The real story of the iPhone 18 Pro's camera

30:15I have not done this yet. I think it is quite smart and it is a very Apple thing to do at the same time where they're like, we're supporting SynthID, which is Google's Google's version of having this invisible watermark, you know, in your image, because the rest of the industry is starting to support SynthID. But we're also running up against that XKCD meme again, where it's like, there are 15 competing standards. Let's make one that universally, you know. But Apple, it's not universal. You have to use the iPhone. And Apple loves living in a world where we don't think about you at all kind of situation.

30:53You know, and so they're like, we're going to do it our way. I think their way is actually pretty smart because the way that they talk about it as a digital negative. And the reason that they talk about it that way is because when film photographers used to take photos and like have them in galleries or whatever, the only way to really prove that you were the photographer is that you had the negative. Right. Like you took that negative, you printed it, whatever. If you have the negative, you're the one that took the photo. That's like very obvious. And it's a similar thing where it's like if it is on your smartphone, which is a thing that is encrypted by your facial data, you know, only you can unlock your phone.

31:28only you can get to the setting where you can tap it and show what came off the sensor that actually in my opinion makes a lot of sense um that is really interesting you know it's a little it's a lot harder to break like c2pa for example it's a good idea in concept where it's like we're going to embed it in the metadata but you know what breaks that a screenshot yeah and this is not breakable by a screenshot. So I think it's a great way for Apple to do it. But it also lives in the world where iPhones are the only things that exist. And I do think that Apple wants to convince people over a long period of time that the iPhone is going to be the only camera that exists, you know, because to your point, Allison, it's a very narrow use case.

32:15This is really helpful for like photo journalists, right? You turn it on when you want to turn it on. If you're out documenting an important event, even in Apple's keynote, they show that the way that they show the Apple reference image is a photo of a politician like doing some movement, you know. And so I think that if you're a photojournalist and you're taking photos with your iPhone, which is the ideal future that Apple wants because they don't see a future where dedicated cameras exist anymore because they convince you that the iPhone is the only video camera you need, the only photography camera you need, then it makes a lot of sense.

32:49But as far as this being a standard that the rest of the industry can use, it's just not. So, you know, I think that's why they had to also adopt Synth ID. Well, and Allison, you mentioned this in your review, like it doubles the file size. Yes. Right. It's a dedicated mode. It doubles the file size. So this is like simply is not a solution for the everyday photo. For sure. This is a proactive. Yes. I need to prove that this is authentic. Which actually, I think, in that sense, is pretty interesting. Yeah. It's an interesting way to think about it. I think, unfortunately, most of the images that get manipulated online, that are either generated with AI or manipulated with AI, were not images that people sought out to make sure people knew the source of or the authenticity of.

33:40however um going out and saying i'm gonna sign all these images for this event is an interesting way to think about it for sure yeah it feels a little bit like a proof of concept to me you know it's sort of like a trial run at like here we're gonna introduce this into the world it's gonna work on this one particular phone um that that is not going to be in the hands of many people you know for a while um and it makes more sense to me to think of it that way because otherwise you're like i have to go into a different mode every time i want to prove a photo is real like that that's wild i want to talk about phone cameras more generally to wrap this up which is the i think the headline feature if you were to ask what the the biggest upgrade to the iphone 18 pro is right like it's variable aperture right that is the that's what that's what apple wants see it.

34:36That is yes. That's actually it. And then in reality, it's actually there's all of these like this world of smaller but impactful camera upgrades underneath. Both of you are big smartphone camera people. I'm curious in what do you actually want to see next? Right. You have carte blanche to change the iPhone camera for the next generation. Like, where is your head at? Is it Is it that we need hardware improvements? Is it more of this software stuff? Because I think all of these companies have this philosophy and they're kind of just chipping away at it. But what actually is it that you are looking for to make these cameras better?

35:16Allison, I'm curious where your head is at here. Okay, how much time do we have? I want a one-inch sensor. One-inch mean camera sensor. They exist. They're in the world. I know it changes everything about how Apple has made cameras and upgraded. The telephoto camera is much better than it was, and I go to it a lot. But just even put a bigger sensor behind that one, you know, why not? I love it. Um, maybe, uh, I, I played with, um, Vivo's, uh, attachment telephoto lenses like earlier this year. They are so fun. They are little tiny, like actual, like telephoto, um, they call them like converters, I think.

36:10And you, you have like a special case that's on the phone. You kind of just attach the lenses to the front of it. I thought it was the goofiest looking thing in the world. And I loved it so much. By the end of the day, I was like, I want every phone camera to have this. So I think I do want hardware. I want like a grip, you know, there are third party options. And maybe there's something good out there. I haven't tried, but I just think of like the Xiaomi photography kit and how that just feels kind of like a sleek, easy to apply kind of solution. And it's just as easy to be like, you know what?

36:53I'm not taking serious photos right now. I'm going to leave all that behind and just have my phone. So that's it. Apple could get working on that right now. I think that is so interesting, though, that like at the end of the day, it's like make the hardware better, right? But hardware is still the biggest single factor in or cheat code to like just level this up. Just level this up. I mean, I look at the photos that come out of those one-inch sensors on some phone cameras and, oh, man, it's night and day. They look very, very good. David, same question to you, but I am going to last second change this on you.

