Bill Gates Has Three Ideas To Fix AI

00:00:00:00 - 00:00:27:03
Unknown
Coming up next, Jeff Jarvis and I check out Apple's new M6 Mac mini and the M5 ultra mac studio. Why are we talking about computers? Well, because they're really good at local open weight models. We also talk about perplexity and Nvidia building a local AI agent that runs with zero token costs, kind of proving why hardware from Apple might make sense even though it's not actually built to run on Apple hardware.

00:00:27:03 - 00:00:59:25
Unknown
But maybe somewhere down the line, bill gates has a warning that AI is more dangerous than Big Tech is admitting. I've found it pretty interesting to read through. Jeff was a little skeptical. We talk about that and a whole block of robots doing crazy things and some scary things. That's coming up next on this episode of the AI Inside podcast.

00:00:59:27 - 00:01:24:01
Unknown
Hey, what's going on? I'm Jason Howell. Welcome to another episode of AI Inside the Show, where we take a look at the AI that is layered throughout the world of technology. I am joined, as always, by author of Hot Type, Jeff Jarvis. Thank you. One weekend. Yeah, yeah. Now you're on the other side. Now you're like, okay. All the excitement led to the launch and now I'm on the other side of the launch.

00:01:24:01 - 00:01:40:04
Unknown
Now I got to get a new launch going. I gotta, I gotta start in the next book. Yeah, yeah, yeah. That's kind of how it feels when I. When I do music, you know, if I release something, you know, it's kind of nice to have the release moment and then I'm immediately like, okay, so what's next? Yep. Like.

00:01:40:06 - 00:02:11:23
Unknown
Yep. Where do we go from here? Well, congratulations. Thank you. Yeah, yeah. It's a big, big accomplishment. And we'll certainly be talking about that a little bit later as well on the show and throughout. We got a lot of news to dive into. We got Apple tons. You know it's more and more what I'm realizing is that I thought that 2026, the big story for AI in 2026 was a kind of like agent AI becoming more mainstream.

00:02:11:23 - 00:02:43:23
Unknown
And while I think that is an important story for the year, I think the bigger story is turning out to be this idea of open wait, local models, weaning ourselves from the frontier cloud, and maybe sooner possibly than later coming. You know, bringing that into our local machines. And that's what Apple has. I think it's related, though, Jason, because I think that that the, the, the agent craze made people see a utility and they wanted to run this on their own.

00:02:43:23 - 00:03:06:05
Unknown
They didn't want to. That's a good point for it. They want to control over it. They didn't want to see their costs rising. So the problem right now is memory chips. We'll talk about this in a second. And can you get a machine that can run these models. And you can buy a spark for 5000 bucks. You can buy the Apple for 10,000 bucks.

00:03:06:06 - 00:03:29:03
Unknown
I mean, it's not not in the reign of the range of most people right now. But I think that's going to change. Yeah. Well, at what level? Because, you know, there are different sizes of these local models. You know, not everybody needs the best of the best. Not everybody needs the massive, massive data set or, or model, you know, installed on their local machine.

00:03:29:09 - 00:03:51:22
Unknown
Not everybody needs the frontier in the cloud. You know, that's another thing that we're seeing is, is like for the last couple of years, a big a big story that that people have been parroting and believing is that or at least the people running the the companies, the. Yeah, companies themselves have been believing we need bigger, better, you know, be steer models and everything.

00:03:51:22 - 00:04:10:21
Unknown
And now, you know, it's kind of like, okay, you don't need that for everything. Like maybe, maybe some companies need that for some things, but they certainly don't need them for everything. Right. And you're holding your event tonight with local business people. And I'm really curious to hear right now it's just a question of how do I use Claude for my business?

00:04:10:21 - 00:04:27:01
Unknown
What do I have to do? Yeah. But, you know, as you've shown in your own use of these models, it's not a big step to then say, okay, this is now part of my process. Oh, yeah. For sure. Now, how do I, you know, make sure this is going to work and I'm not gonna spend a fortune on it and I can control it and so on.

00:04:27:01 - 00:04:46:09
Unknown
So, you know, we should get to we should get to Apple. But I think I think that that signals potentially a big retail chain in how this is going to operate. Yeah. And the more people work with these things in the kind of paradigm in which they begin, there we go, I guess. Paradigm. Five bucks. Please have a drink.

00:04:46:10 - 00:05:13:07
Unknown
Yeah, exactly. The more you use it in that model, then the more it kind of opens up your mind to the other ways that you could possibly be using it. Like, like to your point, as I've talked about a number of times on the show, the last, you know, probably half year, I've gone real deep down the Claude rabbit hole and the co work and integrating it into my business and creating all these processes.

00:05:13:07 - 00:05:37:06
Unknown
And then now we're at this moment where, you know, and then I've gone from the $20 plan to the $100 plan to one month. I went to the $200 plan because that actually made sense. Now I'm back on the 100. But, you know, and so my spending has increased as my capabilities have increased. And now I'm kind of kind of find myself in this, this correction where I'm like, okay, not everything needs to be there.

00:05:37:07 - 00:05:58:29
Unknown
Like it's really useful for certain things and a lot of things. But now that I've like put almost everything there, I'm starting to kind of peel it back a little bit and realize like, no, you know, some of these things I want to have more control over. I don't need it to do absolutely everything. And maybe it would be beneficial to have one of these new Apple machines and just run things locally.

00:05:58:29 - 00:06:22:27
Unknown
And also maybe then I'm also not playing the game of catch up of like keeping up with the Joneses. For every single new update that comes out, I have to, you know, use it enough to learn like what it's really good for. Like, maybe I get to a point or we get to a point to where we just have the model that does all the things that we really needed to do, and we don't feel like we have to continually update it and everything.

00:06:22:29 - 00:06:46:17
Unknown
Yes, I agree with all that. But for the nerds ideas, if you watch what Chief Nerdy or the port's been doing, he has his saddle and he's he's trading in models underneath it because he wants to test them all. So he's going in and out and in and out. In and and out. So. So what it shows is the the models become very much commodified.

00:06:46:19 - 00:07:08:25
Unknown
And we'll get to this too, in terms of the uptake of the open weights, that's a huge factor. Yeah. And so the gating factor, which what I didn't predict, you just did your what I didn't see line I'll do mine. I didn't predict that hardware and memory would be the key gating factor for expansion. Yeah. Yeah. Right. Yeah.

00:07:08:25 - 00:07:31:22
Unknown
So it's interesting. So we've got Apple. Yep. Let's talk about basically dropping some new Macs, some new chip hardware. And really I think it's easy to look at Apple in the last couple of years and continue to say what you know, I know we've said on the show a number of times, which is Apple's playing game of catch up when it comes to artificial intelligence.

00:07:31:22 - 00:07:51:16
Unknown
But not that may be sort of true, but I think what Apple is doing right now is really smart. And maybe when they were starting to develop these machines, they they weren't necessarily thinking of it in these terms. Although in some ways I think they were. Well, the deep sea took a by surprise. People. People bought up mac minis.

00:07:51:18 - 00:08:11:17
Unknown
Exactly. I think that was a big surprise. Exactly. The Mac mini for Deep Sea, the Mac minis for Open claw, like a lot of yes, yes, a lot of signals being sent to Apple as far as what its hardware could be used for in this current moment with artificial intelligence. And and so now you've got these new machines.

00:08:11:17 - 00:08:41:13
Unknown
It's a new Mac mini getting an M6 chip, which is Apple's first chip built on the two nanometer process, which is a very pretty notable and meaningful jump on things like efficiency and density and and capability and everything improved like dual 16 core neural engine. So you get 32 Neural Engine, cause that's significant because Apple is basically saying that's double the AI processing power of the M5.

00:08:41:19 - 00:09:07:09
Unknown
You know, and that's just one example here. But it seems like with this hardware announcement, Apple is saying, hey. Oh, by the way, our hardware is really great to run local AI models on. And I think it's going to be very appealing to people who maybe don't want to spend as much on a spark, but they're willing to spend something significant to get some really solid, localized power for AI models.

