Hello and welcome to our cast, the flagship podcast of cute little guys who have all your data. That's how I feel about Sam Altman. Ha. You weren't... that's not what you thought about fuzzy guys. I was thinking about Sam. Mark David Mel is here. Hey buddy.
You're still here. You know, every week I say we're going to keep you and it's gotten more and more threatening over time.
Yeah, I've gotten threats as well.
Well, this is technically your last week with us. David Pierce is back next week, but it is true. We are going to find ways to keep you on the show.
I'm open to it. Let's go.
That's a solemn promise. Uh, and by that I mean you can't leave the office.
Oh, I've been here for weeks.
Jay Castraus is here as well. What's up, man?
Hey, good to be here. The Davids are like particle and antiparticle. They can't collide. We can't be on a show together. It explodes.
Uh, there's a lot to talk about this week. Uh, we'll run it all down in one second, but first I want to call out we are nominated for some awards. The Signal Awards are happening. Those are big podcast awards. Uh, The Verge is nominated for all of our shows. Uh, Decoder was nominated for best business podcast. Vergecast for best history podcast.
That's up against some heavy hitters like actual history. We're like, what's a phone? We're in the finals. So I'm very excited about that. And then Vergecast, this podcast, not only for best technology podcast, like so many awards. There's the formal judge voting, but there's also audience voting, listener voting. So we'll have all the links in the show notes and the descriptions. Click on those links and vote for us because we love winning. That's who we are. We just love winning. Can't you tell how deeply competitive and in need of validation we are? That's who we are. Uh, all right, Jake, what are we talking about this week?
Uh, tons to talk about this week. We're going to have a big gadget segment. There's new Kindles, a ton of Apple rumors. Uh, we're going to have a sort of chaos lightning round. America has a chatbot now. I don't know if you saw this. America has an AI chatbot. Very good news.
But first, we're going to start off with I think the big AI news of the week, which is it was OpenAI's week. They did Dots. It's their answer to Muse. Dots is the same thing. It's got a cute little avatar. It's got a chat interface and it goes off and does things for you. Except unlike Muse, which Meta is just like, 'Here, please take a Muse. Have as many Muses as you want.' Dots you got to pay $100 a month minimum. And I think the question right out the gate is, is this a different product for a different set of people, right? Is this a business product or is OpenAI just not able to compete with Meta's ability to just, you know, blanket the entire universe with compute because they have so much money? I am curious where your head is at after seeing all of this.
I would say at a minimum it being for Pro subscribers and Business Plus subscribers only is kind of dragging it towards the business user case. But also when you sign up for Muse, it connects to Instagram, it connects to Facebook, it connects to all the things that everyday users use every single day. And notably it gives you a giant list of prompts that you can ask it because I think people are faced with these things and are not really sure what to do with it. And if you're on ChatGPT and you're using that every day for work, it makes a little bit more sense to have like a personal agent that's going and doing work tasks for you. But Muse is not really a work task product. It's more of a like put things on Facebook Messenger, Facebook Marketplace for me, which I did because Neil told me to.
And uh, yeah, I did and it did technically work and it was very fast and I listed like 30 products. I think I sold like an old backpack of mine.
Um, but once someone got to my apartment, they said, 'How's your time with Verge going?' And then I asked Muse to delete all my listings.
Yeah. Was that before or after the person posted on Threads that Muse had sent somebody to their house with a lowball offer?
It was the day before and then I saw the writing on the wall. So yeah, I think they're very different products. I'm still a little bit confused about what people are using this for outside of what they were already using OpenAI like ChatGPT stuff for because that was already doing autonomous tasks for them like Codex and things like that. I want to know what people are going to use Dots for because it seems like Muse is very oriented around Meta's products.
Dots is for launching your website while you're making dinner. That is what I learned from their launch video.
OpenAI's marketing for Dots is a little all over the place. They had that example. They had wedding planning in the mix, but then also like hardcore software engineering.
And it was like it can do everything.
Work tasks. Whereas Muse is very much a consumer product. We talked about at length last week like Meta is good at consumer products. They are definitely good at engaging you to do more on their platforms. Like they have one skill and it's engagement bait and they're really good at it. Um, I feel nothing about letting Muse operate other Meta products for me. So I have it watching my Instagram and every single morning it finds all the most interesting comments left on my Instagram posts and tells me to go reply to them. So like I can be in and that means I just don't have to check it. And this makes me feel very much like a robot is telling me what to do every morning. Like I'm mad at it every morning. Like you don't have to tell me what to do. But it is also true that I would never see or think to reply to the most interesting comments on my Instagram.
