which I didn't believe that you could run one version of Quinn on any old computer. And I don't believe that's the case. I think it was, I think it was fooling people. But this idea that you can pull in pieces of it to get smaller. And, pardon me. One thing I'm anticipating is that your bot, people are buying Sparks with Blackwell chips.
[00:16:25] And Blackwell chips are meant for training. They're very high powered. I wonder when we're going to start seeing more and more computers that are built with inference chips like Grok, which was half acquired by NVIDIA. And we don't need, at the retail level, we don't need the power of training chips. And so what might we be able to use otherwise?
[00:16:50] Will we see the cost of the hardware needed to run some of this come way down? And you look at things like Meta's little cutesy agent, Muse. You're running a virtual machine. Well, that's just really a harness to their big model. But I think that we're going to see more and more efforts to bring AI down to a personal and retail level. Boy, are we. I'm watching for that. Absolutely.
[00:17:20] And I mean, the first half of this show is dedicated to exactly what you're talking about. You got Mistral, of course. And then kind of timed with all of this is another company, a U.S. company in this case, but Reflection AI releasing Beam. This as well is a fully open weights model. 501 billion total parameters. So not quite as chonk.
[00:17:47] It's like I think I said on Daily Tech News show yesterday, le petit chonk. Le petit chonk. Le petit chonk is Beam. So 501 billion total parameters. 23 billion active at inference. Also a mixture of experts designed. It was trained on 6,144 GB300 GPUs, if that matters to you. But it is Apache 2.0 license, fully open.
[00:18:16] And the founding team here is DeepMind. You've got a couple of folks from DeepMind who came over to do this and then, of course, backed by, yes, NVIDIA is in there. Sequoia and others valued at about $25 billion. And they're saying that this is, you know, a very competitive open model coming from the U.S. So you've got now we're starting to see more of these open weight models making their way from U.S. companies as well.
[00:18:41] So giving you even more options outside of the cloud-based frontier. And, yeah, people are getting really excited about this, too. I mean, I think it's just that, you know, choice is good. And in a moment where open is really proving itself, even if it's not 100% as capable as something in the frontier, it's still incredibly capable. And there is the kind of value tradeoff, which is, yeah, but I own it. It's local.
[00:19:09] I don't have to go to the cloud for all the things. So maybe I'm okay with the fact that it isn't pushing, you know, the absolute boundaries. It does enough for me. Yep, exactly. Yeah. And to keep up, I think all the big companies are going to have to release open weight models. Smaller, less powerful, but who cares? Does OpenAI have an open weight model? Google has Gemma. Yeah. Meta's had Llama.
[00:19:37] I don't know what the latest is, but they say they plan to stay open. They must have one. Anthropic and OpenAI. I mean, they do have open models. GPTOSS is what I'm coming back to. I don't hear people running that. No, that's interesting. Yeah. OSS. Open reasoning models designed to run locally on laptops. 120 billion and 20 billion parameters. So definitely smaller.
[00:20:07] But yeah, I do wonder. That's a really good question. At what point do we start to see? Yeah, Pegasus in our chat says it's old, which I do remember talking about this, but it hasn't been any time recently. And so I do wonder, that's a great question. When we start seeing these companies who have really leveraged heavily their cloud frontier model architecture and approach,
[00:20:34] are they going to start releasing very competitive open models? Because, hey, you can't ignore it anymore. This is where a lot of the development is happening. So you've got to be there. Yep. Yeah. Interesting. So yeah, there you go. A couple of real open-weight alternatives not coming from China, which is definitely different from… I'd call it a trend. I would call it a trend as well.
[00:21:04] And yeah, I think really we're going to continue to see people sort of picking sides as more models come from all these different directions, you know, based on where their feelings are at. You know, I want open, but I don't want China's open models or, you know, U.S. or whatever the case may be. Now there's competition and options and they're going to need machines to run them on. Right? Yeah.
[00:21:31] And the other thing is that it opens up the possibility for all kinds of other development and entrepreneurship. On our… That's true. …second interview, I think it was, with the then CTO of Shipstead, we talked about how he built a, with a partnership with the university, an open Norwegian language LLM. This is now, what, Jason, two years ago? Yeah, Sven Sturmer Thaulo, right? Right. And where they do it, they did it on top of Mistral.
[00:22:00] Because the open weight models exist, universities, labs, startups can use them to build all kinds of other things without buying debt to do so. Mm-hmm. That's right. Great. Okay. Well, there we have that. Sorry, my interface is suddenly blurping on me. Okay. We've talked about the models themselves.
[00:22:28] We've got the hardware that maybe you can run some of this. Maybe not all of it. You're not going to be able to throw – you're not going to be able to get the hardware to run Mistral's LeChonk on a $3,500 computer at home. But we're going to talk about a computer that now exists for $3,500 and something, you know, an update from NVIDIA coming up here in a moment as well. All kind of driving to a similar point, which is, you know, and something that you and I have talked about a lot, which is,
[00:22:56] when do we get to a point to where these things become a lot less technically complicated and a lot easier for just everyone to buy a machine and get started? And I think we're starting to head in that direction, it seems like. But real quick, I want to thank our patrons, patreon.com slash AI Inside Show. We have a bunch of folks there who help us do this show and support the show on the back end each and every week. We can't thank you enough.