00:09:07:09 - 00:09:37:29
Unknown
Well, it's more than it's more expensive than a spark, I thought. Yes. How much is it? How much is the spark? A spark is 5000 bucks for, for and almost 49.99 95. Whatever it is. Right. It's okay. Right. I think I'll check on a second. Yeah, that sounds about right. I just I just priced out a mac studio with the M5 ultra chip, 36 core CPU, 80 core GPU, 256 GB of memory, one terabyte of the storage.

00:09:37:29 - 00:10:06:04
Unknown
It goes up to 16TB. 16TB is $7,500. Let's be fair, at one terabyte there is $10,799. Yeah, when you've got it specked out like that starts at 5500. Yeah. That's not going to be enough. Yeah. Right. Yeah. So if I want to if I want to get a. That's true. Good point. How much does ADG spark cost.

00:10:06:06 - 00:10:31:17
Unknown
Come on. I have 46.99 the standalone. Okay. A dual spark is 9449. But then you get a lot more now that has. Right. The base unit has the Nvidia Grace Blackwell one 28GB of memory for terabytes of storage.

00:10:31:19 - 00:10:55:16
Unknown
And you double that up. You get all of that versus one Mac I don't know. Yeah. There's going to be a macho chest eating contest going on here. Yeah. What what amazes me, Jason, is that a next surprise? I think Apple didn't have an AI strategy. Yeah, right. They went they went to Gemini. They put Gemini in into Siri.

00:10:55:17 - 00:11:14:22
Unknown
They, they they got rid of all kinds of thoughts that they might be doing, you know, pendants and all that crap. It may turn out that their big strategy will be local hardware for open weight models. That's kind of what. Yeah, that's kind of where my mind is at to. Right. And that can be where Apple can concentrate and can win.

00:11:14:22 - 00:11:32:21
Unknown
And they can say, oh, we're the one you can control. You can locally. You have all the choice in the world. Maybe they bundle a whole bunch of upper weight models with it. Maybe they come up. Maybe they buy a saddle company. We talked last week about stripe buying open router. That's right. Right. Maybe Apple should have bought it.

00:11:32:23 - 00:11:58:24
Unknown
That would not have been an interesting. Maybe Apple should buy stripe interesting and get both at once right. Where does transactions? Benedict Evans had a piece this week about his. He finally explained the stripe thing to me in a way that made sense, is that basically people are going to be buying bits of compute, and stripe is used to taking little tiny pieces of those transactions.

00:11:58:25 - 00:12:17:10
Unknown
Little you're buying tokens, you're buying compute time or power. And so that's what makes it make sense. But underneath an open router, right, is all of these models. And it's always more there's another mysterious one out today. There's more and more models. So if you can buy one of these powerful things from Apple and you know what you're doing, that's the other gating factor.

00:12:17:12 - 00:12:37:05
Unknown
Yes. That is that is an important factor. Right. Just just getting these is the start. You're not saying to your mom who had trouble right of the Tesla last week, mom for Christmas I'm getting a DG spark. You're welcome. You figure. And suddenly your life is going to get so much simpler because you're going to have these models that you can interact with and everything.

00:12:37:06 - 00:12:56:28
Unknown
Yeah, there's there's a lot of I mean, there is benefit to the cloud and the frontier being you just deal with the messy parts and give me a chat interface and I'll interact with it versus having to install and set up and configure and. Right, right. So we're still not there. We're not there at a retail level. And we don't be.

00:12:56:29 - 00:13:14:12
Unknown
That's why I'm really, really curious what you hear from these small business folks. I can imagine some of them are just starting. That's why they're during this event tonight. By the way, it's already sold out, folks. So you can't go sold out. Yeah. Too late for for all of you that are in the Healdsburg area. I'm sorry you can't go this time, but we'll have another one, I'm sure.

00:13:14:13 - 00:13:38:15
Unknown
But it'll be really interesting to follow them. Somebody who says, I don't know about you in my business. And then they say, oh, I can do that. And they start doing it. Yeah. How long before they say, I think I might buy that Mac? I God, I we've been doing this show now almost three years. I've been working for with AI for much of that time on my own and increasing my, my usage of it.

00:13:38:15 - 00:13:56:12
Unknown
And I only feel like in the last couple of months I've even considered like, oh, well, that that would be interesting. Like, I have a better sense of what I might possibly even do with it now. It's taken a long time to get there, right, that for me. And I think there's other things I think there's going to be.

00:13:56:13 - 00:14:18:08
Unknown
I talked to my friend Bill Gross at Idea Lab and he started 150 companies, and he's he's exited 50 of them. He's brilliant at all this. And I don't want to give away anything that he's working on now. But you know, he's talked about this for some time. The tasks that AI can now do for small companies are many, right?

00:14:18:10 - 00:14:57:07
Unknown
You need something to handle hiring or handle payroll or handle planning or make presentations, whatever. Right? Yeah. And those become identifiable tasks. And right now people are going to the general model to do that. But I think there's going to be there's going to be agent stores like crazy for that. And I think if you can buy the hardware, if it can come pre-installed with this stuff, so you don't have to deal with all that, and then if you have the equivalent of an app store for AI to do, define things that thousands upon thousands of businesses want done, I think that explodes.

00:14:57:07 - 00:15:23:27
Unknown
This becomes the moment. Yeah. Yeah. Right. The apple was the apple was the Mac was failing failing miserably or the apple was I mean was not taken off until Visicalc came along and the Mac was not taking off until desktop publishing came along, which I write about in my book type on sale now. Sorry. But it's that it's that killer app.

00:15:24:00 - 00:15:56:12
Unknown
In this case, I think there'll be killer apps. Yeah, yeah, super, super interesting. Really interesting that Apple I think it's it's really smart what Apple has done here and how they've kind of positioned this right there. They're touting double the AI processing power in their marketing. So they are kind of leaning into the AI capability of this and kind of tackling the because I think the the more known AI story about Apple is, oh, Siri is awful or whatever.

00:15:56:13 - 00:16:21:19
Unknown
You know what I mean? It's it's that side. It's not about the hardware. It's it's about the software side of it. And I don't know, there's the chat, our friend. You can't decide between a spark and a mac said I need one petaflops of AI compute performance. Right. This is this is going to expand. It's going to. It's not going to take over your whole world.

00:16:21:19 - 00:16:43:11
Unknown
Your grandma is not going to be doing it. But yeah. Not yet. Yeah. Well once upon a time we said your grandma's not going to get on the internet. Not going to have a phone, not going to flip phone. Yeah, yeah, yeah. The timeline is long. Eventually that might be the case. Yeah. And then you got,

00:16:43:13 - 00:17:03:28
Unknown
Let's see here. What what else? You you threw a couple, so I threw this one in. So Vercel has a has a AI model leaderboard. Okay. And I don't know who sent me to this, but it was really interesting. And so the open versus closed token volume, I don't know how they measure this. I don't know what this really means.

00:17:03:28 - 00:17:31:19
Unknown
It's a couple charts. Two charts down. Open weights are 52.5% of token volume, closed weights are 47.5. So what they're saying in this, but whatever their methodology is that open weights are already winning in total compute volume, which is pretty amazing. That is wow. This is a very interactive thing they have going on here. This looks yeah, it's pretty cool.

00:17:31:22 - 00:18:01:29
Unknown
It's pretty as well. It's super purdy. So they look at token volume open versus closed requests across the top models. Who's winning there. So they're deep seek version four. Flash is winning GPT five six Luna max. Deep seek version for flash. Something else next GPT five nanotech Gemini next. Gemini next GPT next. Step 3.7 flash. I don't think I've even heard of that one step.

00:18:02:00 - 00:18:27:09
Unknown
Yeah, yeah, I haven't heard that one. Then comes Claude Sun at five. At 2.9%, that adds up to 46% of requests. Again, I don't understand. Yeah. Then the spend clause on top cloud opus five and 4.8 are on top GPT next Claude the next three four spots, then Kimi, then GLM. So it's really confusing to figure out who's on first.

