It's time to panic about AI safety

It's a big week for new ideas about computers, so David and Nilay spend some time on Samsung's vision for foldables and Mark Zuckerberg's vision for AI agents that do absolutely everything. Then they talk about the fallout from last week's OpenAI Hugging Face hack, and whether all this talk about AI safety will actually amount to anything. Finally, it's time for Brendan Carr is a Dummy, the vertical video takeover, and the smashing success of the Ferrari Luce. Further reading: Samsung’s Galaxy Z Fold 8 feels like the future Apple launches ‘Upgrade’ program to lease new devices What’s the catch with the Apple Upgrade program? Mark Zuckerberg is planning a big push into personal AI agents Mark Zuckerberg wrote an op-ed about AI for the WSJ. We’re running out of reasons to ignore AI safety OpenAI reportedly didn’t notice its AI agent hacking Hugging Face until a week later. Nvidia, Microsoft launch open AI security alliance — without OpenAI, Google, or Anthropic OpenAI’s rogue AI agent didn’t stop at hacking Hugging Face AI leaders sign a statement asking the government to do something about automated AI Former FCC chairs blast Carr’s early renewal of ABC licenses as an ‘assault on free speech.’ ABC demands FCC drop its ‘punitive’ early license renewal of its stations Qualcomm is raising phone chip prices starting September 1st The Ferrari Luce reportedly already hit its 2026 sales target Samsung’s smartphone biz posts first ever loss, but profit still soars 1,300 percent on RAM. Facebook considers giving up and becoming TikTok YouTube Premium will include Peacock starting next year HBO Max is putting on vertical shorts Subscribe to The Verge for unlimited access to theverge.com, subscriber-exclusive newsletters, and our ad-free podcast feed. We love hearing from you! Email your questions and thoughts to [email protected] or call us at 866-VERGE11. (Timestamps are approximate.) 00:00:00 Intro 00:03:00 Baby Announcement 00:06:00 Galaxy Z Fold 00:12:00 Apple Leasing Program 00:15:00 Kill Switch Concerns 00:19:00 Zuck Agent Future 00:34:00 AI Safety Week Recap 00:42:00 Regulation Versus China 00:49:00 Winning AI Race 00:53:00 Security Arms Race 00:57:00 Brendan Carr Is a Dummy 01:07:00 Facebook Becomes TikTok 01:12:00 Ramageddon Price Hikes 01:16:00 HBO Max Shorts 01:23:00 Ferrari Luce Buzz 01:26:00 Wrap Up and Plugs Learn more about your ad choices. Visit podcastchoices.com/adchoices

• 0:01Speaker 00

  Company spending is complex. Managing it should be simple. This is AMEX Corporate, a single platform connecting cards, expenses, and payments to streamline the way you work. Flexible controls for when priorities shift, support around the clock, all backed by American Express, so you can stay focused on what matters most. Built to simplify, designed to scale. This is AMEX Corporate. Get started at go.amex slash AMEX corporate. Terms apply.

• 0:35Speaker 01

  Interrupted sleep, headaches, constant fatigue. For a lot of women, these aren't three separate issues. In perimenopause and menopause, they're often the same hormonal story. And you don't have to quietly push through. Midi can help because your symptoms have answers. Visit joinmidi.com with code VOX right now to book your first visit today. That's join M I D I dot com. Code VOX to book your first visit. Joinmidi.com code VOX. Insurance coverage varies. Check with your plan for coverage.

• 1:12Speaker 00

  Real innovation is powered by real people. This is AMEX Corporate. Trusted experts, smarter technology, 24/7 dedicated support from day one to answer questions, help set up policies and approvals, and consult on spend capacity as your business needs evolve. Plus, on-demand resources help level up operations at every stage. That's the trusted service you've come to expect. Built to simplify, designed to scale. This is AMEX Corporate.

• 1:45Speaker 02

  Hello and welcome to The Vergecast, the flagship podcast of the Roomba Band. I'm your friend David Pierce. Eli Patel is here. Hello, sir.

• 1:52Speaker 03

  Hello, how's it going?

• 1:53Speaker 02

  You have your normal background, but I think it's a fake background.

• 1:57Speaker 03

  I'll never tell. A lady never tells.

• 2:00Speaker 02

  Where are you in the world these days? You're theoretically on vacation, we should say. So you just, the lure of The Vergecast is too intense for you.

• 2:07Speaker 03

  I just like being in the chair, David. Um, this is my happy place. Uh, I, we're, I'm, we're at our house in the Catskills. The kids are going to a water park. That's what, that's what we're doing. Uh, and I'm theoretically, once again, working on my book.

• 2:23Speaker 02

  How's it going? Give us an update.

• 2:24Speaker 03

  A lady never tells, David.

• 2:28Speaker 02

  You, if I remember correctly, you and AI were having some discussions about how angry various people were going to be about how late you were on the book. How's that going recently?

• 2:37Speaker 03

  It is.

• 2:40Speaker 03

  I will say I received an email today with the word must in it.

• 2:44Speaker 02

  Cool.

• 2:44Speaker 03

  Mm hmm.

• 2:44Speaker 02

  We're doing great.

• 2:45Speaker 03

  Right, like we've, the language has escalated to, to commands. We'll see if I respond to commands. I, you know, like API error, sorry, ill-formed command. Like, we'll just see how it goes, but I'm getting there.

• 2:56Speaker 02

  Yeah. Historically speaking, when we both have other really important things to be doing, we make very long Vergecasts. So this is going to be a really terrific episode. I'm very excited about it. We actually have a lot to talk about. There's a lot of AI safety and open weights stuff going on. Lots of questions about how we do AI in this country and in the world. Lots of gadgety stuff to talk about. Brendan remains up to stuff. Ariana Grande is suing some people, which we may or may not talk about. We'll see. But first, I have two bits of personal news that I need to share with everyone here on The Vergecast. Piece number one is that I went to Target this morning and I bought the new low-profile PopSocet and the O-Snap. And I am now deep in some real scientific testing of phone grips.

• 3:43Speaker 02

  People have a lot of thoughts about these things. And what actually happened is I was like, the new PopSocet looks good. And I got 100 emails from people being like, O-Snap is better. Use O-Snap.

• 3:51Speaker 03

  I just figured your personal preference comes down to who is sponsoring your YouTube channel.

• 3:56Speaker 02

  Yeah, I mean, I do think it's possible I'm being astroturfed by O-Snap employees. But rest assured, I am doing testing and I will, I will reveal my findings at some later date. The second thing, the more important fact of my life right now is I'm having a baby. Again.

• 4:13Speaker 03

  One more time. Run it back, David.

• 4:15Speaker 02

  One last, in all caps, last time.

• 4:19Speaker 03

  I know, man.

• 4:20Speaker 02

  If you're, if you're seeing or hearing this and your response is like, gosh, David, didn't you just do this? I feel like we just had this conversation. You're correct. All of this is true. 14 months later, uh, here I am. We're just babying it up again. So, uh, I am, my, my wife is due sometime in mid-August, but we are sort of, as you're hearing this or seeing this, we are like entering the, it could kind of just happen any day phase. So I will either have a very graceful exit from the show in a couple of weeks and then I'll be gone for a while, or I will just suddenly be raptured away from The Vergecast. My plan is to be gone for six weeks or so, and then you'll, you'll start to see me again. And I think, uh, we're, we're recording some stuff ahead. So you, you won't be totally away from me for six weeks.

• 5:04Speaker 02

  But Eli, you've got the, you've got the fort held down while I'm gone.

• 5:07Speaker 03

  Yeah, we'll be fine. And the good news is one of your other 14 children will happily join me on the show.

• 5:11Speaker 02

  That's true. One of them now is old enough that like, I just leave the microphone on and he'll, he'll come down and figure it out.

• 5:15Speaker 03

  Yeah.

• 5:15Speaker 02

  It's gonna be great.

• 5:16Speaker 03

  That's great.

• 5:17Speaker 02

  But all right, all that is coming later, I hope. If I suddenly leave in the middle of this show, you'll know why. But hopefully it doesn't come to that. We got a bunch of stuff to talk about, but I think, as I was going through all of the news of the week, trying to figure out what is, what is going on in tech. The first thing I kept coming around to is this question we have asked many times over and over on The Vergecast. It's, it's been a bit of this show for a very long time, which is just, what's a computer? And, and I think we are in a very funny moment of asking that about a lot of things, right? Everybody is making AI devices and trying to ask, what's a computer? Apple is wondering if you might be interested in never buying your computer ever again, and instead just happily renting one from Apple.

• 6:01Speaker 02

  We have a new foldable phone that actually I think is very good. We're getting details about Google's foldables. We're just in a phase of uncertainty in terms of how are we supposed to do gadgets? That I think is very fun. So, um, let's just run through a bunch of these things and see if we can make any of these make any sense.

• 6:17Speaker 03

  Yeah.

• 6:17Speaker 02

  I do want to start with the, the Galaxy Z Fold 8, which we've talked about a bunch. But I am curious, we, the, the review's out. Dom really liked it. The general consensus about this thing seems to be that it's excellent. Presumably the iPhone foldable is going to look quite a bit like this. Are you sort of on board the, this is the next big shape of phone trend?

• 6:38Speaker 03

  I remain unconvinced. I, I, I think that the thing that comes out of every foldable, right? We're eight generations into Samsung making foldables. Many, many years into all kinds of other Android vendors making foldables. And unless you are the kind of person who's constantly unfolding the phone, it's not valuable to you. And we've, I think we've just learned, most people never unfold the phone.

• 7:03Speaker 02

  Yes.

• 7:04Speaker 03

  So am I personally willing to make whatever compromise on camera, whatever compromise on battery for a phone I may not unfold a lot? I, I don't think so. I, I have to use the devices. That's a big caveat. We have to see what Apple does, if they can make a better argument for wanting to unfold the phone. But I have always been of the mind that the flip style is the one that'll get people there. I might be wrong about this. We'll see if Apple ever makes a flip and that changes people's minds. But the idea that your phone is a thing you can put away, I think is very compelling to people.

• 7:39Speaker 02

  Yes.

• 7:39Speaker 03

  The idea that you have a phone that can get much bigger has historically proven not compelling to a lot of people, maybe because of expense. But also, I think, because the reality of those devices is you mostly do not unfold them.

• 7:52Speaker 02

  I think that's true. And I think the thing that Dom Preston and our team found that I find compelling is that he, he's sort of of the mind that maybe this thing has hit a good middle ground of, it is smaller, so it is a little more pocketable. It is a little easier to make it go away. It doesn't offer you the same kind of, you know, finality of like, closing and putting away a flip phone that I think is actually meaningful to people. But, but that this is much more so than a giant phone that turns into an even gianter phone, which I'm, I'm just bearish on as a concept. Like Samsung really wants you to believe there are lots of people out there who want to use two apps side by side. And I know those people exist. We hear from those people all the time. I just don't think there's very many of those people.

• 8:34Speaker 02

  And the idea that this might be in a middle ground of comfortable to hold even when you're not opening it because it's a little fatter and a little smaller, makes a lot of sense to me. Again, the, the problem is none of this matters at all while it's $2,000. Because, like you said, unless there is a really compelling reason to open this thing up over and over and over and over again, there's no way you're talking me into what it costs to own this thing. And we're getting better on the durability stuff. We're getting better on the battery stuff. The camera is still slightly disappointing. But like fundamentally, the argument is just, is this worth twice as much as your phone costs? Even, even if you make all else equal, is this internal screen worth twice the price?