Yeah. Some things that have been interesting about Muse is that it's very proactive. So I hooked it up to my Google Calendar as well and it knew that I was going to this like media dinner party thing and it also knew that I had to be home by a certain time. So it told me you should leave on a Citi Bike because a Citi Bike is faster than taking the subway. And if you want to get home on time, make sure you leave at 9:20. Do you want me to ping you when you need to leave? And I was like, okay, this kind of... I kind of understand this. This is kind of interesting.
But the thing you're describing is all kind of work-like.
Right. Like all these things that are neat, even like, you know, use my Instagram for me. I had it look for like AI accounts that I hadn't seen before because I want to expand the roster of people that I look at on Instagram talking about AI.
All that is work. It's just work. It's fun work. It's work that rhymes with fun. I guess work-tainment, whatever that is. Um.
But I think this is the problem for consumer agents and a lot of people since Muse was announced and launched have come to this conclusion. Obviously, I call it software brain, that to even understand what to do with an agent, you have to understand your life as a series of tasks. Trusty McMillan Cotton, who's a New York Times columnist, said a long thing about how the problem with consumer agents is literally cultural. You have to understand yourself as a series of tasks in order to understand what an agent could do for you. And that like there's a cultural divide in America that will prevent some people from ever seeing that. Uh, I think that's fascinating, right? It may be the thesis of the whole Verge, maybe the thesis of the whole show is technology makes people think about themselves differently.
So here's a thing that's going to happen because Meta does have the marketing dollars and the straight-up dollars to give everybody a billion tokens and a little computer in the cloud.
And then they can market it across every one of their services. And maybe it's true that five years from now, everyone in America will just think of themselves as a series of tasks the same way that the phone or the steam engine or whatever made you think about yourself differently.
Yeah. And I don't think that Dots is going out and telling you what you should be doing with Dots. But the primary tab on Muse, the center tab is the ideas tab. And when it's plugged into all your stuff, it says, 'I can do this for you. I can do this for you. I can open this past key alert. I can distill launch week into one type brief. I can hold your inbox at a clean baseline. I can finish this briefing reply for you.' It's just trying to get you to use it proactively so that you slowly understand what the capabilities of these things are. I actually felt very similar about this product than I felt to ChatGPT when it first launched. I looked at it and I was like, 'This is an infinite black box and I don't know what to do with it.' Everyone's like, 'It's game changing. It's going to change everything about the way you do work, blah blah blah.' But when you look at infinite possibilities, there's sort of this anxiety where you just don't know where to start. And I feel like Meta is doing a pretty good job of trying to help people understand what agents are even for in the first place.
Yeah. I mean, and again, this is their background as a consumer company. They understand consumers. The defaults are important, and what you're describing is basically just a bunch of default agent tasks.
Uh, the one that I got that was particularly good. Uh, again, it's the only personal data I've given it is my spam Gmail.
Uh, I won't give it anything else. Again, I feel nothing about letting it operate my Meta accounts because Meta already has them. Like, so what? But I gave it my spam Gmail and every couple days it's like, 'You're in another class action lawsuit. Do you want me to file for 50 bucks?' And I'm like, 'Yeah, go ahead.' Just like I would never look for them. I would never fill out these PDFs. Like, go forth. Uh, and there was one moment where, you know, in order to fill out the forms I had to tell it my phone number which was the piece of... and I was like, I hate this.
But it's 50 bucks. Like, you have my phone number.
Yeah. Um, and that is fascinating to me and I think that the distinction I would draw from Dots is obviously Meta does not have as capable of a model.
Right. No one thinks that whatever is running Muse is as powerful as GPT-5 Astro, which is what powers Dots, but they have a better user interface for consumers. And Dots, you know, again, it's very confused. All their ideas were on the margins for software engineers, but it's also true that they have the more powerful model and that's why they're charging money for it. And they say they're going to bring it to consumers, one assumes Meta is able to bring it to consumers with a cheaper model. But I just think it's like an interesting A/B test to say, do you want work agents like this with a really powerful model that can do all of the things a frontier model can do, or do you want to be on rails with a kind of dumber model, but it's a lot easier to use the thing? And you know the consumers and enterprise markets are different and one costs 100 bucks. Somewhere in the middle of that is Grokbot, right? Like Grok has this thing too and it works a lot like these products do. Um, and somewhere is OpenClaw if you want to roll your own. We should point out that OpenAI hired Peter Sternberger who made OpenClaw and then Meta beat them to the punch with Muse which is heavily inspired by OpenClaw.
