The Vergecast - Two possible futures for AI

Episode Date: October 29, 2024

Kylie Robison joins the show to talk about the recent dueling AI blog posts from OpenAI's Sam Altman and Anthropic's Dario Amodei. What do these CEOs think the future of AI looks like? Then, Will Poor tells us the story of ShakeAlert, an earthquake alert system that has huge potential and some surprising challenges. On The Vergecast Hotline, Allison Johnson joins Will to figure out whether the iPhone's new Camera Control is really as fast as advertised. Further reading: Sam Altman: The Intelligence Age Dario Amodei: Machines of Loving Grace Anthropic’s CEO thinks AI will lead to a utopia — he just needs a few billion dollars first OpenAI plans Orion AI model release for December ShakeAlert If you live on the West Coast and you have an iPhone, here's how to turn on the "Local Awareness" feature that speeds up WEA messages: Download the MyShake app on for iOS or for Android Ready.gov's earthquake advice: About emergency and government alerts on iPhone Apple iPhone 16 and 16 Plus review: all caught up Email us at [e-mail] or call us at 866-VERGE11, we love hearing from you. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Welcome to the Vergecast, the flagship podcast of decreasing the cost of AI infrastructure. I'm your friend David Pierce, and I am sitting here updating all of my AI gadgets. So like six months ago, give or take a couple of weeks, I reviewed both the Humane AI pin and the Rabbit R1, kind of backed back, didn't like either one. Neither one was any good. And to be totally honest, I basically turned them off, stash them in a drawer and forgot about them. But six months later, these companies are still here. They're still making stuff. Humane has slashed its price in a way that suggests not a lot of people are buying that stuff.

Rabbit, I think, has confused people more than enticed them over six months, but they've made a lot of changes. Rabbit is saying it's getting close to shipping the large action model that it's been promising for forever. Humane does timers now. So I figured after six months, it might be time to pull these things out and see if we're any closer to AI gadgets getting anywhere. That's for a future forecast. I literally, I have spent like hours just sitting here charging and updating software on these things. And I had to find a sim card for the rabbit and I had to redo service for the AI pin. It's been chaos. That's for a future Vergecast. Today, we are going to do two things.

First, we're going to talk about the sort of dueling blog posts from Dario Amadeh at Anthropic and Sam Altman at OpenAI about the future of AI. I think normally I find these blog posts sort of ridiculous, but the fact that there were two of them and they're from these two people at this moment in time, I think matters. So we're going to talk about what's in those letters, what they agree on, what they disagree on, and what we might be able to learn about the future of AI. We're also going to talk about earthquake detection. And this huge, sprawling project underway, especially on the West Coast, to alert people more quickly to earthquakes. Super important problem turns out to be more complicated and challenging than you might think for lots of super

interesting reasons. Will Poor has been reporting this story out for a while, and he has come to talk about it, and I'm very excited. We also have a hotline, lots of fun stuff to do this week, very excited about this episode. All that is coming up in just a second, but literally as we've been sitting here, both of these things just finished an update and now both have another update. So I'm going to go do more software updates, and we will be right back. This is the Vergecast. See in a sec.

Support for the show comes from Retool. Too many companies run critical operations on duct taped spreadsheets, Slack workflows, and whatever else they could cobble together. Not because they want to, but because building internal tools means weeks of waiting on someone else's backlog. That's where Retool comes in. Build custom internal tools just by describing what you need. Prompt something like, build me a revenue dashboard on our Salesforce data. And Retool actually builds it on your company's data and your cloud with enterprise security built in. Go to Retool.com slash Vergecast. We all need to retool how we build software.