37:27If it's not going to be hardware, right, if we're in the realm of software, which I feel like is where your brain often is toying with things. where does that change need to happen to to improve things going forward where i think that apple is falling a little bit behind is in low light at the moment um they can still you know they get a lot of flack for making dark low light photos look like daytime and there are ways that they can change that there are you know you can just drop the exposure in the software if you want but google years ago got really good at computational um computational astrophotography and i know this is a very specific like use case and it doesn't have to be astrophotography but the one area that i felt was a little bit lagging behind was the long exposure night photography modes um because there's like there's like a low light mode built into the iphone and handheld you can do up to three seconds but it is quite noisy if you do that and you have to have a tripod and google has made so many advancements in being able to do handheld astro modes or like super good low light modes that i think if apple spent a little bit more time focusing on that or you know i want to say making the sensor bigger because then you don't have to do as much of that or as in this year they made the aperture wider right and so that helps a little bit with those lowlights yes and that that does help um i think like i don't hate the idea of a variable aperture i think that that's what people are sort of getting mixed up about is like yes i don't think it's incredibly useful where we are right now but everyone wants a 1.0 type sensor like allison said and sony's biggest problem because sony is the company that is using these 1.0 type sensors the most is that they don't have a variable aperture and because of that almost no photo you take is in focus, everything is blown out.

39:28For some reason, they have not put like any work into the software experience of actually like making those take good photos, even though they have the hardware for it. So I think something that everyone has said basically in these reviews is in the future, when Apple makes a 1.0 type sensor in their phone, the variable aperture is going to be really helpful, you know, and like what they're doing and they're engineering with it of like it automatically changing without you having to change it, but you can also manually change it yourself. that's super good i see this as like a test bed that they're currently like playing with because next whether it's next year with the iphone the rumored iphone 20 or you know an iphone ultra or something like that and we could get in and i i know we're kind of like short on time but i just want to mention that the iphone handoff feature that they announced where that lets you use one phone number for multiple iphones in my opinion is because people want to use the iPhone duo, but the cameras are just way worse.

40:27And so now we are entering this world where it's like, just like you should own an iPad and a MacBook and an iPhone, you should also own an iPhone pro as your camera and a duo as your fun phone. Oh my gosh. I mean, I have, I've seen people on the subway carrying two iPhones, but it's usually an old iPhone. It's like a work one. It's not, or a work one. I've seen somebody with like an iPhone, a busted up iPhone five and like a brand new like modern iphone because dude the iphone 5 took some nice photos i mean yeah because of less competition photography it lets shadows like there you're not getting the handoff but i see what you're saying i mean this is it i want i want the emails from the people who are buying the duo and the pro so they can well i know that there's a lot of people that are buying the duo and are just not upgrading their 17 pro and that that is who those people are that that iphone handoff features for those people.

41:22No, this is how we bring back the iPhone mini. It can be your fun little phone and then you have a big phone when you have to have a big phone. And the battery life doesn't matter when you only use it for like 20 % of the day. Exactly. I think this has been one of the most fascinating years for the iPhone camera. I agree. Allison, your review, it's up on theverge.com. It's like mostly an iPhone camera review. Everyone should check it out. I think it's super interesting. Allison, you're going to leave us for a minute. You'll be back for the lightning round. We've got to take a break. When we return, Hayden Field is going to join us to talk about the last couple weeks of AI safety drama.

42:04Support for the show comes from AWS AI. AWS AI is powering transformation across the world's leading organizations, moving AI from possibility to production at scale. Like how AWS AI helps Formula One turn race day data into real-time insights that bring fans closer to the action. Or how it helps Booking.com personalize every trip for travelers around the world. AWS AI is even transforming how things get built, helping Siemens power the production of smarter cars. Across industries, AWS AI is helping businesses innovate faster, lead with confidence, and transform what's possible today. AWS AI is how.

42:47Support for the show comes from Amazon Prime. Some of us are blessed with a natural ability to do gift giving right. You probably have a few in your life. People who can make a mental ledger of what their loved ones like and buy things here and there throughout the year. My mom has always been the best gift giver in my life. She could even make the smallest gift feel incredibly special because you knew that she had really thought about that person receiving it. It doesn't have to be the biggest or the most expensive gift. It just has to show someone you were thinking about them. Well, for the rest of us, here's a tip.

43:21Prime Big Deal Days is coming October 6th and 7th. That means that for two straight days, you can get deals on those kitchen gadgets for your dad, electronics for your brother, fitness equipment for your bestie, toys for the cousins, and so much more. There's new deals dropping every day so you can become one of those expert gift givers that we all know. And it's all happening with plenty of lead time into the holidays. so no more scrambling to come up with something last minute. The hardest part will be finding a good hiding place for it all. Mark your calendars. Shop new deals every day at this prime big deal days, October 6th and 7th.

44:02This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required, compatibility and availability varies 18+.

44:30All right, we're back. The Verge's senior AI reporter, Hayden Field, is here with us for the second time this week. Hayden, how's it going?

44:37Allison Johnson:It's going great. Thanks for having me back. It's been, I believe, two days since you were on this show. That was six, quote, instances of misaligned behavior ago. This is right. Like before we recorded this opening, I was like, ah, there were some more hacks are bad guys. Yeah. I went from the Apple event to Arizona and I was sort of out of service and I came back and then there were so many instances of misaligned behavior. It was it was a lot to catch up on. And I feel like I need you to explain what is going on, Hayden. Next time that I see you guys, I'm going to be like, oh, it's been three misalignment incidents since I last saw you.

45:14Allison Johnson:We're going to stop talking in time. Yeah. It's been so long since I've seen you. Yeah. It doesn't sound so bad when you put it that way. It's just it's a cheeky little misalignment incident. Anything could have happened in there. OK, so it's been a pretty explosive couple of weeks for AI safety in particular. and I wanted to get back into what happened, why it blew up today, and what's going to happen next, if anything. So this all kind of started last week when an anthropic researcher quit his job and posted on X that he thinks AI will kill everyone within the decade. A bunch of AI researchers have said the same thing.