• 9:17Speaker 02

  And, and the answer to that is just still no. Right?

• 9:20Speaker 03

  Yeah, I think the answer is no. And I, I do think the case for foldables has to be something about this gets dramatically better when you open it up. And that, at least in Android world, is, well, you better make a case for Android tablets.

• 9:34Speaker 02

  Sure. Which has been a hard thing to do over the years.

• 9:36Speaker 03

  Right? Like, I mean, that, that's kind of what we're talking about here. It's a phone that's going to open up into a little tablet. And you can watch movies bigger has not dissuaded anybody from watching Game of Thrones on the subway in vertical video and stolen TikTok clips. Like, there's just something about this that I, I suspect a lot of people are going to enjoy it. Dom really enjoyed it, really liked using the phone. But I think, I, it just, someone has to make a really good case for when you unfold it, X gets dramatically better. And I don't think that's happened yet.

• 10:08Speaker 02

  What do you think Apple's case is going to be? The, the immediate thing that comes to my mind is video games, that, that they're going to make a meal out of how great it is to play a game on the foldable iPhone. But I, I don't know if that does it. Is there, do you have a theory as to what it might be in September?

• 10:23Speaker 03

  I, it's funny that you picked games, which is a historical area of failed emphasis for Apple.

• 10:30Speaker 02

  But it can't help itself. They keep doing this.

• 10:32Speaker 03

  That's as politely as I can say that. I mean, they make a lot of money on mobile games. But you know, it's Apple and games and the twain shall never truly meet.

• 10:44Speaker 02

  Except for Candy Crush.

• 10:45Speaker 03

  Except for Candy Crush where they.

• 10:47Speaker 02

  Some overlap.

• 10:48Speaker 03

  Apple's entire services revenue, I will remind everybody, is Candy Crush whales. It's just taken 30% off the top of Candy Crush whales. That's, that's the business. But I think they're gonna lean into their other sort of confusing, we must make this work point of emphasis, which is now it's iPad apps. Or, you know, the iOS SDK and with Swift UI and da, da, da, da, da. Like, we saw, we saw it at W W C when you, right? There's a new middle sized app. There's like auto resizing things happening in the developer kits. You can unfold the phone and the apps just work perfectly on the new screen size, thanks to the magic of the Apple ecosystem. And is that gonna do anything? Is it, are a bunch of pilots gonna trade in their iPad minis? You know, like, that, I, I think that's the case they're gonna make, is you get the best,

• 11:32Speaker 03

  you get something approximating an iPad when you unfold this. But whether that is good or bad is totally down to whether or not you think the iPad is a worthwhile device. Like, talk about, don't unfold a phone. How many people do you know with an iPad that never use their iPad?

• 11:46Speaker 02

  A lot. And I would point out that based on everything we know, the foldable iPhone is probably going to be more expensive than the cost of a pretty good iPhone and a pretty good iPad. Potentially like hundreds of dollars more expensive.

• 12:00Speaker 02

  I, I think I'm, I'm more bullish on the wide foldable, I think, than you are, just because it seems really nice to use when it's closed in a way that other foldables have not. So like, it is weird to make the case for a foldable based on how great it is to use when it's not open, but I do think that's something. And I'm, I'm excited to see, we, we saw, by the way, a Pixel 11 leak. Google, as it is wont to do, just was like, hey, here's the phone we're gonna announce in two weeks. And it looks like Google is staying with the, the taller foldables. So we're, we're getting a real, like, head to head fight here between what the size and shape of a foldable phone should be. Yeah. Which I think is, is kind of fun.

• 12:40Speaker 02

  And also a true disaster for Android, which desperately needs some kind of standardization in ideas about how phones are supposed to look. Because it just makes everything harder on everybody when everybody has all these different ideas.

• 12:55Speaker 03

  That's the magic of the Android ecosystem, dude.

• 12:57Speaker 02

  Yeah, sure.

• 12:59Speaker 03

  Come on.

• 12:59Speaker 02

  There are ironically, a, a couple of other Apple-y things around this foldable phone that is potentially coming. I think, uh, Apple just launched an upgrade program this week, which replaces the iPhone upgrade program and essentially lets you lease a bunch of different Apple devices. I don't know that this is specifically about the iPhone fold, but it is very clearly designed to make it more palatable to buy increasingly expensive Apple devices. A lot has been made of this thing. You, you, you lease it for 24 months and then at the end, you can either upgrade to the latest model, you can trade it back in and go about your business, or you can pay the rest and keep the device. There are, I've seen a lot of fervent pro and con on the Apple upgrade program. What do you make of it?

• 13:41Speaker 03

  I feel like I want to have stronger opinions about it. Besides, everything's getting more expensive and Apple needs something in their store to make a lot of their devices.

• 13:50Speaker 02

  Sure.

• 13:50Speaker 03

  Seem better, besides the usual financing options. But carrier financing is a thing. I mean, I, I looked at some of the rundowns and I looked at the analysis we did on the site.

• 14:02Speaker 03

  And it is basically the same structure as the AT&T plan I've been on for like a decade.

• 14:08Speaker 02

  Yeah.

• 14:08Speaker 03

  Right? Where I just pay the money at the end of, I mean, I get a new phone every year. I'm, I'm in the stupidest position of all, but it has never made sense to me, because I get a new phone every year, to like buy and resell the phones and take the risk. I just like pay some installments and then I get a new phone and I pay some installments and I get a new phone and like on and on it goes. And that, I, I feel like Apple is edging ever closer to taking the relationship away from the carriers, because you're buying the devices from them. And so if you're gonna, you know, admit that the Mac Mini is gonna get ever more expensive forever and you need to set up an entire payment infrastructure to do it, you might as well make it big enough to encompass your most important product.

• 14:50Speaker 03

  The phone, which is also getting radically more expensive and in particular with foldable, potentially getting even more expensive than ever before. So I, I just see it as Apple looking at how people are purchasing their phones from carriers every single day and having effectively the same ability. And the iPhone upgrade program was always kind of like a little under the table secret.

• 15:10Speaker 02

  Yeah.

• 15:10Speaker 03

  That they didn't talk about. And now they have a thing, I, I think they just built out a capability and they're marketing it better.

• 15:15Speaker 02

  So you don't see this as some potentially fundamental shift in how we think about technology ownership?

• 15:22Speaker 03

  I think we discover that everyone is buying their Apple Watches and MacBooks this way, I will feel very differently about it. But the reality is, most people are already buying their phones this way.

• 15:31Speaker 02

  Sure.

• 15:31Speaker 03

  So in that case, in the phone case, it's like, sure, everyone's renting their phones anyway. If we come into a future where everyone is renting their laptop and this rumored kill switch, where if you get behind on your payments, Apple can shut you down, ever comes into play, like, yeah, then I will hair on fire. I will be very upset. It just remains to be seen if anybody wants that for real with non-phone devices.

• 15:55Speaker 02

  Yeah. That Apple kill switch thing is so weird, by the way, because the, people found the code for a feature like this. Apple then did what it almost never does and give a full-throated, on-the-record denial that this is not a thing they're going to do. And, and, but it's like, and so I saw a bunch of posts online that were like, so did you just, you just wrote this code for funsies? Like, what?

• 16:20Speaker 02

  Why, why is it here? Why did someone write this code if this is not a thing you're at least exploring or interested in doing? It's very much like all the weird stuff Meta is pushing to smart glasses but not turning on and then like, this feature doesn't exist. You're like, well, you, you did write all the code for this feature. What are we doing here?

• 16:35Speaker 03

  Yeah. And then you gave interviews to, to other journalists being like, yeah, we're gonna make facial recognition.

• 16:40Speaker 02

  Yeah, exactly.

• 16:41Speaker 03

  This is the new thing. Maybe this is just vibe coding. The power of vibe coding is people are just shipping, shipping features they never intend to use. Like, who knows? You know?

• 16:48Speaker 02

  They don't even know the agent built this feature.

• 16:52Speaker 03

  We should do social credit in the United States. Go forth, you know, like, I don't know. But yeah, I'll feel different if it comes to everything else. But I think the status quo in phones is already, and I think it's because your phone already has a pretty big monthly service plan associated with it.

• 17:07Speaker 02

  Right.

• 17:07Speaker 03

  That everyone's just really comfortable and like, oh, sure, I'll get a new phone every year. If it's laptops, if it's like Apple TVs, I don't think Apple TVs are in here, but if it's like Apple Watches and laptops and everything else, whoo, that's gonna, that's gonna feel like you.

• 17:19Speaker 02

  Yeah, it, it is, it is strange. And I think in particular, I was actually thinking about this the other day, that like my laptop, of, of every device that I own, my laptop is the one that feels the most like my computer, right? Like my, the existence of my iPhone is in some combination of like my relationship with Apple and my relationship with Google and my relationship with my carrier. Like, it is so dependent on so many other things for basic functionality in a way that like my laptop, and this isn't even actually all that true anymore because the internet exists, right? But like, my laptop feels like my computer in a way that.

• 17:54Speaker 03

  Oh, yeah.

• 17:54Speaker 02

  Any, any intrusion in that relationship feels weird and different in a way that I think you're right. Going all the way back to, you know, 20 years ago, when we were getting our phones for quote unquote $200. And that was just a lie and you, you paid the difference in your, in your bill every month to your wireless carrier. These fees have always been hidden. And there was a minute where it started to get a little more transparent and Apple is just like, no, no, no, no, let's, let's go back to, let's go back to pretending everything is cheap. This is great. It's so, it's terrific when we can just hide the price of these things when, and make it easier for you to walk out of the Apple Store.

• 18:27Speaker 03

  Yeah, I mean, and you might hate this. Like, I know a lot of people, they're like, you should buy the phone full price every time and then you just sell it on back market or whatever, you trade it in. Great, you can do that in that. Thankfully, that is more possible now than ever before. But in the early part of the smartphone market, when the iPhone first came out, Apple was just selling it for full price. That is that Steve Ballmer video, where he's like, $5.99 with fully with the plan.

• 18:50Speaker 02

  Right.

• 18:51Speaker 03

  He's like, my phone's $99. And it's like, no, it's not. Your phone is the same price, but you've hidden it in some fees. And it didn't work. Like, Apple had to go back to AT&T and come up with its baked into the price of the service.

• 19:06Speaker 02

  Right.

• 19:06Speaker 03

  And, you know, if you just think about like, would you prefer to buy a phone upfront for $600? And I, AT&T's first plan for the iPhone was like 20 bucks a month. Like, it was a great deal, but nobody wanted it. And they went back and they lowered the price of the phone and they made the price of the plans more expensive and then everyone bought that. So there's just something about the, the market here that maybe wants some fees hidden. I, I can't, it's like, I don't know. I, I can't tell you that that's, I love it, but like, the people have spoken and they would like their fees to be hidden. Thank you.