[00:23:23] It's really important and critical to the continuation of this show, right? Like, yes, you can get it for free. You can also contribute a little bit. You get some bonuses like a daily podcast, which I did just put one up right before this show. So you can see here, Wednesday, October 7th. All of the thumbnails are automatically generated, so you're guaranteed to see me looking real dumb in a lot of them. But I just don't have the time to correct it. So I just keep it. That's my gift to you, me looking dumb in thumbnails.
[00:23:51] And I also want to throw a quick thank you to a few random patrons from our role, our patron supporter role, Corky Garko and Stephen Linthicum. Thank you both for continuing to support us. Patreon.com slash AI Inside Show. Going to take a quick break. And then when we come back, we're going to talk about a few computers that you might be able to buy to install these onto if they weren't already sold out. That's coming up.
[00:24:23] The Wired Newsroom is known for award-winning reporting on how technology shapes our world. On Wired's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people building, challenging, and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday, wherever you get your podcasts.
[00:24:54] All right. So let's see here. Everybody was going nuts yesterday about this new startup. And it is, what is it called? Ghost? Ghost. Ghost launched. Yes. So Ghost is the startup. Core is the AI computer. It's a personal AI computer. $3,500. And actually, the founder of this company, just to mention real quick, because it's always interesting when this happens, 19 years old.
[00:25:24] That's – the founders are getting younger and younger, basically. His life mission was to be a quant who uses math and data to guide his investment decisions. But instead, he is one. There you go. Hey, you know what? When you've got an idea – I mean, clearly, he had an idea that worked because these machines have already sold out on day one. But basically, what do we have here? The machine is core.
[00:25:50] Inside NVIDIA RTX Pro 4000 Blackwell GPU. 24 gigs of RAM of memory. AMD Ryzen 5 7600. 64 gigs of – let's see here – of RAM. Oh, I see. The GPU has 24 gigs. The CPU, 64 gigs. Got it. One terabyte SSD. And yeah, this is basically like a personal AI computer.
[00:26:19] Pre-installed with QN 3.8. Next, QN 3.8. 27B and Gemma 4.31B. You can install any models you want from Hugging Face. And he said – the TechCrunch story said it differs from a Mac Mini in that the hardware and software were built specifically for running AI agents. With the models already installed while mic minis are general-purpose computers that have to be configured to AI.
[00:26:49] You have to kind of go in there and put your own smarts in – This is still a pretty nerdy box. Still pretty nerdy, but it kind of gets you a little closer to the I bought a machine because I wanted to bring AI into the home. I mean, are we at the iPod moment? No. We're not necessarily there yet that anybody can just pick it up and they know exactly what to do with it.
[00:27:14] But it removes a little bit of, I guess, the trick of installing and identifying and everything. Right out of the box, you have this. It has some software, which they say they'll update. And they've also said even if the company goes away, they've made this in such a way that it continues to work. So it's not reliant upon them in order for you to get what you need out of it, which is nice. So the team, the founder, Zion Javed, is 19.
[00:27:42] Other founders are 19, 24, and 23. All right. Young team, go for it. Hey, if you can, might as well. My goodness, that's awesome. You can connect it to all your apps, your calendar. It's the agentic stuff, right? Your smart home, your wearables. It builds context on your life over time. Does some of that action proactively? People are going to wonder, do I want it?
[00:28:11] Monitoring all of my things and building that context. But I mean, the beauty here is that this isn't running in the cloud. This is running on your local machine. Given that you trust how it's reaching out of the machine to do things for you, I mean, technically, you should be able to control all that data and keep it local. It also has an onboard firewall that limits what the agent can actually access. So there's that, given that works as designed.
[00:28:40] But yeah, there you go. Is this the actual? I mean, this is the image and tech crunch. It looks like a gold brick. Yeah. Or the obelisk in 2001 that the apes. Yeah. Yes. Oh, that's a good. Yeah, you're right. That's funny. It's scary. It's AI. Oh, my goodness. It's the computer. Everyone bow to the superpower in the cloud or actually not in the cloud on your desktop.
[00:29:08] So this is really interesting from a pricing perspective, too. $3,500 for this computer. The GPU that's inside it alone is worth $3,200. Wow. The case is estimated at around $1,800. Then you have the terabyte hard drive. You have the 24 gigs of RAM. TechRadar kind of broke this down and estimates the parts to be worth somewhere in the range of $5,000.
[00:29:37] But it's selling this for $3,500. How did they do that? Apparently, they scored a huge batch of GPUs before the prices jumped. And so they were able to get them for a lot less expensive. And then, of course, this sold out in a day. So I'm sure, you know. They've been in skill free here. They got $11 million, I think, from Andreessen Horowitz. Uh-huh. So they want to make a splash. That's a way to make a splash.
[00:30:05] That's a way to make a splash. And, yeah, the math doesn't work out, but only based on today's prices. And, you know, they only bought a certain amount of those GPUs. Are they already out of them? And if so, this price is probably going to go up at some point, I'm guessing. Because they're not going to continue to lose money on it. So it's a costly game to play. I wonder what will happen when the Vera Rubin comes out from NVIDIA, whether there'll be an aftermarket for Blackwell Chips that will come way down.