45:55Many people have pushed back on the narrative, but it went mainstream. It caused this sort of full-scale freakout, and you and our producer, Travis, were on the show earlier this week to talk about that, but it's still going. It's been a big conversation. I think this is really the first time that I can remember that it's sort of broken to the mainstream. I actually just want to start by quoting what that AI researcher said because I think it's pretty striking to hear. So this is from Jacob Coxon who spent three years doing pre-training research at OpenAI and most recently Anthropic. So this is across a bunch of tweets or ex posts, whatever they are now.

46:36He wrote, OpenAI and Anthropic are racing straight to self-improving superintelligence and gambling with our lives. Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. The people building AI earnestly believe that this could kill us all by the end of the decade. This is not a marketing stunt. No other human activity poses this level of danger. Hayden, why did this message hit this time? Because that sounds very scary. But to me, it also sounds like things that a lot of these researchers have been saying for a while now.

47:18Allison Johnson:Yeah, that's exactly what's interesting about this to me. Nothing about this was new to me. I've been hearing people say this for years. Right. You know, there's like a lot of discourse about this right now. And I think it's because in July, when a lot of these large-scale cyber attacks happened, starting with OpenAI and Hugging Face, I think that kind of got people talking about AI systems that go off on their own and go rogue. I've heard some criticism of that term because it implies that, you know, these systems weren't built to go rogue or that the companies aren't at fault for it. So, you know, keep that in mind, obviously.

47:58Allison Johnson:Going rogue here means doing something crazy and bad rather than, oh, the companies aren't at fault for it. But yeah, I mean, I think that these cyber attacks that we heard of starting in July, especially because the companies, OpenAI, wasn't aware of the hugging face incident for weeks. And actually, if you count the groundwork that was laid before that, months. So I think all this stuff going on under these companies' noses without them knowing, despite them kind of parroting a lot of safety talk and saying, hey, like, put us in charge of your enterprise, you know, cyber safety is a little bit ironic.

48:36Allison Johnson:And that is what got people talking for months. And then this happening, you know, it was already in the zeitgeist. So I think it was easier for people to say, oh, well, wait, we've been hearing about all these rogue systems. And And now someone that was working on these systems is saying that they have really horrifying, you know, implications. Maybe we should sit up and take notice. Yeah, there's a couple of things that I wanted to ask you about, Hayden, because I've talked to some of my friends that are not as plugged in as we are here at The Verge. And they all kind of hate Sam Altman and the AI sphere so much that they are convinced that all of this is just marketing hype.

49:18And I think that there is truth to that, probably. But at the same time, if you have these agents that are doing things autonomously that you cannot control, that is a real problem because you don't know exactly what they're going to be doing. You know, right now it's just finding the answer sheet to a test, but later it could be something else. And now OpenAI is saying, oh, actually, they did more than we thought. And every week it's they're doing more than they thought they did more than they thought. And the question I have for you is, if you have a dog and it bites someone's face off, are you not responsible for the dog?

49:51Because it seems silly to me that these systems can go do these crazy things. They can hack companies. If a human went and hacked Hugging Face, they would probably be prosecuted. So why is it that OpenAI and Anthropic are not being held liable? And instead, as we talked a little bit about off the show, they're just saying, oh, it's an oopsie. Like, it's just a little it's a little cutesy oopsie. Like, how is that possible? Why is that the conversation right now?

50:16Allison Johnson:Totally. So, yeah, I think that's a great question. They absolutely should be getting in trouble. And I think that the dog analogy is one I hear a lot. It's like, yeah, these companies built these systems. They're responsible for them. And the fact that they are saying, oh, we're in control of this stuff. Trust us with your enterprise AI. You know, trust us with these large scale models that are, you know, trained to help with or flag vulnerabilities in cybersecurity on large infrastructure. Oh, but also we're losing control of our systems and they're like going all over the internet and seeding random stuff and hacking companies and handling personal user data all without us knowing about it.

51:02Allison Johnson:yeah, they should be responsible. And I think that part of the language around like rogue agents and oh, they escaped and this kind of passive language, you know, people have been telling me that's a huge way that these companies are kind of trying to escape responsibility. Responsibility. Exactly. So I think that's huge here. They should be facing responsibility and they are not. And unfortunately, it doesn't look like regulation is nearby because of the president's posts on social media the other day saying all we need for AI regulation is a high IQ president. So yeah, there's a lot going on there.

51:44Is that a view that is widely shared in the AI industry?

51:47Allison Johnson:No. I was speaking with a lot of sources inside labs, AI VCs, you know, people at tech companies, And they were all, no matter how anti-regulation they were, they were pretty anti that statement. As in, that went even too far for some of the anti-regulation people. They were like, whoa, well, we need more than that. So that really tells you something. I think he was motivated partly by trying to get people to be pro data center because he talks about that later in the post. But still, yeah. As for the marketing hype, I think there's some truth to it in that their businesses do do well when their model looks super powerful to the outside world.

52:30Allison Johnson:It's kind of like a hold me back bar fight situation, you know, like, oh, hold me back. Like, oh, it's too powerful. But I also think in this case, we should be looking at this very seriously because, one, I've been hearing AI researchers talk about their fears about this for years. So it's not anything new. It's not just a psyop of one guy being planted to say this stuff. Secondly, it's not good for their business, for AI labs to look like they don't have a handle on this technology. It's great for the overall level of power they look like they have and maybe for investments. But it's not good for their bottom line in terms of their current business.

53:18Allison Johnson:Enterprise and coding are their main revenue drivers right now. No enterprise wants to work with a company that looks like it's out of control or can't control its own technology, especially if you're paying a ton of money to use that technology behind the scenes to plug up your own cybersecurity vulnerabilities. But then also maybe it'll hack you. You know, that's not a good business model at all. And, you know, the government doesn't want an out of control system either. They want a system that they can control every single aspect of that has its own huge issues and controversies. But, yeah, it's not good for their bottom line in this way.