What's up, y'all? I'm Skyler Diggins, seven-time WMBA All-Star, Olympic gold medalist, and mom. And I'm Cassidy Hubbard, host and reporter for nearly 20 years covering the biggest names and stories in sports and mom. And this is Am Mom, a community for athletes, game changers, and moms of all kinds. Dropping May 14th. Tap in with us. Welcome back. So about a month ago, Sam Altman, the CEO of OpenAI, and probably the single most prominent person in the AI world right now, wrote this blog post called The Intelligence Age, all about

basically his vision for the future of AI, why it's going to be huge, why it's so exciting, why we should all be so excited, when it's going to happen, sort of big flowery stuff about the future of AI. Ordinarily, those blog posts are like, whatever, I think people paid more attention to this one because it's Sam and Sam is like right at the center of the tech industry right now. But still, we wouldn't have spent a ton of time on it. otherwise. But then a couple of weeks later, Dario Amade, who is the CEO of Anthropic, which might be OpenAI's most interesting competitor, wrote his own blog post called Machines of Loving Grace

that is sort of spiritually the same, but structurally very different. He wrote this 10,000 word plus opus full of details and full of thoughts about specific ways that AI might be implemented to make the world a better place. And he lands in, sort of the same place, which is everything's going to be amazing. But I think the structure of the way he talks about it is really different and really interesting. We talked about both of these, I think, briefly on the show before. But what they say together is really interesting. These are two of probably the three most important people in AI. I think the third would be Demis Isabas, who runs DeepMind at Google and is increasingly more powerful in all of the stuff that Google is doing

in AI. He, as far as I know, has not written a many thousand word blog post about about how terrific AI is. But these two have. And so I think what I wanted to do was really dig into these and see what we can learn about where AI is going from these two blog posts, where they're the same, where they're different, where there might be conflict between these two companies and these two ideas, but also if you believe the vision of Silicon Valley, and I think somewhere in these two blog posts is the vision in Silicon Valley for AI, what does it look like? The Verges Kylie Robison has been covering all of this. She has read these blog posts. She's talked to people about the blog posts. She, I think, could probably recite some of these blog posts from memory at this point.

And I figured who better to come on and see if we can make sense of all of this than Kylie. Kylie, welcome to the show. Thank you for having me. Just unrelenting AI chaos in your life. Yeah. It never stops. We do have a little bit of tiny news that we're going to get to at the very end. Some next model stuff, which we should just talk to really fast. but the thing I have brought you here to do is talk about these two blog posts. Yes.

And first, I want to know if you agree with my assessment that most corporate CEO blogs are nonsense. Well, that's the first one. That's, do you agree that most? Yes. Okay, good. Yes. But also, I don't know, something about the fact that both of these happened from these two people in particular, like within a couple of weeks, just felt like more of a moment to me than your average like CEO pontificates about the future.

thing. Has that been your read too? Yes. And at the time when I wrote about it, I had real AI, AGI believers yelling at me because like how could I ever consider? I keep saying this, your messiahs are businessmen. I'm sorry. It's calculated. So yeah, I drew the same conclusions. Okay. So the homework I gave you ahead of this was to come up with three ways in which these two men agree and three ways in which they disagree. Let's start with what they agree on. What's the first thing you feel like these two agree on? So I think that they do agree on that the world is going to be such a beautiful place with AGI.

Of course. I would say to flip on disagree, I think that Dario was a lot more measured, where Altman was a lot more, you know, grateful dead about it. Wait, I would actually put that slightly differently. My read on this has been, I think Sam Altman feels like it's inevitable. That like his whole blog post is basically like, we've done it. It is done. And there's even a line in it that I pulled out that was something like deep learning works and the rest will solve itself.

Like that's not what it says, but that's essentially what it says. Yeah. That compared to Dario, who goes over and over and over, like to the point where it makes the blog post too long of being like, Like none of this is inevitable. It doesn't have to be like this. We all have to do it on purpose. A lot of things have to go right.

Like you just get the sense from Sam's that he is like, problem solved. We just have to sit back and wait for this. Like you're talking about beautiful, incredible future where everything is wonderful. And Dario is like, maybe. Like that future exists, but it is not certain that we're going to get there. And that to me was like tonally the most striking difference between the two. And it sounds like you caught that too. That is 100% a great take.

That is exactly right. No, because it's been a long week, and I haven't thought about these blogs. But yeah, that is exactly correct. Who do you believe? Do you think Sam's right? I really want you to just be like, I think Sam's right. We did it. It's all done. Future's going to be beautiful. I just keep thinking one of these two men has been fired from the job for lying.