[00:30:35] Oh, yeah, that's a good question. Yeah. I imagine so. How far down would they come if there's so much pent-up demand? Probably not far, but they'd be discounted something. Yeah. Interesting. Yeah, I think this is a trend. You know, this is also trend spotting. We're going to see a lot more of this in the coming couple of years. Although it's happening at an interesting time when component prices are so dang expensive and everything.
[00:31:03] But, I mean, people – I mean, I think, you know, the real nerds, the hardcore people who would actually buy something like this, they're going to pay the money anyways because they want the capability. When you were in Hawaii, what did you hear about Remigant?
[00:31:18] Oh, well, I mean, Qualcomm based on – mostly I just heard about it in relation to Qualcomm's new chip and how they were recognizing that, you know, and addressing the high cost of RAM by building more capability into the chip itself on storing models on chip.
[00:31:40] So, thereby not requiring as much of the RAM kind of capability to power those models effectively. So, basically moving some of that onto the chip and away from the memory itself. That was kind of my primary takeaway from being in Hawaii and hearing about Ramageddon. I didn't hear everybody talking about Ramageddon, you know, necessarily.
[00:32:04] I mean, I think everybody, at least in the tech press, is just kind of used to the fact that everything is just way more expensive now. And it's just kind of quickly become the norm. Yep. Yeah. The cost of everything has jumped, you know. All the smartphones and everything. Samsung just announced their latest crop of smartphones, you know, all jumped $100 to $200. That's becoming pretty standard for the smartphone industry at this point.
[00:32:30] So, yeah, it's just part of the murky waters of where we're at right now in tech. So, you got this for $3,500. Almost certainly going to jump at some point, right? Once they have more stock. I imagine these don't cost that small an amount when the GPUs go up in price.
[00:32:54] And then you've got NVIDIA, which has already been selling a similar-ish computer, the DJX Spark. But now they have a new configuration with 64 gigs of unified memory, shipping near the end of this month on October 23rd for $5,000. Which, you know, which lines up with TechRadar's estimation of the cost of the previous computer. Almost exactly. I don't know which is the better one to go with.
[00:33:21] Probably the NVIDIA one because it's coming from NVIDIA and maybe from a support perspective. I don't know. And you can get it from NVIDIA, Acer, Asus, Dell, Gigabyte, HP, and MSI. Yeah, right, right, right. So, take your choice. I don't know what separates those manufacturers. Yeah, I don't know either. Runs models up to 100 billion parameters.
[00:33:48] Grace Blackwell, GB10 chips inside. And you can actually daisy chain these together with a QSFP connection for 128 gigs unified together. So, that gets you in the 200 billion parameter range. You nerd. Nerd. You nerd, you. Love to see what you're making with these things. If anyone is watching or listening and has one of these and uses it regular for stuff.
[00:34:17] I mean, anyone who isn't Dr. Dew. Because Dr. Dew, I know you're doing so much of this. We hear about it in the Discord all the time. In addition, yes. In addition to Dr. Dew. But let us know. Contact at AIinside.show. Let us know what you're doing. But, you know, these are all designed to run those open weight models that we were talking about at the beginning of the show. And out there. And again, I think this is just, this is a huge trend. More. Okay. So, here's a question for you.
[00:34:46] Um, I feel like we started talking about AI computers like a year, year and a half ago on this podcast. I can't remember what it was that kind of kicked that off. But we were trying to, at the time, we were scratching our heads going, what on earth would I need an AI computer for? A year and a half or however long later, do you feel like you'd answer that differently now? Do you think it makes more sense for you now to have an AI computer? Not for me, but I think it would for you. Because you're running a business. Yeah. Yeah. Yeah.
[00:35:15] Or consulting and so on and so forth. And I think that's, that's one. Here's an example where, where I do see it coming our way. Um, I'm working right now on a proposal, a grant proposal with Montclair State to bring AI into a project we're doing with the library newsroom project. Um, which is its own wonderful thing, uh, where we want to bring communities and libraries so they can kind of control their data.
[00:35:39] And my thought was, let's bring AI into that and show communities that they can control that AI. Now, I'm not suggesting that every library is going to have a DGX Spark, not for a while. But Montclair State could have a DGX Spark and can run open weight and local models on behalf of a few of these, uh, community newsrooms and give them the confidence that they're not handing over information to a big company.
[00:36:09] They're not handing over money to a big company. They're not, you know, in for a huge cost ongoing because they keep on buying tokens. Um, I think that's the model we got to show people so that we can present a model to news organizations. Uh, if you're, if you're a newspaper, if you're a real old time newspaper, you should be able to afford a DGX Spark or two to do this stuff. If you're a bunch of very small weeklies or online local blogs, then share the resources.
[00:36:36] Join together and, uh, around, let's say university. I think that's where I see the next phase of this. Hmm. You know, and then the question is, what about small businesses? Um, I don't know. I'm not sure where that goes. I think, I think the, the, well, you tell me you're, you're running a small business.
[00:37:00] I think, I think, yeah, I think these machines start to kind of make more sense for, for small businesses. I mean, I, I already hear, you know, in some of the, like the, the educational stuff that I've been doing with the workshops and everything like that. I hear from people who are running small businesses who recognize that AI can help them run their business better, but it still feels a little out there and it still feels disjointed and whatever.