53:49Allison Johnson:So I think it's important to know, like, yeah, they make money when people think their systems are super powerful. But if their system looks out of control, that's not good for their clients. That seems to be the strange push and pull here, right? Because if they are saying, oh, the only way to prevent these hacks from happening to you is for you to buy our models and then use our cybersecurity tools to create the defense systems because we can't control our models. Like that is a it's such a weird way to market your product, but it also makes sense. So I understand why people would do it. It's well, there's also that's where like the specter of China comes in.

54:27Yeah. Right. Right. It's this idea that some other power that actually is malicious. Right. Like open eye hacking, hugging face. It's cheeky and cute because like, oh, we're all friends here. Right. Like both of our stock prices eventually are going to go up as a result of this technology being so powerful. theoretically yeah if a if a party that actually wants to take down banks has ai well then you do need this for the electrical right and so that is why when we start having this regulation uh discussion you have folks like trump who are like absolutely not right because for him it's it's all about winning right it's all about the stock going up up up up and about us having the best possible national security and so forth.

55:13Right. And this is like there's something there. There's something there. But it's also just sort of this kind of impossible to shoot down kind of argument. Right. Because there's it's the specter of something that currently doesn't exist as far as I'm aware, Hayden. Right.

55:30Allison Johnson:It's this issue where China is used as like a boogeyman. And I think, yeah, there are legitimate concerns about All of that. But there's also this innate, it's a strange assumption that a lot of these people make that there's no chance in even negotiating with China to come to some sort of agreement. That will never happen. It's not even worth trying. Instead of even trying that, let's do all this other crazy stuff. And so a lot of the experts I spoke with the other day were saying we've reached international agreements before for, you know, a ton of different things. Nuclear, missiles, like there's a lot of hard things we've done when it comes to international agreements.

56:18Allison Johnson:And so it's interesting that this has not even really been on the table that we know of. And instead it's more, oh, we got to race, race, race to superintelligence, you know, despite the fact that X, Y, and Z really bad things may happen. and we can't regulate it either because we just have to win the race. I'm going to read a quote. Beth Barnes, the head of Meter, which is like a third-party AI auditor, said it like this. When you've heard it from multiple different labs being like, we're the good guys and we have to beat these other irresponsible people, it's a little bit less compelling. It's a bit of a scary attitude to be like, yes, we'll be making huge decisions for the world without any kind of meaningful legitimacy or participation, but it's all right because we're good.

57:03Allison Johnson:We're unusually well-meaning. So I think that tracks for both AI labs that are competing against each other domestically and the China specter, the China boogeyman, because it's like, oh, well, we have to beat them because we're the good ones. Yet these good ones, quote, quote, said sarcastically, are like saying, oh, but also our systems are so dangerous that they're going to hack everyone and steal your money and do all these other things. So there's just a lot of conflicting narratives here and a lot of really weird power dynamics. So we should talk about the state of the freak out, because what happens is this this bubbles up into conversation.

57:38And then Dario Modi, Anthropics CEO, comes out with this blog post about where he says we need to pace the frontier, which is to say we need to very slightly slow down. Yes. It's actually he really like makes a point of it. He's like, I like he's like, Believe me, we're still going to race toward this. But like, you know, maybe like a week or so. Like it's really wild. Yeah. But he wants to he wants to pace the development of frontier models. And he lays out three steps. And the first step is putting these embedded evaluators, these third party auditors inside AI companies. And the second is having some sort of national agreements.

58:23And the third is these international agreements. And he says, each of these get progressively harder. But we're going to unilaterally, he says, agree to number one. We're going to put third-party auditors inside Anthropic. We're going to give them employee-level access. And Altman then comes out shortly thereafter and says, OpenAI is going to do this too. Hayden, you actually wrote a feature that is on TheVerge.com today about these third-party auditing firms and how they have been seeing these sort of issues coming for a long time and what they're trying to do about it. I'm curious, can you tell me what they are doing and what abilities they actually have inside these frontier model companies?

59:12Allison Johnson:Definitely. Yeah, it's been interesting because I've been tracking some of these firms for a few years now and they haven't had much access to a lot of the labs. Some of them have been working a little bit with the labs to kind of flag problematic or really concerning tendencies from their models. These are the people that kind of expose or in the past have exposed when models are cheating a lot or trying to deceive or trying to like obfuscate things in their chain of thought, which is like their mental scratch bad, quote, quote. And I know we're going to get some people saying deception, cheating.

59:50Allison Johnson:These are like anthropomorphic terms. We shouldn't use them. But I don't have a better way to put it, honestly. It's like, you know, they're not, I'm not meaning they're consciously trying to like, you know, do something crazy. It's that they are cheating, quote, quote, at a task that they were given. So stuff like that. It's really early stage, but concerning research findings that these people have found. And then they've also been pushing for more access to these labs for years because when you find something crazy that models are trying to do that could, you know, be extrapolated to be something really problematic in the future, the more power or the more influence these models have, you want to figure out where it came from.

1:00:32Allison Johnson:You want to look at the whole process from start to finish of building the model and say, okay, we can pinpoint it to this. And here's all the other ways that this could be affecting the model and its potential behavior. But these researchers haven't had much access at all so that they can't go back and pinpoint this stuff. It's like they're so limited on what they can do. And I think that was hugely exemplified by something that happened after the hugging face hack. OpenAI was really flamed for its lack of transparency over that cyber attack. And so after a while of a lot of public outcry, they said, OK, don't worry, we're going to bring in three independent auditors and let them, you know, look at everything for transparency's sake.