Yeah, well said. All right, what's the second thing? What else did they agree on? Wait, actually, let me drill down into the first one before we get to the second one. Yeah. Do you think, like, biggest picture, most beautiful future, they roughly agree on where that might go? Like, in the absolute best case scenario, do you think these two CEOs, and by extension, Anthropic and Open AI, think it might be as good in the same way as each other? I think you can see this in the actions of both companies.

one is absolutely gutting safety and the other is doubling down on safety. So I would say that they disagree or what the future looks like I think Oatman is a lot more optimistic than Dario. And Dario says at the top, like, you know, there's potentially a lot of bad things. I don't want to talk about that here

because we talk about it a lot. So I think that's the difference between the two of them. I think Dario sees a lot of the, what could go horrible. wrong and Altman's like, dude, it's going to be beautiful. Yeah. Yeah, I think that's right. And I think that was the thing I kept having to remind myself reading Dario's blog post. And like, to his credit, he says right up front, I think there are a lot of problems. I'm not going to talk about them here. And like every paragraph

of his thing, I kept wanting to like raise my hand and be like, well, I have 1,000 questions about all the things that might go wrong in service of doing this cool thing that you're imagining. And I'm like, okay, that's not, that's not what we're doing here. I have to, I have to take him at his that he thinks there are problems. We're just not talking about them right now. So, fine. We're talking about the big, beautiful future. What else do they agree on?

What else did you write down? How this is going to transform labor, the economy, work. They both seem to agree that a lot of wrote tasks and monotonous jobs are going to get replaced by AI and that perhaps we should have some plan for that. I think that's something they both agree on. For Altman, he has explored UBI, universal basic income. He ran a test giving people certain amounts of money for an extended period of time to see how it changed their life. Anthropic has not done that same kind of research, but I sense that they both agree that this is going to change the way humans work, and that's important to explore. Were you satisfied by either of their answers as to what we do about it?

And again, to Dario's credit in particular, he's pretty upfront about, I don't have all the answers. this is going to be complicated. But they do both treat it as inevitable that at some point this thing will essentially blow up our day-to-day lives and the overall economy as we know it. And they kind of are just like, yeah, we'll figure it out. No, I'm not satisfied. No, of course I'm not satisfied.

And I was just on Decoder talking about this with Nilai about, you know, there's so many unanswered questions. This is so nascent. It's not a fun answer to hear. You expect the AI reporter to be like, we have all the answers. And I think that's what people expect from these executives, too. But we don't.

We just don't have all the answers. And it is unsatisfying. Like, okay, if AGI was real, which, like, I don't even see that as a possibility necessarily. But if that were to happen, then we could all be out of jobs, but we'll figure it out. Right. It'll be fine. Somewhere between three years and a million years from now, everybody's screwed. But we'll figure it out.

Exactly. Right. Exactly. Hopefully the sun explodes before then. I don't know. Yeah. We'll live on Mars, and then it won't even be a problem. Perfect.

What's the third thing? They both in practice, in not even in practice, in writing, agree on the alignment of these systems. And I'm going to talk to researchers, I think, tomorrow about what alignment really is and how it's such a fuzzy term. But I think that they both agree that broadly these should be safe systems that are aligned with what we value as humans. They shouldn't be running amok in writing. However, as I said earlier, one has gutted something. safety and the other has not. But I think they broadly agree that it would not be good to create world destroying AI. Yeah, I mean, it's a hot take, but, you know, it's, it's a, it's a theory.