00:18:27:10 - 00:19:04:16
Unknown
So spend. That kind of backs up this idea that that anthropic has had real success. Yes, in the enterprise. Enterprise and appealing the high value is spend. Yeah. Which is a good strategy. That's clearly it's working for anthropic. They occupy like six of those slots or 5 or 6 of those slots out of ten. That's impressive. And how that changes over time as we get more and more more by in more adoption of the open local models, you know, that'll that'll be really interesting.

00:19:04:18 - 00:19:32:07
Unknown
So then Futurism has a story saying nobody wants anthropic best AI model anymore. Well, that's a yes. I may lie by the stats we just gave you now that there are way cheaper alternatives, but I think that's where the trend goes and we're going to get much later into into where that goes for anthropic business. Yeah. Yeah. This is based on ramp tracking, which is a corporate expense management company that tracks across 70,000 US companies.

00:19:32:07 - 00:19:59:09
Unknown
And they basically said, you know, fable five plateaued at around 11% of the total money companies are spending on anthropic, specifically their their AI tools. So 89% of their spend is going to cheaper, older models compared to fable five, which is kind of their their their highest publicly accessible. They do have, you know, mythos, but not everybody has access to that.

00:19:59:09 - 00:20:24:11
Unknown
Right. And I think, you know, when I, when I kind of thought about this a little bit that lines up with, with my use as well, even though I'm on a very different scale. But like when fable was first being kind of teased and, and dripped out in the very beginning, everybody was super excited to use it because it was the latest, greatest thing.

00:20:24:11 - 00:20:51:18
Unknown
And it actually seemed like it, it performed really well. But very quickly you realized, like doing everything at that level is just absolutely unnecessary. Yeah, in a very short amount of time, I realized what I felt like it was good for planning things, maybe planning architectures around or processes around something, not necessarily evaluating, helping me evaluate what's current in tech news right now.

00:20:51:21 - 00:21:15:24
Unknown
And so I would say about, yeah, probably like 80% of my usage of the anthropic tools is limited to the sonnet and even 4.6 opus, which is not the, the highest opus, you know, I really reserve my use of, of things like fable, like, I hardly use fable now. Right. Because I just don't have many tasks to use it.

00:21:15:24 - 00:21:38:27
Unknown
And we aren't in the token maxing, you know, that that thing happened and then enterprises realize, oh wow, that's a that's a really horrible idea. We need to pull back. And so and who knows how well Quinn would do for you or Nema Tron from India or any of the other or Mistral. Yeah. I think that that that becomes interesting.

00:21:38:28 - 00:21:58:12
Unknown
So last week there was a story I think we talked about it. AT&T said that a high proportion of their AI software now is they're using open source models to reduce their cost, which I think is where we see this goes. And then today we have Thomson Reuters built its own AI model based on Chinese open source models to slash their costs.

00:21:58:12 - 00:22:30:21
Unknown
And they put out a legal model, something that's tuned to to law firms and such, and they open source part of it so that people can get that. So it's interesting here you have a non it's a tech. Thomson Reuters is very much a tech company. It's really a data company right. But they're there. They're out there creating in essence very high end agent store model store for some of their customers because they've got to provide that full functionality.

00:22:30:23 - 00:23:10:29
Unknown
Yeah it's a it's a quickly changing world. Arvi Rob points out it's all on prem as well. Yep, yep. Make companies a lot more happy. Yes. Very very interesting. And then kind of along the open weight side of the conversation is you remember the old darling perplexity. Our love for them are often forgotten darling perplexity. Actually releasing some news that I thought was really interesting, a partnership with Nvidia on a new product that they're calling portable computer.

00:23:11:00 - 00:23:38:17
Unknown
Of course, they launched computer inside of the perplexity universe for kind of interaction with your machine. Yes, your local files. It's kind of, you know, it's kind of like what they all have now, which is some sort of hook and integration into your local file structure. That's what computer was. A portable computer is essentially not using the cloud models at all and just doing it fully local.

00:23:38:17 - 00:23:57:03
Unknown
So it's basically that capability, but keeping it locked into a local locally installed model. And the other advantage of it being local is besides the fact that you're controlling it and it's you're saving a lot of money and so on, is it's always on. It can be run on something constantly. Right? Which is part of the value of this.

00:23:57:04 - 00:24:15:23
Unknown
Now it's true. Right now it's limited. Only you think, oh perplexity that's I know I use them, I can use them. No, it's limited to a GCS park right. Yeah. So the Apple computers that we're just talking about, you can't so far can't use this so far, at least at this point. And I think windows computers coming soon.

00:24:15:23 - 00:24:39:29
Unknown
But right now spark it's limited to in the six Linux machines that run in vedere RTX GPUs. That's right. Part of the appeal of the Mac mini for AI agents like OpenCL was the bang for your buck and processing power, says Gizmodo. A mac mini with an M4 chip sets you back $900. You're going to need to fork over five times that price for DG spark, so that again, we're still at the high end here.

00:24:39:29 - 00:25:08:12
Unknown
But that's going to change. Yep. And we're going to see models on our phones. So there's just there's just kind of no middle ground. That middle ground is going to get filled up. Absolutely. It will. And at the rate at which all of this moves in other aspects of the industry, of course it's going to move, you know, similarly and something like this, especially as demand and interest grows, which it is at a pretty rapid pace right now.

00:25:08:15 - 00:25:49:03
Unknown
The models that you get on this are pretty limited, at least right now it's Quinn 3.8 27 billion parameters there Alibaba that's from Alibaba. And then perplexity has its own version of Quinn that they've trained at 27 billion parameter called hey Brunson. And so those are there now in videos Nemo Tron 3.5 lightning apparently coming soon expect yeah yeah yeah that makes a lot of sense but so yeah optimized for like you said, always on localized agent tasks and work on your machine.

00:25:49:04 - 00:26:02:09
Unknown
I don't know, it sounds it sounds super appealing. And the other thing was we saw in I think two, I Marked My Life Now by Jensen Wong keynotes. I think it was two keynotes ago.

00:26:02:11 - 00:26:24:19
Unknown
They and vedere invested in slash half acquired a company just to get inference chips because inference was the next frontier of this and on the cloud. I think that was true. A lot of efforts did that. I think some measure of that inference is going to move locally more and more and more. We're going to see that chart go up to.

00:26:24:22 - 00:26:48:17
Unknown
So it's not about trying to get, you know, data center inference. I think you're going to find interests running the models locally. It's going to be a small proportion but it's only going to grow. Yep. Yeah. Totally agree. Fascinating stuff. And yeah I'm sure I'm certain we'll talk a little more about this on the other side of the break.

00:26:48:17 - 00:27:11:09
Unknown
I do want to thank those of you who support us on Patreon Patreon.com Inside Show. You are the ones that keep the ship afloat and we really, really appreciate. You. Also want to throw out a few names thanks to Charlie Hoover and Rick Showers, just a couple of our amazing patrons who support us on a deeper level on the Patreon.

00:27:11:09 - 00:27:34:17
Unknown
So thank you for your support. Anybody interested in hopping in? It's pretty easy to do. Go to Patreon.com AI Inside show. You can find a tier that works for you. Get get a daily podcast with the AI Inside daily. Get ads removed from this podcast behind the the Patreon paywall and discord access and a whole lot more Patreon.com AI inside show.

00:27:34:18 - 00:27:57:27
Unknown
Thank you for your support, everybody. All right, we're going to take a break when we come back. Today's big news, I suppose, would be Bill gates talking a little bit about well, some of the big news is he is he still around. He's still around and he's got he's sharing opinions about how he feels about where we are with AI right now and where things are headed.

00:27:57:27 - 00:28:04:05
Unknown
And got some people talking. So we'll talk about it. That's coming up after this break.

00:28:04:08 - 00:28:23:29
Unknown
All right. Would you say the Bill gates news was big or is it just another person, another tech person shares their their similar opinion on gates is so funny because because of course, he had first didn't understand the value of the internet and then, you know, came on full bore when he realized what it was.

00:28:24:02 - 00:28:42:25
Unknown
And I said he was late to AI. I don't think he was, but I don't know that he's the most forward thinking soul we have out there in technology. Still a voice of caution. Fine. We need that in the discussion. I don't, I don't he's smart. Lord knows he has the experience. I'm not diminishing that at all. But.