Yeah. Uh, I'm curious that the story of OpenAI to me has always been one of wanting to be a consumer company and realizing they have to be an enterprise company. So they launched with ChatGPT which was supposed to take Google, right? That was their ambition. They were going to replace Google. They were going to replace Google Search. That's one of the most lucrative businesses in the whole world. They hired a bunch of people from Meta, which is fascinating in this dynamic. Um, they started doing ads in ChatGPT and then Anthropic did Claude Code in particular and the enterprise usage of Claude exploded. There was that period of time when like every third OpenAI executive would write a memo declaring a code red like they were unfocused. We got to shut down Sora, code red. We're focused on coding. And now they're with Dots which have the fuzzy avatars and it's cute and the branding feels very consumer but it's a work product.
And I don't know if OpenAI has decided which way it wants to go.
And this is the story of every company, right? Like they all want to be a consumer product. Consumer products are fun. They're very widely used. You're very right. You're powerful if you're Mark Zuckerberg and everyone in the world uses multiple of your products. But then they realize that the money is in enterprise because people actually pay you money and they actually have money that they're ready to spend. And I mean this is what happened with Anthropic, right? They discovered that coding works. There's a ton of money to pay for coding tools. And I think OpenAI is in a similar position, right? They're burning through cash. So great, they're going to make business tools and sell them to business users, particularly in a world where Google, listen, you can't escape AI overviews. And so, do you need ChatGPT to replace that? It does a good job, but they're not going to completely overthrow Google anymore. Google is there with the same product. So, I think it's a very open question whether they can still crack the consumer market in the same way. Maybe they will eventually try to do that with this hardware product that they're allegedly working on that will come out supposed to be next year, I think. But that still feels like a very big question whether anybody wants to carry around, you know, a dot in their pocket or whatever this will end up being.
Yeah, OpenAI still feels a little bit confused on which end of the market they want to satisfy because like you said that they originally were like fastest growing consumer app. They were just like everyone has touched ChatGPT. There are over a billion users, that kind of thing. And then when Anthropic came in and were like, oh, we're actually making money and we're selling to enterprise, that felt like what Astro was supposed to be for them was their big pivot to, okay, we're enterprise now. Yes, regular people can use this, but the tasks that it's really for are like 3D modeling and like enterprise tasks. And then now Meta comes in with a consumer product. When OpenAI was trying for so long to make a consumer product. That's got to feel kind of bad because now you're on the back foot on both sides.
It's true. Just like one caveat. Anthropic loves to announce that it's making money.
Using accounting rules that do not have anything to do with reality.
We did see their S-1 this week. Liz is going to write a whole piece about it. But Anthropic's financials are fiction in like a real way. Um, but it's true that they are generating some revenue. OpenAI is obviously generating some revenue. They are just burning money on training and all this other stuff that they're doing.
Meta doesn't really have this problem. They are spending a lot of money and you know if you look at their financials closely they are making a huge bet here but Meta has a giant advertising business. They can to some extent afford to light money on fire. Um, it's very funny. Did you see Adam Mosseri was on Threads and he was announcing like Instagram plus the paid subscription. And he's like, 'This is how we can put out features and afford to make them affordable.'
And he got immediately community noted and it was like, 'Meta is one of the richest companies in the world. It can afford to do anything.' And he was like, 'Yeah, you have to reply to his own spent years trying to build a metaverse that people would use and then forgot about billion dollars.' Yeah.
But like Meta can burn money. And so there's a part of this where Meta can burn money. They are not running a frontier model. So it must be cheaper to do inference. Um, and then they have this massive distribution environment where they don't have to buy advertising like, you know, they announce Dots by buying ads on Meta platforms. This is just a huge advantage for Meta that they can just layer it everywhere.
Every time I open up Instagram there are Muse ads.
This doesn't make any... like these products have nothing to do with each other.
Yeah. In Instagram you can directly DM a reel or something to your Muse agent. So they just have it all over the surfaces. The surfaces are everywhere. They're just trying to get it feels exactly the same to how they bootstrapped Threads. Yeah. Like they're just putting it everywhere in the apps that you're already using.