1:01:15Allison Johnson:They can release their own report, et cetera, et cetera. They only gave them access over six days and it was three people to analyze countless transcripts. They also couldn't go before the incident really, more than like a week, I think, or after. They were really limited in their scope on what they could look at. And they were super limited on the questions they could answer in their report. I think they could only answer seven predetermined questions that were approved by OpenAI. It couldn't go. Yeah, it was like a really. How is that an audit? That does not feel like an audit. That feels like we are telling you what to say.

1:01:53Allison Johnson:It was it was not really telling them what to say, but more, yes, they could only honestly answer certain things. So it was super limited. And so Like that was, of course, led to even more controversy. And that's when Dario came out and said, okay, guys, don't worry. We are going to give Meter or similar companies full access starting with eight weeks and then can expand from there. We're going to let them have as much time as they need. So he was kind of directly insulting OpenAI there, like in a shady way, kind of being like, hey, we're going to give them a lot of access. But the thing is, even if these companies all do a line and kind of give all these embedded evaluators a lot of access to look at their systems and interview employees, things like that, there's still other parts that they likely will hold close to the vest because it's IP.

1:02:46Allison Johnson:And so, I mean, this is a step forward. They're going to let embedded evaluators in. But it's also, you know, just another example of who's in control here. It's the companies. They're deciding to do this. You know, they could just as easily decide not to or change the scope of what those evaluators can look at once they're inside. So that's a tough part of this. It's like still at the end of the day, it's voluntary. And so I think that's going to be a problem. Yeah, I kind of want to get your take on this because you kind of have a prisoner's dilemma situation happening here, right, where everyone is saying we're going to pace the frontier.

1:03:21We're going to slow down. Don't worry, we agree. But you also have this opportunity for any of these labs to basically be like, OK, now that everyone else is slowing down, we can secretly be speeding up. Do you think that there is a lot of merit to all of these people saying because it wasn't just open AI and it wasn't just anthropic. It was also Elon Musk. It was like multiple of these labs saying, let's pace, let's slow down. But it feels to me that all of these people are trying to get ahead of each other because there's this idea that whoever wins AI, whatever that means, because it's a very nebulous idea, wins all of the money.

1:03:57Right. It's sort of like there's a jackpot to be had. Everyone's racing after it. do you think that these people that are all agreeing with each other or are actually agreeing with each other? Because it's to their benefit that the president and Jensen Wong come out and say, don't worry about it. Don't worry about it. Even Zuckerberg is coming out and say, saying, I don't want to praise the frontier. I want to speed up the frontier. And as soon as you have one of the prisoners in the prisoner's dilemma disagreeing, that breaks down the entire system. So what do you think about their statements here?

1:04:26Allison Johnson:Yeah, I think that it's absolutely a prisoner's dilemma because even if they do agree the second one of them breaks it they're all gonna race and they may all be secretly breaking it like you said at the same time the problem with this whole situation is they're just these are just tweets you know like they're just saying yeah he's right which is like a great pr move you know i mean it's like uh you know i think elon musk's post was literally just dario's right right he wasn't like yeah here's all the things we're going to do and here's how we're going to pace it. Even pacing the frontier already is super vague.

1:05:01Allison Johnson:I mean, a huge chunk of lab employees signed this letter a few months ago saying, please, please, let's pace the frontier. Let's slow things down. But even then, there's not any real specifics there. So I think, I mean, it's almost a moot point to even think about it right now because there's no reality behind this. Right now they're saying, hey, we should slow this down. okay, how, when, who, and also what's going to happen if you don't? There's no consequences right now, which I think why, yeah, we love, you know, voluntary stuff is great. Internal stuff is great, but you need external accountability as well.

1:05:42Allison Johnson:You need regulation because, you know, at the end of the day, again, who's in control here? The labs. They are deciding to slow down. OK, they can just as easily decide to speed up. So it's like I think the power dynamics here are the whole reason I became an AI reporter, because it's just crazy that like five guys are in control of this entire thing with no external pull, really. And the incentive structure of Mark Zuckerberg being very behind in the AI race and wanting to just speed ahead while everyone slows down, or even NVIDIA, who just wants to keep selling pickaxes during the gold rush.

1:06:16Like, unless there is regulation, I just don't see how any of this happens. And I think your quote of they're just tweets is probably the best way to summarize this entire situation.

1:06:27Allison Johnson:It's crazy. And the other thing about it that's funny to me is like, Anthropic has a reputation for being like the more safety minded one, right? That's something they've really painstakingly built up, this reputation. But I saw a great criticism of them the other day where, you know, they're kind of the ones that are speeding ahead in this race sometimes more than others. You know, one of their execs said the other day, you know, you can't do safety from second place. So they kind of use their safe reputation to be an excuse to push things ahead and say, oh, well, we're the ones that has to win because we are the best and we're the safest.

1:07:07Allison Johnson:which some people may agree with that who knows but the point is it's still pushing the race so i think that's the interesting part of this whole thing it's like like you said it's a prisoner's dilemma and it's just you know how much do these people trust each other who is keeping track of this externally not a lot in both situations it's such a fascinating tension and i do get the impression that there are folks even like high up at these companies who genuinely believe that the AI safety stuff is a very serious issue that needs to be dealt with. And policy is not going to happen. Congress is gridlocked.

1:07:43They're not going to go anywhere. Trump has outright said he doesn't want to regulate these companies. And so you look at what is the mechanism of accountability here, right? We're going to embed some third party AI evaluators. The mechanism of accountability at this point is basically just bad press, which is maybe a win on some level, right? Like if you get enough of this and you genuinely believe that you need regulations, then you get bad press for everyone. You hope it's not you. You hope it's the other guy. But you need these incidents to happen and you need someone to show just how bad they are.