[00:37:27] And I can kind of foresee in the next couple of years, more efforts like this being marketed and created in such a way. Also, you know, combining things like these proactive agents, like, even though Muse happens in the cloud, but things like that, but on a lower, on a local level, making things a lot more easier for, and, and maybe more.
[00:37:51] More streamlined and specific to running a business because, um, I feel like I go in a few different directions, but like, as I've talked about many times, like I use Claude a lot for running my business right now, but I'm also kind of an architect around how that gets built out. And I know that I'm imperfect in how I build it out too. It's, it's running a lot of things, but it's also not entirely efficient and certain things do fall off the map.
[00:38:15] And I'm constantly thinking like, man, someone's going to come along and they're going to create something that does all of the potential of this in a way that is, that doesn't drop the, the ball in certain ways that really effectively collects that information and presents it to small business owners. As one example, in a way that is kind of like a no brainer way where it's like, you know what, get this box.
[00:38:43] And it already knows the functions of running that particular business or whatever effectively. And you don't have to be the architect of it. You just kind of like install the box. It has this powerful model. It already has kind of the routing and the understanding of what it does with that information and where it puts it and how it brings it back into all the other work that you're doing so that you don't have to use it and be like a creator of the system. You can just use the system.
[00:39:13] And yeah, you know what I mean? And I do see that as being something that will probably happen. I mean, I could be wrong, but I can totally see benefit for someone like me who runs a business to just be like, I don't want to have to create all this shit stuff. Sorry. I don't want to have to create all this stuff. I just want it to take all the benefit of what it's capable of doing and to do it right. And I could totally see that. Here's another example. It's my friend, Beth Simone Novak, who I knew when she was at Rutgers. She's now at Northeastern University.
[00:39:43] She wrote an op-ed in the New York Times. It's down toward the bottom of the rundown about how AI can be good for democracy. And her example is that when she was in New Jersey and working with the New Jersey Office of Innovation, which she began leading after its founding in 2018, they built an AI tool to match records across systems and find eligible children who hadn't yet been reached for federal food benefits.
[00:40:11] And the benefits were then extended to 106,000 more children in the first year and 96,000 in the following year. No family filled out a form. The state used information it already had. And so I think this is a huge opportunity to take these things and because of worries about privacy, to be able to say we're going to run this locally. We're not going to be sending it up to OpenAI or Anthropic. Right.
[00:40:35] But we can take the data that the state, that local governments already have and use it. And that can happen inside the government. It can happen outside the government in library newsrooms, in old-fashioned newsrooms, in lots of places. So I think there's a big opportunity here to show best practices around these kinds of activities. Mm-hmm. Yes. Yeah. Journalistic, small business, government, education.
[00:41:02] We got to start seeing these examples where it's useful. Mm-hmm. Mm-hmm. And the big cloud-based frontier labs continue to – I mean, Cloud does this a lot where they'll release some new segment or plug-in set for Cloud. That's for small business or for this area or that area or whatever. But it's still firmly in the cloud. You still have to be okay with sending all your stuff in the cloud.
[00:41:30] There's probably some extra kind of latency in the experience as a result of that versus being local and you lose that sense of control and everything like that. And I think that's only going to appeal to a certain subset of people. When you're talking about business, when you're talking about government, education, all that, yeah, there is a real need for knowing that you've got control over that information. You're not just parsing that off for them to train their models on future – future models from.
[00:42:00] So, yeah. I'm curious to see how it all develops. And I hope that through this, the whole thing improves. Like, the process of things like running a business off this stuff improves because the more I work with it, the more I'm like, man, there's just so much potential here. And I am doing a lot with it, but I can see so many ways in which there's room for improvement. And I feel like someone's going to figure that out.
[00:42:30] And maybe that someone should be me. No, I don't even know how to tackle that. But someone's going to figure that out, and it's going to be a real eye-opener for people who are entrepreneurs and small businesses and everything and be like, oh, my goodness, this is the tool set I always needed. And it covers so many facets of my business through this one interface. So, I don't know. We'll see. Yep. Did you – have you had any time to play around with Muse? No, I haven't yet.
[00:43:00] I haven't. I've got to spend some time to do it. I've got to find the right tasks for it. Yeah, that's kind of my thing too. I did install it, and I did get it set up and pointed it at my – kind of my Gmail accounts.
[00:43:15] And I pointed it at my folder structure, everything that Plot has been kind of creating for me over the last however many months and everything, which arguably is a huge data set at this point because I've just been adding to it and refining it and everything. What I noticed on the other side is like, Muse, make sense of all this stuff. I want you to learn everything that you can learn about me from this folder.
[00:43:42] And it makes me realize how incomplete that folder actually is because there's certain things that it's like, oh, well, this is an open loop that you still need. And it's like, no, I closed that loop. That happened. But maybe it wasn't updated there. So I was kind of hoping that it would get that and immediately we'd be off to the races. But what it came back with was very incomplete. And so I think where I'm at with it right now is just keep an eye on my inbox and look for things that I may miss.
[00:44:11] I just want to know from collaborations or sponsorship opportunities to something that I've overlooked. I mean, it's kind of basic, but I want to see from its proactive sensibility if it actually starts looking out for me in ways that I feel my other systems I have to execute on it myself. Because it's always running. Yep. Yeah, we'll see. I haven't gathered that yet.