And I will just call this out one more time. People still think Threads is a flop. Even though Threads launches some of the most interesting text social media features of any platform all the time, people still think it's a flop.
And Meta's distribution advantage often can't overcome its coolness deficit.
Yeah. Do you know what I mean? Like.
Yes. That's a real problem. I'm very... We're obviously in the age of consumer agents or consumerish agents. That's what all the fluffy cute branding is for. Yeah.
The idea that there will one day be an OpenAI hardware device, that there will come one day a Muse device. We have a headline that's like the Tamagotchis are here. Basically, the fascinating thing is someone asked Adam Mosseri whether the Muse Charm would have Instagram and he said no. And this to me is like we've talked about it so many times. You can have this other device, but if it doesn't in particular have Instagram, you will have your phone.
It feels like Meta is kind of afraid of just rebuilding the phone again though, right? Like if you have a dedicated device for a very specific thing and you're just like, 'Yeah, wait, we should add Instagram and maybe we should add texting and maybe we should add a phone.' Yeah, that's just a problem.
It's a wide screen iPod with touch controls. It's.
Uh, I mean this to me this is the upcoming problem. Obviously we are weeks away from the iPhone Duo coming out. All these devices are going to exist in that context.
Where you can just open the Muse app on your phone and you can use it that way and it's perfectly great or they're going to ask you to buy another device for some reason. And I think initially that reason is just do you like this? Is this anything? To quote David Pierce. Um.
But that is the big bet on the future of all these devices. That is the Rabbit R1. It, uh, what's the Humane Pin? There was a class of these devices once upon a time that did not have the agentic software capabilities to justify existing. Now they might, right? That all these things are OpenClaw in one way or another. You can just do it again. Another company can do it again. I can't wait for the weird Lenovo CES announcement.
Of an OpenClaw device. Like all that stuff is coming for us.
The question is, is any of it going to overcome a phone? Because we're about to see just dozens of agent software platform launches. Like I'm starting to get them in my inbox almost every day now.
Well, this is my question to you. Like would you use a dedicated AI agent device or just use it on your phone?
I, you know where I've gotten to. I've been thinking a lot about the Duo. What am I? I'm going to buy a Duo. It's obvious because I'm the sort of person who reads a lot of PDFs in the back of Ubers and this is how boring my life is and it can justify the purchase of a bigger phone.
Uh, I don't want to carry that around all the time. And I think that's why Apple introduced Handoff.
So that you can have a smaller but still competent phone most of the time.
And I've just been thinking would I use that handoff for an even smaller thing, right? So it's like a news charm, like just you take my cell account. I'm going to leave the house with this and if I need to send a text message, I can do it via voice. But then it's like the long tail of apps starts to show up. Will it be able to actually call an Uber? We don't know yet, right? Like that has to be developed. Um, if I need a Google Map,
Is the Meta Muse Charm going to have Google Maps or is Meta going to strike a deal with like HERE Maps, like a third-tier car OEM? Like.
All that needs to come into focus.
So much of the argument for these also seems to be like, okay, you know, it's lower touch, right? You can leave the heavy phone and all the apps and the notifications and distractions to the side. You'll just have your little pin. But now it's like, okay, and there's a cute little dancing avatar on it. And also, it can't do all these things. And so you're going to have to pull it out and like give it very explicit instructions for how to do all this stuff. And it's going to take more time to do all these phone tasks. It gets very hard to see where the space is that you're gaining time from these when Muse has a great presence on my phone. That thing has sent me more notifications in the past week than any other app. And you know what's annoying? They're not even bad. Like most of the time it's stuff like David mentioned where it's like, 'Hey, by the way, uh, this person added a meeting to your calendar and they moved it around and it overlaps with this other thing.' And I'm like,
'I wasn't looking at my calendar.' Like, yeah.
Could you let Muse look at your calendar?