00:28:42:27 - 00:29:15:12
Unknown
So you thought it was huge? Well, no, I just I saw a lot of people responding to it. And his, you know, his his perspective is basically not anything that we haven't heard before from others. Right. You know, he's he's basically saying, you know, we're facing turbulent times as, as he says jobs vanishing. He he thinks that, you know, the big risks are that those jobs will impact younger people first, many of those jobs disappearing forever, as we've talked about.

00:29:15:12 - 00:29:51:28
Unknown
Sure, that's that's true. But what does it get replaced with? So that's the other side of it. But he's basically saying, you know, your typical like white collar jobs, customer support, software engineering, paralegal stuff. He also said that construction and hospitality would follow by the end of the decade. And I don't know, really like construction jobs and hospitality being well, I think to some extent soon to me for something like that, I think we're going to I think we could see robot made modular housing and building 3D printed housing, but I don't think that's going to be huge.

00:29:51:28 - 00:30:08:02
Unknown
I think it's going to take a long time, especially the way local zoning regulations work. I mean, we have a housing shortage in this country, so it would be a good thing if that happened. We found ways to make that more efficient. And of course, the fact that in the United States, a lot of the labor involved in construction is now being deported.

00:30:08:04 - 00:30:35:10
Unknown
But that's that's a big ramp. I don't see that happening. No, in terms of hospitality. So, you know, a robot delivers your lunch. Okay. But yeah, I don't think a robot is going to cook it for a long time. Yeah, that's true, that's true. Yeah. Considering that we pay people terribly in that field, then someone, a human being making your bed in the in the hotel is probably going to be less expensive than a robot doing it.

00:30:35:10 - 00:31:03:18
Unknown
So yeah, I don't I don't agree with them. Okay. That's that's one of one of his risks is that that kind of block. He's also talking about criminals, empowered criminals, basically saying even criminals with very limited skills will be able to target victims at every scale. Talk and fraud, talking, deepfakes, disinformation, bioterrorism. Is that just fear, uncertainty, doubt or I think this is the most important part.

00:31:03:20 - 00:31:28:02
Unknown
I think it is worthwhile to, you know, you look at what happened with the water plant recently, which is just a shot across the bow. And my friend Craig Newmark is concentrating on cyber security more than anything else. Greg, Greg Newmark concentrating on cybersecurity and warning that that people aren't taking this seriously. And I think we are vulnerable.

00:31:28:04 - 00:31:50:10
Unknown
And so I don't think AI can, on the one hand, make us more vulnerable because it's easier to find vulnerabilities, on the other hand, because less vulnerable if we use it wisely. So let's get around to use it it wisely. Let's figure out where the vulnerabilities are. There are some things that you know, I'm not sure why a water plant is hooked up to the internet at all.

00:31:50:12 - 00:32:13:05
Unknown
Yeah, right. Good point. And I think that there's a lot to be thought through, worst case analysis here. And to do that and to protect us also individuals, you know, your grandparents being spoofed by AI is more of a problem. So yeah, I think I think that's legitimate warning.

00:32:13:08 - 00:32:43:23
Unknown
And then third, stunted child development, he says AI companions that are always available and never get mad at you would in many cases replace the the experiences that kids would normally have if they didn't have that as their next, you know, their next destination. If they didn't have that standing in the way of the actual experiences as acting as a replacement upon them.

00:32:43:26 - 00:33:08:21
Unknown
Such interesting dialectic going on there. I just got off doing a frustrating channel for UK podcast about the meta settlement $17 billion because of harm to kids, blah blah blah. And part of the settlement is is to max out time online to two hours a day. We see Australia has tried to cut out young people from online, but more than 80% have come back anyway.

00:33:08:23 - 00:33:36:15
Unknown
UK is about doing at the same Canada's, I think, thinking about the same thing. So there's a lot of effort to say what do we take away? Because we have to protect children. Part of what we're doing is we're also protecting them from education, from learning lessons about life, from learning what to do. And part of that is to be able to, frankly, see some junky stuff and decide on their own that it's junky and build their own sense of judgment and taste.

00:33:36:21 - 00:34:09:12
Unknown
So on the one hand, we're coddling them. I think too much in this, in this techno panic that we have around that. So now you switch to AI and what gates is saying, he's oddly making the same argument I'm making. If AI makes life so damned easy, what lessons are children spare that they should get through experience? One of the books that I hoped would have about sign up for my my AI book series for Bloomsbury Intelligence.

00:34:09:15 - 00:34:35:08
Unknown
Is about that is about when when AI makes everything so easy. What do we lose in experience? I think back to Mark Zuckerberg's 3000 word gates beat him to it with 6500 words essay the other week. And I thought about this afterwards, Jason, that oftentimes we see these guys say, you'll have your own tutor, you can learn anything.

00:34:35:10 - 00:34:52:29
Unknown
And that leads the fact that learning is hard. I can't tell you how many hours I've spent trying to learn Spanish and I didn't get anywhere. I can't do it. I just I'm too old. My brain's not built that way. And my German is bad. It's not as if you just give me this thing in front of me, and suddenly I'm going to magically learn.

00:34:53:00 - 00:35:16:04
Unknown
Right? Right, right. Yeah. And whether that's a school lesson or a life lesson, I think gates, in a way, has a point here. But we should listen to what I think is the hypocrisy in the rest of our dealings about technology and youth, where we're saying danger, danger and trying to restrict them from those lessons. Sorry, that's a weird, kind of convoluted rabbit hole.

00:35:16:06 - 00:35:39:12
Unknown
No, I appreciate that. I yeah, it's it's interesting because one of the things that stood out for me in this, a couple of things that stood out for me is that and I hope that I'm able to retain them enough to to remember both of them. But on one of them, one of the points that stood out is, this is a guy who's been in technology for as long as he has.

00:35:39:14 - 00:36:11:18
Unknown
I mean, since, you know, since the basically the birth of the, the desktop computer, desktop PC era. And so he's got a lot of perspective. And one of the things that he shares in this is for most of his career, he's felt like technology didn't move fast enough. But now something's different. Now technology is moving too fast, faster than we have prepared for, and he finds himself in a different position now than he's ever been.

00:36:11:18 - 00:36:31:21
Unknown
Which is. Wait a minute. Let's let's pump the brakes a little bit. We haven't prepared enough for the for the speed and rate at which technology is moving. So I think that's an interesting reflection to for for someone to make from his career in technology is kind of acknowledging that, oh wow. You know this is this is interesting.

00:36:31:22 - 00:36:55:14
Unknown
It never felt like it was moving fast enough. Now very suddenly it feels like it's moving too fast. And so I think there's there's something to that. And then what was I knew I was going to, I knew I was oh yeah. The other, the other side. My brain isn't firing on all cylinders. Maybe I need an AI to think for me today, which is just the access that he has to other people in the industry.

00:36:55:15 - 00:37:20:02
Unknown
You know, he's he's in that top, that upper echelon of of tech tech people for whatever that means. Maybe that's worth nothing. But he has access to the leaders of all of these these companies. He certainly had conversations with them. And he's sharing and here that, you know, whether they say it publicly or not, all of these people are intimidated by the speed at which this is moving.

00:37:20:02 - 00:37:46:09
Unknown
They just don't feel like they can say anything about it, because the value of the company and and the progress that they're making financially is, is such that it doesn't behoove them to actually speak their discomfort publicly. And so, you know, and I don't know, like that that kind of makes sense to me that, you know, that I could I could totally I can I feel like I can feel that in that's a really good point.

00:37:46:10 - 00:38:05:03
Unknown
And I think it goes I'll go back to Craig once more is that he's talking about cybersecurity more than I think some of the people in these companies are. But the other thing that I'll praise gates for now is he's talking about real safety. He's not talking about and real job security and real issues of education. He's not talking about all the Duma crap.

00:38:05:05 - 00:38:28:24
Unknown
He's not talking about AGI taking over life and all that. No, you're right, that's a good point. Yeah. At least I didn't read every word of the thing. But the summaries I've read. Interesting. He also. So the Wall Street Journal being the Wall Street Journal, picked up on these three things about what to do about this. And first, he calls for taxes on AI tokens and bots, saying that this could delay companies rush away from human labor.