[00:44:40] But I'm interested in this story that Wired wrote because people are, I guess, pushing back or Wired has discovered, looked further into how MetaMuse actually works on a proactive basis.
[00:44:54] And says they were able to, this is Lily Hay Newman is the writer at Wired who wrote this story, able to get the internal instructions from Muse and learn a little bit about how it does what it does behind the scenes. Meta says it deliberately made those files accessible for full transparency. So who knows if that's true or not.
[00:45:18] But basically that instruction set was extracted and how it does what it does. You know, it creates a page for every person in the user's life that includes partners, friends, coworkers, people they follow. It creates kind of like these data points around all of these people. It is Facebook after all. Totally. I mean, some of this is kind of like, oh, so this is like the Facebook knowledge graph.
[00:45:47] Or whatever that's called. Social graph. There we go. You know, where someone lives, their work, their birthdays, their goals, their past agreements and arguments. How close those particular relationships feel. And all of that is, you know, meant to get the Muse assistant to get closer to understanding how those connections actually relate to the user's life.
[00:46:15] And I think some people look at this and they're like, oh, I don't want Muse to have all that information. I mean, I think there's something legitimate to explore there about like building a profile on someone who didn't agree to have a profile built about them. But. Yeah. I mean, I think people are going to react to that. I don't know that I necessarily agree to that because it's just data and information about my life. So, two things.
[00:46:43] One, Zuckerberg, I saw an interview with him somewhere. I don't even know where it was saying Facebook being Facebook, the stuff you put there is not meant to be put in a vault. I've said this for years when I wrote my book, Public Parts. You're putting things there to share. The last place to store something private is Facebook. Yeah. Right. A. B, Facebook already has that profile. Now you got a crack at it, which is more interesting. Right?
[00:47:12] Find me the right gift for somebody or whatever use may come out of that. You know Facebook already has that information. They already have that ability. Should you have it? I don't know. Third, I didn't mean to plug this, but I just wrote a piece for Project Syndicate. It's the bottom of the rundown where I was asked to write about revisiting questions of privacy in the age of AI.
[00:47:35] And I am provocative, as is my want, and said, well, we have to consider – oh, sorry, you've got the subscription block there. I asked a provocative question. Who wants ignorant AI, wrong AI, stupid AI? Right.
[00:47:56] And if privacy advocates and copyright advocates and regulators and crusaders all succeed in starving AI of quality and reliable and authenticated information, or worse, or worse, in my view, poisoning it as a matter of civil disobedience. And if we're all using AI, don't we all suffer? So we've got to have a two-sided discussion here when it comes to data.
[00:48:23] And the one side is what obligation might we have as journalists, as academics, as companies, as governments, as individuals, as society, to these tools that everyone are going to use in areas like medicine? Mm-hmm. But second, of course, obviously, is what obligations do those toolmakers have in turn to us? So we need a two-sided discussion about responsibility.
[00:48:51] And it's not as simple as, I don't want any of my stuff up there. Right, right. To some extent, yes, I get that, and I agree. And should we trust these people? No. But we've got to look at the larger implications of what happens if AI is an idiot. Mm-hmm. Mm-hmm. No, I think this is spot on.
[00:49:11] And this is exactly how I felt about kind of this Wired article is you got people pushing back for the collection of all this and integration. At the end of the day, though, AI is only as good as the context that's fed into it. So if you're going to use a proactive assistant, if you're like, oh, that sounds really interesting. I want my assistant to be proactive because that would be very useful. You want that information in there.
[00:49:40] Otherwise, it's not that useful. Right. The only way it gets useful is understanding these things about you. It's kind of like, you know, again, it's kind of like I hate to put it in human terms, but it's hiring an assistant that follows me around in my business and understands all of my relationships so that my job as the leader of the company or whatever is easier because they understand me as well as I understand myself.
[00:50:04] And so they know how to fill in some of the blanks and do some of these things for me that I don't want to do or whatever that might look like. And so, yeah, at the end of the day, sometimes I get the feeling that people want to have their cake and eat it, too. They don't want to give the information, but you have to give the information to get what they do want, which is a useful AI. And you can't have one without the other, really.
[00:50:30] Let's acknowledge fully that there are good reasons not to trust big tech companies with your data because of what they screwed up here. And I've written books about that and I've screamed about that. We blew it early on. We blew it at the simple level of advertising cookies. Yeah. And made it mysterious and didn't give people control over them and didn't let them know what was being done and why. And now that carries over into AI, but that could affect us all. If you come to AI and you're asking a medical question, you're better off if people have anonymously shared their medical data.
[00:51:01] And we need to have a discussion about whether that's a social responsibility. And people are going to hate me saying that, but I'm trying to be provocative here so we can have a discussion about what those rules are on both sides then. Yeah. I think that's absolutely fair. Makes a lot of sense to me. At the same time, I don't really want to give Muse my email yet. Yeah. You know, this is where I realized how far I've come. Because I was like, oh yeah, no problem. Here you go. Whatever. Here you go.
[00:51:30] You know, will I regret it someday? I don't know. But yeah. I mean, a part of me is kind of like, but I got to know. I got to know so I can talk about it, you know? So, but maybe that's not the case. Maybe I don't. Maybe I can just hide behind, you know, and basically say like, no, I'm not going to. It may be cool, but I'm not going to. Or at some point you get a local model that has the smarts. Yeah. But does the analysis locally. Yes. I would love that. That's where we need to go. That's where we need to go.