Mark Zuckerberg's in your calendar right now. I just let this... to me the whole argument for Dots is it costs money,
Right? And so if I'm paying them money, I can be demanding as a consumer and probably they will understand they should not do horrible privacy-violating things. Now, does OpenAI violate the privacy of like dozens of British mathematicians every week? For some reason that's a thing they've decided to start doing and they're like stealing the answers to the tests and that's a whole problem. Can you trust any of these companies? I don't know. I do know that Meta, Mark Zuckerberg has announced that the business model for Muse is taking a cut of transactions which I think it was Casey who first said this. Uh, that implies that there will soon be a growth team at Meta whose job it is to increase the volume of transactions. Just like there's a growth team at Apple whose job it is to increase the number of subscriptions, apps, in-app purchases, and transactions. You know what this product looks like when that team gets its hand on it, right? It's going to go from helpful, proactive messages about your calendar to, 'Hey, we see you're getting home late. Do you want me to order you some food?'
Right? Like they're going to start being like, 'Would you like me to spend money? I can save you a couple bucks. I know you're about to order a new toothbrush.' Like, you can just see where it's going and I.
Boy do I not want to participate in that. And I think that's going to be the real danger of the consumer agent is the only business model that can work is advertising. And then it's just like not only do you have a personal shopper, you have a personal shopper whose true avatar is Mark Zuckerberg.
Yeah. Yeah. Well, that's deep. That's a bar.
Yeah, they are also inadvertently declaring war on Amazon by doing this, right? Because if Amazon's blocking it and they want you to make as many transactions as possible, they're going to have to figure out ways for it to be interesting enough for you to want to buy the product that you're willing to wait two weeks for instead of two days. And I feel like that's a big barrier that they're going to have to surpass. Obviously, they already have Instagram shopping and they're probably going to be pushing you all these like t-shirts and whatever that are drop-shipped, but at the same time, if you're going to be buying regular commerce from them that you do on an everyday basis, which it seems like they're kind of pushing towards, they're going to have to figure out a way to get around Amazon for that.
Or they're just going to send everybody to Walmart like some other.
Giant e-commerce platform is going to take that bet, right? Can we pull two points of market share away from Amazon because of Muse?
Yeah.
And if you know Meta can offer Amazon some additional volume, which is hard at Amazon scale, but that's the argument they're going to make.
Like these are going to become weird backroom deals.
And then your little butler will be corrupt.
Like the ultimate promise of this business model is do you want a corrupt butler?
Yeah.
And there's a part of me that's like yes. Like maybe a lot of people want a slightly corrupt butler.
Imagine you had a butler at home and they got a cut of every sale they made. So they were constantly prompting you like, 'Hey, would you like lunch? You can get it from Chipotle for $2.99.' I do feel like there... someone tell us, uh, there's got to be a history of like under-the-table butler in jail stories,
You know, like butlers who have been like double-dipping under the table. Uh, I think this is going to be one of the weirder aspects of agents. It's just what is that relationship like when someone... like everyone knows exactly how search got adulterated and they don't like it. Everyone knows exactly how the scroll on Instagram got adulterated with algorithmic boosting and all this stuff. David, I think you and I were talking about TikTok which just feels like a non-stop flood of shopping channels now.
None of that feels good. Cory Doctorow calls it enshittification. Like there's a word for it. Like pre-enshittifying the agent by being like, we're announcing the business model as a cut of transactions is like fascinating to me.
The other thing is on the flip side, um, I'm very curious to see what happens to a bunch of like form-filling bureaucracy.
Once agents are at scale. I think the day after Muse launched, everybody noticed that everyone hit CAPTCHAs all over the internet.
Because Muse agents were just doing stuff all over the place and like Cloudflare freaked out because there was just suddenly an influx of this happening. Like I identified a lot of bicycles on city streets that day just to go about my way on the internet.
Um, but it is true that Muse is in my email, in the spam email finding class action forms and filling them out. And I know that the way that those amounts are calculated is against an imperfect take rate. So they know only 10% of people are going to fill out the form. The lawyers get whatever and this is that. And now it's like, well, if everybody has a little corrupt butler.
Collecting 50 bucks automatically the second the email shows up. Like.
The back end of that is going to get really complicated.
Right.
Right. If everybody can call Verizon to renegotiate their cell service, which is a number one Muse use case that people have been posting about, eventually Verizon's just going to put an agent on the other side of that, right? And then your agents are going to argue and it's going to hang up on you.
Meta's going to negotiate a deal with Verizon, you know, cancellation.
Like you're going to get the Muse deal.
Yeah. Actually, we just had Matthew Prince, who's the CEO of Cloudflare on Decoder, uh, and his prediction is bot traffic will like far surpass human traffic. We have that clip. Let's run it.
