Mark Zuckerberg has an Instagzam

Have you seen Instagram's new logo? Meta debuted it right before we started recording, so of course David and Nilay needed to talk about it. Then, the hosts turn to Mark Zuckerberg's 6,500-word missive on the future of AI, and the problem with his idea that giving AI to everyone will fix everything. After that, it's time for Claude watermarks, Spotify AI personas, Brendan Carr, Flock, and much more. Further reading: ⁠This is Instagram’s new logo | The Verge⁠ ⁠The Pepsi Universe PDF ⁠The Future is for Everyone⁠ ⁠Four takeaways from Mark Zuckerberg’s massive AI manifesto ⁠ ⁠Mark Zuckerberg doesn’t understand how to live ⁠ ⁠https://www.meta.com/design-at-meta/blog/⁠ ⁠Claude will apply invisible watermarks to AI text and images ⁠ ⁠‘If it was opt-in, nobody would opt in.’ ⁠ ⁠Spotify says it won’t recommend music from ‘AI Personas’ ⁠ ⁠Apple could help you prove your iPhone photos aren’t deepfakes | The Verge⁠ ⁠Guitar company D’Addario admits that AI music was used in a promotional video | The Verge⁠ ⁠Fender’s CEO seems to think your bandmates are just analog AI | The Verge⁠ ⁠Suno Unveils Download Song Caps for Free, Paid Tiers⁠ ⁠Brendan Carr is one firing away from an unchecked FCC⁠ ⁠YouTube is making it harder to earn money on YouTube ⁠ ⁠Flock CEO: ‘We got this one wrong’ ⁠ ⁠The first rival Android app store just arrived in the US Play Store ⁠ ⁠ChatGPT and Gemini both just passed 1 billion users ⁠ ⁠The sub-$100 phone market is disappearing. ⁠ 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. 0:00 Welcome and Parental Leave Announcement 00:43 Instagram Redesigns Their Logo 01:04 Why Companies Change Logos 03:37 Meta Design Blog Quotes 07:22 The 2008 Pepsi Universe Presentation 09:41 Blanding 10:37 Zuck’s AI Manifesto 13:29 Collective Action Problems 18:49 Superintelligence for All 20:57 Cybersecurity and AI Risks 22:22 China Chips and Distillation 27:48 Write the Law Challenge 32:30 Claude Text Watermarking 35:33 How The Watermark Works 39:34 Limits And Evasion Risks 41:54 Apple Marks Real Photos 46:03 Provenance Standards Fail 46:58 AI Music Backlash 49:39 Fender CEO AI Scandal 53:08 Where AI Fits In Music 59:13 Twitch Opt Out Training 01:01:35 The Lightning Round 01:02:55 Brendan Carr is a Dummy 01:04:55 Bans Don’t Work 01:05:56 Made in America Assembly 01:10:19 Gemini Claims a Billion Users 01:15:40 YouTube Partner Paywall 01:20:56 RAMageddon Cheap Phones 01:23:52 Flock Camera Backlash 01:27:19 Goodbyes and Plugs Learn more about your ad choices. Visit podcastchoices.com/adchoices

• 0:00Speaker 00

  Company spend can be complicated. Managing it shouldn't be. Amex Corporate connects cards, expenses, and payments all in one place. That's less time managing work and more time doing work that matters. Built to simplify, designed to scale. This is Amex Corporate. Get started at go.amex slash Amex Corporate. Terms apply.

• 0:28Speaker 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.

• 1:05Speaker 00

  Growing teams need a corporate card built for more, something that doesn't sacrifice speed for control. Introducing the American Express Corporate Cash Back Card. Get 1.5% cash back on eligible business purchases, advanced spending controls, and real-time visibility, plus intelligent expense management. Get started at go.amex slash Amex Corporate. Terms and cap apply.

• 1:36Speaker 02

  Hello and welcome to The Vergecast, the flagship podcast of Instagasm, the new hot social network that launched this week. I'm your friend David Pierce. Neil Patel is here.

• 1:46Speaker 03

  What's up?

• 1:47Speaker 02

  This is just a hot start to The Vergecast for us. We're doing great today.

• 1:49Speaker 03

  It's real. It's either the best one or the roughest one we've ever had.

• 1:53Speaker 02

  So this, I'm frankly very excited that I'm here because I'm going on parental leave any minute now. Tomorrow is my last possible last day. I am just disappearing into the wilderness for a while. There has been so much wild crap on the internet this week that I was going to be very sad if I left before I had a chance to have a lot of feelings here on The Vergecast with you about AI.

• 2:15Speaker 03

  Yeah.

• 2:15Speaker 02

  But we're here. We're gonna do this. But we have to start with Instagram. Instagasm.

• 2:20Speaker 03

  We were putting together the rundown and trying to organize all these AI ideas. And then Adam Mosseri announced the new Instagram logo. And I was like, well, there's a full hour of our show. And the only reason that he gave for why there's a new Instagram logo was it's been about 10 years. It's time for a new one.

• 2:37Speaker 02

  Okay, I'm glad you started with this because this is a question I have actually for you in particular. I have had many jobs in media. I have never actually been at a job in the middle of a logo redesign. I left The Verge right before one redesign and came back kind of right after one. You have done a bunch of redesigns. And the only thing I can think of in these moments when you are at a company for a long time, you're doing one thing, is just maybe you get itchy. Like the people at Gap were like, I've been looking at this logo for too long. Let's change it. And they do that at Cracker Barrel and they do that at Jaguar. And like, it's mostly all bad. But there is, is there just something that happens when you're like, I've been staring at this logo for too long. I need to change it.

• 3:21Speaker 03

  There's a little bit of that. It's a great way to get attention.

• 3:24Speaker 02

  Sure, fair.

• 3:25Speaker 03

  And the attention can be good or bad. The CEO of Cracker Barrel got fired.

• 3:31Speaker 02

  That one goes in the bad column.

• 3:33Speaker 03

  Roll the dice, Adam. Let's see what happens next.

• 3:38Speaker 03

  Coke changed their logo and some of the design of, I think, the Coke Zero cans. And everyone's like, dude, that's just the Marlboro logo. Like, what are we doing here?

• 3:46Speaker 02

  This was like very recently.

• 3:47Speaker 03

  Very recently. That's all very funny. This one, I can't quite tell. Our redesigns, we've always done around format shifts. So the original Verge logo had a bunch of serifs in it. It was designed for like a desktop website. Then we went to mobile. Everything got small. The fine details were getting obliterated. So we made a change. And then we made another change recently with this last redesign because that old logo was really blocky. It was hard to lock up in all these places. We wanted to go all these other directions with our design and colors. And so there was some reason. There was some external force that was pushing you on the old logo can't do all the things you want to do with your design system in the future.

• 4:31Speaker 03

  There is also the general, you can't bring back the old logo in a flurry of happiness if you never take it away.

• 4:40Speaker 02

  For sure.

• 4:41Speaker 03

  This is just a thing, right?

• 4:43Speaker 02

  It can't be the classic logo until it's gone.

• 4:45Speaker 03

  Yeah. You can't, there's no, you can't have loud parts if you don't have quiet parts. And sometimes you just gotta take the old logo away. I don't know if any of that is happening for Instagram. Adam has talked about, you know, the Instagram app is gonna change. It's gonna be more flexible. It'll be more AI. But none of that appears today. There's no external force pushing Instagram to be different. It really does appear that they just got a little bored. They haven't said anything else about it.

• 5:11Speaker 02

  Yeah, there's a really sort of flowery post on the Design at Meta blog about this that I expected to be the place that is like, as we lean into new paradigms, like all this stuff you're talking about, none of that.

• 5:24Speaker 03

  I have to make this work on AI glasses.

• 5:26Speaker 02

  None of that. They just say it's been over 10 years since we refreshed our brand identity. That's the explanation. And then it goes into a bunch of flowery nonsense. The thing that is actually most striking about all of this to me, the logo is hideous, by the way. For anyone who has not yet seen it, I encourage you to really spend some time with it. Well, I'm sure you're seeing it here. We're gonna talk a lot about it.

• 5:49Speaker 03

  It says Instagasm.

• 5:50Speaker 02

  It says Instagasm.

• 5:54Speaker 02

  Shaquille O'Neal starring in Instagasm.

• 5:56Speaker 03

  And there's some part.

• 6:01Speaker 03

  That was, who's the other guy in the Mandela effect version of that?

• 6:05Speaker 02

  Oh, Sinbad.

• 6:07Speaker 03

  This is the Sinbad version of Instagram. This is the bad timeline.

• 6:12Speaker 02

  Exactly.

• 6:13Speaker 03

  Yeah. Go ahead. You react first.

• 6:16Speaker 02

  But the theory behind this is like, I keep looking for the reason that they did this, right? And I think even if you land on logos that not a lot of people are gonna like, everybody's gonna have issues with change, especially to something, I think Instagram's script qualifies as like a genuinely iconic wordmark. Like, I think, I think most people of the Instagram generation could like close their eyes and write the Instagram wordmark relatively successfully. It's that, that cursive lettering has been around for forever. It looks really great. I fundamentally disagree with the premise that it looks old, except that it's cursive. Like, fine, whatever. I don't know. But anyway, they go through, and I just pulled two quotes from this blog post that struck me

• 7:00Speaker 02

  as ridiculous, frankly. It says, "Our new wordmark elevates simplicity, creativity, and connection. It was important to craft a mark that still makes Instagram feel like Instagram while introducing something to carry us into the future." That sentence, word for word minus the proper nouns, is what everyone says every time they change their logo for any reason. "We want Coca-Cola to still feel like Coca-Cola while introducing something to carry us into the future." This is, this is the only reason anybody changes a logo is because they're like, what are we, is it the future? Drives me crazy. Anyway, here's a quote from Jasmine Probst, the director of product design at Instagram. I do not feel bad making fun of this quote because there is just absolutely no chance Jasmine Probst actually said this. It says, "Script reflects individuality,

• 7:45Speaker 02

  so referencing that felt right. The new wordmark is bolder and simpler with touches of unexpectedness. It cues individual expression, a personal voice, both of which have always been core to Instagram." What? Like, it's, we took boring sans serif font and made a couple of the letters bad. Individuality.

• 8:05Speaker 03

  We made this harder to read. I agree, the existing Instagram logo is iconic. It means a lot of things to a lot of people. When you change your logo, you change all those relationships. Everyone is always mad at you. And every designer in the world has now been trained to be like, that feedback doesn't matter. We know it's right in our hearts.

• 8:22Speaker 02

  Yes.

• 8:23Speaker 03

  And that's 100, and Mosseri put this on threads and was like, show me your hot takes. I, again, absent some reason to do this, which they have not said, this feels like an unforced error.

• 8:33Speaker 02

  Yes.

• 8:34Speaker 03

  And it honestly feels like this entire new Instagram logo is the human designers at Meta getting one over on Mark Zuckerberg before he replaces them with AI. Like, right before they get clipped. They're like, all right, we're gonna, we're gonna set up one time bomb and then you're gonna, your robots are gonna have to fix it, Mark. Muse Sparked is gonna fix it. You go ahead, buddy. You have it. The other thing I want to say, which is very important to me, is that it doesn't matter how good or bad your logo redesign is. It doesn't matter how good or bad your brand guidelines are, your design system is. Every single design presentation is fundamentally the Pepsi Universe from 2008. Do you remember, do you know about the Pepsi Universe from 2008?

• 9:17Speaker 02

  I think so, but I'm gonna Google this as you talk.

• 9:20Speaker 03

  There is a PDF that has just been floating around the internet for years, since 2008. And it is like a logo redesign and brand strategy document for Pepsi that costs millions upon millions of dollars. This is the rumor. And it is the most insane description of Pepsi with charts and graphs that you have ever seen in your life. And every time I see any other design document, oh my God, it's the Pepsi Universe.