[00:52:00] And real quick, just so you know, if you are interested in building your own Muse gadgets, you can. Apparently, Meta launched Muse gadgets, what those are called. It's an open source firmware and dev tools offering for building custom AI hardware. And what this made me think of was Mike Elgin and his son who created for many years. I don't know if he's still creating this educational tool, this assistant for kids to build for themselves.
[00:52:28] I wonder if this is something like an evolution that he could move into. Yeah. Be interesting. I was trying to remember the name of that project and I could not remember it before showtime. So if anyone in chat knows, tell me so that we can give a solid kind of call out for that.
[00:52:47] So Xreal finally releasing details about the arrival of its Aura smart glasses or kind of XR glasses. This is the glasses solution that has a puck. It has kind of like a wired puck to the glasses. It runs Android XR and looks a lot like normal glasses, although the hardware doesn't at all.
[00:53:16] Like when you actually look at it, it looks very similar actually to a lot of the other hardware that I get from different companies in the smart glasses world. Like Rayneo, these are the GT, I believe. And they have like an internal like reflective kind of glasses. I'm trying to show it off so that it's easy to see. In other words, it's not just like a lens and everything goes into the lens.
[00:53:45] It projects down into this like slanted kind of... Like Grammy glasses inside the glasses. Kind of. Yeah, yeah, yeah. It kind of looks like that. Yeah, that's a good way to put it. There you go. But it's kind of like a projection system as opposed to like a wave guide that goes into the lenses. And that's how these work as well. But I've tested these now three separate times. The last time was in Maui a couple of weeks ago. Spent like a solid 45 minutes with them.
[00:54:16] And, you know, it was always like, okay, well, how much are they going to cost? When are they going to come? Still don't have a firm date, but sometime by the end of this year, probably this fall. And it looks like $1,279. So, which is kind of the price that I was expecting for something like this. You know, I don't know how popular these are going to be, but it's a notable moment for these to finally kind of make their way. So, any interest?
[00:54:46] Not really. Or are you completely done with smart glasses after Google Glass? We'll see what Googles are. Yeah. But I think, again, there's corporate use. So, was it Amazon is going to supply? No. Amazon smart glasses will allow delivery drivers. Yes. Well, I thought I saw stories about this. About Amazon doing what?
[00:55:16] They have smart glasses. Oh, that they have smart glasses? Yeah. I mean, I vaguely remember something like this, but I couldn't tell you right off the top of my head. To have more in like a workshop environment or something. Or not workshop. Kind of, you know, their fulfillment. Yeah. I think they're drivers. So, the environments. I guess they can take a picture of this is where I dropped your package. Oh, I see. They can prove it to you. Yeah. These are definitely not that.
[00:55:46] This is more of like an entertainment. You're not walking around in the world with these on. Let's put it that way. Got it. This is kind of like VR in a glasses form factor. Yeah. But it's XR. And so, there is a little bit of that kind of augmented reality kind of connection to the real world sort of thing. It's kind of like a lower. You don't have to put the huge headset on in order to use these.
[00:56:16] But you still get that AI, you know, kind of integration with the world and the cameras that see out and all that kind of stuff. Meta had their event a couple of weeks ago and showed off the Meta VR glasses, right? And that's what I think these kind of compare. These overlays. Compare these two. You know, Metas are just slightly more expensive. I think they're $1,300.
[00:56:46] And so, price-wise, they're comparing pretty closely. But, you know, one is running Android and has access to the whole suite of Android apps for good and bad. And then the other is Meta, which has a history with the Quest and Oculus hardware. And so, they have a whole library to tap into. So, from an ecosystem standpoint, I'll be curious to see which one is the better play.
[00:57:14] I will say from my time with the X-Real Aura, it promises a lot. And in some ways, it's really impressive. And in other ways, like this last time that I had some demos with it, it was a little... There was a little jank in there. It was kind of like, okay, that's not really quite working the way it should. You know, but it's all early. So, it's hard to judge. So, you've tried almost all these glasses now. I mean, I've tried Aura. I have not tried the Meta VR glasses. Okay.
[00:57:43] Yeah, I wasn't at that event. I would love to try them. I think they're super compelling. And definitely, you know, to compare them between the Aura. But I don't even know that this is something that people... You know, the category is still proving itself. Yeah, exactly. A decade and some change later, it's still kind of trying to figure out, like, who wants this? Do people want this enough to spend $1,300 for it?
[00:58:11] Apple hasn't even proven it, right? Apple Vision Pro has not been a huge success. One time, I sat on a plane next to him, right after it came out, with a guy wearing Vision Pro. It was kind of silly. Yeah. He's enjoying his movie, whatever. And I wanted to almost go... But the fact is, you can see outside the Vision Pro, right? You have some sense of it. So, I couldn't... If you got it set up to do that, to let through, yes, you can. Yeah. Yeah.
[00:58:38] Well, finally, before our break, real quick here, apparently, Nala Health is the first company to have AI autonomously prescribe medication in the United States. Which freaked me out at first. It's very specific. It's for acne. You take images of your face. It uses the AI to say, okay, here's your severity and here's what to prescribe.