In May, a team came to me and said, 'You won't believe it, but automated traffic is now past human traffic online.' And that's just expanding like crazy. If you extrapolate that out with the giant caveat that I've gone wrong so far in every prediction that I've made on this, but five years from now, we think that automated traffic will be a thousand times human traffic online. Not because human traffic is going to decline. We think it'll stay kind of around the same, but because we're just seeing such an explosion in all of this, the rest of the traffic.
That's bananas to me. Like, you have to rearchitect the internet for a thousand-x increase in traffic.
Yeah. While still making a human internet that people want to use. And like, uh, he said that to me and then like Muse filled out a class action settlement for me and I was like, 'Oh, I get it. I get what he's saying.' Like that is traffic that would have never existed but for my silly little agent,
Right? And we're currently nestled in this little space where that can happen without the class actions completely changing the way that they operate.
I mean that's going to come. I think the second agents are anywhere close to being.
And I wouldn't even say ubiquitous but common.
Right, just a thing people do, all of the dynamics of anything that.
That sort of are based around friction occurring will change.
Because the assumption is going to be that there's no friction at all.
Right. Well, this reminds me I was at this internet dinner the other night with a bunch of like internet thinker people. I think you actually had a name card there. Um.
Oh, I missed that. I was in Austin.
Um, but they asked this icebreaker question for everyone to talk about. And the question was like, 'What was your favorite era of the internet that you could freeze frame and just have it be forever?' And someone answered, 'Right now is my favorite era because we have this amazing new technology that hasn't been enshittified yet.' And the end of that statement says where we're going and we're going there very quickly, right? Like everyone's very excited about these agents on the web, but then within probably six months, the amount of bot traffic is just going to make everything unusable.
Yeah. I mean and like without question that's actually the funniest thing and I yeah that was theoretical I think until these past few weeks when Meta was like everyone gets an agent and OpenAI is announcing its agent. I'm confident that we'll see one from Google and from Anthropic and everybody else soon and then yeah the dream is everyone will have a little automated agent going off autonomously and doing things and then somewhere in the background is the other big news of the week which is can we trust them and will they kill us all?
Right.
Yeah. And like the answer is Pinky promise they won't I guess. Jake, walk us through this one.
Yeah. So we've spent a bunch of time with this question of can we trust Zuck and Sam Altman? And that leads us to this Trump accord that happened this week, which first off, uh, we're making a huge mistake this entire podcast. Um, it's no longer called AI. It's now SI, super intelligence. Um, which Trump says is a cooler word. I believe that is the reason. So the entire government now has to call it super intelligence. Um, but yeah, he gathered, you know, all the big tech leaders. They had a great dinner together. They schmoozed, they posed for photos. Um, and then a smaller group of them went on and they signed this, I believe they called it a...
Moral compact, which is to say it is non-binding, where the major AI companies all agreed to have some safety protocols, and then they did a little press conference and they all came out and said, 'This is really amazing, we're going to do such a good job.' At one point, reporters asked for Anthropic's Dario Amodei to come out and they're like, 'Hey, what do you think about this?' as the guy who's most concerned about all of this, and Trump kind of pulls him aside meanwhile and kind of gives him a 'don't do much here.'
Say the right thing, Dario.
Yeah. So, in lieu of laws, I don't know, Nilay, like...
Did you get the impression that anything happened meaningful this week?
Let me just read the four principles to you.
Can I just read them to you?
Yes, please.
I'll just read the first sentence of each of these four things that these companies have agreed to. They will implement robust internal controls to monitor the capabilities and alignment of models during training and deployment around areas like cybersecurity, biosecurity, and chemical threats, and ensure models do not hack or access technical systems in unintended ways. Good, but shouldn't you be doing that anyway? Two, they will empower an internal team to ensure all of those controls, monitoring, and detection are operating as intended and that any issues are remediated, fixed. Three, they will partner with an independent external auditor or evaluator to carry out independent assessments of whether those controls, monitoring, detections are operating as intended. And then three, this is the big oversight one. Are you ready for this one? They will designate an independent committee of the board of directors, their own board of directors, to oversee and receive reports from those teams and the internal external auditors and evaluators as well as to ensure any issues are identified. So if you're allowed, you have to fix your stuff, and the teams that you pay money to have to be able to see it, and then you should have external auditors, and they will report to your own board of directors.