• 9:46Speaker 02

  This is pretty spectacular. The title of this document is breathtaking, in all caps.

• 9:53Speaker 02

  The Pepsi DNA finds its origin in the dynamic of perimeter oscillations. We'll link to this in the show notes. It's fabulous and everyone, everyone should experience this document.

• 10:02Speaker 03

  So this is a page from the Pepsi Universe document called breathtaking. Creation of identity gravitational pull. And it says, attraction theory, the Pepsi proposition. And it begins with typical light path, which is a star shining on the earth. And then it goes on to show you the gravitational pull of the Pepsi logo, pulling the light away from earth. And that's, that's why you should have a new Pepsi logo.

• 10:27Speaker 02

  This is insane.

• 10:29Speaker 03

  It is incredible.

• 10:30Speaker 02

  I'm sorry, the bottom of this page, on the left side, it has relativity of space and time. And on the right, it has Pepsi propositions. What are we doing?

• 10:39Speaker 03

  When you're in the Pepsi aisle, the new logo will bend the light, will bend you towards it. I'm confident the Instagram people. We're, we're doing something different. This entire document is worth this scroll through. It's one of my favorite design documents in the entire world.

• 10:56Speaker 03

  But every time I see one of these logo redesigns, I think about the Pepsi Universe and how you have to just dress up any logo redesign with the most obtuse nonsense in the world. So props to Instagram for not doing that, for just saying generic stuff. No props for taking an iconic logo and making it say Instagasm.

• 11:14Speaker 02

  Yeah, it's not great. It also makes me think, was it the Airbnb redesign a while back? That made everybody think, oh, we're getting back to an old era of design. We're going to do a little more skeuomorphism, a little more physicality. We're entering into an augmented reality world. We need things that feel like objects again. I was sort of excited about that era of design and boy, is this not that. I think you could argue this is, I mean, and we've been on a run in the tech industry for like a decade now of just everybody taking interesting, colorful, cool logos and just dumbing them down to a sans serif font. Like this, every company has done this.

• 11:49Speaker 03

  It has a name. It's called blending.

• 11:52Speaker 02

  Is it really?

• 11:52Speaker 03

  Yeah.

• 11:53Speaker 02

  I like that. That is a good name.

• 11:54Speaker 03

  Very good.

• 11:54Speaker 02

  Uh, and it is a bummer that this continues apace, except in Instagram's case, they just ruined one of the letters and that's apparently designed. It's bad stuff.

• 12:02Speaker 03

  Have an Instagasm on Meta Platforms Inc.

• 12:06Speaker 02

  Don't give them that idea. I don't, you, this is not branding I'm ready for. Okay. The other weird thing that happened at Meta this week is the other thing I want to talk about, which is Mark Zuckerberg's 6,500 word manifesto about the future of AI. It's called, the future is for everyone. Have you read this?

• 12:23Speaker 03

  I have read this.

• 12:24Speaker 02

  What, I have a few specific things that I want you to react to. But first I do, I just want your overall thoughts on Mark's thoughts on AI.

• 12:32Speaker 03

  Liz wrote a piece on this, this was really good this week. And they were searching for a headline and my suggested headline was, Mark Zuckerberg should have finished Harvard.

• 12:42Speaker 03

  Cause if he had just completed his, the course of his studies in the humanities, he would have maybe realized that he was having like third year social level thoughts about the world. There's an argument that you gotta write the stuff to be dumb now, right? The kids can't read anymore. So you gotta dumb it way down to be like, vote's good, dictator bad. Like maybe that's, maybe he wrote it for an audience.

• 13:09Speaker 02

  China frowny face emoji.

• 13:10Speaker 03

  They can no longer read, but that's how it comes across. Yeah. Like it's a lot of big ideas that basically are like magic genies will change the structure of the world. What will we do now? Weird, that, that's weird just as a thesis. And then the, what should we do about it? The political philosophy underneath it is freshman year. And I, I think that's very strange. And I think it's symptomatic of the tech industry right now that they have not done the reading.

• 13:40Speaker 02

  Say more on that, cause this, this is actually the, the big picture thing. I think if you, if you were to boil Zuck's manifesto down to one big idea, it's that he believes that giving everyone access to the best possible AI is a good thing on its face, right? He talks about checks and balances. He talks about democratization. He has this idea that no one should be in control. And if everyone is in control, it will, it will be better because everyone will check against each other. Everybody gets to personalize things for themselves. And this is like, I also had this moment of like, as a political science major, you, you read a bunch of books about how that fails. And we have spent the last 20 years of Facebook watching how that fails.

• 14:22Speaker 02

  And it just like, it was so striking to me that this guy is still so convicted of the idea that if you just give everyone access to everything and to each other, it will, it will inevitably universally be a good thing. And I don't know if that's what you're thinking about as you say, I wish he'd done the rest of the reading. But that was my reaction to, God, I wish he had done the rest of the reading.

• 14:42Speaker 03

  The reason that we didn't go with Mark Zuckerberg shouldn't have finished Harvard as a headline was furious pushback from our team that it does not track that completing a course of education at Harvard makes you a good person. Like those aren't one to one.

• 14:56Speaker 02

  Sure.

• 14:57Speaker 03

  As, as we can see, you know, like by many elites in our society. But I share the frustration. I feel the same frustration that, that you're describing here, which is the entire history of like political philosophy in the West, in the East, like everywhere is, boy, there are these things called collective action problems, where if you act in your own interest and I act in my own interest, actually we will lower our collective results. And, and the collective action problems just exist across society. USBC is a collective action problem. Right? It took government regulation to make Apple put USBC on its phone

• 15:36Speaker 03

  because they could get more individual benefit to Apple by selling lightning connectors to, to you, the consumer. And actually we had to be like, no, you have to do whatever everyone else is doing. And that increases our collective benefit. There's like, they're everywhere you look in society, collective action problems. And mostly the past 20 years, particularly on the right, has been an absolute denial that collective action problems exist. Vaccines are a collective action problem. They only really work if everybody gets them, but I don't, I, I'm crazy and I'm just going to eat steak and sauerkraut instead. And whatever is going to happen is whatever is going to happen. This is a description of a political philosophy from Mark Zuckerberg that is in denial of collective action problems. It says, if everyone is maximally free, then we will all have the best outcomes.

• 16:21Speaker 03

  We talked about this, there's some parts of this manifesto that he first published in the Wall Street Journal. And his big, you know, his big example that always drives me crazy is if I have a super intelligent lawyer, I'll win every case, but if we both have super intelligent lawyers, justice will be more efficient. And none of that is true.

• 16:37Speaker 02

  No.

• 16:38Speaker 03

  Just flatly, that is not how lawyers work, especially extremely intelligent lawyers. Actually slow everything down. That, and Mark Zuckerberg's lawyers do that all the time. And so there's just a part of this where it is ignorant of the entire history of how do you organize and govern a society? Because the main thing you are trying to do when you organize and govern a society is get people to work in their collective interest to solve collective action problems.

• 17:05Speaker 02

  Right.

• 17:05Speaker 03

  And that, I realize as a tech podcast and that's deeply in the weeds. That is what all these guys are charging at with AI. Who should regulate AI is fundamentally a collective action problem. Right? If we all act in our interest, we will destroy the world. So someone has to make us act in our collective interest. Who will that be? And Mark's answer is, well, it should be someone, but I don't want it to be anyone. And I just, I can't square that at all.

• 17:31Speaker 02

  Yeah, he's, he's very clear, like Dario Amade, the CEO of Anthropic. His name is nowhere in this, but it is everywhere in this.

• 17:39Speaker 03

  Yeah.

• 17:40Speaker 02

  Because Dario has been out there very loudly saying essentially, we've got it, we'll take care of it. Don't worry, we can be trusted. No one else can be trusted, including the government. Uh, and that has made a lot of people feel a lot of feelings. Mark is very clearly arguing against Dario. And that whole theory of the case that the people who make the AI are the ones who should be able to control how it is regulated and, and used and the guardrails on which we put it. Mark sort of wants government intervention, but not to, like just the right amount. And he wants everybody to just kind of be cool. There's a lot of like, everybody should just be cool about this and it'll be fine in this letter. But I just wanna, I don't want to dwell on this for too long. I actually encourage people to read it.

• 18:23Speaker 02

  It's, it's relatively readable for all the reasons you said, Neelai.

• 18:26Speaker 03

  And it's, it's written for sixth graders.

• 18:27Speaker 02

  It's, it's written for sixth graders and there are chunks of it, I think, to be charitable to Mark Zuckerberg, that it immediately becomes very clear are not his area of expertise, like scientific discovery. No offense to Mark Zuckerberg. He, he, this does not come off like somebody who understands the process of scientific discovery. But there are things that he is deeply immersed in and I think knows very well. Um, and I just want to read a couple of them to you and, and see how you respond. He writes, your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies and more. It will free up time for the things you enjoy and help you accomplish more than you could otherwise. Right next to that, my daughter loves to bake, so my agent plans personalized recipes for us to make together each weekend, orders the ingredients and then offers suggestions as for baking. How does this make you feel, Neelai?

• 19:11Speaker 03

  I have mixed emotions about that one. I, you know, there's a lot, there's a lot of baking that happens in our house. And it's, we, it's not hard. Americans have been baking for years.

• 19:26Speaker 02

  Yeah.

• 19:26Speaker 03

  Without the assistance of data centers. It's, it's just like, not that hard. The idea that you're baking an infinite series of personalized recipes is interesting on its face. The idea that the ingredients will show up if you're on a busy household, interesting.

• 19:39Speaker 02

  Yep.

• 19:39Speaker 03

  None of this requires the level of investment. None of this justifies, we're going to build data centers without permission across the country. And that's the disconnect here.

• 19:49Speaker 02

  Yeah, you're, you're describing software, not the, the future of humanity. I will say, I find this much more palatable than the idea that Sam Altman tweeted a while ago about the personalized podcast that'll tell your kid about the day that they're going to have in the car on the way to school, which I find, I mean, truly like 10 out of 10 dystopian. That like, don't talk to your kid about the day, have it play their personalized podcast instead.

• 20:12Speaker 03

  Every parent responded to that with, no, the kid in the backseat of the car, who just starts talking to you is actually one of the great joys of parenting.

• 20:20Speaker 02

  It's the whole thing. Yeah. Okay. Anyway, next one. We talked about this a little bit already, but uh, here's how he leads into his lawyer example. People in institutions with competing interests naturally check and balance each other to lead toward positive outcomes. The best and most realistic path to building a positive AI future is by delivering super intelligence to everyone. Do you buy this theory?

• 20:40Speaker 03

  This is flatly wrong. And I, I don't mean to say it's flatly wrong that not everyone should have a, a magic genie that can do whatever they want, improve their relationships in some way. You go ahead and, and put your magic robot in everyone's relationship and let's see what happens. The idea that. Necessarily giving everyone the same capability will lead to checks and balances is wrong on its face. I know this because Instagram exists and Facebook exists and everyone has the same broadcast capability to millions upon millions of people on the social platforms that Mark himself runs. And his entire life is a series of moderation controversies,

• 21:21Speaker 03

  of CSAM scandals, of misinformation scandals. He had to set up a Supreme Court to deal with all the moderation scandals and he hated it because they keep telling him that everything is more complicated than he wants it to be. And he has essentially sidelined it because he'd rather be the dictator. And that's just that. That's just social media, which is an intractable problem no one can solve. If you give people offensive cyber capabilities, you're like, naturally, this will lead to checks and balances. You are completely off your rocker. Some people need more capabilities than others and some things should be flatly illegal. That's fine. Like, that's, this is what I mean about a sixth grade level. This is sixth grade philosophy. If everybody has a gun, there will be no more gun violence in America.