1:08:20But you're not going to necessarily put all that out there yourself. Maybe you are. Maybe you are if you're Dario now. We'll have to see, right? He has announced this yet. I think we don't know the details. Sam Altman agreed to put in evaluators, but just said, we'll have more details later. So I'm very interested to see what becomes of this, but this feels like baby steps, right? Because, you know, Hayden, you've been tracking this. It seems like we had this freak out. We had this big conversation. Everyone said, we need to do something. And then it kind of just hit a dead end. Is that where we're at?

1:08:56Allison Johnson:Yeah. I mean, I haven't seen a lot of movement since. It seems like behind the scenes, A lot of these lab leaders are working together to establish some type of standards body in the industry to regulate them. Apparently reports say they've been meeting with government officials trying to make this happen. But at the end of the day, who's setting this up? The labs themselves yet again. So I think, you know, it's just, it's tough to watch not a lot of checks and balances here. And like you're saying, I think bad press, bad PR is the main check that's having any type of effect here, which, you know, great that we have at least that.

1:09:45Allison Johnson:But it's, I can't believe at this point in the industry that that's still the main thing we have. I mean, there's truly nothing else like this where the people in charge are like, we got to go, go, go, go, go. But also this is going to be a big problem. Like, sorry, sorry, everyone. Yeah. I'm very interested to see more from these companies as they actually detail this stuff. Whether that actually is to accountability, I think remains a wide open question. But I do think we're going to see more of these kinds of freak outs as this stuff happens. I think it is very good at a bare minimum that AI safety has sort of broken containment and people are taking it seriously.

1:10:24Hayden, thanks so much for coming on. I've got to let you go. When we're back, we're going to have the lightning round. Support for the show comes from Amazon Prime. Some of us are blessed with a natural ability to do gift giving right. You probably have a few in your life. People who can make a mental ledger of what their loved ones like and buy things here and there throughout the year. My mom has always been the best gift giver in my life. She could even make the smallest gift feel incredibly special because you knew that she had really thought about that person receiving it. It doesn't have to be the biggest or the most expensive gift.

1:11:00It just has to show someone you were thinking about them. Well, for the rest of us, here's a tip. Prime Big Deal Days is coming October 6th and 7th. That means that for two straight days, you can get deals on those kitchen gadgets for your dad, electronics for your brother, fitness equipment for your bestie, toys for the cousins, and so much more. There's new deals dropping every day so you can become one of those expert gift givers that we all know. And it's all happening with plenty of lead time into the holidays. So no more scrambling to come up with something last minute. The hardest part will be finding a good hiding place for it all.

1:11:36Mark your calendars. Shop new deals every day at this prime big deal days, October 6th and 7th.

1:11:46This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome? That's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required, compatibility and availability varies 18+. push your limits train with precision see the results at equinox that's high performance loving everything you need to lock in and unlock your potential at equinox start today at equinox.com

1:12:29okay allison is back and is once again time for the lightning round verge cast producer josh kahas in the control room do we have a sponsor this week we do This week's lightning round is presented by AWS, how leading businesses use AI for next level innovation. Okay, I want to talk about Waymo because some news came out this weekend about an arrest related to a ride in a Waymo. So a couple of teens got into a Waymo with a ghost gun. And these teens were, number one, under the age in which you're allowed to be riding in a Waymo. So they, in that way, violated Waymo's terms of service. They also had a ghost gun.

1:13:08can you explain what a ghost gun is ghost gun pretty printed uh there are various types of ghost guns i don't know the exact type of this one but generally they are 3d printed guns that are off the grid they are not traceable okay um but it is it is an active functional weapon and uh yeah it is a thing that uh you know they're in california and so i do not know the state of gun laws everywhere but i'm fairly certain that you're not allowed to have an unsecured gun in the back, an unsecured ghost gun in particular, in the back of a vehicle in California. So there's this fascinating situation where these teens are in this car, clearly violating the terms of service of Waymo, very probably breaking some gun laws here.

1:13:59The Waymo calls the cops on them and they get arrested for this. And I think this just is a can of worms here. Because on one hand, I think it is very important to enforce gun laws. On the other, I don't think people know that Waymo might call the cops on you. And I don't think we know the mechanism by which they do that. Right. Have you both ridden in them? Because my understanding is there's, I have not. My understanding is there are cameras all throughout them, but that you don't necessarily see them. Yeah, they're not obvious. It's similar to like in a Tesla, how they have like the camera in the top front.

1:14:40And it's kind of just always tracking to make sure you're paying attention when you're in autonomous mode. But they're not. It's not like there's all these cameras everywhere. I haven't ridden the new Waymo hardware, but at least in the Jaguars, that's what it was like. There's also like all these like past words, people like hooking up in like autonomous vehicles. Yeah. And they're just like, I think there's cameras. I don't really know. And it's like. I would assume. there's cameras. Please, there are cameras. You should just assume there's cameras. And someone is possibly monitoring those, which like, and this is the thing, and I think this is where it gets so messy.

1:15:15Like we don't entirely know how this works, right? We don't know if it is an AI system. We don't know if Waymo has some control center where there's a million screens up and they're like watching all the writers to see what's going on. We don't know what their options are of like, Like, you know, if you, you know, light up a joint in the backseat, right? Is that something that is going to end that way? Are they going to pull over and say like, hmm, get out of the car? I think that this is just like a very new world for everyone. This is just not how cars have worked in the past. I mean, to be clear, you were never supposed to light a joint in a car.

1:15:54I think that was always generally frowned upon, if not forbidden. Yeah. But you generally didn't have cameras connected to some sort of control center, potentially connected to police. So what happened with this ghost gun Waymo? I don't know the actual outcome, but but ultimately, like the the the kids got arrested. Right. Because Waymo contacted the cops and the cops came and picked them up. And, you know, I think this is this speaks to, I think, how complicated this is. You know, there was another incident earlier this summer where, again, it was two teenagers in a Waymo. They were drinking.