00:38:28:27 - 00:38:49:26
Unknown
He said that societies should set aside jobs for humans. I don't know about this like public lands where governments agree to forbid development, societies can choose not to deploy AI for certain positions either, because the people this is the journals summary who hold those roles will be difficult to retrain or because the position requires human empathy and care.

00:38:49:28 - 00:39:08:18
Unknown
So do you put a robot into an old folks home? Right? Well, yeah. But on the other hand, right now a lot of the old folks home experts are Haitians who are being deported, and there's a huge labor shortage there. So what do we do about that? Yeah, I think it's as easy as saying, okay, we're just not going to do this.

00:39:08:20 - 00:39:34:10
Unknown
Is it better to have some solution versus no solution that becomes the question that particular or does having the tackle, does having the technical solution make it too easy to then deport the people you need? Yeah. Right, right. Oh, boy. Yeah. That's that's the flip side. Yeah. And then third he this is again the journal summary. He urged governments globally to collaborate on regulation specifically say in the US and China will need to work together.

00:39:34:12 - 00:40:04:29
Unknown
And he's right that this is what's done with aviation ozone to some extent with nuclear weapons. And I don't think AI is equivalent to nuclear weapons, but it's always easier said than done to imagine global governance of this stuff. Yeah. So quickly. And and that's the kind of thing that in their rush to regulatory capture, the boys at AI and anthropic say, oh, we should create a federal, you know, an international commission for this and that.

00:40:04:29 - 00:40:23:23
Unknown
And that's a form of them saying that they'll sit on those commissions and they'll have them slow down the regulation rather than the development. So I don't know. But yes, gates matters. His voice matters. This matters. I still don't think it's a big story of the week. Yeah, yeah. Not sure that I feel it's the biggest story of the week.

00:40:23:23 - 00:40:43:22
Unknown
It was certainly near the top. Attack me when I opened it up earlier this morning. Oh yeah. Oh yeah. It has been superseded by Meta's settlement, as you mentioned, a couple of couple of minutes ago. But that doesn't really have a whole lot to do with AI. But anyways, that happened. Well, I think I think the panic that there's a delay here.

00:40:43:24 - 00:41:06:29
Unknown
So the moral entrepreneurship that led to that led to this happened with social media. Social media was the was the thing to argue about what, four years ago? Yeah, five years ago. Right. So it's a preview of what I think is going to happen with AI. There you go. Ding ding. Yeah, I think you're right. You're totally right.

00:41:07:02 - 00:41:27:11
Unknown
Okay, so then we've got the Bill gates story, and then we've got a bunch of robot stories, which I just was like, you know what? My brain needs a break, my brain needs a break. And robots are always, always enjoyable to track here. So let's see here. What do we got? Let me pull up. I have a couple of videos to show you.

00:41:27:11 - 00:41:59:12
Unknown
First of all, if I can put away the little the little pop ups here. Okay. There won't be any audio on this. So apologies to those of you listening. We'll just have to describe it to you. But there was a Robot Olympics at the world humanoid robot Games in Beijing. I call it the Robot Olympics. But you know, it was not actually affiliated with the Olympics, but is basically robots doing Olympic type things, including Tiangong Ultra, which is a Chinese robot that ran a 100m in 9.39 seconds.

00:41:59:12 - 00:42:18:06
Unknown
Look at them go, and then they can't stop themselves, so they run into a wall and fall down. But I mean, you know, hey, they ran pretty damn fast. I love that they got to spray them off because they're just so overheated. They got a card them away on a on a stretcher. It's just cool that we're at the like.

00:42:18:08 - 00:42:41:20
Unknown
I just I see video like this and I'm like, wow, this is the stuff that when I was a kid on my Commodore 64 and I was playing video games of robots playing human sports, like, this was all, you know, sci fi video game stuff. And now we're at a point to where you can go to a tournament of robots running, like beating Usain Bolt.

00:42:41:22 - 00:43:05:18
Unknown
But the good thing is, it's never going to replace Hussein Bolt because of course, a machine can outdo a human. Yeah, yeah. So yeah, it's not surprising that a machine could could do. So it is it is kind of the John Henry thing. Sorry for this. This warning book plug coming up now in Hot Type, I write about this wonderful story of speed type centers.

00:43:05:18 - 00:43:26:25
Unknown
They would they would have contests people would pay for to watch them put little tiny letters in a stick over an hour. They paid to go see this. It was called the golfing or the bowling of the day that way. Yes, yes. That's fascinating. They were called the Swifts, and they went down to town and they had tournaments and they won money and they won silver composing sticks.

00:43:26:26 - 00:43:51:27
Unknown
Hold on. I mean, of course you have, you have them within reach. So this is a composing stick into which you put all these letters, right. Yes. Click, click click. Anyway. So the International Graphical Union forbade these contests to continue. Why? Because they knew that the latter type, such as my book, was coming and there was no winning against the machine.

00:43:51:28 - 00:44:17:27
Unknown
And they didn't want to have that John Henry V machine contest they didn't want to lose. And it's interesting right. Yeah. So there are certain things where I think the roboticists will say look what we can do. And the question is whether they take away dignity from certain tasks as, as a role as a result. Yeah. I mean, well I don't know, the typesetters were doomed anyway.

00:44:17:29 - 00:44:38:10
Unknown
So they took over the machine, which was their, their key to success. Right. I mean, I don't look at that race and think, oh, human races are over now we're just going to care about robot races. I see them as their own distinct camps, but it's damn cool to watch robots run that fast and in many ways emulate the human racers.

00:44:38:11 - 00:44:50:23
Unknown
You know that they were kind of built to emulate. There's also this video that I have of AlphaGo. Sorry, not AlphaGo of a gal bot playing tennis.

00:44:50:26 - 00:45:10:24
Unknown
There are human players on on there, but it's just, you know, you don't watch them. You watch the robots doing the run up and and hit. And this makes me think like, how far are we from like someone with probably a lot of money having their own tennis court and having like, they want to practice tennis, they just practice with their robot.

00:45:10:26 - 00:45:26:22
Unknown
Like, that's so cool. Like that's basically what we're watching here. That's true. You don't have to hire a what do you call. What do you call? You don't call the sparring partner. You call them something else. Yeah. I don't know what it would be in tennis, but, I don't know. It's just this stuff blows me away that we're okay.

00:45:26:23 - 00:45:48:29
Unknown
So since since you're the nice guy about robots here, I'm going to counter you now. Okay? So we also have video, which I moved up there of an Amazon drone misfiring on delivering a package. Sorry, I should Jason, can you find it? They're now finding it. Okay. And is the video embedded? Yes it is. Okay. Little scare pop up here.

00:45:49:00 - 00:46:18:10
Unknown
Get rid of that play. So here's boom. It's very quick. A package went straight into a pool. Oh no don't don't don't do it. They couldn't detect that splash. Oh poor package actually actually went into the hot tub even more so. Yeah. Right. But I wonder if it was on a more serious note. Obviously we have drones becoming the future of warfare.

00:46:18:12 - 00:46:52:27
Unknown
Yes. And so smuggled and vedere computers are powering Moscows AI drones, which is also part of the the drone the robot future. We have Shaw for us. So we've got the. Ha ha funny playing tennis. And then we've got the frightening future of not just drone warfare, but it's directed. No. Yeah. Sorry. That's the point, right? It wasn't just a no humans autonomous.

00:46:53:04 - 00:47:14:20
Unknown
Yeah. In the loop on that. Yes. And that is frightening. Like, I can't even imagine being on the receiving end of something like that. That would be absolutely terrifying. So sorry for that Donnie Downer moment, but that's okay. I think I can bring us back. I think I can bring bring back a little hope because where is that story?

00:47:14:21 - 00:47:43:28
Unknown
There is also a robot that is basically it's a Dutch company called VTC that just got FDA authorization for a robot that draws blood. And the device itself is called the Aleta. It's authorized through the FDA's de novo pathway, and it is the first standalone robotic device that's clear to draw blood from humans with without a human hand on the needle itself.