• 22:04Speaker 02

  I will tell you, I have thought a lot about the idea that the only way to stop a bad guy with a gun is a good guy with a gun in the context of the way a lot of people are talking about AI. And it is, I mean, that's Zuckerberg's entire argument. That's not even a disingenuous reading of the argument. That is the argument. That if you have AI, the only way to defeat you is if I have AI. Like, literally, okay, I'm just, I will, I will use Mark Zuckerberg's words to prove my point. He says, "Over time, I expect that widely deployed AI models with strong cybersecurity capabilities will lead to systems that are more secure, not less. This will be definitively true once superintelligence enables most of the world's code to be verifiably secure."

• 22:48Speaker 03

  Verifiably secure is doing an awful lot of work.

• 22:51Speaker 02

  A lot of work and also the tacit assumption that there will be no more holes for the, we're going to get AI that is so good that it can patch all the holes, but not AI that is good enough to find new ones. It's just, it just, that doesn't make any sense.

• 23:05Speaker 03

  It doesn't make any sense. It also doesn't account for social engineering, which is an actual exploit vector that many AI systems have been able to use, including the ones that have broken out of their sandboxes in various ways in recent weeks, including Meta's own AI, which you can believe whether it did or it didn't, or if Meta just felt left out. You know, like, everyone else's AI is breaking out of the sandbox, like ours did too. There's a little bit of that going on here. But the idea that it's just the code that is the vulnerability and not the systems or structures or human beings around the code that are the vulnerability, it doesn't even factor into the analysis here. And that makes no sense. It's just fundamentally no sense.

• 23:45Speaker 02

  Yep, totally agree. Okay, I'm just, I'm just gonna do two more. I have, I have 15 here. But I'm just gonna do two more. This one, I just chose for you specifically, Neelai. It says, "Export controls on silicon have been successful for slowing the progress of foreign labs during this critical period, so it is the right strategic move to continue those. Any policy that slows American model releases even by a month could add significant risk to American leadership while letting foreign models race ahead." China, bad, Neelai.

• 24:14Speaker 03

  It's funny, because we're gonna get to this later. Our government here in the United States is touting the fact that banning Chinese drones has resulted in a surge of drone manufacturing in the United States. And we will come to the truth of that later in the show, in America's favorite podcast, the podcast. Export controls have been successful so far, but it's not like China's gonna sit there and not invest in chip making and chip design, or steal our chip IP, or find any number of ways to make their own chips. And so, we should continue those. But anything that slows down America, creating these new offensive capabilities that might break containment and cause havoc all on their own, well, we

• 24:57Speaker 03

  can't have that. Because even though we know that China will react to us trying to slow them down, we must proceed. Like, there's just something confused in there where, again, it's a collective action problem. If what you actually had was cooperation between rivals to make sure that problem doesn't escalate out of control, you, you might have good outcomes. You might have nuclear non-proliferation. There's like, right, there's like dozens and dozens of years of political and military strategy here to not kill everyone in the world that is just not even mentioned here. Okay, there's just a part of this where it's,

• 25:38Speaker 03

  if, you know, tech people love first principles. And it's like, great, you did first principles. You know, there's like 100 years of principles built on top of them, right? And of how great powers should interact. He should have finished Harvard. That's all I'm saying.

• 25:53Speaker 02

  According to Mark Andreessen, looking backwards is not useful. You should, no introspection, always look ahead. Okay, here's the last one, and this is the purest Neelai bait in the whole manifesto. He's talking about distillation, which has been very controversial, the idea that you can essentially use a model to train another model. It's how a lot of people think the Chinese models have gotten so good so fast. Very hot topic in this space. Should distillation be legal? Mark Zuckerberg writes, "The ability for models to learn from other models is an important principle of how the open source ecosystem works. All AI models are derived from human knowledge. Some have tried to frame distillation as harmful, but I think it is important to protect the principle that you can learn from anything you can observe. This is how the world works, and the US will not be able to lead if we

• 26:36Speaker 02

  restrict ourselves on this front." This is Mark Zuckerberg saying, "You have to make it legal for us to do everything that we've been doing with AI models."

• 26:46Speaker 03

  The distillation discourse is very funny to me. It just fills me with a kind of joy.

• 26:54Speaker 02

  Sure.

• 26:55Speaker 03

  Because everyone is mad about the thing they did. Right? If you're anthropic and you're mad about distillation, well, you stole all the books in the world. What, what do you want? Fine, you got nothing to say. The principle that you need to learn from what you can observe is just a bedrock. Applies to humans. Computers that retain perfect copies of everything that they've ever ingested, and in many cases with these models, can spit out perfect copies of everything they've ever ingested, totally changes the game in terms of what, who should get compensated for what, and who has permission to ingest what. And Mark doesn't care. Like, "This computer is the same as me" is also a fundamental thought that exists throughout this entire framework.

• 27:40Speaker 02

  Yes. And throughout the entire AI industry, right? Like, the fundamental goodness of computers and bonkersness of humans is a real thing that is just burbling up.

• 27:50Speaker 03

  "You leave my robot girlfriend alone," said Mark Zuckerberg.

• 27:55Speaker 02

  I'm sure she's very nice. Okay, I have a bunch more. He goes on a whole tangent about RSI, which essentially describes AI that can-

• 28:01Speaker 03

  Repetitive stress injury.

• 28:05Speaker 02

  Exactly. It's, it's, it's when AI can start to train itself, essentially. You set it loose and it is able to train itself on how to get better. There are a lot of things that become possible when that happens. These things can theoretically improve very quickly. That's also how we get to like Terminator stuff. So he goes on this whole long discursion about how we should allot compute to the RSI models so that they can get better, but we have to give humans more compute because that's, it's like, all of this is insane. You should just read it just to understand his theory of the world as much as anything, which is that humans are fundamentally all great, and if you just shove

• 28:49Speaker 02

  them all together, everything will be fine. Like, meta.

• 28:52Speaker 03

  I went to a school for nerds, extreme nerds. The motto, the unofficial motto, the University of Chicago, literally, they make t-shirts, is "Where fun goes to die." I loved it there. Could not be happier with my education there. And a number of professors had two cliches in all my political science classes. One, "Where's your evidence?" This is what they would just say. You would say some nonsense, and they'd say, "Where's your evidence?" And then you would have to be a big nerd and you have to like produce your books and do it. Mark has no evidence for any of his ideas. The second one, much more actionable, an exercise you can do if you, like me, are screaming in your car at Mark Zuckerberg. Try to write the law that puts any of this into practice. Write the rule.

• 29:35Speaker 03

  That says, we have to somehow slow down China by, by limiting chip exports, but we can't let American models fall behind in any way, shape or form. We have to allow distillation, but we have to protect the value. Everyone should get as much superintelligence as they want to do as much baking as they can, but they also have to use their superintelligence to make sure the bad guy next door is somehow not doing offensive cyber attacks. Just write it down, just write down your, your law, pass it. Go to the United States Senate, get a number of dead people to show up, you know? And be like, this is, here's the rule, I'm giving it to you. Mark Zuckerberg's lobbyists are certainly trying to do that today. He has not shown us what they are proposing in any way, shape or form.

• 30:17Speaker 03

  He's given us this freshman year nonsense, but write the rules down.

• 30:21Speaker 02

  Yeah.

• 30:22Speaker 03

  That's actually my, my number one response to this was to rewind my brain to when I was in college and I would have professors say, write the rule that you're describing, make it work.

• 30:32Speaker 02

  I think this is why everybody is so obsessed with the idea of government review of frontier models, because that is at least in theory, one thing you can do. And there are so many things in this space that are not actually things you can do. But when does the government review models? How do they review those models? What does that review look like? Who's in charge? That is all actually tractable stuff in a, in a way that is very unusual. And the AI industry stance is, we'll do it voluntarily. Don't worry about it. We're all just going to be cool. Everything's going to be fine. We don't need laws. We will, we will volunteer to do this and that will be, that will be enough. That will be sufficient because I think they're deathly afraid that the minute somebody tries

• 31:14Speaker 02

  to write a law, it is going to accidentally or intentionally cause them vast quantities of pain that no one even saw coming.

• 31:21Speaker 03

  This was the Biden administration, which said, you're bleeding edge models. You have to tell us what they're capable of. There's some transparency around them. It was not a lot and they all fought against it. So they all went to Trump's inauguration because he took away all the rules. And the Trump administration has quickly realized it's going to impose some of the exact same controls the Biden administration imposed, if not more in additional controls.

• 31:43Speaker 02

  Yep.

• 31:43Speaker 03

  And again, the, the, when I say write a law, it's not frontier labs should be forbidden from doing offensive cyber attacks. That's not a rule. You can't, you gotta write it down. You gotta define a frontier lab. You have to define offensive cyber attacks. You have to say what it means to not do them. Write that, it's very, very challenging. So all of these platitudes are great. All these predictions about how it's going to be amazing when there's superintelligence having a relationship for you. Amazing. I couldn't be, I love reading science fiction. Show me the rules, Mark, because I don't think they exist.

• 32:19Speaker 02

  And I think his argument generally is maybe they shouldn't because we'll just give it to everybody and it'll be fine. Don't, don't even worry about it.

• 32:24Speaker 03

  And that has never caused any problems ever.

• 32:27Speaker 02

  Including on meta platforms. How could he know? He's never seen this before.

• 32:31Speaker 03

  I'm having an instagasm. Sorry.

• 32:35Speaker 02

  All right, we should take a break.

• 32:36Speaker 03

  That was it. That was, this is, we, it's your last episode before you have a baby. It's my last episode ever.

• 32:43Speaker 02

  I'm having an instagasm is the title of this episode. It's going to be fantastic. I can't wait. We're going to take a break and we're going to come back and unfortunately for all of us, we're talking more AI. We'll be right back.

• 32:55Speaker 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.

• 33:33Speaker 04

  This episode is brought to you by Michelob Ultra. Good times mean competitions between you and your friends, like seeing who can throw some pigskin the farthest or rush the field the fastest. Winner gets an ultra because it's not just a superior like beer. It's also giving fans superior access to their favorite sporting events. Superior is worth playing for. Visit michelobultra.com/superioraccess to enter. Enjoy responsibly. Copyright 2026 Anheuser Busch InBev Michelob Ultra Beer, St. Louis, Missouri.

• 34:01Speaker 05

  Are your ad campaigns lighting up the dashboard? But not the pipeline? That's bull spend and marketers are calling it out in dashboard confessions.

• 34:11Speaker 06

  My boss asked for results, so I opened my dashboard for the only positive sounding metric I had, impressions.

• 34:18Speaker 05

  Cut the bull spend. See revenue, not just reach. LinkedIn delivers the highest return on ad spend of major ad networks. Advertise on LinkedIn. Spend $250 on your first campaign and get a $250 credit. Go to LinkedIn.com/campaign. Terms and conditions apply.

• 34:34Speaker 02

  All right, we're back. Neil, I, we have, we have more AI shenanigans to talk about today. The unfortunate theme of this show is everybody has bad ideas about AI this week, which is fun for me because I get to get all of my feelings out and then go on parental leave. But for you, you get to just stew in everybody's bad ideas about AI. But I want to start with an idea that I have not decided whether it's good or bad yet. And it's Claude's new watermark for AI generated text. The idea here is, is Anthropic is rolling out a new tool by which it will invisibly watermark text created with Claude.

• 35:16Speaker 02

  I have, I have done more research than was necessary on this and I find it fascinating. We can get into it. The idea is that it will be then detectable by the, the AI detectors. They're also going to embed information in images. They're, they're trying to be more transparent about what is and is not generated by AI. This seems like a good thing in the way that all of these ideas seem like good ideas. We've also, once again, it seems run into the ex KCD comic of there are 14 standards.

• 35:45Speaker 03

  Yes.

• 35:45Speaker 02

  Let's come up with a new standard. Problem, there are 15 standards. Is this just, are we just doing this again?