1:16:34So there's that. They're also firing off Orbeez from the backseat. So like there are water pellets. I am too old for this. Yeah, absolutely. I mean, as am I. Right. Again, I don't know the legality of of shooting. water pellets from a Waymo. It's probably from or inside of. I believe from. Okay, that's not cool. Again, this does not, none of this seems good. No. None of this seems good. Yeah. And I think this is where it's so interesting because like these behaviors are clearly bad, right? Like we do not want people firing things from a moving vehicle. That is a bad future for autonomous cars.

1:17:19I do think people need to understand that they are being monitored and what can happen with that monitoring when they get into an autonomous car. And yeah, it feels very clear that people don't know that. I mean, maybe these are teens and that's just the case when you're a teenager. But I think that this is probably a societal shift that is going to need to happen as these kind of expand. It's very weird. Yeah, there's a lot of questions here. I mean, they probably have a lot of this stuff in the terms of service, which nobody reads, but they also should maybe have some sort of pop up that just says, FYI, you're being monitored.

1:17:56You know, the whole thing about security, the whole Steve Jobs sort of quote about security is that the people around you should know that they are being recorded. And I think the same goes for people that are getting into these vehicles that are owned by these giant corporations. So I agree with you. You shouldn't shoot things out of a car. but uh i also think it's probably a right to know when a camera is recording you even though there are cameras everywhere all the time and you're always being recorded at this point in time we've got a story up about this uh and how complicated and naughty this is on the verge.com this week waymo did tell us in a statement that they use automated technology uh to help their teams identify potential violations like the presence of a firearm they said quote if waymo detects a firearm in the vehicle, we may reach out to local law enforcement.

1:18:46So again, I think all of this is still very vague and fuzzy. And I suspect that eventually we're going to have to learn more about how this works. Okay. That's it for Waymo. David, what have you got for us? All right. So David Pierce is not here, but I am still a David. And David Pierce may have kicked off one of the biggest e-reader revolutions of the decade with his obsession with books or books. We call it books. I think they say it's books, but I prefer books. They've got it wrong. The books Palma, the books Palma was sort of this device. There was a small e-reader device that was the shape of a smartphone, which was not something we were used to because the Kindle was always sort of this like squared off.

1:19:29It had like a little area for your fingers to go so you could hold it easier. But now suddenly you've got an e-reader the size of a smartphone. It got very popular. A lot of people got really into it. Then the Books Palma 2 Pro came out. And then all of a sudden, this company called XTE Inc. entered the fray. And they created these devices that are these really tiny e-readers that are even smaller. They can slap to the back of your phone via a magnet. There's all these things. That created, I don't want to say it's like an AI race, but for e-readers. Where XTE Inc. and Books are now in this race to create the ultimate e-reader device, which I think is good.

1:20:06That's called competition. and uh so now books has started making a bunch of new ones they have the books pico that's coming out which is basically the same four factor as the xte ink x3 and x4 which they have been making xte ink now has a pro version of theirs anyway the main point of this story is i'm sorry i give too much context sometimes but the books palma 3 has just dropped and the reason we like this is because it is made of aluminum it is so much more sleek it is so much more nice yes then the books palma 2 because the books palma and the books palma 2 pro were like plastic backs they were sort of like a cheap android phone kind of situation this kind of this kind of cooks this kind of looks good i'm gonna say i tell you the the biggest upgrade that i care about here i i have looked into and very very strongly considered buying these repeatedly yeah like they run the oldest version of Android you can legally ship yeah for whatever reason I think that is probably why they've been so cheap yeah and this one Android 16 yes which I'll take it it's only one version behind which is a pretty big deal that's basically the current version yeah I think in Android world yeah in the Android world it's practical it's what most Android phones are going to be running on possibly it is more expensive it is launching for$339 which the Palma 2 launched for$279 everything's getting more expensive though for an e-reader it has 128 gigabytes of storage which is like probably half the cost if not more of that money.

1:21:41It has expandable storage which you're not going to be able to afford because microSD cards are too expensive but you know it has a 3950 million power battery which seems pretty big for a body that seems quite slim and aluminum and very thin looking. And I just think that for people that want the more premium books experience, this is a very exciting thing. It also supports their stylus, which they launched with the Books Palma 2 Pro. I don't think you really need it. It's an extra like$46 too, which is a lot of money. I would just, you know, I don't know. It's just nice. I love this like premium e-reader race that we're in right now.

1:22:21And I just had to bring it up as the current David on the podcast. I love it. Allison, did you get in on the book type or what type of e-reader person are you? I, in the past, have used an e-reader. I really do not. Like when I was a bus commuter, I was using the Libby app on my phone. I'm an analog book girly right now. But these things all like appeal to me so much. like an adorable little e-reader I can slap on the back of my phone oh my god yeah yeah I support that fully like I'm I'm also if it's if I'm reading a novel like physical all the way I the thing I dream of and the reason that these are so appealing is like it's Android right like I can read the New York Times in this I can read the version this I can I don't necessarily want to scroll x in this but you know I I think that is the advantage like I don't want to have this this flashy beautiful premium screen in my face.

1:23:19I want to have this slightly ugly black and white screen in my face. I think a big part of the sales pitch of these devices is that they run Android. So you can do the things that you would do on an Android phone in it. But because they're e-ink, they are a terrible experience for the dopamine generators that we have on our phone. So you're not really going to be watching TikTok on these. It's mostly just for reading. It is wild how much of modern tech we're like, and it's a little bit bad. And we're so excited about that. Fantastic. That is so much of technology right now. It's kind of crazy.