• 19:35Speaker 02

  Yeah. The one last piece of this I've been thinking a lot about is a thing Emma Roth on our team included in her story about this, which is something like more than half of people who buy something through Klarna miss a payment. It's just like, is this predatory? Not exactly, but it often plays out that way. And so like, be careful out there is all I will say. All right, we have a couple more to get to before we get out of the what is a computer questions here. And we should probably talk about Mark Zuckerberg, who is, just loves, loves a weird idea about computers. Nobody does weird computers more than Mark Zuckerberg. And there are kind of two different moments this week where Mark sort of laid out the way that he thinks about computers going forward. And we're gonna split these because one of these is just perfectly aimed at me and one

• 20:20Speaker 02

  of them is perfectly aimed at you. Does this, does this sound okay?

• 20:22Speaker 03

  That's right. And by perfectly aimed, it means, oh, I was alone in the woods screaming at the sky.

• 20:28Speaker 02

  Yeah. I, I punched a hole in my basement ceiling after reading the other one, so we're doing great. Okay, so this, this is a quote from Mark Zuckerberg on Meta's earnings call this week. He said, "Soon we will have agents that can work 24/7 on your behalf to help you achieve your goals and improve your life, your health, your relationships, your finances, whatever you want. The first domain that agents have really taken off in is coding, but engineers are more technical and willing to spend time making those agents work. So to build great personal agents, this needs to be a great consumer product that just works out of the box and is easy enough for billions of people to adopt and use." All like this, this is such a like, Mark, you, you're almost getting it.

• 21:03Speaker 03

  Yep, they're so close, man.

• 21:04Speaker 02

  You almost understand that actually, everything we've built only works for coders. And this, this is the most like, Neelai sitting in his basement screaming that people do not yearn for automation.

• 21:14Speaker 03

  This man has software brain.

• 21:16Speaker 02

  Yes, yes.

• 21:17Speaker 03

  So it's pure software brain.

• 21:20Speaker 02

  Yes. And, and there is this idea that because this works for people who like to write and work with and operate code, we are one small step away from everyone else being able to manage their finances and relationships through agents. Is this, is just this like idea that has taken hold in Silicon Valley that drives me absolutely insane? Because those two things could not be further apart from each other in, in what it takes to make them work, in what people want from them. They have no relation to one another. And Mark is actually being like, well, we figured out how to code, so we're this close to fixing your relationships with AI. And it's like, brother, no, you aren't.

• 21:59Speaker 03

  They're so far away.

• 22:01Speaker 02

  But their big idea, and this is all around a, like, clear product roadmap from Meta that is about agents, right? Like, this company is deep into the idea that this should come to your profiles on various meta platforms. It should be part of how you interact with their computers, be part of how you interact with technology. Like, they are going to brute force weird AI agent dating ideas onto the billions of people who use their products every month. And I, it's just going to be such a disaster.

• 22:29Speaker 03

  I just want to point out, this is a company that cannot meaningfully solve the problem of listing something on Facebook Marketplace and having 10 billion bots say, is this still available?

• 22:39Speaker 03

  Have you ever tried to give something away in a pay nothing group?

• 22:41Speaker 02

  Like, a true disaster.

• 22:44Speaker 03

  Is it still available? And it's just like robots all day long. So you'd solve that problem, Mark, and I'll let you do health next, you know, like.

• 22:50Speaker 02

  Seriously.

• 22:51Speaker 03

  But yes, you're right. Like, I read this and I think I had the same reaction as you, which is the whole world is not databases, no matter how much it looks like that from your position, running a giant platform.

• 23:04Speaker 02

  Yeah. You know what's super not a database, by the way? Your life, your health, your relationships, your finances. He picked four incredibly not databasey things and is like, oh, it's fine. We'll just, we'll just AI agent it.

• 23:15Speaker 03

  No, that's, that's software brain. Your life is contained in a series of databases, right? You're, you're just on Instagram all day, which is basically just a database of photos and videos. And you're doing whatever, you're communicating that and you're planning trips because no one can resist planning a trip. That's all anyone does in their life. Your health is just a series of medical records in a database.

• 23:37Speaker 02

  Yep.

• 23:37Speaker 03

  Right? You could just go down the list and you can see how all of these dudes think the world is a series of databases to be manipulated and for AI to go traverse them and manipulate them and whatever you're going to do. And that will somehow be useful or valuable to people. And they all, it all falls apart when you point out that anyone who's ever run a database knows the database and reality do not actually line up. And it is way easier to change the database than it is to change reality. But if your agents are stuck in the database and they can't make that jump and they can't actually see what's happening in the world, then you're left with nothing. And I think that is generally the experience a lot of people are having with AI as it exists today.

• 24:18Speaker 03

  It's, it's in a hermetically sealed like container of only the context you have given it. And if you don't give it everything, it's going to get real confused about everything all the time.

• 24:28Speaker 02

  Yep.

• 24:28Speaker 03

  And most people are never going to give it anything, everything. And then if you're Mark Zuckerberg, you have to overcome the fact that people don't trust you or your company with any of their information. And you're never going to get enough personal context to make this thing useful. And so you're right back at software engineers who work at your company doing agentic loops, which by the way are verifiable because you can compile the code and test it, which is not something consumers can do in their lives.

• 24:53Speaker 02

  Yep. Which is why they're just going to brute force all of this into Instagram and hope that you funnel the data they need into it more or less by accident. But Neelai, can I offer you a counter theory, which runs into the one of these that I want you to really get in your feelings about? Which is just, what if everybody had great databases? Maybe the problem is we don't all have good enough databases. And if I can get you a better database, I can solve all of your problems forever and ever.

• 25:16Speaker 03

  You're going to make me read the lawyer part.

• 25:17Speaker 02

  I'm going to make you read the lawyer clip.

• 25:19Speaker 03

  So Zuck claiming that soon we'll have agents that work 24/7 on behalf. That's on the earnings call, right? He's trying to tell investors that all the money he's spending on AI is going to be worth it. Right next to that, he published an op ed in the Wall Street Journal, calling us all to be more optimistic about AI, which is really a push for more open weight AI and less regulation of AI in the United States, because we're scared of China. But to sell that idea, you gotta sell a grand vision of the future.

• 25:45Speaker 02

  Right.

• 25:46Speaker 03

  And everyone has realized we're going to fire all the white collar workers in America is not a good enough grand vision of the future. And it is in fact causing data centers to be protested. So everyone is now promising super intelligence will save us all. So here's Mark's thought experiment, and he literally calls it a thought experiment. I'm just going to read you a quote. This is from the Wall Street Journal. "As a thought experiment, imagine only one person had a super intelligent lawyer. He would have an unfair advantage in court, even if his position was wrong in the merits. That would lead to a worse society. But now imagine everyone has a super intelligent lawyer. Justice would be carried out much more fairly and efficiently than it is today.

• 26:28Speaker 03

  When people are empowered, they naturally check and balance each other and the power of large institutions."

• 26:35Speaker 03

  Let's run this thought experiment to ground. Let's just, let's just take this thought experiment and put it into practice. Mark Zuckerberg is a very wealthy man who operates a very wealthy company. When Meta goes to court, it has the fanciest lawyers you can buy with infinite resources and infinite time, because they can employ infinite first year associates to do whatever it is that they want their first year associates to do. I would, that's a version of having a super intelligent lawyer, I think. Meta loses the cases.

• 27:09Speaker 03

  Right? A bunch of like impoverished state attorneys general show up without Meta's resources or money or polish or free use of AI tools to write the briefs. And they have been consistently winning child safety cases against Meta. Right? Like, in reality, the price or intelligence of your lawyer meets the actual justice system, meets the actual jury. And you cannot predict what will happen. And Zuck should know this, not only because he goes to court with his armies of expensive lawyers,

• 27:44Speaker 03

  but dude runs a court.

• 27:48Speaker 03

  The Facebook Oversight Board exists to adjudicate decisions about content moderation that he thinks are too hard. And that thing is slow, it is confusing, it is made up of like former prime ministers who don't agree. Zuck doesn't want to pay for it. Like, they get things wrong or in confused ways. You have all the, you literally run a justice system for content moderation. And you're like, I'll just throw AI in this and that'll make it more fair or efficient. I don't know, man, like, this, when I see software brain, it always comes out. It's always pointed at the legal system first.

• 28:27Speaker 02

  Yep.

• 28:27Speaker 03

  Because the formalism of the legal system, like, we're going to use these words to say to this judge at this time, makes people think it's code. Like, literally, Larry Lessig wrote a very famous book that was like, law is code, code is law. Like, the idea that the structures of the software create rules and regulations that map to the law, like, it's so tempting. And you're like, if I could just make a robot lawyer, everything would be better. And it's like, dude, you take your super intelligent lawyer to your next child safety case and be like, your honor, instead of my fancy law firms, Muse Spark will be defending me today. Let's just see how that goes.

• 29:09Speaker 02

  Ironically, the, the thing they keep losing a lot of these cases on is A, the merits, which Mark Zuckerberg says have no bearing on anything. And B, what a lot of the opposing lawyers do in these cases is do things like read Mark Zuckerberg's words back to him in court. Things that he says in emails that are extremely incriminating, which is a thing that if you gave Muse Spark perfect access to all of your things, it would also be very well equipped to do. I usually think you could make a really great case that having an AI lawyer is like a really easy way to just like nuke your own chances in discovery.

• 29:44Speaker 03

  I mean, this is the game. There's a lot of companies chasing this dream.

• 29:49Speaker 03

  But the idea that like two super intelligent robot lawyers would lead to justice being carried out much more fairly and efficiently really implies, one, that the two robot lawyers do not just enter a version of procedural warfare to delay any outcome forever. Which is basically what rich lawyers can do for rich clients.

• 30:10Speaker 02

  Yeah. They discover they can bill more if they don't end the case. This is the paperclip problem, but it's the infinite.

• 30:16Speaker 03

  Yeah. You're like, you, you want to sue Elon Musk? I'll tie you up in court forever is a real thing, rich people say all the time, right? So like, great. So you have to somehow make everybody's AI be like, what we want is efficient justice. Good luck.

• 30:30Speaker 02

  Yeah.

• 30:30Speaker 03

  That doesn't seem like a thing. And then it also implies that the judges will agree that the robot lawyer lawyering should be handled more effectively and efficiently. And judges by and large are like, get your AI out of my face. Get out of this courtroom right now. Like, literally, it's like, as a thought experiment, it's like, Mark, have you thought about it?

• 30:50Speaker 03

  Ridiculous.

• 30:51Speaker 02

  Yeah. Can I interest you in robot judges?

• 30:55Speaker 02

  Which famously would go super well.

• 30:57Speaker 03

  People love a robot judge. There's, there are, there's academic data suggesting that people believe the regular court system is biased and they don't, and they think the robots won't be. People love a hidden feet, man. That's all, you can't, you can't account for taste.

• 31:13Speaker 02

  Yeah. All right. This is, this is a good segue into some of the, what do we do about all of this AI stuff? That we should talk about. So let's take a break and then we're going to dig into some AI safety stuff. We'll be right back.

• 31:27Speaker 01

  Interrupted sleep, headaches, constant fatigue. For a lot of women, these aren't three separate issues. In perimenopause and menopause, they're often the same hormonal story. And you don't have to quietly push through. Midi can help because your symptoms have answers. Visit joinmidi.com with code VOX right now to book your first visit today. That's joinmidi.com, code VOX to book your first visit. Joinmidi.com, code VOX. Insurance coverage varies. Check with your plan for coverage.