• 35:50Speaker 03

  The most important thing to know is that all of the labs have realized people don't like AI content and they especially do not like AI content when they're not told it's AI content. And you can see it in, on your own feeds. Somebody posts something and the immediate response is this is AI. And it doesn't matter what that thing is.

• 36:10Speaker 02

  Yeah.

• 36:10Speaker 03

  And they all knew this was coming. They did essentially nothing about it. And now the backlash is here. Everyone hates them, right? They're polling worse than ice. No one wants your data centers. You gotta do something. And so we're going to do lots and lots of watermarking ideas so that you can see the AI content and hopefully you can see that you like some of it, which is the unremarked flip side of all of these efforts.

• 36:35Speaker 02

  Oh, that's interesting. They're like, no, it's AI and it's good.

• 36:39Speaker 03

  You loved it, didn't you?

• 36:41Speaker 02

  Hmm.

• 36:42Speaker 03

  We'll see about that.

• 36:42Speaker 02

  That's really interesting. So you don't think this is sheer sort of corporate responsibility of these companies trying to be good guys? You're like, they, they want, they want you to be tricked and then have your mind blown by the fact that you've been tricked.

• 36:54Speaker 03

  It's like a reveal. It's like they pull the cloth off at the end and you're like, it was Claude all along. Some of that is in there, I think. I, I think there's also just the notion that maybe you'll opt into a feed full of all AI content, which is a dream they all have, right?

• 37:10Speaker 02

  I'll tell you right now, no.

• 37:13Speaker 03

  Half of them have tried it and mostly they have failed. But that's the meta structure of this, right? The, the reaction to AI content in almost every case is negative. And we need to label it because everyone, there's a disinformation crisis across the world and a misinformation crisis across the world. And so the labels will help us get there. How are we going to do this is not a problem that was solved yet. I'm dying to know if Anthropic asked Claude how to watermark Claude's text, because this system seems so complicated to me. Did you get all the way down to, to red lists and green lists?

• 37:49Speaker 02

  I did.

• 37:50Speaker 03

  Can you, can you, can you explain this to me? Cause I'm not sure I quite understand it.

• 37:53Speaker 02

  I can try. So first of all, shout out to our commenters who immediately found this, uh, this great video from a couple of years ago that links to a University of Maryland paper from a couple of years ago, digging into this idea. So the idea, as I understand it is watermark is actually sort of the wrong term. Watermark connotes that there is some thing invisibly embedded in the text, like a special character or some kind of underlying metadata that says this is generated with AI. This is not that, to the best of my understanding. What this is, is basically if, if you think of an AI system as next word prediction, which a lot of people don't like to, but it is fundamentally what's going on. It looks a bunch of words and says, what's the next one?

• 38:37Speaker 02

  It is essentially going to up and down rank certain answers to what the next word might be. And the way that they talk about this in this paper is you have a red grid list and a green list. So like Neil, I went to the, there are a bunch of words that could come after that, right? So like store, football game. And then there's a bunch that it would never be like, I don't know. Taylor Swift concert.

• 39:05Speaker 02

  You love Taylor Swift. I'm just kidding.

• 39:07Speaker 03

  That's sad.

• 39:08Speaker 02

  But so the idea is then you essentially break these things into all the things that could possibly be. And you up rank some and you down rank some in order to create a certain kind of slightly slanted version of the output. So there are going to be words that the text doesn't use. It's going to use certain words more often, but it is basically up ranking and down ranking certain answers to the question. And then it's able to, over time, the way that you use red and green is like you take all the possible answers, put them into red group and green group, and you only use the green group, or you mostly use the green group in the actual text that you're predicting. Then if you go back and you say, oh, that's so weird. This is four paragraphs of it always picked the green one.

• 39:53Speaker 02

  Suddenly you have an incredibly high likelihood that this thing was generated by AI. And that's essentially the idea is rather than embed some extra character. This is why Anthropic is able to say, if you copy and paste the text, it will carry with it. Because it is the text itself, the way that it was predicted and the order in which, and the types of words that are actually used are the ones that reveal that it was created by AI.

• 40:22Speaker 02

  The paper is fascinating. It's deep and weird, but it seems to work. And it especially works on long texts. And that's a big deal because one of the things that Claude even will do, in this case, they'd say, is not do a great job with short text. One of their like cautions is if the text isn't long enough, it may not be reliable. But with like multi paragraph stuff, you're going to get to the point where statistically, it is incredibly unlikely that a human would have written this same set of things in exactly the way that we have slightly geared Claude to do.

• 40:55Speaker 03

  Did you feel the same glimmer of recognition when you read this description that I did?

• 41:00Speaker 02

  May, no, tell me.

• 41:02Speaker 03

  This is how people detect that things are AI.

• 41:06Speaker 02

  I mean, yeah.

• 41:07Speaker 03

  Yeah. Why would you use M dashes? It's like literally the thing that everyone online is always doing. Like, you know, it's, that's not AI. That's just common sense. Like whatever that thing is that you are always doing with AI content now, where you see the fake AI influencer on TikTok.

• 41:23Speaker 02

  Here's my honest answer, colon.

• 41:25Speaker 03

  They're too shiny. Yeah. Like, you know, like AI video people, like AI avatars are, they're too glossy in like a very specific way. It's like, oh, this is the thing that I do. Like there's something about this that I just find fascinating.

• 41:38Speaker 02

  Yeah.

• 41:38Speaker 03

  It also means it won't work unless the model providers sign on to it. So here's another collective action problem. Claude's list of green words is made by Anthropic. They probably won't release them. Right? This is, Anthropic is particularly closed. They're probably not going to tell you what, what words or what tokens Claude has.

• 42:00Speaker 02

  They'll tell like the pangrams of the world, but they, they're.

• 42:03Speaker 03

  Maybe.

• 42:04Speaker 02

  They're very unlikely to share it with everyone.

• 42:07Speaker 03

  That's, even that to me feels like a hard maybe.

• 42:10Speaker 02

  Fair.

• 42:10Speaker 03

  So only Claude text is detectable and ChatGPT text is not detectable. Now you have an imbalance of power and it is an open eyes interest to never ever use Claude's detection system.

• 42:20Speaker 02

  Right. And if you take text out of Claude and put it into ChatGPT, it immediately breaks the whole thing, right? Like this is, these are not hard systems to get around, unless everyone agrees that these are the systems.

• 42:31Speaker 03

  So yeah, I mean, I applaud Anthropic for being like forthright that mountains of text spamming every text box on the internet is a problem. And you should be able to have a detector that can just disregard it. Straightforwardly disregard it or say, are you, you're a real person. You've been shoving a lot of AI text into LinkedIn lately. Would you like to pay us some extra money? Cause that's gonna happen.

• 42:56Speaker 02

  Yeah. I also wonder, I mean, to your point about the, the already overused words in AI tools, it's going to be very hard for them to do this in a way that is actually imperceptible. And they, they say that it is imperceptible. The, the paper it's based on says it is imperceptible that you can do this with very little extra compute, very little extra time, and very little degradation of the output. Everything we see from AI generated text suggests that that is probably not the case. That what this thing is going to do is it's going to develop a new set of tics that are going to make it even easier for humans to recognize AI written text that are going to make it weirder and harder for Claude to know what to do with that.

• 43:39Speaker 03

  Yeah. I think walls of AI generated text are so inherently disrespectful that I, I kind of don't care about this one.

• 43:46Speaker 02

  Fair.

• 43:47Speaker 03

  Like you, you can see it. It's not hard to see it. I don't want to read it. No one wants to read it. Fine. Like hopefully Anthropic is successful with this and people learn to not do it because everyone's looking for detectors. It's the multimedia stuff that is way more dangerous, way more widespread in a lot of ways, and way more fractured in terms of detecting it, how the watermarks work, and in particular, how the platforms display that information. So there's code, I think in the latest betas of iOS called Apple Reference Image, where if you take a photo with an iPhone, it will mark it as the reference image and say, these are the actual pixels, which is already funny because there's so much computational photography going on in iPhone.

• 44:33Speaker 03

  And whether or not they're the actual pixels is challenging.

• 44:39Speaker 03

  W W C, when Apple started showing off their AI edits, background remover, changing poses, all the stuff you can do. I asked them if they were going to use some of the watermarking standards that we already have. C2PA, Synth ID. I think they said they were going to use Synth ID, which is Google's. But it sounds like they're, they're going at it the other way. They're also going to mark the real image, which is not an approach many people have taken thus far. But you want that real image to travel around and be marked as real. And you want the AI edited image or the AI generated image to also travel around and be marked as AI. Somewhere in there are some real fuzzy lines, right? Even the noise reduction on iPhone now is like an AI powered situation.

• 45:21Speaker 03

  So these are very fuzzy lines of where the AI takes over and what makes an AI image or not. But everyone is labeling and no one is picking the same labeling standards. And none of the platforms have agreed on how to display the information.

• 45:34Speaker 02

  Right.

• 45:34Speaker 03

  And this, when you say, this is just segments of bad ideas by AI, the reality that people don't like this stuff has just like hit them in the face and they weren't ready. And they did not have a unified plan in any way, shape or form to handle it.

• 45:50Speaker 02

  Yeah. And I think from Apple's perspective, the idea of we're going to mark the real one makes a lot of sense to me because that in a certain way might be the more controllable outcome for Apple. Right? Like it's actually, it's a bleak state of affairs, but it makes a certain kind of sense. If you think we're going to get to the point where you should assume that everything that's not marked real is fake. And if we're in a moment where no one is going to agree on how we mark things fake, it's easier to be the one who says, well, we can be the arbiters of what is real. And that is just a weird societal switch that I feel like we are making. And again, it makes sense for Apple. Apple can actually verify whether this thing was made on an iPhone

• 46:33Speaker 02

  is probably vastly easier in many cases than trying to verify whether something was generated with AI.

• 46:40Speaker 03

  Yeah.

• 46:40Speaker 02

  Weird state of affairs. Like maybe, maybe we're not going to have an AI feed. We're going to have a real feed. Do you know what I mean? Like this is, that's the, the mental shift we're all having to make. And I think are kind of making in real time from all of this is probably real to all of this is probably fake.

• 46:56Speaker 03

  Oh, I mean, we wrote that story ages ago. Sarah John wrote it for us. The headline was, no one's ready for this. We'll link to it in the show notes. And the, the mental shift in society to, I'm probably looking at something fake when presented with a photo or a video is totally destabilizing. And it, it already is destabilizing.

• 47:14Speaker 02

  I listened to a fantasy football podcast. Shout out the Ringer Fantasy Football Podcast. And last, training camp. So a thing that happens at the beginning of every NFL season is all the teams go to training camp and you see all the players for the first time in several months. And last year, everybody got fooled by a picture of Debo Samuel. Doesn't matter who Debo Samuel is, but somebody had, had edited the photo to make him look really fat. And it became, it became a news story before everybody realized it was fake. This year, everyone assumes every picture of every player is fake. It's the most, it's just the most fascinating shift that like in 12 months, they've gone from instinctively believing the photo they saw on Twitter to assuming it's fake and actually having to go back and verify before discovering that, oh no,

• 47:58Speaker 02

  he's actually fat and out of shape. And this is going to be a big problem for the NFL season. And it's just the speed with which we have all had to make that change just kind of continues to blow my mind.

• 48:07Speaker 03

  Yeah. And this is, I mean, there's market demand for provenance information embedded. It's the time of capture from the sensor. Great. So Apple's going to do it and it's going to go up to private cloud compute and generate an ID. And someone is going to have to respect the ID that Apple embeds in the photo and display the information. Do you think that that is X, the everything app?

• 48:29Speaker 03

  Right. This is, this is where it falls down.

• 48:32Speaker 02

  Yeah.

• 48:33Speaker 03

  And Jess Weatherbed has written about C2PA, which is the competing standard here over and over again. C2PA doesn't work very well. It can be defeated pretty easily. And the other problem with it is that no one respects it.