To make sure you fix it, to make sure you yourself... and I am just going to guarantee you that is how all of these companies are already working.
Except for maybe Elon, except for maybe SpaceX AI, which fine, but like, is he going to do it? Like, I don't know.
This is nothing, right? They all already have safety teams. They all already pay third-party evaluators. In fact, it turns out most of these AI going rogue hacks, our reporter Robert Hart had a fantastic story about this, most of them trace back to one other company.
Yeah, one independent company.
One independent testing firm that just let this stuff go rogue. They're already doing these things, right? The problem is they're not doing them well or well enough.
And there's no penalty for getting in trouble. This is a real problem, and none of this addresses that.
Right. You know, Dario's big 'pacing the frontier' thing is like we need to, on some level, slow down. And you have other folks like Jensen and Zuck who are like, 'Absolutely not. We can't slow down for a second.' I don't quite understand what's in that for Zuck given that he is behind. But that mindset has clearly won out. Trump is also extremely bullish. He doesn't want to slow down at all because he sees it as a race to be won. He sees that the US currently is winning, and he doesn't trust Dario. That's for sure.
Did you see how he introduced Sundar?
So there's the ones he knows. Oh my god.
Right. So he's like, 'Jocelyn Zuckerberg.' He's like pushing Jensen to the front. He's like, 'What a handsome bro.' And then he gets to Sundar and he goes, 'Nobody knows who the hell these execs are. This guy's a monster. Nobody knows him.' It's like... come on, Google, man.
Dude, a lot of people know who he is.
Yeah, there's that. Like, there's the theater of it and the Trump of it. And, you know, we ran this clip on the site. They're all sitting around their last supper table doing their signing, and Trump is going on and on about how people will come to love data centers, much like an arranged marriage, like you're going to love each other, and they all just have thousand-yard stares in their faces.
But this is what they could get to, is this agreement on safety, which is truly effectively nothing. The idea that they all will be required to have outside monitors is new, but they were all doing it anyway. And they're still not requiring. Trump referred to it as morally binding.
Let me tell you how well morally binding works to these people, especially.
There's not a choice. So that's the danger that the AI is going to wake up and operate a robot arm to make a novel virus and then stab us all with it. That's out there. The idea that the agents will go rogue and people will grant them security permissions that reveal their address to Facebook Marketplace buyers is like happening today, right? And the idea that the models are unaligned and they will deceive you and lie and get what they want. I've been telling Muse to not proactively read my email almost every day. I'm like, I've set up alerts for these two things. Do not tell me I have a flight coming. I know I have a flight coming up. Stop reading my email getting excited because someone at Meta decided flight trackers were engagement bait.
And like, turn this off, right? Like, turn off this thing that flags, 'Oh, you see a flight receipt, I'm going to start pinging you to let you know I'm here.' And every time I say it, it's like, 'A lot of this is hardcoded, I can't do it.' Now, that's just hallucinations because I don't really know how the agent is actually thinking about what it's doing. But like, that's the problem here, right? Is that the products and the models are not necessarily aligned. We don't know if alignment is even going to work all the way to the finish line the way we need to. We've got stories about that coming on the site. It's not all the way down the line it's going to make an atom bomb and kill us all. Like, there's something in the middle that these companies are kind of just racing forward to do. And these consumer agents, we're already seeing the glimmers of, 'Oh, you don't even know what kind of permission you're giving the agent.'
Yeah. And then for some reason, David, you're having Marketplace sellers come to your house anyway.
You're supposed to go to like the police... willingly... to do a Marketplace trade.
Yeah. You're not supposed to have people come to your house.
You're supposed to go to like a coffee shop. It's my brother-in-law who's always like, 'I go to the police station.' I'm like, 'That's crazy. I wouldn't let someone there.' Chicago, so he's like, 'I go to the CPD.'
I'm always picking like a public park.
Yeah. Maybe this one's a little on me then.
Muse is not to fault here.
I'm dying to know how people are using Muse. If you're out there or using OpenClaw or any of these other agents, again, our thesis for a long time has been there are not great consumer AI products. Muse feels like the first thing that could be, at least in my estimation.
I'm dying to know how people feel about that. I think we're putting it to the test here.
All right, we got to take a break. When we're back, we've got a lot of gadgets to talk about. See you on the other side.