00:47:43:28 - 00:48:07:27
Unknown
And basically, I wish I had a photo of the machine. I guess I don't in this article, but basically it uses an infrared light and Doppler ultrasound to find the vein, and then it inserts the needle. It, you know, fills the tube, withdraws, and all that takes about a minute and 49 seconds per draw. And that means that's that's faster than a phlebotomist could normally do.

00:48:07:27 - 00:48:33:04
Unknown
So in essence, a phlebotomist could supervise up to three of these devices at once. So it's kind of like 3 to 1 time impact for a phlebotomist, which is just kind of remarkable. So I was going to recommend to our nephew, who was working in hospitals, that he should take phlebotomist training. Maybe not now. I mean, you'd just be directing phlebotomy, phlebotomy, phlebotomy.

00:48:33:06 - 00:48:36:02
Unknown
That's it. That's.

00:48:36:04 - 00:49:00:10
Unknown
And if they're not called phlebotomy, they really should be because or phlebotomy maybe if you want to just go even shorter you could. But yeah. So robots doing the darndest things I guess some of them better than others, some of them downright frightening. And we're not seeing this so much in the US, but oh, by this this is this is interesting.

00:49:00:10 - 00:49:32:10
Unknown
And I hope that I've got the whole article to show here, yes I do. Okay. So this was just this was just a fun one to, to flip through, which is a New York Times article that very visual. Lots of video that's AI is everywhere in China. See for yourself. And it just kind of it kind of is a day in the life sort of thing where, you know, it takes you through the streets of China to a jaywalk device and shows you kind of the technology that's there, that's scanning.

00:49:32:10 - 00:50:05:22
Unknown
And you know, what it what it actually sees and everything. It's just a lot of different examples of how technology is intertwined into culture and society inside of China. And yeah, I mean, here in the US, if we're concerned about the the impact and the integration of AI, you know, there's probably some clues here as far as like what that would look like if you live in a country that just kind of seemingly just embraces it and says, let's just go with it.

00:50:05:24 - 00:50:30:08
Unknown
And, yeah. So anyways, it's fun to flip through and kind of see the different, the different pieces of technology and, and integrations of AI that we just don't have over here. Well, it's possible there because they have these government supported companies and applications that everybody has. Everybody has WeChat. That's just that's just cash now. So it's easier, I think, to make this happen there.

00:50:30:10 - 00:50:55:28
Unknown
Is it better like like are we missing out on something by not having this or are we protected from something because we don't AI trash cans. Okay, there we go. Scan error. You do. We miss out on some things. I think we if we if we were better at at recycling, if we had the convenience of being able to buy, you know, do we eliminate 7-Eleven?

00:50:56:00 - 00:51:02:12
Unknown
No. But can we buy something through in the morning if we really need it? Yeah.

00:51:02:15 - 00:51:28:05
Unknown
Convenience store robot. You could buy any number of things from the robot. Like it will. Great. Yeah. Totally. This this feels more like a, oh, a robot that knows how to reach onto these very sparsely stocked shelves so that it can definitely grab that one item or maybe not grab did not grab it. It's not perfect. Look, robots are like humans.

00:51:28:05 - 00:52:00:24
Unknown
They're not perfect. Anyways, I think it was kind of fun to scroll through and and also just to kind of ask like, okay, so things are seemingly things are very different in China and you point out the reason why that is. I just wonder like is, you know, in many cases technology in, in China or in Japan or whatever has been years ahead of where when we see it here, I think the example that I always kind of go to is like text messaging.

00:52:00:24 - 00:52:24:02
Unknown
Wasn't that in like Japan for like many years before we even started? Just kind of see it almost thought it would never happen here. Yeah. Yeah, totally. It was just very far ahead of the times. A $5,900 humanoid robot. Okay, I think you'd rather have an egg work. Yeah, I think so. This would be neat. For a little while, you'd show your friends, they'd go, okay, cool.

00:52:24:02 - 00:52:47:18
Unknown
And then it would stop being neat. Probably doesn't do a whole lot, I'm guessing. I think it'd be cool to have a walk with you in the streets of Petaluma. For now. A $42,000 humanoid robot. If you could afford something like that, that would be better. Empty the dishwasher? Suppose so. Anyways, it's a fun one to to scroll through and definitely fell into the robots thing.

00:52:47:23 - 00:53:14:12
Unknown
Before we take a break, let's get off of robots and onto jalapeno. Just to mention that we had been talking about jalapeno for a little while, which is essentially OpenAI's first custom chip, and they have now officially presented this released some benchmarks. They say, hey, it's really good. It outperforms video and runs. You know, how does it outperform Nvidia?

00:53:14:14 - 00:53:37:04
Unknown
I, I don't have that detail. This seems to be more of an inference chip than a training ship. That's true. It is it is inference. It is not training. And to compared it they said it would be, but it beat Nvidia's training chip rather than Nvidia's new inference chip. So it's a little more chest beating by the AI boys.

00:53:37:06 - 00:53:55:09
Unknown
Yes. Let's just get more chips. Make make more chips. I think we're all going to get affected by the the shortage, but it's not these fancy chips we need. We need memory chips. Yeah that's true. That's a memory chips. OpenAI doesn't care that you need memory chips. No, they don't know. They don't want us to. They don't. More interesting not to have them.

00:53:55:11 - 00:54:16:03
Unknown
Exactly, exactly. Real quick, if you enjoy the show, leave us a review on Apple Podcasts. It is a huge help. We haven't had new reviews in a while, and I'm sure there are some new fans of the show who've been listening for a while and haven't given a review. So go on Apple Podcasts, leave us a review however you feel we would really appreciate it.

00:54:16:04 - 00:54:34:19
Unknown
It is helpful to get other people to pay attention to, because there are an insane amount of podcasts out there, and then there are an insane amount of AI podcasts out there. So hopefully you like what you're getting here. We're going to take a break and come back and talk about a few rapid fire stories that we have all lined up and ready for you.

00:54:34:20 - 00:54:37:25
Unknown
That's coming up here in a moment.

00:54:37:28 - 00:55:22:16
Unknown
Okie doke. That story that we talked about last week with spirit being spirit data being bought by Google, it's on hold. And basically so this was a $10 million deal. Google was going to pay $10 million to spirit in bankruptcy court. Now apparently the bankruptcy proceedings are on hold. The Association of Flight Attendants is officially and formally fighting this in court, saying, you know, it adds insult to injury, that while there are a lot of privacy protections baked in around consumer facing data, the actual data is very much and disproportionately so employee facing.

00:55:22:16 - 00:55:45:07
Unknown
And so the union is saying, hey, we got to protect the workers because much of their information and their identity potentially is tied up in this data. So Google doesn't have it yet, but I don't know. They're probably gonna get it I mean, regardless. Right. So I'm grappling with this right now. I just got asked to write a piece for Project Syndicate on privacy and AI.

00:55:45:10 - 00:56:03:14
Unknown
Since I wrote about privacy in the internet and social media in the past to kind of update this in my argument, which will piss people off, will be that at some level, who wants ignorant AI? Who wants stupid AI, who wants the wrong AI? It's in our interests as users of AI and we're all going to be users.

00:56:03:15 - 00:56:24:11
Unknown
I mean, I have a smarter AI, and that means it's in our interest for the AI to have data. Now, we may not want that to be open AI or anthropic, but that's where the open weight stuff comes in. So in this case the data would go to Google. But Google will get smarter about how a company operates, how airlines operate, and so on.

00:56:24:13 - 00:56:48:29
Unknown
Is that better in the long run? If we know that AI is going to be used by airlines? Well, spirit may be a really bad model for this. Possibly so. Not the airline, maybe the old Pan Am, but but yeah, I think that it raises interesting problems here is that I understand wanting to protect. And I think there are ways to build on the protections for the employees as well.

00:56:49:01 - 00:57:14:29
Unknown
But do we stop it? Kind of the point? Yeah. No, I don't think we do. Yeah. So what? So when it comes to privacy protection, in the case of this, if it's, you know, potential, potentially identifiable information inside of the data set or whatever, and meanwhile Google like maybe this is a stupid question, but does Google care that much about the identity of anyone in this data set?