• 48:46Speaker 02

  Right.

• 48:46Speaker 03

  And, and I don't just mean they don't respect it because it's bad and they, they look down on it. I mean, no one displays the information. You have to go to another website and like paste an image into it and run the detector check. And this is all nonsense and no one's ever gonna do it.

• 48:58Speaker 02

  Completely defeats the whole purpose.

• 49:00Speaker 03

  The platforms that are doing it are, are kind of interesting. So Spotify announced it's not going to recommend music from AI personas this week. Because they know that their customers do not like it when AI artists show up in their feeds and their recommendations. And they want music from other real people.

• 49:16Speaker 02

  This goes back to exactly what you were saying too, about people not wanting to be tricked. There, there is, I think, a lot of people have had the experience of having AI music appear in their algorithms for one reason or another. It shows up in a playlist. It shows up on your Discover Weekly, whatever. You get 30 seconds into it and then you start to get that kind of tingly feeling that this is AI. Or worse, you like it and then you discover it was AI. And I have watched a lot of people go through that experience in various ways where like, oh, this song is a bop. And then you're like, oh my God, it's AI. And that feels bad on so many levels.

• 49:53Speaker 03

  I mean, we just had an entire news cycle in the music industry about rubbers.

• 49:57Speaker 02

  Oh, yeah.

• 49:58Speaker 03

  By Phoenix Flexen. He's, he's a real person with, it's a, obviously an AI track. And first they denied it was AI and then it was obviously AI. And now they've been admitting that more and more of it is, and they're gonna get to the end, which is, yes, we just opened Suno and told it to make a Smith song.

• 50:17Speaker 02

  Yep.

• 50:17Speaker 03

  And then, and I pretended to sing a Smith song.

• 50:20Speaker 02

  Wait, have you been tracking this, uh, D'Addario, the, the guitar company thing?

• 50:25Speaker 03

  The guitar industry is in full turmoil.

• 50:27Speaker 02

  Can you explain this one to me? I have not followed this one as closely.

• 50:31Speaker 03

  Uh, D'Addrio, which is, they make, they make all the guitar strings.

• 50:35Speaker 02

  Uh huh.

• 50:36Speaker 03

  If you have ever played the guitar in your life, you have ended up with a pack of D'Addrio guitar strings. They made an ad and guitar people thought the music in the ad was AI. And then their first denial was that it was a real song and they just recreated it with AI. Which is incredible.

• 50:58Speaker 02

  That doesn't even mean anything.

• 50:59Speaker 03

  Sure. And then they just issued a full walk back and apology and said they believe in real music. This is an incredible problem for the company that makes guitar strings. I'm confident they make other things. Someone is gonna correct me and be like, they're the biggest manufacturer of violas. I don't know what it is. I've played guitar. And they, they're the guitar string company and they should probably lean into being the analog guitar string company. And not have AI anywhere in their process. They threw like someone under a bus. Like, we didn't want to do this. It just happened that the music was recreated with AI.

• 51:30Speaker 02

  Got it.

• 51:31Speaker 03

  This is right next to the CEO of Fender describing having bandmates as analog.

• 51:39Speaker 03

  And then having to do a full walk back. You were following him much more closely.

• 51:43Speaker 02

  So, okay, this guy's name is Bud Cole. He's the CEO of Fender. Unbelievable name for the CEO of Fender to be named Bud Cole. I imagine him sitting on a stage playing blues guitar. He gave an interview, I believe in May, to, uh, the, the magazine T3. And he said, among other things, I think AI has existed in music as long as there's been recorded music, which, what?

• 52:07Speaker 03

  Sure, bud.

• 52:08Speaker 02

  And then he says, I actually believe cover music has been sort of analog AI for a long time. I listened to a lot of R E M, U2, The Smiths and The Cure, and at some point I got sick of just listening to them. I wanted to play it, so I learned to play guitar. To compare that to AI is so fundamentally insane.

• 52:25Speaker 02

  And this goes, I mean, this goes back to what you were saying about the difference between humans and computers, right? Like, it's, it's one thing to say, I listened to a bunch of Smith songs and so I learned how to play a Smith song on the guitar. The idea, just, just at a very simple way, the idea that I play the same notes and am thus playing the same song is bananas. And anyone who has ever heard a song, a song, knows that to be true. But then this guy, I think, without meaning to, is basically denigrating every music artist ever in the history of music, which is a wild thing for the Fender CEO to do. But he, he basically is like, we are going to free people up

• 53:06Speaker 02

  from the nightmare of having to learn other music and play with bandmates and just create whatever they want to.

• 53:13Speaker 03

  Yeah. This is, by the way, the same thing that the CEO of Suno has said over and over again.

• 53:17Speaker 02

  Yeah. But only one of them is trying to sell me guitars. Like, Suno's at least trying to sell me AI software. Like, I get why they would say that. Bud Cole has guitars to sell me.

• 53:28Speaker 03

  If you, uh, listen to a lot of R E M, U2, The Smiths and The Cure, and you want to play it, it's, yeah, it is true that it's important to learn to play the guitar. You also definitely need to buy a delay pedal. And so I think that is a probably a better expansion opportunity.

• 53:42Speaker 02

  Yeah.

• 53:43Speaker 03

  For Fender, you know, like here's, here's the Johnny Marr sound from How Soon Is Now? No one has ever been able to figure it out. AI will figure it out for you. That would have almost made sense. This does not.

• 53:54Speaker 02

  No.

• 53:54Speaker 03

  Suno, by the way, uh, also under fire and they've unveiled download song caps now. Uh, because they know that there's mass exports of Suno music flooding platforms. They're under fire. Actually, Ashley Carmen, uh, former star Verge reporter Ashley Carmen, now at Bloomberg, just did an entire profile of Suno, including interview with the CEO over at Bloomberg with Luca Shaw. That piece is great. They know that they're in trouble. Like, they have to meaningfully participate in the music industry in some way. They just signed a deal with BMG. They've got other deals in the works. They're in copyright hell because everyone's suing them. But the music industry secretly loves Suno.

• 54:32Speaker 02

  Right.

• 54:33Speaker 03

  They're every, they're, they're percolating all over the music industry all the time. Because artists who do know how to play their instruments, who do know stuff, still find it interesting and easy to prompt Suno to give them a sketch. And then sometimes that sketch just ends up on the tracks.

• 54:49Speaker 02

  Right.

• 54:50Speaker 03

  And the, I think the trick is, when is that deception? When is that garbage? When is that slop? And when is that something meaningfully new and interesting?

• 54:58Speaker 02

  Right.

• 54:58Speaker 03

  And it's, it's, I don't know that there's an answer across the board. And all of these companies have just rushed forward to be like, this is the future without accounting at all for, are we deceiving people? How will they know? And are they gonna hate us for it?

• 55:12Speaker 02

  My most maybe controversial opinion on this front is that I actually think the line is somewhere in the middle. Like, we're, we're gonna hear from a lot of people watching and listening to this who are like, the, the answer is clear. As soon as you use any AI tool, it is tainted and ruined and I'm out. And I respect that opinion. I just don't agree with it, right? Like, I, I think, I think to the extent that AI is software, which is a thing that I believe, can you use software to make music? Sure, right? Like, people, people freaked out when we started making computer music because you could make a drum kit in GarageBand. Like, I don't think that makes it not music anymore. It's, if you want to disagree with that, that's fine. But I do think that is the interesting conversation in AI to have.

• 55:53Speaker 02

  And in a funny way, like, you talk a lot on this show about how music always is at the front of every technology question. It is so perfectly suited to have this discussion about like, I think, I think most reasonable people will agree that typing a sentence into Suno and generating a song is not an artistic expression. Some people think that it is, Suno's CEO certainly thinks that it is. I do not, and you will not convince me otherwise. Somewhere in between there and you are not allowed to open your computer is the right answer. And I, I think it's, it's actually going to be really interesting to watch the music industry from all sides, the, the creatives, the platforms, like Spotify on this front. The, you mentioned the personas thing. They're very clear that this is not about AI generated music. This is about AI personas.

• 56:39Speaker 02

  People, artists who are wholly AI creations. In part because Spotify can't figure out how to detect or litigate what makes AI music. Because that would require Spotify to make a decision about when a music becomes a piece of AI generated work. And nobody knows and we should really figure it out. We should really figure it out. Like, as, as people, as societies, as businesses, like, that is the question in front of so many people. And it really does feel like the music industry is just the deepest in it right now, but it's kind of coming for everybody.

• 57:11Speaker 03

  The music industry is, is always at the vanguard, like I'm always saying, for a few reasons. One, everyone always listens to songs over and over again. So Hollywood, streamers. You might have watched Infinity War 50 times. That is the most anyone has ever watched Infinity War. Do you know what I mean? Like, that's the max. Most people watch that movie once or twice and that's it. And then you watch clips on it, bootlegged on TikTok for the rest of your life. But you've consumed that movie once or twice. Your favorite show. Even, even if you're a massive weirdo like David and you're just streaming The Office 24/7 in the background of everything you do all the time. Like, you're only watching that so much.

• 57:52Speaker 02

  Yeah.

• 57:52Speaker 03

  Music is infinitely consumable. I have probably listened to London Calling by The Clash 10,000 times. High Fidelity, uh, is on Netflix, so I keep watching it by accident.

• 58:02Speaker 02

  By accident, for sure. That's not like a perfect Neelai movie. Definitely not.

• 58:06Speaker 03

  You just like watch to, it's in, it's in Chicago. I can't get enough of it. Nick Hornby, who wrote High Fidelity, uh, in one of his other books, he wrote that he has listened to Thunder Road by Bruce Springsteen every day of his life.

• 58:19Speaker 02

  Wow.

• 58:19Speaker 03

  Infinitely consumable material. Like, there are songs like this for lots of people, right? There's music like this for lots of people. So you have, over and over and over again, you have the opportunity to sell the same product over again. It is also mostly a closed ecosystem. There's only a handful of record labels. There's only a handful of distributors, a handful of big agents. At this point, there's only one artist in the world. Her name is Taylor Swift. And they all are repeat players. So they're all doing deals with each other all the time. And if they overdo it, then they won't get a deal the next time. So they're motivated to like make long-term deals. And then the music industry is so litigious and so protective of its value because of these other two, uh, qualities.

• 59:02Speaker 03

  That they sue people if you screw them in a way that Hollywood doesn't.

• 59:05Speaker 02

  Right.

• 59:07Speaker 03

  And so you just have this machine that figures out how to make music worth money over and over and over again, because it, the catalogs are worth something. If you buy one of the big catalogs from Neil Young or some, someone, that's worth a lot of money to you in a way that buying the entire back catalog of like Warner Brothers, when it eventually crashes into the ocean, might not be worth anything to you.

• 59:31Speaker 02

  Right.

• 59:32Speaker 03

  And so all of this means that it, these issues will get figured out in music first in like, maybe annoying ways, maybe the wrong ways, but there will be some set of answers. And I think one of those answers looks like Suno has to put the labels on it. If Suno generates a song with vocals, the vocals will be tagged as not a real person, because people particularly hate that. On and on and on, and Spotify will take pressure from Universal and BMG and whoever else to label the stuff generated by Suno in very specific ways, because a human artist won't participate with Spotify. And on and on and on, because it's a closed ecosystem that knows how to make money.

• 1:00:13Speaker 03

  And then whatever happens there will come to YouTube and Disney Plus and TikTok, because the model will exist in music. And this is a prediction, I'm not saying this will definitely happen, but there's a reason why everything happens in music first.

• 1:00:26Speaker 02

  It's a good theory. Do you remember that phase of music in like the mid aughts, when the first 25 seconds of every song was all the producers saying their producer tag?

• 1:00:35Speaker 03

  Yeah.

• 1:00:36Speaker 02

  Like, and you just get six of them in a row and it would always end with like DJ Khaled just being like, DJ Khaled, like, that was, it was great.