• 32:04Speaker 04

  This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required. Compatibility and availability varies 18 plus.

• 32:29Speaker 05

  If you've got an insurance question, you could talk to the butcher at your local grocery store. He'd probably talk about trimming the fat, but it'd be about your brisket, not your insurance policies. Or you could talk to your local GEICO agent. They offer personalized assistance in finding the choicest cuts of coverage for all your insurance needs, which means more money for filet mignon, or if you're a vegetarian, tofu le mignon. To find a GEICO agent near you, visit geico.com/ local.

• 33:02Speaker 02

  All right, we're back. Neelai, I struggle to figure out how to summarize this week in AI safety. But let me just very briefly try. So last week, there was a lot of news about OpenAI hacking Hugging Face because one of its AI agents essentially went rogue, found a way out of a sandbox onto the internet, did a bunch of wild stuff in order to game Benchmark. The details of that are sort of fascinating, but kind of immaterial to where we've actually gone. This has become, over the last, I don't know, 10 days or so, sort of an increasingly large conversation. Right next to this has also been a bunch of new AI models coming out of China that appear to be very competitive with the the big models coming out of Anthropic and OpenAI and Google.

• 33:47Speaker 02

  And a big thing that the Chinese models by and large have going for them is that they are open weight, which is not the same as open source, which Robert Hart has to keep writing on our website like every eight hours. A reminder that open weight and open source are not the same thing. But what this does is give you a lot more access to understanding how the model works, which means you can tune it and tweak it and just understand it better. They also run much cheaper and more efficiently. So these two things are happening, right? Where all of a sudden, there is this new, cheaper, more open competition. And also, these closed proprietary models are running rampant. This has caused a panic. Panic might be a little strong. It has caused a lot of people in AI to feel a lot of feelings about AI, all the way up to and including Sam Altman,

• 34:32Speaker 02

  who has just been out there giving interviews, sort of wringing his hands about what's going on with AI.

• 34:36Speaker 03

  Even though OpenAI is at the center of one very important controversy at the heart of all this.

• 34:41Speaker 02

  Yes, very much so. Couple of things that happened this week. Last week, Jensen Huang, the CEO of Nvidia, put out an open letter basically calling for more openness and more sharing among open weight models in particular, in support of them because the US government is trying to crack down on how AI gets released. It became a whole kerfuffle about who did or didn't sign that letter. Nvidia, Microsoft, and a bunch of other companies then announced an OpenAI Security Alliance. As far as I can tell, sort of meaningless conglomerate of companies designed to say nice things about AI. AI companies like OpenAI, Google, and Anthropic are not part of that.

• 35:20Speaker 03

  Sure.

• 35:20Speaker 02

  It's become this whole thing. So I'm not even exactly sure where I want to start, except to say that I had a mild panic attack on this show last week about how big a deal this was. And I feel more justified all the time in thinking that this is this is a big deal and we need to do something about it. What's your read on on kind of how the industry has responded to this over the last week or so?

• 35:45Speaker 03

  I think there's a lot of competitive posturing going on here. It seems obvious that you need some sort of regulatory framework to at least make the people who make the mistakes pay money.

• 35:57Speaker 02

  Sure.

• 35:57Speaker 03

  Right? Like, at the very least, you need to be able to say, well, you're you're going to be liable for your agents going rogue and hacking to other companies. Like, we can invent that in the courts through some lengthy robot lawyer process. Like, we need some statutes and some frameworks and some like ways to account for this is bad and you should be punished for it. And that should provide you some incentive to not do it again, which OpenAI.

• 36:20Speaker 02

  I think the robot judge says yes to that.

• 36:21Speaker 03

  Yeah. Presumably, that's where we need to go. Now, are the people in charge of AI regulation in this country capable of coming up with that plan and then passing it? Like, no, absolutely not. Like, without question, that is not the case. And, you know, if you go back to like the Biden executive order, which was pretty toothless, they railed against things like, you have to tell us how capable your models are before they go out. And that was like way too much. Like, we're going to lose the space race, guys. And now we're kind of at like, what if we shut it all down? And you have like, half of the companies on the, we should shut it all down, but we can't do it on our own because no one else will be motivated or incentivized to shut it down if we shut down. So we'll just lose.

• 37:05Speaker 03

  And then you have, I think, this other set of companies are like, the open weight Chinese models are here. You can't do anything about it. The only thing we can do is build more data centers and make sure our models are the best so they can protect everything from this universe of rogue models. And the answer has to be somewhere in the middle. Right? This is what a government is for. I find it particularly galling when we hear big tech leaders say things like, you have to trust the democratic process.

• 37:31Speaker 02

  Right.

• 37:32Speaker 03

  But like, in this case, this is what it's for. It's just our government, as presently constituted, can't do anything. Right? Like, physically locating senators is not a capability of our government right now.

• 37:48Speaker 02

  Sure isn't.

• 37:49Speaker 03

  So like, how are you going to get a framework that makes everybody comply and like play along and keep everyone safe? Which is fundamentally a collective action problem that laws are meant to solve? You've got to have a person in the government who's like, here's the framework. It's happening. We're going to pass it through Congress. Can you identify that person? Can you identify Congress's ability to pass laws? Can you identify who in this motley cast of characters is going to complain? Who's going to see that as a lock-in advantage for them? Which I think a bunch of big companies would like. I think Anthropic would love tightly regulated AI in this country because they think they're going to win in that game. So there's just a mess that is meant for politics to solve. And our politics are so broken that I think we're just going to keep having hacking incidents

• 38:33Speaker 03

  until something goes so wrong that everyone is forced to stop and contend with it.

• 38:37Speaker 02

  I think all of that is true, but I also keep thinking about when we did this with social media last decade, that you had the same sort of setup, right? You had a run of incredibly powerful companies having an incredibly consistent set of bad things happen on their platform. And one by one, they all came out and were basically like, please, dear God, regulate us. We don't want to set the rules because if we set the rules for ourselves and they turn out to be less profitable than the rules somebody else sets, then we'll lose and we can't risk that. So we need every, we need some external party to set the rules for everybody so that we all have to play by them. Because because competitively,

• 39:19Speaker 02

  none of us are incentivized to make good, rational decisions about the world. That is exactly where we are with AI right now. There is, they, all these companies are like, we desperately want you to slow us all down because we can't slow ourselves down because it's bad for business. And and so you get people out here being like, you, this regulation is good. Maybe there should be a pause on some of this stuff. Even Sam Altman is like, maybe we should slow some of this stuff down to speed up the process of working on safety and that that we should do that more. And there are a lot of external people who are like, we should actually pause all model development until we have built some of these rails. Do you know who's not going to do any of that? Is any of the companies making models? Because they're all losing money at record pace.

• 40:04Speaker 02

  There is still a belief that this is going to be some kind of winner takes all industry. And they are like, we can't possibly slow down. The only way we can slow down is if someone bigger than us demands that we all do it. And I think to some extent, that is just offloading responsibility onto some other body that you also know is not going to get it done for all the reasons you just said. But is also like, in the world in which we live right now, where the entire economy is tied up in AI, it's all these like circular deals with Nvidia, where when somebody gets worried about the weird deals that Nvidia is making to invest in everybody else, like, the whole economy panics. And so we're just at this point now where I kind of don't know what the middle ground

• 40:47Speaker 02

  you just described even looks like. Because the government is like, we need more control, but we also have to beat China. And the AI companies are like, we are moving the fastest, but we also desperately need to be regulated. But if you regulate us, we'll lose to China.

• 41:03Speaker 03

  That's great.

• 41:04Speaker 02

  It's just like, everyone has tied themselves up in knots that I'm not entirely sure how to unravel here.

• 41:11Speaker 03

  So let me distinguish this from social media. I agree with you that it looks the same. And in fact, several years ago, when Biden put out the AI executive order, President Obama was on Decoder and he made the exact same comparison.

• 41:24Speaker 02

  Oh, really? See?

• 41:25Speaker 03

  And.

• 41:26Speaker 02

  Me and Obama, I always say we have a lot in common.

• 41:28Speaker 03

  Yeah, you and Barry. And his point was like, we screwed this up. We can't screw it again. And then, you know, we talked for a while about how it got screwed up. And the difference with social media is that social media is people's speech. And writing speech regulations, generally not allowed. Now, we'll come to Brendan Carr in a minute. He's trying his hardest. But government speech regulations in this country are generally forbidden. So if you want to write a bunch of content moderation rules for Meta and TikTok and Snap, they're just going to say, well, the First Amendment prohibits you from doing this. You got to find some other way into it, which is essentially how we've backed into a bunch of kids' safety regulations.

• 42:11Speaker 03

  Right, there's, you can make the argument in some way that in order to protect the kids, we must restrict everyone's speech. And maybe that's gonna work. Right now, there's a bunch of data saying that would actually help children or children are being harmed. And we're gonna see how this all plays out. These lawsuits are ongoing, but that was the solution. And it took over a decade to land on this being the solution. It also took a whole bunch of kids who got harmed by the ills of social media.

• 42:36Speaker 02

  Yeah.

• 42:36Speaker 03

  So this is not great. AI does not have this problem. There's no First Amendment protection for code hacking other systems autonomously.

• 42:47Speaker 02

  Sure.

• 42:47Speaker 03

  It just doesn't exist. So in the social media era, you know, Facebook would have billboards all over DC. They would put up ads at the top of Politico being like, we welcome regulation.

• 42:57Speaker 02

  Yeah.

• 42:57Speaker 03

  And you, you, I would just always look at that, be like, but you know, they can't write any. Right, like asterisk, but the First Amendment will keep you from ever doing it. Like, we're just gonna pay your bills and you're gonna say Facebook is working with us, but we know the truth. In this case, there's no First Amendment protecting the output of an AI system or protecting the things an AI system does. And we are now very much in a place where the computers are doing things. And who should have liability for the things the computers do? And whether all of that stuff should be subject to export controls? Or whether you can classify some of it as weapons and put it under that? Like, there's a whole universe of stuff you can do that is no longer blocked by the First Amendment, because we're not really talking about people's speech.

• 43:40Speaker 03

  Maybe in some cases we are, but the AI system's doing things, like taking actions. That's, that's not speech that you have to worry about the First Amendment. So I think there's a critical difference here. I just don't think that the Trump administration, it is presently composed, or our Congress, it is presently composed, can come up with a bunch of ideas that they then have to muscle through, because they will have to muscle through whatever framework they put into place. And there's enough division and enough money and enough competitive fury and enough China phobia all in the mix here.

• 44:15Speaker 02

  Yep.

• 44:15Speaker 03

  That the political will to muscle through a framework almost certainly dies immediately. And I, I think that's really bad. Like, I, you talk about trusting the democratic process. That's a pure failure of an actual democratic process, because you look around and AI is still polling bad. We have a bunch of great coverage on the site this week of protests at data centers and the, how bipartisan those protests are. Gabby Devai was in Florida at a very MAGA tinged data center protest. Not even tinged, just like full MAGA data center protest.

• 44:43Speaker 02

  Yeah.