00:57:14:29 - 00:57:39:22
Unknown
They're really they're really going after, like the communication or the style of communication or the things people ask for, whatever. They're not buying this data set, as far as I know, because they want to out anyone or, you know, whether intentionally or not. But is that kind of the fear, I guess then? Well, so. Right. So I think no, Google has no motive to out individuals.

00:57:39:24 - 00:57:57:22
Unknown
But as experts will point out, often whenever you say anonymized data, there's no such thing as fully anonymized data because you can find inferences that you may not have otherwise known or there and focus on someone in a way. Right. So.

00:57:57:24 - 00:58:17:13
Unknown
That's always an issue. So it's not whether Google would do that. Would someone outside do that. What's Google's interest here? I think it needs to know that employee two is the same throughout. So if they try to understand scheduling flight crews, they need to know that that's the same person all along. And the as the objections were arranged.

00:58:17:18 - 00:58:36:05
Unknown
Here's what they signed up for. You can't really do that if you don't know that that person is the same person who did this over here. Yeah. And could someone go in and say, well, there were only so many flight attendants who were based in Tulsa and we know this or that about them. So what do you learn about that flight attendant?

00:58:36:07 - 00:58:56:12
Unknown
That's where you get to the question of harm. What could this flight attendant have done that would be harmful to be known? Their employer do whatever it was? I don't know, but on a principle level, I understand the desire. We don't have the discussion on the basis of harms. Too often we have the basis of that could go wrong.

00:58:56:13 - 00:59:15:17
Unknown
Okay, well tell me about the. Yeah. Totally interesting. Okay. That makes a lot more sense. Yeah. And as AV Rob says in our live chat, there are doubts that data can actually truly be anonymized anyways. Yep.

00:59:15:19 - 00:59:39:28
Unknown
Let's see here. Oh, I knew you would have an opinion on this one. The New York Times testing AI generated summaries on its own readers. It's a new search page feature with a small test group, at least for now. And yes, the summaries draw exclusively from New York Times content their own, their own journalism. It's not using outside sources, however.

00:59:39:29 - 01:00:05:29
Unknown
Apparently the text is not being reviewed by a journalist or an editor, which is the first time for the newsroom. The News Guild is not very happy about this. The union wants human oversight requirements as well as a revenue share pool for AI licensing, I don't know. New York Times saying saying on one hand, AI generated summaries are not okay.

01:00:06:00 - 01:00:21:05
Unknown
On the other hand, saying we're testing it. I mean, maybe they're just doing that because that's just the sign of the times, I don't know, what do you think? Yeah. I'm just trying to see if I if I got anything to do. I don't have any. They're not giving me those kind of summaries right now. A few things here.

01:00:21:05 - 01:00:49:04
Unknown
One is that, like USA today is on every story putting three bullets up. A lot of places are doing this now. You know, they're generated and they say it's run by an editor, but who knows what it really is. So that already exists. Number one, .1.2, my most frequent use of Gemini, now that it's on my browser and on everything else, is that when I come across a Ben Thompson Street 7000 word treatise about something, I got to be honest.

01:00:49:04 - 01:01:08:28
Unknown
I go to Gemini and I say summarize, and sometimes I read the whole thing, but sometimes I see enough to say, okay, I get the point. I don't need to. Which, of course, is what sets everybody when it comes to Google search and has been for years. But my time is my time and that's how I choose to use it.

01:01:08:28 - 01:01:28:15
Unknown
So the summaries are not that bad. So I think that as long as we know that it is what it is, I tend to think it's okay if it's on the search page. The searches are specific, so I don't see how. I think this is why the time is doing it. There is because you can't have an editor pre-approved everything because you don't know what all the search is going to be.

01:01:28:17 - 01:01:51:22
Unknown
That's the other thing if you put it on the top of an article. But that's the guild is also going to complain about that, but that you could have absolute editorial oversight of that, like USA today. The other interesting thing, just while while we're on this is that last week, a, an op ed in the and the Wall Street Journal was written by AI, and the writer acknowledged it and said, I'm busy and I'm not a writer and I'm an executive.

01:01:51:22 - 01:02:10:07
Unknown
So, yeah, I wanted to say this, and I had the AI do it BFT and the Journal said, fine with us. So journalists are raising storms. Get used to it. I see more of that. You're going to see a lot more than I talk to somebody yesterday. I don't want to reveal anything about them. But he had a big project.

01:02:10:07 - 01:02:33:19
Unknown
He had to do a final thing and said, you know, I just while I was there, I said, well, the AI to a draft. And I looked at it and I said, well, crap, it's good. I'll edit it, I'll change it. Yeah, but it's okay. And we talked about watermarking and I said, well, you know, if you have a paragraph in there that you didn't edit, then yeah, you better be very transparent about which he would be.

01:02:33:22 - 01:02:51:13
Unknown
Absolutely. But if you could imagine someone else along this line, they could be accused of AI in the whole thing. When there's one paragraph they copied and pasted into it. Yeah. What are you going to do? You're gonna go in and you're going to play watermarking in reverse. You're going to change all kinds of words to synonyms, just so you can try to think you're fooling the watermarking.

01:02:51:13 - 01:03:15:25
Unknown
It gets absurd because it's, you know, all AI has cooties. Better not touch it. It's here folks. So whether it's a summary in search that makes the search better, no different from what Google is doing and their search. So for the quality, I'm okay with it. How about you. Yeah. No, I mean I, I, I think very much the same as you sometimes I do read through the articles.

01:03:15:25 - 01:03:31:10
Unknown
Of course. I'm very you know, I've done a lot of that over the years for my podcast work. And sometimes, you know, again, I don't have the time or the summary that I read, I feel gives me enough of the context that I'm kind of like, okay, I think I've learned enough from this. Like, I can, I can move on.

01:03:31:10 - 01:03:58:19
Unknown
So yeah, it's just it's interesting to watch the opinions on this and the business practices on this shift over time as, as it's been around longer and longer, you know, people, companies and outfits that were once vehemently opposed suddenly kind of coming over and understanding, like, okay, you know, probably understanding what you're saying, which is it is what it is.

01:03:58:19 - 01:04:16:17
Unknown
It's here. You know, I guess we just got to adapt and be okay with it. It's just interesting to see that happen with a company like New York Times, you know, they've had they've had strong opinions on this. So and yet here is their product. And they're only testing with a small group of people. I'm sure that will widen I'm sure it will.

01:04:16:18 - 01:04:46:05
Unknown
But yeah. Anthropic expected to tell investors that it sees over $30 trillion in potential revenue. The entire US GDP is $32.4 trillion. So, you know, it's just absurd. It's BS all the way down. I understand what they're saying. This is our potential. Market and space had a potential market of almost 30 trillion. So they're doing one element here.

01:04:46:07 - 01:05:06:26
Unknown
But it's BS. It's complete BS. And if they come out with a $2 trillion valuation before having been profitable, granted Amazon was not profitable, profitable for years, but that was by choice. They had a valuable service. You can see where they could make it profitable. In this case, it's not yet clear where they'll be profitable. There's a lot of danger ahead.

01:05:06:27 - 01:05:27:27
Unknown
We talked to the top of the show was all about that. These open weight models are a huge threat to them. Yeah, big threat, big time locally run models are additionally a huge threat to them. They're related. I know, but but the point is that those are those are two vectors of consumer demand that could hit a hosted frontier model on both sides of it.

01:05:28:04 - 01:05:41:29
Unknown
Yep. Yeah. So yeah, 33 is just BS. Dun dun. Oh and just look he's so proud of himself. Yeah well he's the owner of the Brooklyn Bridge. So you know.

01:05:42:01 - 01:06:07:04
Unknown
We've got China's Z AI basically behind. That's who made it. Oh okay. Because the Ox Alpha came out and nobody knew who made it. Exactly. It was a stealth AI model and showed up on open router. Open router. There we go. That's where where it was open router called it the biggest launch in the marketplaces history. And there were there were assumptions and suspicions that it was.