• 1:00:42Speaker 03

  Another one.

• 1:00:42Speaker 02

  Another, right. We're gonna get that, but with like AI labels at the beginning, where it's just at the very beginning of every song, it's just gonna be like, Suno. And that's how every song is gonna start.

• 1:00:52Speaker 03

  A Suno producer tag is actually great. I'm actually, I'm, one, I'm shocked there's not a DJ calling himself Suno already.

• 1:00:58Speaker 02

  True.

• 1:00:58Speaker 03

  Just take their credit for it. And the second, it was producer tags and for a hot minute, my friend, it was face tattoos. If you recall, the SoundCloud rappers and their face tattoos, just to stand out from the crowd, were a real thing. And whatever the AI version of face tattoos, it's coming for us.

• 1:01:16Speaker 02

  Unfortunately, it might be face tattoos.

• 1:01:19Speaker 02

  We should take a break and then get to the lightning round, but I, I do, I just want to leave us with one more piece of evidence in the, everybody knows this is a problem bit of the, the story here, uh, which is from Mike Minton, the chief product officer at Twitch. Twitch just rolled out a new thing where they basically are starting to train Amazon's models from Twitch content. They made it opt out, which means that by default, everyone on Twitch is being used in this program unless they go to opt out. He was asked, why not make it opt in? What if, a very simple thing to do is just let people choose to be part of this or not? And his answer is, if it was opt in, nobody would opt in.

• 1:02:00Speaker 03

  This is the, the true nature of all opt in.

• 1:02:04Speaker 02

  Yeah, I mean, yeah.

• 1:02:05Speaker 03

  Will you opt in a targeted advertising?

• 1:02:07Speaker 02

  You want this bad thing? Yeah. They, like, this is my point, they, they, they know, right? Like, it is increasingly clear to all of these companies that people do not want this. And they are in a position of either having to make hard decisions or run afoul of Wall Street and the economy in a very real way. And I, like, I don't envy the people in that position, frankly, but boy, are they just making user hostile decision after user hostile decision.

• 1:02:35Speaker 03

  You know, I'm not up on the nature of the relationship between Twitch creators and Twitch, the platform. I'm guessing it's as testy as the nature of all platforms and creators. We're gonna come to some YouTube changes in the lightning round.

• 1:02:52Speaker 03

  My prediction is that the music industry figures out some labeling and attribution stuff first. My prediction right next to that is YouTube undergoes some sort of AI training creator crisis first.

• 1:03:05Speaker 03

  Right? Because they have basically opted all their creators into trading and they're making it harder to make money. And somewhere in there is a very angry swirl.

• 1:03:15Speaker 02

  Yeah. Every, I mean, it is, it is now, I think, basically beyond doubt that YouTube was a foundational part of the training data for every AI model everywhere.

• 1:03:26Speaker 03

  Yep.

• 1:03:27Speaker 02

  And boy, have creators not had their full say about that. Well, let's get to the sum of that stuff. Let's take a break, then we're gonna come back. We're gonna do the lightning round. We'll be right back.

• 1:03:38Speaker 04

  This episode is brought to you by Michelob Ultra. Good times mean competitions between you and your friends, like seeing who can throw some pigskin the farthest or rush the field the fastest. Winner gets an ultra, because it's not just a superior like beer, it's also giving fans superior access to their favorite sporting events. Superior is worth playing for. Visit michelobultra.com/superioraccess to enter. Enjoy responsibly. Copyright 2026 Anheuser Busch InBev Michelob Ultra Beer, St. Louis, Missouri.

• 1:04:06Speaker 05

  Are your ad campaigns lighting up the dashboard? But not the pipeline? That's bullspend and marketers are calling it out in. Dashboard, confessions.

• 1:04:16Speaker 06

  My boss asked for results, so I opened my dashboard for the only positive sounding metric I had. Impressions.

• 1:04:23Speaker 05

  Cut the bullspend. See revenue, not just reach. LinkedIn delivers the highest return on ad spend of major ad networks. Advertise on LinkedIn. Spend $250 on your first campaign and get a $250 credit. Go to LinkedIn.com/campaign. Terms and conditions apply.

• 1:04:37Speaker 07

  College football is back. So Hilton called to me, the superstition concierge, to make your fan rituals a reality. Need a room to match your lucky number? We got you. Wanna make sure our team doesn't wash your lucky jersey? Oh, that smells lucky. Hilton's unmatched hospitality can keep up with any superstition. Even a marching band wake up call at 5:55 in 55 seconds. Hit it! When you need a team that will do whatever it takes on game day, it matters where you stay. Hilton, for the stay.

• 1:05:09Speaker 02

  All right, we're back. It is time now for the lightning round, beginning with, I get to escape this, I just realized, I don't have to do this for a while, but you have to every week and this week. America's favorite podcast is in a podcast. Brendan Carr is a dummy.

• 1:05:23Speaker 02

  Brendan Carr is a dummy. Brendan Carr is a dummy. Brendan Carr is a dummy. Brendan.

• 1:05:30Speaker 03

  Brendan Carr is a dummy. Brendan Carr is a dummy. Brendan Carr is a dummy. Brendan Carr is a dummy.

• 1:05:40Speaker 02

  All right. That's from Rich Strafalino, who sent this in with the caveat that it is, quote, "bad in a distinctly human way."

• 1:05:47Speaker 03

  It does seem like you drank a lot and opened a soundboard, but in like a fun synthwave way, you know?

• 1:05:53Speaker 02

  Yeah.

• 1:05:54Speaker 03

  David, I got a surprise for you. Um, you know, you're having a baby, I hired you and you night nurse. His name's Brendan. He's, he's coming to your house.

• 1:06:01Speaker 02

  He's just, he's just gonna sit with my baby in one arm and on the phone with like a Comcast and he ever, just from 8:00 PM to 4:00 AM.

• 1:06:10Speaker 03

  He's just gonna look at your shoulder, Meg, you can't say that every single day.

• 1:06:15Speaker 02

  He's gonna yell at the late night TV while he sits there with my baby. That's not funny.

• 1:06:23Speaker 03

  I mean, it's pretty good.

• 1:06:24Speaker 02

  I'm down, I'm ready for this.

• 1:06:26Speaker 03

  He's coming to your house.

• 1:06:27Speaker 02

  Oh, God. No, thank you. What do you do this week?

• 1:06:30Speaker 03

  Brendan Carr this week is, it's a complicated, it's not the usual, we are banning speech and being a traitor to the First Amendment Brendan Carr segment. So I'm just letting you know, it's a switch up. This week, the FCC has taken to tweeting proudly about how shotgun banning different things has led to a resurgence of manufacturing in America.

• 1:06:53Speaker 02

  Oh, sure. That's exciting.

• 1:06:54Speaker 03

  They tweeted, "Made in America," following the FCC's action to ban imports of new models of foreign-produced drones. There's been $5 billion in US drone investment, thousands of new American jobs, a new US factory square footage. Read more. And then, actually, this is particularly good. They, uh, retweeted RapidResponse47, which is an official White House account, that posted a bootleg clip of Fox News visiting a Michigan drone startup called AG3 Labs.

• 1:07:20Speaker 02

  Perfect.

• 1:07:20Speaker 03

  Perfect. Just a little copyright infringement thrown in the mix there. Love it. Can't be happier about it. They also tweeted, um, "Economic independence. Under Brendan Carr FCC, we have taken action alongside the executive branch to ensure that America is not reliant on hostile foreign nations for drones, routers, advanced robotics, or power inverters." They tweeted a link, uh, to a study about the manufacturer of power inverters in America, which is gonna hit some high number. They forgot to note that the research note that they tweeted also pointed out that they would be vastly more expensive than the Chinese power inverters we were on for. This is all very good.

• 1:07:57Speaker 02

  Fabulous.

• 1:07:58Speaker 03

  So this is on its face, great, right? We've banned Chinese products or Chinese components. That has led to a resurgence in American manufacturing, which is one of the biggest policy goals of this administration. And then you scratch at it and you're like, oh, none of this, none of this is actually happening. And it's right there, it's right up there with tariffs. I think this rhymes with, we're going to put tariffs on all this stuff and that will move manufacturing back to America, which has basically not happened.

• 1:08:25Speaker 02

  No, instead it's just made everything more expensive.

• 1:08:27Speaker 03

  It made everything more expensive than the tariffs got knocked down. So as Sean Hollister on our team has been pointing out over and over again, these bans are first of all fundamentally ineffective. If you want to buy a DJI drone today, you can. These manufacturers are finding ways to get these products onto shelves. Sean is basically the compliance officer for the United States of America right now. He's finding stuff and sending it to the FCC and saying, isn't this illegal? And the FCC keeps not responding on the record.

• 1:08:53Speaker 02

  Perfect.

• 1:08:54Speaker 03

  And I'm just pointing out that if they had any stones, Brendan, they would respond on the record with a name spokesperson, per our background policy, instead of just saying some stuff we can't print. Pointing it out, that is the state of the reporting right now. Sean Hollister every day is like, I bought another illegal Chinese camera. Do you have anything to say about this?

• 1:09:13Speaker 02

  Look at all these drones in my house now.

• 1:09:15Speaker 03

  And every day, the FCC has nothing we can print to say about it.

• 1:09:19Speaker 02

  Perfect.

• 1:09:20Speaker 03

  And then the things vanishingly disappear from Amazon shelves. This is a real dynamic that is happening. We're going to report it all out. I'm just putting a little pressure on our nation's government to respond on the record. The other thing is that even these companies that claim to be doing all this investment and manufacturing all these devices in the United States are not actually doing it. Because what they're really doing is importing components through other channels to get to the United States and doing final assembly here in the United States. I know this because we just had Evan Smith, the CEO of Altana on Decoder. Altana is a startup that does massive supply chain tracking and data. They just make it easier for goods to move around the world. They've built huge platforms that everyone uses that can track everything.

• 1:10:02Speaker 03

  And I asked him specifically if manufacturing is moving back to the United States and he specifically responded with respect to drones. Here's what Evan Smith said about drone manufacturing in America.

• 1:10:13Speaker 08

  The Chinese control roughly 90% of the, you know, relevant value chain components for autonomous systems. So even if you have a US startup that's saying, hey, we're making everything here in America. No, you're not. You're assembling component parts to make, to make, you know, an air or land-based drone and, you know, you're saying made in America.

• 1:10:37Speaker 03

  Yep. Here's lying. And here's our nation's government being like, look at these cool lies.

• 1:10:41Speaker 02

  So what we have in America is like, we have screwdrivers. Everything else comes from China. Here in America, we have the screwdriver.

• 1:10:49Speaker 03

  Yeah, and look, you can say it is our goal to move component manufacturing and component supplies closer to America or in America specifically. There's a lot of that going on. There's a lot of shipping to Mexico or Canada, which have different relationships with China, and then shipping to the United States, a slightly more finished product. And maybe you can get around it. And the Trump administration wants that to suck too. Like, there's all this stuff happening, right? There's, there are rules and the company's trying to get around the rules and all this is fine. Like, this is the nature of the system. None of this has resulted in true American manufacturing end to end of these products. There's, is some investment occurring, but we haven't actually achieved the goal.

• 1:11:31Speaker 03

  And we can tout the investment, but the shotgun approach to banning things left and right. The FCC just banned Roombas. And they didn't actually notice until Sean noticed that the rule is that they banned Roombas.

• 1:11:45Speaker 02

  Yeah.

• 1:11:45Speaker 03

  And like, this is where we are with Brendan Carr's FCC. We're doing a big First Amendment violations with speech. We're issuing laws that make no sense in regards to broadcasters and how much broadcasters could be owned by a single company. We're doing corrupt favors for Trump cronies left and right. And then over here, we're issuing blanket bans based on national security that are not achieving their outcomes and then openly lying about the outcomes being achieved. And you don't have to look too far to see if the outcomes are being achieved. You just have to look on Amazon to see if you can buy a DJI Phantom. And the answer is you can. And it is, it boggles the mind that Brendan thinks everyone is as stupid as he is.