[00:59:03] The first couple hundred cases, a doctor will review it to make sure that the AI is not wrong. But then they're going to let it loose. And... Still, that's notable. I mean... It's very notable. Right. And what else might they... If insurance companies are going to love it... Well, yes and no. They have to pay for more medication. They won't. But if they can get rid of doctor visits, that's a huge savings for them. Yeah. Yeah. No kidding. And they get the data.
[00:59:35] Um... So, what else might they do this with? And maybe you're not allowed to have a doctor visit for certain prescriptions anymore. And then the AI becomes the death panel. Dun, dun, dun. Yep. Well, it's a notable kind of first. Um... I don't know if this opens the doors for other ones where we scratch our heads a little bit more.
[01:00:05] But, uh... Yeah. There you go. And I mean, maybe this is a good kind of lower... I don't know. Lower impact or lower downside thing to do it with. But, I mean, anything dealing with someone's health is, you know, high importance of getting it right. So, I've seen a lot of pieces by radiologists who have a vested interest saying all the predictions that my job was going to be eliminated are BS. It didn't happen.
[01:00:31] That was thought to be the first frontier for AI is to analyze x-rays. But, um... I've seen a lot of writing. Again, it's by people who have a vested interest saying, no, they're not that good. It doesn't replace us. It's not reliable. So, I don't know. We'll see how many zits AI can cure. Yep. Yes, we will.
[01:01:01] Hey, Daniel, thank you in chat for answering my call. Chatterbox was the name of my... Mike Elgin's son, who's named Kevin Elgin. That's right. I couldn't remember that. Hello, Chatterbox. Jason's getting old, you know. Yeah, I know. My memory's fading. Apparently, he works at Kagi now. Yes, he does. But, um... V2 might be coming out soon. So, there you go. It's a cool thing that he's been working on for quite a while now at this point. Very, very cool project.
[01:01:31] All right. Going to take a break. First of all, just a quick reminder. If you enjoy this show, please. Give us a review on Apple Podcasts. It's a huge help. We really appreciate it. You'd be great, folks. You could use some new refreshed reviews. So, even if you've left a review before, I think you can update it and put some new... Put a new spin on it. But actually, what we would love is more reviews. So, go to Apple Podcasts. Leave us a quick review. Take you 30 seconds to a minute to do. And we'd super appreciate it.
[01:01:59] It helps raise the awareness around the show. But we're going to take a quick break. Come back and do a few quick stories to turn out before we get you out of here. Hold on. Hang tight. We'll be back. This episode of the AI Inside Podcast is brought to you by Outshift by Cisco. That's Cisco's incubation engine for frontier technology. Multi-agent AI is on every enterprise roadmap right now. And for good reason.
[01:02:27] But here's what most people aren't talking about yet. Agents can pass messages. They can coordinate tasks. But they can't actually think together. And when they don't, you get what you might call cognitive failures. Misreadings. Unverified claims. False consensus. The kind of quiet underperformance that conventional monitoring doesn't even catch. That's the problem Outshift by Cisco is building the fix for. And they're calling it the Internet of Cognition.
[01:02:55] It's an open source foundation for building multi-agent systems from design all the way to production with shared context. Shared memory. Guardrails. To drive the results you actually expect from AI. If you're building in this space, this is worth your time. Read the paper. Experience the demo. Grab the code at outshift.com. That's O-U-T-S-H-I-F-T dot com. We thank Cisco for their support of the AI Inside Podcast.
[01:03:25] The Wired Newsroom is known for award-winning reporting on how technology shapes our world. On Wired's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people building, challenging, and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday, wherever you get your podcasts.
[01:03:56] All right. Google has released Foresight. This is a Mac app. It's a note-taking app that runs entirely on device. It uses Embedding Gemma 2, which is a 740 million parameter model. Takes up around 191 megs of RAM. So real low footprint, basically. And it's an open source app under Apache 2.0. It's part of Google's AI Edge initiative.
[01:04:26] So it's meant to show what on-device models can do. I think what this made me think of is this trend that we're seeing with the Apple Watch and with a few other things where it's ambiently picking up context from conversations and doing something with that.
[01:04:45] And I wonder if this is kind of like tapped into that as a destination, like putting this on smaller devices so that, you know, it with a very small footprint, it can gather and create that context without having to like, I don't know, record outright the conversations that are happening. Yeah. Kind of seems like a step in that direction. Seems to be. Yep. I can't use it because I don't.
[01:05:14] Well, I do have a Mac. But I don't use it for that. So I couldn't test it. Yeah, I haven't. I haven't tested it out or used it. But there you go. We've got OpenAI rolling out TextGrain, its own solution for invisible watermarks embedded into ChatGPT and Codex output. The register called it WeakSauce. WeakSauce watermarking for AI text. But we can say to the EU checkmark, we're doing it. Yeah, right. Right. Exactly.
[01:05:42] This is to get that pesky EU offer back. This is opt-in for API customers outside of the EU. So it isn't just in the EU. It is available to everyone if you so choose. Yeah. OpenAI says they'll open source the technology so others can build on it as well. So, okay. Okay. There's that. OpenAI also putting ads in their chat. They were already doing some ads, right?