• 44:44Speaker 03

  There's a lot going on here that democracy should actually take a hold of. And like, our representative government should do something about it. And I think what's gonna happen is it's just gonna accelerate and get crazier and crazier until the midterms. And then a bunch of people will be elected or not. And those people, I think they're, they will come into office on the back of a political groundswell around AI. And they will almost certainly have to do something.

• 45:13Speaker 02

  Maybe, but then you, you run the very real risk of that something being essentially tanking the economy. Like, there, there's just walls everywhere you look, right? And I think, to your point about the kids thing, I, I'm increasingly convinced this is why everybody uses China. Like, we, we talk all the time about the only way to get things done in the United States is either to save kids or fight China. It's like, those are the only two bipartisan beliefs in, in the government right now.

• 45:35Speaker 03

  Yeah.

• 45:35Speaker 02

  And because they can't really talk about kids with AI, although there's some of that happening. They're, they just yell about China. And one of my favorite things is every time we cover this, we talked about it on The Vergecast last week. We, we talked about it all week. Robert Hart's been writing some really great stuff for us. I think he's gonna be on the show next week talking about some of this stuff, too. Over and over, the response from our commenters in our audience is, what, what does it really mean if China wins in AI? And there's like, there are people who would love to tell you all of the things that happen if China beats us to drug discovery and all these things and like giant disasters. But I think fundamentally, this is a weird race that people don't totally understand or believe in at all levels.

• 46:16Speaker 02

  And, and so you get the government yelling about China and you get the AI companies yelling about China for a completely different reason. They're not worried about national defense. They're worried about the fact that Chinese companies are gonna undercut their costs and destroy their IPOs. Like, everybody's mad at China, not for the same reason, but just because it's a really useful boogeyman whenever you wanna get your way with AI. And it's working and it will continue to work, I think.

• 46:42Speaker 03

  Yeah, I mean, I think a really good exercise here is just going down the list and being like, what do all these companies and people think winning in AI looks like? Demis Hassabis believes we're in the foothills of the singularity.

• 46:54Speaker 03

  Is that winning in AI? Is it the singularity? And then we'll never know what happened before, you know?

• 46:59Speaker 02

  Sam Altman said this week that, uh, he, that I think he said something along the lines of, we're on the verge of having a genie that can grant any wish, which is like the most you can just say things of anything that has happened to me this year.

• 47:11Speaker 03

  That is the purest software rain I've ever heard in my entire life.

• 47:14Speaker 02

  Unbelievable.

• 47:15Speaker 03

  You can just talk to the computer and things will start happening. It's like, well, is everything else connected to the computer?

• 47:21Speaker 02

  Insane.

• 47:23Speaker 03

  I would love a robot genie that could grant you a wish. This'll do it. Okay, that, is that winning in AI?

• 47:28Speaker 02

  Yeah, I, I guess.

• 47:29Speaker 03

  You know, and, and Microsoft will call it superintelligence, but they won't say AGI. Like, literally winning in AI is what? And it is a useful exercise to just go through and be like, all of these companies at the, at the top are not really talking about economic winning. They're talking about getting to AGI or whatever first.

• 47:47Speaker 02

  Right.

• 47:47Speaker 03

  And then something is gonna happen that's gonna rule.

• 47:52Speaker 03

  What is it? Is it Mark Zuckerberg that's dating for you? Like, you, you tell me what that is.

• 47:58Speaker 02

  Yeah.

• 47:58Speaker 03

  And usually the fallbacks are, yeah, we're gonna do drug discovery. Yep, science will accelerate in whatever way it's accelerating. But it never comes back to if China starts doing those things first, there's some reason why we, we will never do them. It, it, I mean, it's a different, it's a different comparison, but I mean, China was also the reason that we had to overbuild 5G, right? We, we had to win the race to 5G. And it was always like, and what will happen if we don't? Will all the people in Kansas move to, to Shenzhen? Because they've got great 5G there? Like, what, what do we think is happening that is not already happening? Because the 5G networks don't exist. And I never got a good answer to a question I ask everybody.

• 48:38Speaker 03

  We should start, maybe in our reporting, spending more time being like, tell me what winning in AI is and what happens if we lose.

• 48:45Speaker 02

  It's a good one. I like it. All right, before we get to the lightning round, I, I am curious how you feel about the OpenAI hugging face thing in particular. They've done a bunch of new excavation of what happened. There's some really interesting, like, technical deep dives on how all this happened, including the, the discovery that one of the things OpenAI's agent did was go and find login credentials to other services. They haven't named which ones, but four other services where just usernames and passwords were apparently available on the web, which is cool and fun, uh, and went and hacked into services that way. This brings a whole new weird angle to all of the data breaches everywhere that all of a sudden you're gonna allow AI agents into every service because they can just get in the front door now? That's wild.

• 49:27Speaker 02

  But I, I think what I've been trying to figure out is whether the OpenAI hugging face thing starts a really interesting, broader conversation about AI safety? Or if there's something about that particular thing that is gonna end up being really important here? Like, do you think this comes back on OpenAI in any kind of important way? That it is the sort of instigating factor for this scary thing that everybody's talking about?

• 49:48Speaker 03

  I think the idea that these companies need really intense, both safety guardrails and research guardrails, and there needs to be a punishment if you screw up, will come back on OpenAI. This will be the poster child for that thing. There was reporting last week that OpenAI hadn't even noticed this for several weeks.

• 50:04Speaker 02

  Yeah.

• 50:04Speaker 03

  And then there's further reporting this week that, you know, there's safeguards in the testing environment were insufficient to begin with, which is how the model got out. Like, it wasn't supposed to get out.

• 50:11Speaker 02

  If you built a sandbox that something can use to access the internet, you built a bad sandbox. Like, it's, it's very simple. Build better sandboxes.

• 50:19Speaker 03

  I mean, that's straightforward. Like, you can just be like, you were negligent, now you have to pay money. And I, I, I think you can apply a bunch of existing laws and frameworks to this situation in that way. But this will be the poster child of some subcommittee hearing where we draft new regulation to punish out of control AIs and pin the fault on the company that makes the AI. And there's gonna be a lot of arguments on that, because the AIs are not deterministic and the gargoyles can fail and they can do all kinds of stuff that you wouldn't predict. But someone has to be at fault and it is almost certainly an after the companies that make and deploy the AI systems.

• 50:55Speaker 03

  And if, we can't, we can't agree on that. We can't get there. And this incident can't be the catalyst to do it. I'm not sure what worse thing has to actually happen.

• 51:05Speaker 02

  Yeah, I mean, this is, this is a ultimately sort of low stakes thing, which all of these, everyone involved has liked to remind everybody that this is fundamentally about a benchmark test and it had very little low stakes for what actually happened. But you, you change a couple of the nouns around in this story and you have actually an incredibly high stakes problem.

• 51:25Speaker 03

  Yeah.

• 51:25Speaker 02

  And there is nothing preventing it from being an equally high stakes disastrous problem. So I think you're right. And, and it'll, I think one of the reasons we've seen OpenAI and hugging face go out of their way to be like, we're taking this seriously, but it really wasn't that big a deal, is to avoid exactly the thing that you're describing. And I think it'll be interesting to see if it, if, if that gambit works or if this conversation comes for them in that way.

• 51:48Speaker 03

  Yeah, I just, I, I think if you sketch this out all the way, right? If, when I was a precocious eighth grader, if I had access to these tools, like, my school would not be safe.

• 52:00Speaker 02

  Sure.

• 52:01Speaker 03

  Right, like it wasn't safe when we were running on Emacs, those little like molar shaped Macs that were running what, OST 8? Like.

• 52:10Speaker 02

  Good old days.

• 52:11Speaker 03

  Sure. Like, I caused a lot of problems in that school.

• 52:17Speaker 03

  The big trick was to install the keylogger, but change the extension icon so it looked like a normal extension.

• 52:24Speaker 02

  That's pretty good, actually. I like that.

• 52:25Speaker 03

  Don't, I should be in jail. Don't be like me, kids. Like, what are we gonna do? Are we gonna have a bunch of kids trying to change their grades, burning tokens in their bedrooms at home while the school IT administrator is burning tokens on the other side, although accomplish nothing? Like, this can't be the outcome. I mean, it's a great outcome if you're at a AI company, because everyone's just burning tokens to accomplish nothing. But that can't be the endpoint of computer security, is everyone is constantly using compute to create and defend attacks. And it's just like where we're headed, inextricably. And I'm saying this because I would have done it.

• 52:59Speaker 02

  Burn the power grid.

• 53:00Speaker 03

  Without question, I would have done it. Yeah.

• 53:03Speaker 02

  Yeah. Yeah. It's, listen, the world is more filled with Neelai Patel's than anyone's willing to admit. You know what I mean? It's a real problem we're all having. All right, we should get to the lighting round. But first, we're gonna take a break and then it's time to talk about our man Brendan Carr.

• 53:19Speaker 03

  He did it again.

• 53:20Speaker 02

  We'll be right back.

• 53:23Speaker 04

  This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required compatibility and availability varies 18 plus.

• 53:48Speaker 05

  If you've got an insurance question, you could talk to the butcher at your local grocery store. He'd probably talk about trimming the fat, but it'd be about your brisket, not your insurance policies. Or you could talk to your local GEICO agent. They offer personalized assistance in finding the choicest cuts of coverage for all your insurance needs, which means more money for filet mignon, or, if you're a vegetarian, tofu le mignon. To find a GEICO agent near you, visit GEICO.com slash local.

• 54:18Speaker 06

  USAA knows dynamic duos can save the day, like superheroes and sidekicks, or auto and home insurance. With USAA, you can bundle your auto and home and save up to 10%. Tap the banner to learn more and get a quote at USAA.com slash bundle. Restrictions apply.

• 54:37Speaker 02

  All right, we're back. It's time now for the lighting round, beginning with America's favorite podcast within a podcast. He did it again, folks. Brendan Carr is a dummy.

• 55:06Speaker 03

  That's good. That's good.

• 55:10Speaker 02

  He's just so dumb.

• 55:11Speaker 03

  Yeah.

• 55:12Speaker 02

  It's fabulous.

• 55:13Speaker 03

  That is right up my alley.

• 55:15Speaker 02

  Can I tell you what that makes me think of?

• 55:17Speaker 03

  What's that?

• 55:17Speaker 02

  This is very, like, early 2000s MTV, that starts playing and, like, Neelai, your game show host, comes bounding out onto the stage and, like, high-fiving everybody in the audience. And then he's like, "What's up, everybody? It's Neelai. Welcome back." That's that theme song. I love it.

• 55:34Speaker 03

  That's good. I caught a little, like, you know, there's some lounge punk in there. I'll take it. There's the cramps are in there somewhere. I'll take it.

• 55:43Speaker 02

  Yeah. That theme is by Ricky Hopper. Thank you, Ricky. It's good stuff.

• 55:46Speaker 03

  It is good stuff.

• 55:48Speaker 02

  All right, what do you do this week, Neelai?

• 55:49Speaker 03

  By the way, just to say it out loud, because it makes me laugh every time I say it, the best cramp song is called Bikini Girls with Machine Guns.

• 55:58Speaker 02

  That is very good.