• 1:12:29Speaker 03

  Because you can just see that he banned foreign-made routers and a bunch of router companies said, whoa, hold on, can we get an exception? And he gave them exceptions. And the mechanism of that, still a little unclear, but based on everything I know about the Trump administration, suggests only one of two possibilities and neither of them are fully above board.

• 1:12:48Speaker 02

  No.

• 1:12:49Speaker 03

  Anyhow, Brendan, if you can answer how any of this works or how you're going to move component manufacturing back to America or whether the router ban is effective in the slightest, please, join the show. We'd love to have you on. I have a long list of questions for you at this point. My AI will ask them politely of you.

• 1:13:06Speaker 02

  I mean, he's upstairs. I can just go get him.

• 1:13:07Speaker 03

  He's watching TV.

• 1:13:10Speaker 02

  He's just hanging out.

• 1:13:10Speaker 03

  Yeah. Also, if you have babysitting availability, I would love, I'll pay you above market rate. Anyhow, Brendan, that's been Brendan Carr is a dummy. America's favorite podcast, when the podcast.

• 1:13:20Speaker 09

  Brendan Carr is a dummy.

• 1:13:28Speaker 03

  I will say, we have the other Brendan Carr's dummy theme song that we love, that I have asked that we license. And there's a big debate on our team because it's obviously AI-made.

• 1:13:39Speaker 02

  True.

• 1:13:40Speaker 03

  You want to have it like a real emotional breakdown. There's a hot Brendan Carr is a dummy theme song.

• 1:13:46Speaker 02

  Yeah.

• 1:13:49Speaker 03

  I feel nothing about it. I think we should play it again.

• 1:13:52Speaker 02

  All right. I have a lightning round item for you, Neil.

• 1:13:55Speaker 03

  Yeah.

• 1:13:55Speaker 02

  This is a segment that we're going to do one of because I'm leaving for a while. It's called Numbers Are Lies and everyone should stop it. As you may recall, I had a long obsession with my running thesis that everybody's view count is nonsense and that views don't matter and they're all lies and you should never believe them all. I had a similar experience this week with two companies, OpenAI and Google, both declaring that they have reached a billion users. So Sundar Pichai tweeted this week that Google Gemini now has a billion users. Big, huge, incredible number. Last week, OpenAI kind of buried in a blog post that ChatGPT had hit a billion users.

• 1:14:37Speaker 02

  But I went, found like, reported that out, managed to confirm that like, yes, they have in fact hit a billion users. This has started kind of a firestorm all over the internet because, does it, does it make like emotional sense to you that a billion people use Gemini every month? Like, does it, does it feel popular in that sense?

• 1:14:59Speaker 03

  It just depends on how US-centric your worldview is.

• 1:15:03Speaker 02

  Okay. Explain.

• 1:15:05Speaker 03

  Because Google is just available in many, many, many more countries than ChatGPT.

• 1:15:09Speaker 02

  Yes. I mean, that's definitely true. But I think people's immediate reaction to this, which I think is largely correct and also just really interesting, was there is no way Gemini is as popular as ChatGPT. It's just impossible. Even their numbers are different. Google reported a billion monthly users. ChatGPT apparently hit that number a while ago. They wouldn't tell me more specifically than a while ago. But they hit a million, a billion weekly users a couple of weeks ago in July. So monthly versus weekly is meaningful, but I would say historically speaking, not like orders of magnitude different necessarily. But it just, it feels wrong to people in this space

• 1:15:52Speaker 02

  that a billion people use both of these products. And so everybody starts essentially accusing Google of cheating. There's this, because like, a lot of these people come from, one, one way you might use Gemini is by owning an Android phone, right? If you have a recent Android phone, you have used Gemini by accident. It just happens. It happens all the time and you count as a user. Gemini is in Chrome constantly asking you to do stuff. There's the Gemini app. So like, Google has gone way out of its way to say that, no, the Gemini that pops up to help you do stuff in Gmail does not count, but an Android phone does. And on the one hand, that's fair enough, right? Like, you're using the assistant as it is intended to be used. But everybody is like, this doesn't count.

• 1:16:36Speaker 02

  All of these billion people had to go get ChatGPT and use it on purpose. You're cheating.

• 1:16:43Speaker 02

  And like, to me, it's like, well, kind of, right? Like, they certainly didn't have to work as hard, but you're also describing all of Google's strategic advantage in this space all at once.

• 1:16:52Speaker 03

  Yeah, Google's distribution advantage was a huge advantage, is not cheating.

• 1:16:56Speaker 02

  Exactly. And yeah, so I have just been, I have been so caught up in this that like, essentially what Google's saying is a lot of people on Android phones.

• 1:17:05Speaker 02

  Which is true, but it also makes Gemini really popular. Like, it, both of those things are true at the same time. And I think, you know, we talked a lot about Demis Hassabis and Jeff Dean and this sort of shakeup in the AI industry. You did a decoder this week. The headline was basically like, does Google even want to win in AI? And this, this feels right next to that to me, that it's like, actually, Google isn't even playing the same game that OpenAI is playing. And I think that drives everybody crazy who like, really wants to watch the, the model benchmarks compete with each other. And Google's just over here being like, buddies, we got the third best model and Android. And we are going to win.

• 1:17:44Speaker 03

  Yeah, and we have everyone's training data and we own YouTube.

• 1:17:46Speaker 02

  Yeah.

• 1:17:48Speaker 03

  Uh, Bloomberg ran a piece this week that's just like, Google is killing the web. Right? Because there's like, the search traffic has gone down. And one, it's nice to be right. It's nice to be validated in this way.

• 1:17:59Speaker 02

  Did they name you specifically?

• 1:18:00Speaker 03

  I actually, I didn't read it. I just saw the headline. I was like, well, here we go. Like, at this point, this headline exists every single day.

• 1:18:05Speaker 02

  Yes.

• 1:18:07Speaker 03

  And it is just abundantly clear to me that Google has moved on from wanting to fight the same fights as everybody else. Right? Like, publishers can be mad at Google all day and all night. They've got YouTube. And eventually YouTube creators will be mad at Google all day and all night, but where are you gonna go? Which is what they basically said to publishers for 500 years.

• 1:18:28Speaker 02

  Yep.

• 1:18:28Speaker 03

  And ChatGPT can say they have more users. Google's like, we have Android. And this is just how Google works. It's probably the reason Google doesn't always appear to have its foot on the gas. Right? And they don't feel that competitive pressure in the same way. But they've also won.

• 1:18:44Speaker 02

  Right. They don't have competition in the same way. It's, it's not, OpenAI, ChatGPT is not competitive with Google in the way that I think OpenAI would like you to think that it is.

• 1:18:54Speaker 03

  Yeah. I mean, again, how long ago was it when Satya Nadella said he was gonna make Google dance? On Decoder? No one danced.

• 1:19:04Speaker 02

  Nope.

• 1:19:05Speaker 03

  There was, there was, I mean, there was some dancing in the sense that like, all the executives are different, but not in the sense that like, one point of market share moved.

• 1:19:13Speaker 02

  Yeah. Yeah.

• 1:19:15Speaker 03

  I'll actually connect this to my lightning round item, which I hinted at earlier. YouTube has made it vastly harder to get into the partner program and make money on YouTube. So we've said this over and over and over again. The vast majority of creators do not make their money from the platforms. Wait, they make their money from brand deals. The platforms are notorious for not really paying anyone for content. The gold standard of paying creators for content is YouTube. And the YouTube Partner Program and the rev share. And you can have a really big debate about if YouTube started today, would they have the same payment terms to creators? Because when they started out early, it was a little bit more idealistic at the time. It wasn't clear that everyone would win in the same way, or that they'd be so dominant in the same way.

• 1:19:58Speaker 03

  And so the, the partner program has been historically quite generous compared to what TikTok has, or what Instagram has, which is nothing.

• 1:20:05Speaker 02

  I was gonna say, not, not actually generous, but certainly relatively generous.

• 1:20:09Speaker 03

  Yeah, against that floor. Yeah. Instagram is nothing. There's some amount on TikTok. And then YouTube is like, compared to that, the most generous any platform has ever been.

• 1:20:17Speaker 02

  Yeah.

• 1:20:18Speaker 03

  Until now. Right? Because there's a surge of creators and YouTube is paying out more money from its bot to more creators. And so they've just made it harder to get into the partner program. So you now need at least 1,000 subscribers and 8,000 qualified watch hours over the past year. These are almost impossible numbers for a small creator to hit at the gate.

• 1:20:40Speaker 02

  Yeah, 8,000 watch hours is a, is a big number.

• 1:20:42Speaker 03

  This, this is a double. Or, this is my favorite one, 20 million Shorts views in the last 90 days.

• 1:20:50Speaker 02

  Good lord.

• 1:20:52Speaker 03

  That is, the only way to do that as a new creator is to pump so much garbage into the feed that you collect as many views as possible.

• 1:20:59Speaker 02

  Yeah, that's like hundreds of thousands of Shorts views every single day, always, forever.

• 1:21:05Speaker 03

  Just 90 days to get into the program.

• 1:21:07Speaker 02

  No, but I mean, but then, but then you have to, you have to maintain a super high number like that, usually to stay in the program.

• 1:21:12Speaker 03

  You have to maintain 10 million Shorts views over 90 days to earn money from the Shorts creators pool, which is a different pool of money, but if you don't hit that, you don't fall out of the program.

• 1:21:22Speaker 02

  Okay.

• 1:21:23Speaker 03

  All, all I'm saying is, if you're a new creator, getting into the money now basically requires you to front a bunch of money. Right? You have to, you have to make a bunch of videos for extremely free for quite a while to hit any of these numbers. 8,000 watch hours is, is a lot. Right? If you're a new creator, getting to 1,000 subscribers is already challenging. 8,000 watch hours is even more challenging. You, to do the math, every one of those people has to watch you for eight hours.

• 1:21:52Speaker 02

  Yeah.

• 1:21:54Speaker 03

  Uh, it, there's just something here that suggests to me, YouTube is not just making it harder to make money with them. They, they're trying to radically shrink the amount of money they actually pay out, even though the amount of content coming out of the platform is going up. And there's a lot of analysis here that says this is because of an influx of AI content, right? Under the previous requirements, you could probably just flood the platform with a bunch of AI content, hit the previous numbers and go.

• 1:22:20Speaker 02

  And that has in fact become like a noted public strategy of a lot of creators.

• 1:22:25Speaker 03

  This is an arbitrage, arbitrage opportunity all day long. And by the way, the number of people automating Instagram in this way is like, off the charts.

• 1:22:31Speaker 02

  Oh, yeah.

• 1:22:32Speaker 03

  Everywhere, all the time. But Instagram doesn't pay out any money, so Instagram doesn't care. YouTube pays out money, so they have to care. And their response is to make it vastly harder to make money on YouTube. Now, existing creators aren't gonna be affected by this. Like I say, every creator gets their wings when they make a video about how mad at YouTube they are, and they all do it. They all turn into beautiful business butterflies when they make the video about how YouTube is a monster that they can't get away from. But something is gonna come here. Something is gonna happen when a cohort of new creators shows up to make money on YouTube and they discover that it's effectively impossible. And actually, what they should do is go get brand deals.

• 1:23:12Speaker 02

  Right.

• 1:23:13Speaker 03

  Right? The, the pressure to go get brand deals and be a niche creator is gonna go up because the actual programmatic money you can get from YouTube is now harder to get.

• 1:23:21Speaker 02

  Aren't we kind of already there, though?

• 1:23:22Speaker 03

  That's, that's gonna turn YouTube into Instagram in a more meaningful way.