[01:06:10] Like, I don't see them because they've got a $20 a month account. They insist that they're earning a billion dollars in advertising, and I don't see it. I don't see how they're racking up that kind of money, but they say they are. Yeah. Yeah. Apparently. Okay. Well, in this case, things are coming in that look a lot more like an ad than I imagine. It would be pretty difficult to ignore and not see this if this showed up in your ChatGPT interface. They're very visual. They call themselves out as an ad.
[01:06:38] I'm sure if there are other ads, they probably do that too. But OpenAI does say that these will not influence answers. The ads are contextual, but they are not profile-based, and they won't influence any of the answers that they're putting out. So if Heirloom Grocery, as the example in this screenshot, appears on your screen, it's not going to suddenly form and morph the output of a question that you have to Heirloom because they're buying ads.
[01:07:07] So that's good to know. And see the prior discussion about the cash flow these guys are going to need and the price competition provided by OpenWeights. We could well see AI delivered for free with advertising, like social media, like search, and then we're in the same bucket we were in with the internet. The attention economy, for good or nil.
[01:07:36] You may get more free things, but you're going to get it with advertising, and the advertising is going to be targeted, and the data is going to be gathered so they can do that. Scooped. Yep, yep. You are the product, right? And it's going to be only better at targeting. The thing about targeting right now is it's pretty damn stupid. I already bought the damn boots. Would you stop trying to sell them? I hate that. Yes, we've all seen that. I hate that. When your agent buys the boots, it will know to stop selling you the boots because you've already done that. It will sell you what you want next. Yeah, it'll sell you socks.
[01:08:07] Well done. Socks that come just below the boots or maybe above them, depending on your preference. See, it'll know your preference about how you wear your socks. How hip you are, how cool you are, how fashionable you are. Yes. Yes, indeed. And then finally, I'm seeing so many of these. Google's Nano Banana 2.1 image generation model is now official. And let me guess, better image quality, better text rendering, better character consistency.
[01:08:39] And that is true. That is what it offers. Nano Banana has been kind of the leader in this. You know, I felt that way. I get the best results out of Nano Banana. But I know a lot of people, more and more people that get really great results out of ChatGPT's image editor, image generation. I think they're all kind of getting into that really good range at this point. Or when I say all, you know, the big players. But yeah, I've constantly used Nano Banana.
[01:09:09] It's just the one that I go to and I know how to use it to get what I'm looking for. But I think there's competition there for sure. You're going to find this in Gemini, search, creative tools all around the board. So there you go. And then finally, I don't know why I left this in here just because I think it's funny. But Elon Musk has confirmed, lest you be too worried about it, that SpaceX AI will now be renamed SpaceX SI.
[01:09:40] Because here in the U.S. it is no longer artificial intelligence. It is super intelligence, apparently. Jason, ask me whether I care. I already know you don't. I don't. I don't really care either. I saw this and I was like, of course he did. Is OpenAI going to change their name to OpenSI? Open Sports Illustrated. Yes. To which page? I don't know. That is it.
[01:10:09] That's all we got for today. JeffJarvis.com for everything Jeff is working on, including the link to go to the Haverhill. Haverfield. Haverhill. Haverhill. Haverhill. You don't say the H. Haverhill. Okay. Got it. It's like Melbourne versus Melbourne. Yeah. Haverhill, Massachusetts. Go do on this upcoming weekend. This Saturday. Spend the day with Jeff. I will give a talk story at 1030. And then you can have lunch in Haverhill.
[01:10:39] And then we go up the hill, not far away, to the Museum of Printing, which is really quite wonderful, where you can see a working linotype, which I'll happily explain how it works. In addition to a great collection of presses and computers and all kinds of dirty stuff. Love it. Well, have a great time. I wish I could join you, but I will not be able to. Well, come on. Fly out, Jason. I will be driving my daughter to and from volleyball clinics all weekend. So, yeah.
[01:11:09] I can't make it this time. Next time, though. Next time, though. I'll hold you to it. Okay. Thank you, Jeff. JeffJarvis.com for everything. Jeff is up to PodTuneUp.com for podcast consulting or podcast commiserating. Hey, I'll commiserate with you if you've got a podcast and you just want to talk to someone who's experienced in the world of podcast. You're not even looking necessarily to solve problems, or you are. Either way, I'm happy to talk with you.
[01:11:38] PodTuneUp.com. I'm having a lot of fun consulting right now, and I want to talk with more people about their podcast. So hit me up. You can get 30 minutes free just to talk about your podcast and see where it leads. PodTuneUp.com. Do it, folks. Yay. Yeah, I hope to hear from some of you. AIinside.show for everything you need to know about this show. All of our recent episodes, a store for merchandise with a wonderful shirt that I love wearing,
[01:12:07] and everything can be found there. Live information, Patreon link. It's all there. And then Patreon.com slash, dang it, AIinside.show is the place where you can go to support us on a deeper level.
[01:12:50] And let's see here. Three podcasts, five days a week. All of it can be yours. And we hope that we'll see you there. A t-shirt at the executive producer level. So thank you, everyone, for your support. Thank you to Anthony Lemos, aka Amos, and Daniel Croft for video support behind the scenes. And I think that's it, Jeff. Thank you, as always. We will see you all next Wednesday on another episode of AI Inside. Peace out.
[01:13:20] Bye, everybody. The Wired Newsroom is known for award-winning reporting on how technology shapes our world. On Wired's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly
[01:13:48] with the people building, challenging, and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday, wherever you get your podcasts.