• 55:59Speaker 03

  It's a classic of the genre. I don't know what else to say about it. All right, Brendan. Brendan is getting himself in real trouble across the board by actually following through with his threats to speech. So Brendan, you know, if you're keeping track of wins and losses, Brendan was winning because he would make these, like, proclamations that he was gonna punish speech, and then he would chill speech, which is the whole point of not allowing government speech regulations.

• 56:30Speaker 02

  Right.

• 56:31Speaker 03

  It doesn't matter if you do them, if you have government officials fulminating that they will punish you for speech, you will chill the speech. People just won't say things because they're afraid of being punished.

• 56:41Speaker 02

  Yeah, he wasn't winning, like, in court, but he was winning in sort of a mafia intimidation kind of way.

• 56:46Speaker 03

  He was doing the bad thing.

• 56:47Speaker 02

  Yeah.

• 56:48Speaker 03

  He was issuing the vague threats, and the speech was going away. And there's evidence of this, right? ABC wasn't booking congressional candidates on their shows because of the equal time rule that he was threatening, but he's actually doing it with Disney, right? With, he's putting up ABC stations for license review early, breaking with years and years of tradition, and he's saying they're not acting in the public interest, all the way to, you didn't run the president's speech in prime time. Obviously, every time the president's speech is in public interest, our nation's private broadcasters must carry this. That's not in public interest. This is a reason to pull your license. I feel like that is very dumb, and no judge in America is gonna fall for that, but ABC has to spend the money,

• 57:31Speaker 03

  lawyer up, go to court and do it. Now, here's the problem with actually doing the speech regulation. No one likes them.

• 57:41Speaker 03

  No one likes them. Do you know who hates speech regulations the most? Conservatives. When you body up against the First Amendment in this specific way, and you do government speech regulations, it's not just a bunch of unhinged liberals, it's everyone. And in particular, conservative groups are filing letters.

• 58:02Speaker 02

  Conservatives who, by the way, have spent basically the entirety of the internet era screaming about how they're being censored on the internet.

• 58:09Speaker 03

  Yeah.

• 58:09Speaker 02

  Like, a lot of feelings about free speech coming from just as far right as you could possibly go on the political spectrum.

• 58:15Speaker 03

  So a bunch of conservative groups, including the Center for Individual Freedom, Citizens Against Government Waste, and Americans for Tax Reform, wrote a letter to Carr last week urging the chairman to avoid setting a precedent that could hurt conservatives in the future, because if you start allowing government speech regulations, the other side will use them, which is a thing that everybody has known since the beginning.

• 58:36Speaker 02

  Right.

• 58:37Speaker 03

  Right? If you give the government new powers, it won't just be your team that uses them. Inevitably, the other team will come into power, and then the other team will use them to punish you, which is a great reason to not have government speech regulations, and in fact, the reason why it is the First Amendment to the United States. The literal first one. Congress shall make no law respecting free speech.

• 58:58Speaker 02

  It's also a good, it's a real, like, mutually assured destruction thing, that it's like, if you go too stupid, eventually you'll lose an election and they'll do the stupid right back at you.

• 59:07Speaker 03

  Yep. So here is the letter, which was published by Fox News Digital. Always a good time here on Brendan Carr is a Dummy when Neelai Patel and Rupert Murdoch are on the same team. But here's the letter. First, they start by sucking up to Brendan. "For media and broadband to the space economy, your deregulatory record is helping unleash investment and growth, and we are grateful for your leadership. Getting government out of the way, it's in that same spirit. We respectfully urge the FCC to preserve the well-established routine predictable and presumptive nature of broadcast license renewal. We ask the FCC avoid steps that would transform renewal from the largely ministerial process Congress designed into a discretionary checkpoint that would invite the politicization that conservatives, free market advocates, and others have long condemned. Our concern is

• 59:50Speaker 03

  that establishing precedent non-routine renewal creates a powerful lever that future FCCs, including ones hostile to conservatives, religious or otherwise if they are broadcasts, will be all too willing to pull." This is the argument. You start doing government speech regulations, someone else will start doing government speech regulations. This comes on the heels of a bunch of former FCC officials represented by two lawyers on wildly different sides of the spectrum, calling the early broadcast license renewal process an assault on free speech, and also say it's a broader pattern in a campaign to punish speech disfavored by the Trump administration. This is over a dozen bipartisan former FCC officials, again, represented on, just, the wild bipartisan nature of thinking Brendan

• 1:00:32Speaker 03

  Carr is a dummy cannot be overstated. We all agree on a few things here in America. One, the truck should ever get ever bigger and jump over more things. Two, nobody likes data center. And three, Brendan Carr is a dummy. Like, it's just bipartisan, that we should not have speech regulation in this country. And Brendan, I think, he has gone from fulminating and sort of winning by saying he'll punish you, but the rubber's gonna hit the road here. And when it does, when this goes to court, I think you're gonna see a lot of judges be like, no, this straight up looks like you're punishing speech because the president told you to. And something will fall apart. Now, I can be wrong. I have

• 1:01:12Speaker 03

  no idea what the courts are gonna do. I have not deployed a super intelligent robot lawyer to fight these battles on our nation's behalf.

• 1:01:20Speaker 02

  Not yet.

• 1:01:21Speaker 03

  Who knows what's gonna happen? As I keep saying that the courts are not deterministic or predictable, but I do think I know that when this comes to a head, Disney is going to be a much more favorable position. You know, Mickey Mouse is gonna say, "The government is trampling my free speech." And that's gonna come out faster and farther than Brendan saying, "You must carry the president's speeches, or I'll punish you."

• 1:01:50Speaker 02

  So, okay, my, my only pushback on this, it would be that Brendan seems in many cases to be very happy on an island by himself, as long as President Trump is a fan. That's like,

• 1:02:04Speaker 03

  Sure.

• 1:02:05Speaker 02

  Through the whole Jimmy Kimmel thing, he's basically like, I don't care what you guys think. Trump doesn't think Kimmel's funny, so I'm gonna keep picking this fight. Uh, but I guess what you're saying is maybe he is going to run this stupid fight one tick too far, and he's actually gonna get punched in the teeth for it.

• 1:02:21Speaker 03

  It depends on what you think the goal is. And if your goal is to just make people stop saying things you don't like, he was winning. I, I really think that his war on, on free speech in America was, was notching wins. People were not, the speech was chilled. People were not saying things. They, they were afraid to talk. People criticizing Charlie Kirk were getting fired. This is all of peace, right?

• 1:02:44Speaker 02

  Yep.

• 1:02:45Speaker 03

  Okay, if that's the goal, then he, he should know he's winning and he should stop. If your goal is to test the ultimate power of the executive branch of this country, and to push it to the max, you have to keep going. Cause you need someone to say, you actually have the power to punish speech. You actually have the power to yank broadcast licenses. And if you actually have the power to yank broadcast licenses, you can take that next step and say, well, cell phones are on the internet on public spectrum. We can regulate cell phones. Like, you can get there. I know that that is Carr's goal, because he wrote the chapter on telecom regulation in Project 2025, and he just said it out loud. So, if

• 1:03:26Speaker 03

  the goal was just to chill the speech, he should have stopped. If the goal was to maximize power in the executive branch and actually get authority to regulate content on the internet, well, you gotta keep going.

• 1:03:39Speaker 02

  Yeah.

• 1:03:39Speaker 03

  And I do think that's the goal. And it's, it's of a pattern with the entire Trump administration, right? It's, I, I, I keep, they're, they're all just like in the library at Hogwarts, like finding ancient spells and seeing if they work, you know? And that's, that is the approach to the latent power of the administration. Like, does this statue let me do tariffs? No? What about this one? And it like, we're just gonna see what happens. And that's very much Carr's approach to telecom regulation, and particularly speech. Anyway, Brendan, if you wanna come on the show, or Decoder, or anywhere, I'll read you Project 2025. It's available. And you can tell me if you're doing what it says. You wrote it, so you should know. Uh, but I, I welcome the conversation, as long as speech in this country is still legal. That's been Brendan Carr as a dummy, America's favorite podcast within a podcast.

• 1:04:33Speaker 02

  It's beautiful. See, it was like, it was, it was a, we, an upper in and a downer out. Now we're level again. It's like, it's like it never happened. This is great. Um, okay, I have three lightning round items for you today, all of which relate to what I would call deep video silliness. Uh, the first one is largely just a shout out to Jess Weatherbed on our team, who wrote my favorite headline that I've seen on The Verge in quite some time, which is, "Facebook Considers Giving Up and Becoming TikTok." It's just, it's, it's perfect. Um, this comes from a blog post that Tom Allison, who is the head of Facebook, wrote that includes this sentence. "So later this year, we'll begin testing a reimagined experience that puts a subset of people who we think want

• 1:05:15Speaker 02

  more video on Facebook into full screen video the moment they open the app." Facebook has realized, Facebook is awful. And at the same time, has realized, people watch reels. Do people like reels? Debatable. People watch them. They sure, they sure super do.

• 1:05:34Speaker 03

  You can accident your way into scrolling through reels in a completely arbitrary and frankly insulting way, like, all the time.

• 1:05:43Speaker 02

  Yep. Also, on Meta's earnings, they talked about how Instagram engagement continues to climb really, really fast. That is all about reels. Like, I cannot overstate the extent to which reels is the primary product of Meta at this point, in terms of like, actual user-facing product. It is basically a reels company at this point. So, in that sense, it sort of leads Facebook to say, well, Marketplace is over here, we're doing weird stuff with groups. Nobody's on the newsfeed anymore, except for weird AI slop and insane political nonsense. Anyway, what if we just do reels? It's just the reels company, we're just putting it forward, literally giving up and becoming TikTok. Like, I wrote, I

• 1:06:26Speaker 02

  wrote a thing for the site last week about how, uh, vertical video has, has won in sort of unprecedented ways, which is going to be a theme of my next lightning round item, spoiler alert. But this would be, I think, the true final nail in that coffin, right? Like, we, we've talked a lot about, is, is text-based sort of town square social media over? This would be one big point in the column of, yeah, it's over. It's just video now.

• 1:06:51Speaker 03

  I, wait, I can, I can argue against that in one way.

• 1:06:55Speaker 02

  Okay.

• 1:06:55Speaker 03

  The best part of social video platforms is the comments, which are often not videos, right? Like,

• 1:07:02Speaker 02

  Agreed.

• 1:07:02Speaker 03

  All we're really talking about is the, the era of the, the prime unit of social being text might be at an end. But the part where there's massive amounts of comments, and everyone is still participating in text, that is not dwindled. It's only gone up.

• 1:07:19Speaker 02

  Yeah. I mean, yeah, "social networking," quote unquote, is now videos with comment sections.

• 1:07:25Speaker 03

  Yes.

• 1:07:26Speaker 02

  That's, that is what it, that is what it has been. Like, in, in every real way, that is what it has already become. And I think, to an extent, Facebook making this change would be late, almost, like a recognition of what the world actually is, rather than holding on to this idea of what Facebook was supposed to once be. But do you remember not that long ago, when Mark Zuckerberg was like, "We need to bring friends and family back to Facebook. We're gonna do the OG Facebook again." Lies. We're doing video.

• 1:07:49Speaker 03

  I cannot wait for my local parents groups from the school to be video-first.