• 1:23:26Speaker 02

  That's interesting. I guess that's true. And then it becomes, it actually rests some of the control away from YouTube. Because a big part of YouTube's appeal for a long time has been, it is a place you can actually make money, right? Like, we've, we've talked about this many times on the show. A lot of people use a lot of platforms, but they feel like YouTube is their most important one because it is the thing that pays the money. And, and even if it's not all of your money, that thing is still meaningful. And I, I think you're right, that even, because one of the pieces of feedback I've seen to this all week is a bunch of people being like, well, I don't care. I don't make any money from like, 10 million Shorts views, like that and a nickel will buy you a soda. You know what I mean? But I think we're, we're in a position now where like, it actually, that may be YouTube testing the where are you gonna go theory in a, in a pretty powerful way.

• 1:24:11Speaker 03

  Right. And it's, and it's right next to eventually a bunch of creators are gonna find out that Google trained on their content and they're gonna be unhappy about it because they will want payment. These platforms just don't pay creators. They don't pay creators enough and it's turning everything into advertising. And I can rant and rave about that, like I usually do, but everyone's heard it before. What you got, David?

• 1:24:31Speaker 02

  Okay, I have a brief Ramagathan update for you.

• 1:24:34Speaker 03

  Oh, nice.

• 1:24:34Speaker 02

  Things are bad.

• 1:24:37Speaker 02

  They, prices continue to go up. We, we talked about the Pixel 11 prices on the, the livestream on Wednesday. Everybody should go watch that. But there was a really interesting bit of data from a, a research institution called Counterpoint about cheap phones. One of the things we've talked about a bunch on the show is the possibility that the market for inexpensive phones is essentially going to evaporate because you just can't make cheap phones anymore. And what Counterpoint found is that sales of phones under $100, uh, which are crucially important to lots of people all, all around the world, um, dropped 64% year over year in, in the second quarter of this year.

• 1:25:17Speaker 02

  And essentially, the, the main thing that happened is everybody but Samsung and Motorola just can't be in the market anymore. So the, the others portion of the, the chart here, it's basically like Samsung, Motorola, everybody else. The everybody else portion of it dropped by about half because there, there are just aren't companies that can afford to float these things anymore. If you're Samsung and Motorola, you can shrink your margins, you can make your money somewhere else, you can do things to stay in the cheap phone market. But if you are a manufacturer of cheap phones who mostly makes cheap phones, there is essentially not a path forward for you at this moment in time. It's, it's just, it's bad on so many levels, but we're at a point now where like, if you,

• 1:26:02Speaker 02

  if you make a phone that costs $100, most of that sales price for not, not just like material price, most of that sales price is going to be eaten up by even the like, meager amount of RAM in the phone, if you can get your hands on it. So like, this, this is an industry and a price tier of gadgets in general. We have wondered if it would survive this, and increasingly, the answer just looks like no.

• 1:26:26Speaker 03

  Yeah, I think it's gone.

• 1:26:29Speaker 03

  And by the end of the year, I think these are gone.

• 1:26:31Speaker 02

  Yeah.

• 1:26:32Speaker 03

  I think there's probably still some stock on some shelves, but none of these companies are motivated to go figure this out right now.

• 1:26:37Speaker 02

  No. And it's, and it's even, it's hard to even like, be mad at them about that. It's just, it's a horrible business idea to go spend $70 on RAM and try to charge $90 for your phone. Don't do that. It's called bad business. But it's bad out there, dude.

• 1:26:52Speaker 03

  If anyone wants to report out a story about like, weed dealers and the crisis in burner phones.

• 1:27:01Speaker 03

  Just saying.

• 1:27:02Speaker 02

  That's pretty good.

• 1:27:03Speaker 03

  It's probably something.

• 1:27:04Speaker 02

  Yeah.

• 1:27:04Speaker 03

  You're a young reporter on the make. Give me a call.

• 1:27:07Speaker 02

  The thing like at the counter at CVS is now $300.

• 1:27:10Speaker 03

  Those like blue phones. What are you gonna do? And you can't get a burner?

• 1:27:14Speaker 02

  It's pretty good.

• 1:27:16Speaker 03

  There's a lot.

• 1:27:16Speaker 02

  All right, what's next?

• 1:27:17Speaker 03

  I guarantee you there's a young reporter in New York City who can get that story.

• 1:27:20Speaker 02

  Oh, absolutely. Go buy every burner in New York City. Neil, I will pay for them personally.

• 1:27:26Speaker 03

  I got a good one. Lauren Finer reported this one this week. She interviewed the CEO of Flock, which makes the licensed white cameras.

• 1:27:32Speaker 02

  Yeah, everybody's least favorite company on earth right now, I would say.

• 1:27:36Speaker 03

  And so Flock has the same problem as the AI companies, the data center companies. They need to, they need to go make everyone love them. Unlike the AI companies, the data center companies, they can't just like show up in your local community and be like, we'll, we'll buy your electricity for you. Just like, let us have a data center. They're not flush with cash the same way. So the CEO of Flock said to Lauren, we thought that there'd be some regulation around surveillance that would tell us what to do. And we were wrong about that. So Garrett Langley.

• 1:28:10Speaker 02

  With, by the way, with the, the parenthetical implied there being, so we went and did a bunch of surveillance.

• 1:28:16Speaker 03

  Yeah.

• 1:28:18Speaker 02

  So we went super hard at surveillance because no one was here to stop us.

• 1:28:21Speaker 03

  Garrett Langley to Lauren, he's seen firsthand how slowly the regulatory process can move. It's such an obvious way to curb abuse, he said of the tool that lets agency administrators require case numbers before searching the worldwide surveillance track that he built. And I said, fine. You know what? I think that's right. We got it wrong. We should make case numbers a requirement. This is, this is, he's gonna make people love him this way.

• 1:28:47Speaker 02

  This is so bonkers, dude. I think the whole flock thing is, is absurd. Like, the idea that a company came along that made Ring look sort of benevolent and neighborly is bananas to me. And this is what Flock has done. They're like, look, we have horrible dystopia cameras. This is fine, right? And everybody's like, no, thank you. This doesn't even like tell me when there's a package at my house. It just sucks. And so to come in and be like, oh, we didn't do anything because we were waiting for somebody to tell us to stop. And so we never did is so, like, even, even his apology is so grossly saying the quiet part loud that it just really bums me out.

• 1:29:29Speaker 03

  He's like, American democracy didn't stop us, but you all hate us.

• 1:29:32Speaker 02

  Right.

• 1:29:33Speaker 03

  So we've added some fields to the database.

• 1:29:36Speaker 02

  You want an argument against capitalism? Like, there's a damn good one. There it is right there.

• 1:29:40Speaker 03

  The ACLU released an analysis of the changes and it says, quote, "They appear more focused on addressing a perceived PR problem than the significant harms these products create."

• 1:29:50Speaker 02

  Yeah.

• 1:29:51Speaker 03

  Very good. Langley has, he's been a fighter in the past, but everyone, I mean, he's called D-Flock a terroristic group.

• 1:30:01Speaker 02

  Woof, really?

• 1:30:01Speaker 03

  And they just have a database of where flock cameras are. So he's, he's walking back. He's gonna be a real cuddle bug now. He got a bunch of things wrong. But I had a friend text me and say, you're all about data centers being the, the thing that realigns left and right in America and brings people together in midterms. Flock cameras and surveillance is the other one. Keep an eye on it. And I think that might be true. There's a political article about it this week. And the headline was something like, "Work boots and Subarus come together in a protest flock." It's a great headline.

• 1:30:32Speaker 02

  That is good. It is really, Flock, I feel like just in the last couple of weeks, has really broken a certain kind of containment. And I feel like everyone is aware of what it is doing now, in a way that I think, I think your friend might be right.

• 1:30:44Speaker 03

  Letting the government track your movements without permission is just, what are you doing? That, that, that doesn't seem right in any way, shape or form.

• 1:30:52Speaker 02

  Well said. All right. On that note, I got a baby to have. We need to get out of here. Neil, who's on Decoder next week?

• 1:30:59Speaker 03

  Decoder next week is Ben Smith, the editor in chief of Semaphore. They're launching a big new thing called Silicon Valley in the world. He keeps describing their event series as Davos with follow up questions, which is very good. It's very good. We talked about publishing. We talked about the army of teenagers working for free on platforms. Semaphore uniquely, big website, spends a lot of money, one of the few without a paywall. So we just talked about the, I've had a lot of media people on Decoder recently.

• 1:31:28Speaker 02

  Ben also not short of opinions about the future of media.

• 1:31:31Speaker 03

  Ben is absolutely not short of opinions. That was a really fun conversation.

• 1:31:34Speaker 02

  It's a good hang. Looking forward to that one. Also, go listen to the Decoder from this Thursday with Hayden about the Google stuff and AI. Hayden's done a lot of really good reporting on what's going on with Google's AI teams. She has really good perspective. That's a good episode. All right, we need to get out of here. Neil, you're, you're, you're alone for six weeks. Are you, are you gonna, are you gonna be okay? Jake's gonna be doing a lot of stuff. We have a really fun run of guest hosts. Don't worry, it won't just be Neil.

• 1:32:00Speaker 03

  I think I've immediately answered that question. He's like, Jake will fix it.

• 1:32:03Speaker 02

  Jake, Jake will be here to keep Neil I company.

• 1:32:08Speaker 03

  That'll be great.

• 1:32:08Speaker 02

  I'm, I'm gonna, yeah, it was super fun last summer. It's insane that it was last summer that I did this already.

• 1:32:13Speaker 03

  I was also gone last summer.

• 1:32:15Speaker 02

  Yeah, you were also gone last summer. We had Hot Girl Vergecast Summer. It went fabulously. And now we're gonna have, I don't know what this one's gonna be called.

• 1:32:23Speaker 03

  I'm just gonna crash out with other people who are like, what are you doing? People listen to this show?

• 1:32:29Speaker 02

  I will, I will tell you, if they launch a foldable iPhone, there is a 100% chance I am going to bully my way onto that live stream with a baby.

• 1:32:37Speaker 03

  It's gonna be great.

• 1:32:38Speaker 02

  It's gonna be rad. I'm excited about it. Until then, a reminder, as always, the best way to get in touch with us about all of this and everything, if you have knocked down a flock camera recently and you want to tell us about it, email us, [email protected], call the hotline, 866-VERGE11. And the best thing you can do to support all of this is to subscribe to The Verge. Theverge.com/subscribe. It gets you all of our newsletters, all of our coverage of Google and Flock and insane things that Mark Zuckerberg says and everything else. Theverge.com/subscribe. Thank you in advance. The Vergecast is a Verge production and part of the Vox Media Podcast Network. The show is produced by Josh Kahas, Eric Gomez, Brandon Kiefer, and Travis Larchuk. We will see you next week. I'm gonna be gone, but we've recorded a bunch of stuff. So things that we have made already are going to keep appearing. We actually have a couple of really

• 1:33:23Speaker 02

  fun episodes coming up at the beginning of next week that I think folks are really gonna like. I am just going to sit and listen to them loudly in my living room. And I'm very excited about it.

• 1:33:33Speaker 03

  We're gonna miss you, buddy.

• 1:33:34Speaker 02

  I know. I'll miss you too. Give me, give me a good one so I can leave.

• 1:33:38Speaker 03

  Rock and roll.

• 1:33:42Speaker 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.

• 1:34:19Speaker 10

  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.

• 1:34:50Speaker 11

  This week, it pays extra to be a Milo's Rewards member. Through October 14, members get $10 Milo's money when you spend $200 or more on an eligible purchase during Milo's Money Days. Also using your Milo's Rewards credit card? Even better. That's an additional $20 Milo's money. Use it to save on your next Lowe's haul. Exclusions, restrictions, and more terms apply. See lowe's.com/milosmoneydays for details. Milo's money expires 30 days after issuance. Subject to change. Credit offer subject to credit approval.