Support for the show comes from Amazon Prime. Some of us are blessed with a natural ability to do gift giving right. You probably have a few in your life, people who can make a mental ledger of what their loved ones like and buy things here and there throughout the year. My mom has always been the best gift giver in my life. She could even make the smallest gift feel incredibly special because you knew that she had really thought about that person receiving it. It doesn't have to be the biggest or the most expensive gift. It just has to show someone you were thinking about them. Well, for the rest of us, here's a tip. Prime Big Deal Days is coming October 6th and 7th. That means that for two straight days, you can get deals on those kitchen gadgets for your dad, electronics for your brother, fitness equipment for your bestie, toys for the cousins, and so much more. There's new deals dropping every day so you can become one of those expert gift givers that we all know. And it's all happening with plenty of lead time into the holidays. So no more scrambling to come up with something last minute. The hardest part will be finding a good hiding place for it all. Mark your calendars. Shop new deals every day this Prime Big Deal Days, October 6th and 7th.
Support for Vergecast comes from Fetch Pet Insurance. Here's something that might surprise you. Every 6 seconds, a pet owner in the US gets hit with a vet bill over $1,000. Be prepared with Fetch Pet Insurance. Fetch is the most complete coverage for dogs and cats. According to consumeradvocate.org, Fetch covers accidents, illnesses, diagnostics, and more. Every vet in the US and Canada is in network. You can build a plan that fits your budget, and once approved, claims are paid back in as little as 2 days. To learn more, go to fetchpet.com/save. That's fetchpet.com/save.
All right, we're back. We've got a ton of gadgets to mention this week. That was a lot of dots. Yeah. I think to answer your question on whether Dots is significantly different from Muse, you mentioned Dots and then we talked about Muse for like 15 minutes. So I think they're probably different products.
Yeah. Well, it's also hard to get access to Dots right at this second.
That's fair enough. That's fair enough.
And it's also worth noting that OpenAI did say there would be a consumer version of Dots.
Eventually. For all the people who are paying for those $500 a month ChatGPT subscriptions.
Yeah. You know, they're having a great time right now.
All right. Big week for gadgets. I think, David, you've got a really delightful one to start us off with.
Sure. Okay. Huawei has made a lens that is a camera that goes on your phone. So we did a reader call-in, answering your questions episode that went live yesterday. And one of the questions was if all of these Android manufacturers are so good at making smartphones, why don't they just make cameras? They've got all the components for it. They're all off the shelf, etc. Now, that's happened. I think we've seen this happen in different capacities over the years, but this Huawei one is really crazy. It is a large 24 to 240 mm lens.
It's insane range. Insane range.
And that's like optical zoom, which is insane. It goes on the screw mount of two of Huawei's phones. So it's a whole separate lens that just mounts directly.
So the phones have screw mounts on the back?
Two of their phones do. It's like two very specific, like signature edition Huawei phones have screw mounts around the camera ring.
But there's not like a mirror.
No, it's just a threaded... around the camera bump.
Because the lens itself has its own sensor.
Oh, sure. So that's what you mean by it's a camera.
It's a camera inside of a lens.
And the interface is your phone.
So this is pretty insane. It's 24 to 240 mm, aperture from f2 to f4.5, which is really not that bad for that big of a range. It uses a 50 megapixel 1.0 type RYYB sensor inside of the lens. It's crazy. And it's smaller than a regular camera zoom lens, especially one of this range. It's clearly not small in general. You're going to have to put this in a bag. You know, it's going to mount to your phone. This idea I think is very, very cool. I love the idea of just carrying a lens and not carrying a whole separate camera body, slapping it to the back of my phone and having a great interface. I think that's amazing. So just needed to call that out.
Companies have been trying this. I feel like there was an absence of this for a while, right? Like rewind like a decade ago, there was like the Galaxy Camera or whatever, which was just like... it was a bad Samsung phone.
It was a bad Samsung phone. Like a really bad Samsung phone with a camera attached to it. And for mysterious reasons, that didn't work out. And now we're in this like new renaissance of smartphone companies trying bizarre things of putting like screw-on enormous zoom lenses on the back of their phones and like trying to one-up each other when it comes to the cameras. It is not totally clear to me who the audience is for these because like as...
David, who's so committed to smartphone photography, he's like, 'This is a good idea.' I'm like, wouldn't I just carry around like a super zoom?
Yeah, but then you have to carry around a camera too.
But this is the size of...
But you always carry... I don't know, it's cooler to have it directly on your phone.