• 1:07:56Speaker 02

  It's coming, dude.

• 1:07:56Speaker 03

  It's gonna be great. Here's how I know this is a real problem for Meta. And forgive me, because I'm always looking at ads, because I'm always thinking about the money. But I'm always, like, the internet is just ads, right? The, the whole social internet is just a big advertising machine, and Meta is at the heart of that machine. The video products, and I think the ad products in the Blue app, the, the Facebook app, they're under such pressure, that they're putting crappy text ads over the videos in Facebook. You would never expect this, or accept this, on Instagram, or TikTok, or YouTube Shorts. But if you open one of those videos, you just see the chum box ad sitting over the bottom of it. That's just like, here's an ad for some other crap. Like, here's one vegetable you shouldn't

• 1:08:41Speaker 03

  eat. And it's like, something's wrong, right? When we're making this money, we've either over monetized the experience from driving people away, or the whole thing is not making enough money, and we're desperate for cash, so we've added chum box ads to the social videos, which are just gonna make people flee and go to Instagram.

• 1:08:59Speaker 02

  Yeah.

• 1:08:59Speaker 03

  And then the pressure to turn Facebook ever more into Instagram will just increase, and we're just gonna keep doing this. I don't know, man. Like, I, I, in many ways, Zuck's focus on the metaverse, and then now his AI glasses and building super lawyers, he's lost sight of his products.

• 1:09:16Speaker 02

  Yeah, oh yeah.

• 1:09:17Speaker 03

  Like, in huge and important ways. And I, you just look at that chum box ad, and you're like, oh, this is not good. Like, something is in decline here. These, these ads are a symptom of decline.

• 1:09:28Speaker 02

  Yep. Agreed. Um, all right, what's your next one?

• 1:09:32Speaker 03

  Sadly for you, I have a Ramageddon update. It just keeps coming, brother.

• 1:09:35Speaker 02

  What can't I afford now, Eli?

• 1:09:37Speaker 03

  Uh, Qual, I was just gonna read you the list.

• 1:09:40Speaker 02

  Okay.

• 1:09:41Speaker 03

  Qualcomm is gonna raise its phone chip prices starting September 1. The rumor is that they will go up by double-digit percentages, and these are significant.

• 1:09:49Speaker 02

  That makes everything more, like, immediately everything more expensive.

• 1:09:52Speaker 03

  Everything's gonna get more expensive.

• 1:09:53Speaker 02

  Yeesh.

• 1:09:54Speaker 03

  Samsung's smartphone business posted its first-ever loss, but its RAM business saw profits soar 1,300%.

• 1:10:02Speaker 02

  Good lord.

• 1:10:04Speaker 03

  Uh, and Samsung is finding ways to make more RAM for Apple, so Apple doesn't go elsewhere. The Framework Laptop 13 Pro, that price went up by like $800, right? We already talked about this, and the configurations are all crazy.

• 1:10:18Speaker 02

  I was so annoyed, by the way. Antonio de Benedetto put up his review this week. I really desperately hoped that thing was gonna be like, pretty mediocre. But it's like, oh, okay, this is actually not the Windows laptop I've been waiting for forever. Turns out it is exactly the Windows laptop we've all been waiting for forever, except it costs $1 million.

• 1:10:36Speaker 03

  Yeah.

• 1:10:36Speaker 02

  It's awful.

• 1:10:37Speaker 03

  And the comments in that piece are so mad. I think Sean published a, basically just like a, a cry for help is a headline. It's like, I love this laptop, but it's already too expensive because of the stupid RAM situation.

• 1:10:49Speaker 02

  Yep.

• 1:10:50Speaker 03

  And then Logitech, basically, is like, we're not gonna make a handheld because of Ramageddon. Like, we're done. We're out. And so, your prediction that we're gonna stop seeing entire classes of gadgets every day is coming true.

• 1:11:03Speaker 02

  Man, that Qualcomm thing is worrying. Not surprising, but worrying.

• 1:11:07Speaker 03

  Yeah.

• 1:11:08Speaker 02

  Because it's, it's like, if, if we're bleeding out of, uh, we can't just make this particular kind of memory, to actually, most other kinds of component supply is going to be constrained, as everyone shifts all of their everything to building that kind of stuff that these data centers are looking for, uh, this could, this could cascade in some pretty ugly ways.

• 1:11:29Speaker 03

  Yeah. And I, again, I, we've had a lot of politics on this episode of The Verge Cast, but I'll just point out, I don't think- The AI companies are making a good enough case for what they will give to people to account for the cost. And some of the cost is you can't build a gaming PC. Some of the cost is logic is canceling products. Some of the cost is, I think, everything with a smartphone shipped from Qualcomm is about to get more expensive. And I think regular consumers can see all this stuff is getting more expensive. It's already a time of great anxiety. And it's getting expensive because of that data center over there on their community that they hate. And no one has made this connection in a way that allows them to go contend with it. We're still talking about drug discovery.

• 1:12:12Speaker 03

  But you're not talking about, well, you can't play video games anymore.

• 1:12:15Speaker 02

  Right. And what's, what's funny is I think people at a, at a mass level have made that connection you're describing.

• 1:12:22Speaker 03

  Yeah.

• 1:12:23Speaker 02

  Between my, like, no one is mad at Framework for the fact that its computer got more expensive.

• 1:12:28Speaker 03

  Yeah.

• 1:12:29Speaker 02

  No one. Everyone is mad at the AI companies for the fact that Framework got more expensive. And I think that's not like a perfectly accurate corollary, but it is very, it's close enough that everyone is pointing their anger in the correct directions. And especially when the, the juice is not worth the squeeze on the other side. And, and so I think like this, the things that these AI companies are being held responsible for, and the ways in which data centers in particular are becoming the totem of all of that anger, is only going to increase as every single one of these things keeps happening.

• 1:13:04Speaker 03

  Yeah.

• 1:13:04Speaker 02

  And, and I don't think the, I think you're right that the AI companies have not yet figured out how aggressively all of that is coming for them and how quickly it's growing.

• 1:13:13Speaker 03

  Yeah. I mean, these are just real consumer outcomes. You can't buy a computer. The iPhone will be $10 million. Okay. Why? Oh, because of the data center I hate?

• 1:13:25Speaker 02

  Yep.

• 1:13:25Speaker 03

  What's it gonna do for me? And the answer is like, have you thought about super intelligent lawyers battling it out? That doesn't seem cool.

• 1:13:35Speaker 02

  Uh, okay. My next one, you just immediately made me think, God, I would watch an HBO show of super intelligent lawyers battling it out. It's actually called Suits, by the way, and it's Neelai's favorite show on television.

• 1:13:45Speaker 03

  No.

• 1:13:48Speaker 02

  Anyway, my next one is, uh, it's, it's about vertical video, um, which is now coming to HBO Max. Because what is a streaming service without a vertical video feed at this point? Netflix has one, Paramount Plus has one, Disney Plus has one. They all call it clips.

• 1:14:06Speaker 03

  True.

• 1:14:07Speaker 02

  Which I'm fine with. HBO Max decided to call it Shorts, which is rude and unhelpful, and I do not appreciate it. But, um, in HBO Max, the idea is essentially that it's, it's a, it's a feed of clips and trailers and like behind the scenes content and stuff. And if you watch something that you like, you can press a button and immediately go actually watch that title on HBO Max. This, to me, is fascinating because just that description, that, that two sentences.

• 1:14:33Speaker 03

  I'm sorry. It's actually called HBO Max Shorts.

• 1:14:36Speaker 02

  It's HBO Max Shorts.

• 1:14:38Speaker 03

  Not just shorts.

• 1:14:39Speaker 02

  Max Shorts.

• 1:14:39Speaker 03

  It's called HBO Max Shorts.

• 1:14:41Speaker 02

  Do you know what Max Shorts are called, Neelai? They're called pants. This is HBO Pants is what this is called.

• 1:14:49Speaker 03

  I didn't realize it was the formal name was HBO Max Shorts.

• 1:14:52Speaker 02

  It's, it's all very upsetting.

• 1:14:54Speaker 03

  It's, that's very good.

• 1:14:55Speaker 02

  There was a time, not that long ago, one name change ago, this product would have just been called Max Shorts.

• 1:15:00Speaker 03

  Yeah.

• 1:15:01Speaker 02

  This is, this is, we almost got Max Shorts.

• 1:15:03Speaker 03

  It's good. It's good stuff.

• 1:15:04Speaker 02

  Almost. But anyway, so HBO Pants, uh, this is, this is a thing that in like an abstract way makes a lot of sense, right? You're like, okay, we're gonna give you a, a feed of all the stuff you're watching on all these social platforms anyway. We're gonna give you all the clips, we're gonna give you all the trailers. And if you are intrigued, we have solved the discovery problem because you can press a button and go actually watch the thing that we're showing you the clip from. Neelai can go find the episode of Suits that this clip comes from. Everybody wins. My question for you, Neelai, is have you ever once, one time in your whole life, engaged with one of these vertical feeds in one of these streaming apps?

• 1:15:39Speaker 03

  No.

• 1:15:40Speaker 02

  Me neither. I like, this is the thing.

• 1:15:43Speaker 03

  But I might.

• 1:15:43Speaker 02

  Maybe there is some huge behavior that I'm not aware of. But my, my interest in watching clips from Netflix shows is so much larger than the number of times I'm willing to go to Netflix and then watch clips of Netflix shows. That is just, when I go to Netflix, I would like to watch Netflix. Do you know what I mean?

• 1:16:00Speaker 03

  No, wait, I can, I, no, I disagree. All of these platforms have massive discovery problems, right?

• 1:16:07Speaker 02

  Totally.

• 1:16:07Speaker 03

  And the fact that people are watching the clips on TikTok, like, I think you and I have both at this point, watched all of A Few Good Men on TikTok in clips.

• 1:16:14Speaker 02

  Many times.

• 1:16:15Speaker 03

  I don't know why that keeps happening to me. At this point, I'm, I'm basically halfway through a full Game of Thrones rewatch on TikTok. Why is that happening to me? And I am of the opinion that HBO Max is pirating its own content onto TikTok to promote what amounts to a pretty lackluster season of House of the Dragons. They're like, would you just like to watch regular Game of Thrones? Like, here it is. Here's random scenes from that. Who knows how this, this promotional cycle works?

• 1:16:38Speaker 02

  This is just the official version of that.

• 1:16:40Speaker 03

  Yeah.

• 1:16:40Speaker 02

  Like, it, and it does, it makes, it makes intellectual sense.

• 1:16:43Speaker 03

  Can I read you their description of how this works?

• 1:16:46Speaker 02

  Oh, God, sure.

• 1:16:48Speaker 03

  It's so bananas.

• 1:16:50Speaker 02

  Okay.

• 1:16:51Speaker 03

  The clips for this feature are powered by an AI tool built in-house, which helps human editors select and verticalize content by using machine learning algorithms that utilize scene-level metadata to comb through thousands of hours of titles and suggest the most compelling, high-impact scenes for discovery.

• 1:17:07Speaker 02

  You already lost me at verticalize, but keep going.

• 1:17:09Speaker 03

  Scene-level metadata. What is scene-level metadata that, that tells you if something is the most compelling and high-impact scene? Is it literally metadata that's like, this scene rules?