I like that. Right. I caught that too. And I think the idea that that is a thing we have to do on purpose is sort of encouraging, right? Like, again, you can read what you want into the actions of these two companies. And I have a lot of thoughts that are a lot less optimistic than these two blog posts based on what they do. But the sense that the way we build these tools and the way that they get used and the way that we manage them together, like one of Dario's big ideas was that we should have,

I don't think he called it this, but essentially like a United Nations of AI where like a bunch of countries bands together to build and manage AI in a way that promotes liberal democracy around the world. Yes. Like, I don't know if that makes any feasible sense technologically, but that's the kind of thing we should be thinking about, right? that it's like, okay, we are, if this thing is going to be anywhere near as powerful as we think,

that's the kind of stuff you have to start doing and you have to start building the tools at the beginning to be like that. And I think they both agree on that. Whether both companies are interested in taking the time to do that in a crazy gold rush of AI, the answer seems to be no mostly so far. But like in the long run, that at least strikes me as slightly encouraging. Totally. This is something I've been digging into recently. Just about, for the listener, researchers who want to do safety research at these big companies require a certain amount of compute to do these studies to figure it out. And kind of a block of important safety researchers at OpenAI left because Altman promised them about 20% of the companies compute to do safety research, and they did not get that, and it really pissed them off. So I think as Dario said, it is you have to continue making this choice to align to do that work.

And I think that's important to watch to see how seriously they're taking it because their researchers will tell you. They will yell about it. Yeah, they're not shy. No. All right. So any other agreements before we move to the stuff they don't agree on? No, I think that's it. All right.

So that's all the stuff they agree on. What are they disagree on? What's first on your list? I would say that Dario is a little bit more scared of hyping up. AI than Altman is. And I think that's kind of what we discussed already is that he's like, I don't know if I should be doing this. Like, I don't know. This is not inevitable. And Altman's like, in a thousand days, like hits the joint in a thousand days. This is all going to be so different, you know. So this is actually a good thing I was going to ask you about, which is why these two

people wrote these two blog posts at this particular time. And you have a strong theory about why, and I want to know what that theory is. I have a strong theory. And people like to talk about this topic. And they're like, you know, motivations aside. I'm like, well, aren't motivations like the biggest piece here of why they're writing this at the time that they're writing this? So Altman wrote his just around the time he was closing $6.6 billion in funding for OpenAI, the largest round for a private company in history that we know of. And then when I read this blog, the first thing I was thinking I was like, Anthropic is also trying to raise money. We don't know how much yet. But Anthropic is also trying to raise buttloads of money.

And I think that's the timing of this is, you know, investors are kind of trepidious because there's this problem of you need to sink so much money into these models. And there's a lot of promise and hype involved that this will change the future. But at the moment, it cannot even count the number of ours in strawberry. So I think 14,000 words is a good way. to say, look, I promise. I mean, like, I guess he doesn't promise. He says it could or could not happen. We have to make those choices.

We have to do the work. However, it is such a beautiful future if you give us billions of dollars. Yeah. Do you feel like spiritually, this is not that different from, like, Travis Kalanick back in the day who was just like, once I have, you know,

cars on the road with an app, but eventually robot cars are going to transform the way that every... And it's like, I feel like that is, to some extent, story. Like, you have to tell a story that is 10 times bigger than your actual story in order to get people to give you all this money because of, like, the structure of venture capital demands that you maybe be worth that much money. It's like, it's just all this weird stuff that

eventually you sort of have to lie about the possibility for the thing that you're building in order to get the money to even try and maybe get to half of that over time. Yes, I 100% agree. So you can imagine both of these blog posts as like hostage. negotiations, right? They're like, they're like, I know, I hear you. I'm so sorry, I'm so sorry, let me just, I'll put this out, and then you can give me money and everything, it'll be fine. Exactly.

No, I think that it is, it is exactly the same. And luckily, I work at the verge with a bunch of great people who have lived through a lot of bubbles. And they're like, hey, I remember the CEO saying almost exactly the same thing. Like, Sergey Brin once said that their new health division at Google was going to cure death. I forgot about that. Yeah, that's in the article. Go ahead and read it.

Oh, and I put this, this is even a bit in Silicon Valley, the TV show, where one of their caricature for a tech executive, he says, I don't know about you guys, but I don't want to live in a world where someone makes it a better place than we do. Yeah. And that is, I mean, it is just a tale as old as time. And unfortunately, the people who really believe in AI are like, this is different. I get it. I don't want to slam AI. I do think that it could prove to be really useful, and it is useful. in limited use cases today.