1:23:52Oh, my God. Actually, speaking of which, Allison, what do you have for us? Yeah. Speaking of it's a little bit bad, I come to you with two tiny, adorable cameras. So the first one is from Camp Snap, sort of known for their their lo-fi camera experience. This is the 110D, and it is that kind of long and slim form factor. I don't know how I would describe it. This looks like a Viewmaster. What is this? This is a 110 camera. A lot of people have not seen these before. They take 110-type film. This does not take 110-type film because the D stands for digital. But Lomography, last year or the year before, released a new 110 camera that looks exactly like this.

1:24:44and they're very trendy because they're really weird looking. They've just got a viewfinder on them, like the mammography one. Like it's not a rangefinder, it's just a little piece of glass that you look through and they can fit in your pocket and they're so like weird. And it seems to me that Camp Snap is just trying to get on the trend of making a weird film camera looking device. But it's digital, which is a very popular thing for manufacturers to be doing right now. Can I call your attention to it? So it comes in two colors. um one is the kind of you know traditional retro black and silver there's a cover there's a color called dreamsicle which is white and this like orange and i want it so bad i think it looks great it looks kind of stupid but it looks great wait i have to say i was just on camp snap's website the other day and their like more recent mainline camera, the Camp Snap 2, they did a translucent body like edition of them.

1:25:46They have like hot lime and orange with like you can see the components inside. It looks really good. They're on one right now. They are. David, do you use these things at all? Not currently but only because I don't own them and because I shoot a lot of film. But yeah, I would be interested. I mean, they're pop like I have gone to a number of parties recently where I have seen people like a couple of years younger than me just like going crazy with these kind of things. My impression is that the newer ones, I think, hit like a better balance between the image quality being not so bad that you don't actually want to use it, but but not too good that it feels retro.

1:26:35I enjoyed the photos that I've seen from the camp snaps. There's a middle ground where part of the reason people were buying Digicams was the aesthetic of owning a Digicam and shooting with a Digicam. That is part of the reason. The other part is that the images do look more retro and aesthetic. But these companies can take the cheapest sensor they can possibly find out of China and get the same effect. So it's pretty easy for them. And this is a lot cheaper than buying some of those retro Digicams. Yes, some of those go for like$500. now so it's insane that's so insane what's your other one i'm looking at these translucent camp snaps now god dang it i want one you're gonna buy one now aren't you ah shoot this lightning arm is very expensive to me personally um okay the other one the other tiny bad cute little camera is from fujifilm the instax pal 2 it's a wide narrow boy also these are both weird long cameras Yeah, this one looks a little tinier.

1:27:35They're both modeled after 110. Yeah. Yeah. So this one has an LCD on top. There's a lot going on with this. Top level viewfinders are awesome. Yes. Everyone likes it. Yeah, there's a little optical finder, it looks like, on the back. But you can also, yeah, do the top down, look at the LCD. The rendering of this LCD screen, it makes it look like the world's best little screen. And I feel like that's probably not the case. Anyway, also interesting. This thing has a little USB-C thing built in so you can just like pop it onto your phone after you took some photos and like just transfer it to your phone that way.

1:28:21I that's very compelling to me because I will end up not taking something out if I'm like oh well then I got to go home and like or if I have to Wi-Fi to it or use an app or some nonsense like that I'll be like I'm just non-starter. Notably, Fujifilm has redesigned their camera apps like four times in the last like 10 years. And every time they're still bad and get like 1.4 stars on the app store. So having a flip out USB port, I feel like is kind of a response to that to just be like, look, just put it on your phone. Like just put it directly on your phone. If you want to feel fury like you've never felt before, download the app that you're supposed to use with your camera.

1:29:10Any camera. Literally any of them. The Leica one is the only one that's any good at all. Oh, well, that makes sense. I also want to say they clearly did the top level viewfinder. There's a trend in film photography with something like a Hasselblad 500 CM, which has a top level viewfinder or a Nikon F3. you can take the viewfinder off the top and then you just got ground glass on it. And so the light that comes in through the lens bounces off the mirror and up into the ground glass and you can see the world through that. And so there's a trend on Instagram or on TikTok where people just film directly down onto the ground glass because it sort of diffuses the look of the world around them and makes it look kind of dreamy.

1:29:54And you get that look of what you would normally see on film. and so I would be willing to bet a lot of money that that is the reason why they put the LCD on top. That's fascinating. I'll also say like as someone who is like a little more familiar with those types of cameras, like the top, sorry, top level viewfinder. Like they just look very cool. They do. Right? Like the use case for them, as far as I'm aware, is like pretty specific in film photography. In digital photography, it's like maybe not there in the same way and so they don't exist and this like doesn't necessarily make sense, but it looks really cool.

1:30:30And you know what? I'm into it. I'm into it. All right. I think that's all we've got today. That's it for The Verge Cast. Remember to subscribe to The Verge for ad-free episodes of our podcasts, exclusive newsletters, and so much more at theverge.com. We'd love to hear from you. You can email us, vergecast at theverge.com or call the hotline, 866-VERGE-11. Verge Cast is a production of The Verge and Vox Media Podcast Network. Today's show is produced by Josh Cajas, Eric Gomez, Brandon Kiefer, and Travis Slarchuk. We'll see you next week. Rock and roll. Push your limits, train with precision, see the results.

1:31:06At Equinox, that's high performance loving. Everything you need to lock in and unlock your potential at Equinox. Start today at equinox.com. This episode is brought to you by ChatGPT. Hey, it's Bill Simmons from the Bill Simmons Podcast. Have you guys heard about ChatGPT work? It's the new way to use ChatGPT for bigger multi-step projects. And when you need more than just answers, give ChatGPT work access to your apps and files, and it can create real work documents like spreadsheets, slides, and structured reports. Get started at ChatGPT.com by selecting work mode available on Plus and Pro plans.