• 1:17:20Speaker 02

  Yes. Is there, is there fighting?

• 1:17:22Speaker 03

  Robots in this one.

• 1:17:24Speaker 02

  Yeah.

• 1:17:27Speaker 03

  What?

• 1:17:28Speaker 02

  This scene has two swords. This scene has five swords.

• 1:17:31Speaker 03

  But like, you could do that with an AI, right? Like, is it just the captions? Is it just like, literally, like, it says Tom Cruise? Tom Cruise talking now.

• 1:17:43Speaker 02

  It's measuring the decibels of Tom Cruise's voice. The louder it gets, the more they're like, this one.

• 1:17:48Speaker 03

  It's scene-level metadata. I was like, what is that? All right. So they're gonna, they're gonna have machine learning algorithms utilize scene-level metadata to suggest the most compelling, high-impact scenes for discovery. Once the tool suggested those clips, HBO Max editors will decide which best represent the story's viewer love and capture jaw-dropping moments, pivotal plot twists, and visually stunning sequences. So we can't just trust the AI. The AI is gonna use scene-level metadata to make a list of the stories viewers love and jaw-dropping moments. And then humans will be like, the robot's got it right or wrong.

• 1:18:29Speaker 03

  And then those clips will then be rendered in a new vertical video experience to ensure seamless playback.

• 1:18:37Speaker 02

  Uh huh. Okay.

• 1:18:39Speaker 03

  This is the most overcomplicated thing I've ever heard. I'm like, are you guys watching what's working on TikTok? They're just cutting the whole movie into two-minute increments. Like.

• 1:18:51Speaker 02

  And then showing you a random one, and then you're fine. It turns out it works.

• 1:18:56Speaker 03

  Why are you doing this in the hardest way possible?

• 1:18:58Speaker 02

  Seriously.

• 1:18:59Speaker 03

  If you work, uh, at, at WD Disco, and you can tell us what scene-level metadata the machine learning algorithms are using, I would love to know.

• 1:19:10Speaker 02

  I also, I'm just now reading this as you're reading it. It, it says, yeah, these clips are then rendered in a new vertical video experience to ensure seamless playback. What? A new vertical video experience is just vertical and seem, like, we've been doing seamless playback for some time now.

• 1:19:27Speaker 03

  Is this not a video streaming application? It's very, it's, it's very good. The whole thing is like, it's, it's AI slop in like a different way. Like, they needed to dress up. We're gonna cut the movies into thousands of little pieces and stream them like TikTok in our app. Like, that's all it needed to say. But we had to back our way into a human AI hybrid of scene-level metadata. Again, if you can tell me what it is, I'm sure there's a reasonable explanation that isn't just, we're reading the captions.

• 1:19:55Speaker 02

  No, I bet it's what you think. I bet it's, it's like, this is where everyone is yelling.

• 1:19:59Speaker 03

  Yeah.

• 1:20:00Speaker 02

  We'll do that. Also, these companies should have something like perfect data about what people watch and when they tune in and when they tune out. And they're all like obsessed with how data-driven they are. This seems like information they should all already have about this.

• 1:20:13Speaker 03

  Maybe. Or you could just go on TikTok and be like, what parts of our movies are stolen the most? And just do that.

• 1:20:18Speaker 02

  Also very true. Uh, all right. What's your next one?

• 1:20:21Speaker 03

  I gotta go down a round of sap. Some good news for once, David.

• 1:20:24Speaker 02

  Okay.

• 1:20:24Speaker 03

  The Ferrari Luce has reportedly already hit its sales target. People love it.

• 1:20:29Speaker 02

  Okay.

• 1:20:29Speaker 03

  I saw a video of one driving, and I was like, god damn, that is the ugliest car in the world.

• 1:20:33Speaker 02

  I have a really important follow-up question.

• 1:20:35Speaker 03

  Yeah.

• 1:20:35Speaker 02

  What was the Ferrari Luce's sales target? Was it two?

• 1:20:39Speaker 03

  It was, uh, just under 500. Which, for, yeah, for Ferrari, it's a number.

• 1:20:44Speaker 02

  I was just about to say, actually, is that a lot or a little for Ferrari?

• 1:20:47Speaker 03

  It's, it's not a lot, but it's also a little, like, you know, the game with Ferrari, there's like a reason crypto bros all have Lambos and not Ferraris. Like, it's very hard to buy a Ferrari. And in order to get a good Ferrari, you often have to buy bad Ferraris.

• 1:21:03Speaker 02

  Right. You can't just go to the Ferrari dealership and buy a Ferrari.

• 1:21:06Speaker 03

  Yeah. They don't like that.

• 1:21:07Speaker 02

  You have to like know a guy who knows a guy.

• 1:21:09Speaker 03

  I mean, you can buy bad Ferraris, but nobody wants the bad Ferraris.

• 1:21:13Speaker 02

  Okay.

• 1:21:13Speaker 03

  So to get an allocation of like a 12-cylinder or whatever, you have to like buy some, like, five bad Ferraris. And I think the Luce, particularly in China, is like, yeah, there's like rich people in China are like, yeah, whatever, like, give me, give me one of those things. I'll, I'll keep it, but I want a good one, too. And I, Ferrari insists that they're not playing this allocation game with the Luce, but like, every car YouTuber is like, of course they are. We'll see.

• 1:21:36Speaker 02

  I love that the, isn't this the most expensive Ferrari, too?

• 1:21:39Speaker 03

  Yes.

• 1:21:40Speaker 02

  They're like, okay, so you're like, I'm, I'm a, I'm a super rich guy. I'm like, I'm gonna buy a Luce, and then I'm just gonna sort of push it off a boat. And then you're gonna allow me to buy the much cheaper Ferrari I actually want.

• 1:21:52Speaker 03

  Yes.

• 1:21:53Speaker 02

  Love this.

• 1:21:53Speaker 03

  Yes.

• 1:21:54Speaker 02

  If I'm Ferrari, I'm like, hey, just write me a check for a million dollars and everything will be fine.

• 1:21:58Speaker 03

  No, you, again, they, they, they don't want your money. They, what they saw was exclusivity in real, whereas like Lamborghini is like, what, what, what, how many do you want? Do you want lime green? This one's lime green, like, whatever it is. Um, uh, Ferrari is, Financial Times says Ferrari is trying to reach a new generation of wealthy buyers, particularly in China and Silicon Valley. And apparently, US tech entrepreneurs have had positive feedback on the Luce. So maybe it's working. Maybe we're all wrong.

• 1:22:25Speaker 02

  You can't buy taste is a phrase that keeps coming into my head as we have this conversation.

• 1:22:32Speaker 03

  If you see one on the road, like, or just a video of it rolling, you're like, what is going on with this car? It just does not look good. I truly am as befuddled by this car as any of the cars that we often talk about that we are befuddled by.

• 1:22:48Speaker 02

  Yep. And Ferrari's also had a really weird rollout of this where they keep giving demos to car YouTubers and influencers and stuff. But there are all these weird on rails canned thing that sort of make it look like the creators are just being put on a like roller coaster made by Ferrari. It's very weird. The whole, this whole thing does not suggest that Ferrari thinks its car is awesome and everyone should have one. Do you know what I mean? But I guess congratulations to the 500 people who spent a bunch of money on them.

• 1:23:18Speaker 03

  Yeah, they're doing great. They hit the target.

• 1:23:19Speaker 02

  I hope you, I hope you enjoy it. Are you gonna, was one of them you? Be honest.

• 1:23:23Speaker 03

  No, no, no, no, I, no, no.

• 1:23:26Speaker 02

  You didn't, you didn't sell the Raptor and get this instead?

• 1:23:28Speaker 03

  I mean, I'm up here, which is where I had the Raptor and I think about that truck every day, David. But that's the one that got away.

• 1:23:36Speaker 02

  All right, we should get out of here. We, Neelai has more vacation to take, more books to write, more, more Luches to save up for. When, when it happens, you need to preorder the books so that Neelai can get a Luché. That's our new, our new goal as a group.

• 1:23:49Speaker 03

  It's like I'm putting that's a negative, negative reason to finish this book is like I have to buy a Luché at the end of it.

• 1:23:56Speaker 02

  Yeah, it's good stuff. Who's on Decoder next week, Neelai?

• 1:23:58Speaker 03

  We have Tony Schneider, the new CEO of Blue Sky.

• 1:24:01Speaker 02

  Oh, the like official now.

• 1:24:03Speaker 03

  Official, he was the interim, he's the new CEO. We got into it, that was a fun one.

• 1:24:07Speaker 02

  That's good. I think we might do something Vergecast-y with that next week too, but I have, you were excited about that one coming out of it, if I remember correctly.

• 1:24:15Speaker 03

  Yeah, they've got, again, yeah, I'm enthusiastic about open social media and they've got new stuff coming out in the protocol. He's aggressive. I liked it.

• 1:24:24Speaker 02

  Love it. We're between seasons on Virgin History, but all of the smart home stuff is up. And we're also in the middle of making very fun plans for next season, which I'm very excited to do after I get back from parental leave. This is some, so I'm gonna go have a baby and then we're gonna talk gadgets again. I'm very excited about it. That is it for the Vergecast. Remember, the best thing you can do to support all of this is to subscribe to the Verge, verge.com/subscribe. It gets you all of our podcasts, including Decoder and Virgin History and the Vergecast with no ads. It gets you all of our newsletters, all of our coverage of everything. Go read all of the stuff Robert Hart has been writing about China and the US and AI. He's very clear-headed and it's making all of our comments just really angry and I love it very much.

• 1:25:05Speaker 02

  Also, if you have thoughts, feedback, questions about any of this, you can send us an email, [email protected], call the hotline, it's 660011. As always, we genuinely, truly love hearing from you. The Vergecast is a Verge production and part of the Vox Media Podcast Network. This show is produced by Josh Kahas, Eric Gomez, Brandon Kiefer, and Travis Larchuk. We will see you next week. Neelai.

• 1:25:24Speaker 03

  Rockin' well.

• 1:25:28Speaker 01

  Interrupted sleep, headaches, constant fatigue. For a lot of women, these aren't three separate issues. In perimenopause and menopause, they're often the same hormonal story and you don't have to quietly push through. Midi can help because your symptoms have answers. Visit joinmidi.com with code VOX right now to book your first visit today. That's joinmidi.com, code VOX to book your first visit. Joinmidi.com, code VOX. Insurance coverage varies. Check with your plan for coverage.

• 1:26:05Speaker 07

  Your heart can tell you a lot about your health. Apple Watch Series 12 measures your heart rate every five seconds with the most accurate heart rate sensing in wearable. So your vitals app, now with heart rate variability, can tell you when something is off. And your readiness score can let you know when to rest and when to push. Hear the story in every heartbeat with Apple Watch Series 12. The features described are for wellness purposes only and not for medical use. iPhone 11 or later required. Based on Apple conducted study of heart rate accuracy, August 2026. Visit apple.com slash apple watch series 12.

• 1:26:35Speaker 08

  Everyone's got a hard question about AI. Your job, your kids, where it's going. Anthropic was built to surface those questions and share what it finds along the way because there's hope in hard questions. Ask yours at claud.ai/spotify and keep thinking. That's claud.ai/spotify.