However, it is not... We were talking about the spreading democracy thing. That's something I read in Dario's blog. I was like, okay, come on, man. You know, they do have to make these grand proclamations. It's the same as decades before them. Well, there's a certain amount of... Like, you know how in TV shows, when they get to, like,

season five or six? They've made the stakes so high because you have to keep inventing new things to do to keep people interested. And you're like, well, we've already exhausted all of the interesting stuff. So, like, Fast and Furious Nine is literally, like, in space. Like, it's just the stakes creep is so real.

And I feel like AI in general is such an incredible case of stakes creep where we have hit this point now where, like, everything has to be the biggest thing. It's Soonerperchai said it's fire. Like, you can't, you can't come back from fire. And so, like, this is all I read now is I just see these things and I'm like, this is Fast and Furious in space. Like, this is, this is, like, die hard where he jumps a car into a helicopter. Like we've just lost our minds on how big all this stuff has to be because we're so desensitized to the rest of it. That is so good.

I'm mad. That's so funny. Yes, this is fast and furious in space. Yeah. This is, I've spent a lot of time the last couple of days lying on the couch watching TV. So this is, this is, it's for work now. I get to expense all that time that I spent on the couch watching TV.

What else? What else did they disagree on? What's next? Weren't you supposed to tell me one? All right, fine. I'll give you mine. The timing they disagree on in a way that I found actually very surprising. And I think this is one way in which, and again, this is just, it's two blog posts.

What these men actually, like, are saying in meetings and what they believe in, even what they've said publicly sometimes differs from this. Sam Altman is less convinced that this is all going to happen tomorrow than Dario was, which I was very surprised by it. Like, his thing is basically were a few years, this is Dario, a few years away from what he calls powerful AI because he doesn't like AGI, which fine. And then after that, all this stuff he's describing, all the stuff we've been talking about, all these changes to democracy, and we're going to cure all diseases. He puts that on a five to 10 year time frame.

Like, that's like, we're looking at like, I don't know, 2040 at a conservative guess for Dario as like everything about the world has completely changed forever because of AI. Sam is comparatively more restraint. Like, he says the next couple of decades, which again is like, it's soon, but it's not. the same sort of immediate, like we're a couple years away from the thing and then five more years away from it changing everything. And then, yeah, what is it? He says a few thousand days,

which is such a like delightfully meaningless thing to say. I should check, but I feel like, yeah, maybe he does say a few thousand days because I thought he said eight thousand days. But yes, I told an opening eye source of mine recently because they hadn't read that blog post. And I was like, well, Sam says in about a thousand days that we'll get this AGI. And he was like, really? I think it's going to be, well, he thought he was like, it's going to be a lot sooner than that. He's like, I know. Oh, boy. Which is just really interesting. And I really want to know what the listener thinks, but like I think most people think that this is bullshit. And then there's these people, I live in San Francisco. I'm talking to these people all the time. I'm, you know, having drinks with them

going to dinners. And they're like, yeah, it's inevitable, you know, next year. And I'm like, what the? Where am I? So I don't know who to believe because I, I'm, you know, I think that these people are really smart, and they are building really smart systems. Like, I believe that, but also we're going to cure death, you know. Yeah, and we are all going to be in robo taxis. But, like, this is the game that we play. And I think, like, what's the Bill Gates thing? Everybody underestimates what they can do in a year or no, overestimates what they can do in a year and underestimates what they can do in 10 years.

I think there's probably a lot of that coming. Yeah, definitely. Like, I would put a lot of money on the over of one year from now. Certainly. Yeah. So what is the next one on your list of how they differ? So I think that just generally these are completely different people. For instance, if you guys don't know, Dario left Open AI because he's like, this is not safe and calculated.

And we need to make a more safe public benefit corporation that became anthropic. So I think the difference is just these two people. Dario is safety, slowness, which is why I thought this point. blog was so out of place, and that's what made me think about funding. And Sam is more like hitting a joint in space, fast and the furious, this is going to happen, it's happening soon. I think that these are categorically different people attacking the same problem. They both want to build AGI. They're both building it in basically the same way. They both have big cloud partners, but they're just completely different people approaching it much differently.