01:06:07:05 - 01:06:35:12
Unknown
The AI is now confirmed, also known as Zippo, which is a Beijing based company. This is essentially the latest in their GLM series. So Ox Alpha coming from that firm, they announced plans to open source the weights. So, you know, just another powerful Chinese model coming out. Just another thing we said at the beginning. Yep, yep. Exactly. Tie in right in on the music tip A couple of things to to note here.

01:06:35:12 - 01:07:11:15
Unknown
Australia's Aria charts, which happens to be their official music charts. Now banning wholly AI generated music outright. Basically they're saying if music was created entirely with AI, it cannot chart. If human artists are using AI tools in production, that's fine. This is all about fully AI generated material. And this basically came about because there was an AI cover of Like a Prayer by Madonna that hit number four on the charts in July without Madonna's permission.

01:07:11:15 - 01:07:39:19
Unknown
So it was generated. It was a generated AI cover of Madonna's Like a Prayer, and yet people still apparently liked it a lot because it hit number four, you know? So I don't know. I don't know what you say about that, but there you go. And then Apple is adding labels made with AI to Apple Music songs that it detects and identifies when an AI model was used to make that track.

01:07:39:19 - 01:07:48:24
Unknown
So they're going to label it. So let me let me ask you, since you are my Bruce Springsteen, okay. Oh, it shucks.

01:07:48:27 - 01:08:09:03
Unknown
If we talk about producing, we've talked about this with text all the time, right. How do you say that? It's not AI produced when you use spell check, when you use translate, when you use Grammarly and so on. Right. Same thing with photos. If you use a lot of the tools out of Adobe, they're going to be AI based or even Google Photos.

01:08:09:05 - 01:08:32:27
Unknown
When you take the photo with your phone, there's AI going into that. Okay. So those areas I think are already so infiltrated with AI. To my mind, it becomes impossible to say that it's not made with AI at some level. What about music? What about music? Have AI tools begun to infiltrate music? Enough or not yet? It's not there.

01:08:32:29 - 01:09:05:15
Unknown
In my experience, there are plenty of music production tools that have AI as part of the process, be it, you know, processing the sound quality or mastering the music or even full on generating pieces or whatever. But I don't know, like, you know, I've only ever used these tools really as like collaborative partners. That's that's the only way that I feel in my own experience that it's worthwhile.

01:09:05:16 - 01:09:24:00
Unknown
Like, like I don't quite enjoy just saying generate an entire section for me and I'm just going to use that because I'm also too picky. Like when I do that, the music that it generates always sounds off to me. I feel like it can always hear that kind of AI glossy quality to it, and I just want it to be pristine.

01:09:24:00 - 01:09:46:27
Unknown
And so I would rerecord it, so I might might do that. I do think that there are a lot of AI tools integrating in into music production software and the process and everything. I think a lot of people in writing music, whether they're admitting it or not, are actually using services like Suno for inspiration if they're not using it for anything else.

01:09:46:29 - 01:10:12:17
Unknown
Oh, so does that count as AI use? I think it's AI use, but I don't think it counts for something like label. Label. You know, like when I think of AI generated music, I think this is a song that was created entirely by a model, you know, in the same way that like we use nano banana to create an image, we just say, give me an image of this.

01:10:12:18 - 01:10:33:01
Unknown
If I take that and I, you know, it has some sort of metadata on it that says this was generated by Gemini, you could strip away that metadata, I suppose, or you could alter it yourself, and then you're introducing your own kind of or you could, you could hear a melody that pseudo gave you and not use it digitally, but just use it as inspiration as you just said.

01:10:33:03 - 01:10:48:22
Unknown
Yes. And that's that's exactly how I've used it. When I've used those services, that's generally how I use it. It's like, oh, that's an idea that I hadn't considered. I'm going to work on that. And then I figure out how to do it, and I see how it sounds, and sometimes it sounds great and sometimes it doesn't work.

01:10:48:22 - 01:11:14:13
Unknown
But by going into pseudo and saying, you know, record this or make a version of this song and then going, great, that's a great version of the song, I'm going to release it. I don't know. I personally, I think a label for something like that deserves to be there, at least for now, but I but I do think that at a certain point, these things are probably going to get good enough that it becomes really, really difficult to know.

01:11:14:13 - 01:11:31:15
Unknown
And then at the end of the day, it's kind of the question that we talk about, which is, if you like it, does it matter if at the end of the day it produces something that reads well or sounds good to your ears or whatever? How much does it really matter that we watermark these things and identify them?

01:11:31:17 - 01:11:49:12
Unknown
Yeah. Yeah, yeah. So I guess that's my my version is this. I can see both sides. Everything that touches right now has cooties. But I think we're going to find more and more and more. It's going to be impossible to tell that which. Yeah, just touch it. Yeah. At a certain point it becomes next to impossible to tell.

01:11:49:13 - 01:12:15:23
Unknown
Yeah. And then I don't know, like there's not a whole lot you could do at that point. Right. It just is is what it is, I guess. But yeah. Slack now has its own vibe coding tool called slack code. That sounds appropriate. You can tag your agents so, you know, cloud code. Devon, GitHub, copilot, all those things directly inside of a slack group chat.

01:12:15:23 - 01:12:43:11
Unknown
And then it has the ability those agents have the ability to like, create channels, pull in people if it needs to build the code, you know, do the HTML, show you a preview, whatever. And then humans just have to sign off on it. So yeah, that's available on all slack everywhere. It's everywhere. That's right. And that's what we got.

01:12:43:17 - 01:13:07:11
Unknown
Thank you Jeff. Fun times as always. And again big time. Congratulations on the release of Hot Type last week. And that can be found at Jeff Jarvis. Go there and you can. You don't even need to preorder. I'm asked by son to take that out. He's in charge of my page. You can just get it soon. I will have a link there if you want to get an autographed copy.

01:13:07:13 - 01:13:32:01
Unknown
Oh, awesome. I did a deal with a local bookstore, so they're nice enough to fulfill. And I'll go running over there and sign the copies, but it's not quite ready yet. Right on. That's cool. We all have to mention that on the show. Once we once once that happens, please do please bring that up. As for me doing on occasion some podcast consulting with folks, also some AI consulting.

01:13:32:01 - 01:13:52:23
Unknown
I've got my workshop happening tonight and Healdsburg. Maybe next week. I'll tell you how that goes. Oh, you're going to tell us next week? Yes, I'll, I'll give you a report on how my first day I workshop went next week but also do podcast consulting. So pod tune up if you've got a podcast or you're thinking through challenges with a podcast, you know that's me.

01:13:52:23 - 01:14:20:04
Unknown
I can help pod tune up. I can help you think through those things. And how many years are you doing a podcast? Jason I think 2021 now 21 is what he's talking about, folks. Thanks, Jeff. Thank you. I appreciate you and patrons. You can always support us. Patreon.com inside show. You can become an executive producer and you get all sorts of extra perks, including hearing your name each and every week.

01:14:20:04 - 01:14:40:18
Unknown
Doctor do Jeffrey Marikina Radio Asheville one, two and 3.7 Dante. Saint, James. Botto, Derek. Jason. Jason. Brady, Anthony. Downs, Marc. Starker and Karsten. Would love to add another name by the way. So I think I'm going to go in. I'm going to change the order just to throw Jason off. Oh you should. Yes. I don't know why I don't change the order.

01:14:40:18 - 01:14:53:27
Unknown
I'm a very much a you know, I get I get stuck in my patterns I suppose. So my orders is just doesn't sleep sometimes. Why did you say doctor do.

01:14:53:29 - 01:15:15:16
Unknown
Doctor do. Oh, boy. That was a that was a dream. I was in the discord. Doctor do was saying, like 20 things I couldn't keep up. That's actually reality. Doctor do is very prolific, by the way, in the discord AI inside show for everything that we do for this show. Thank you to Victor Bogner and Daniel Croft behind the scenes for all of the video support that they give us for the show.

01:15:15:17 - 01:15:27:12
Unknown
Couldn't do it without you. Actually, I could, but I would be insane by the amount of work. So thank you for helping and thanks for watching and listening, everybody. We'll see you next time on another episode of AI inside. Take