Totally. I mean, I think the funniest thing about looking at these two blogger, post side by side is how sort of embarrassingly underthought Sam's seems in comparison. I agree. Like it really is like, this is like a long tweet from somebody who got like a little high and had some thoughts about AI. And Dario is like, my man just like wrote an academic thesis about AI. And they might both be totally wrong. They might be wrong about different things.

But like the just the sheer like difference in level of thought here is just crazy. Like I just went back and found this. There's such a funny moment in Sam's blog post where he says there are a lot of details we still have to figure out, but it's a mistake to get distracted by any particular challenge. Deep learning works and we will solve the remaining problems. That's like Dario says like 11,000 words about those two sentences. Yes, he does. It's just, it's just crazy to me. And I do think, again, to some extent, they're like trying to do different things.

It's fine. But reading Sam's after reading Dario's was like, this is like a kindergarten level of thinking compared to that, which was. striking to me. That's funny, the order that you read them because, yeah, of course. And I think that's why it's hard for me to compare because this isn't like the only piece that Altman has wrote about AGI, but it is like timing wise, easy to compare them. And I agree. It is a thesis about, you know, I think he wants to genuinely inform Dario, wants to genuinely inform the public of his thinking. And he talked about how he's been writing this since August. It's gone through a bunch different versions. And yeah, it definitely seems like Sam just fired this off and here we are.

Yeah, and that's fine. Who among us hasn't just fired off a blog post that too many people read, right? Like, it's a occupational hazard. I will say one thought that I had in this was that it makes a lot of sense to me why Anthropic is building better products than Open AI right now. Damn girl. Like consumer products. I mean, you go through and like Dario's is just full of ideas about things a person might do with AI and ways that they might be used. The only one Sam, mentions is like it's going to be a personal assistant that will help you accomplish tasks. And like, fine. Everybody has that idea. We've had that idea for decades. But like it's a bit of a galaxy brain take to just read that out of two blog posts. But like I think Anthropic is just

destroying open AI when it comes to actually building good products for people to use. And you can sort of see that from these two. I don't even think that's that hot a take. Claude is better than Chad GPT right now. It just is. I agree. I agree. Um, I keep name dropping podcasts I've been on recently, but I said this on Pivot. They were asking me, what's my favorite? What's my favorite? And I was Claude.

Obviously, Claude is so much better than the competition. It's crazy. Yeah. Yeah. And I think the question of what is the underlying technology is going to be like a forever game of leapfrog between a bunch of companies, principally these two. But right now, like, Anthropic is just a mile ahead

in terms of actually building good consumer products. And you can see it in Dario stuff. Like, he has ideas about how people might use it in a way that I'm not sure. There's not a lot of evidence that Open AI does. Right. So you're counting Gemini out? Gemini is just so its own thing. That it's, they're just like, Google is just like over here. Kind of like winning Nobel prizes. And I still like my true galaxy brain take is I think Google is just going to win all of this and it's not going to matter. And we're going to have to reckon with that sometime very soon. But that's for another day. I have a lot of, I'm also like, I've been fighting that fight for too long now. So I have to die on the Google. Google is further ahead than anybody realizes Hill.

Yeah. I think I'm okay dying on that hill, but we'll see. All right, real quick. Yeah. Some news, and then I'm going to let you go here. Yes. We had two scoops last week about upcoming new big stuff from both OpenAI and Google. Tell me what's going on.

Yeah, so my scoop with Tom Warren, our Microsoft reporter, was about how by the end of the year, OpenAI is planning to release Orion, which you can think of as GPT-5. it would be the successor to GPT4. So just in September, we wrote in the article that the researchers threw a party for finishing training the model. At Microsoft, they are preparing the compute to host Orion. Microsoft gets it first. And then they're going to release it to limited partners so they can find their own use cases, find the features that they want to build with Orion, which is the codename of the project, Orion. So that's supposed to happen by the end of the year.

Altman, I had my first real run-in with tech executive hits joint and takes to Twitter. He called it fake news, which it's not. And I said this at the top of the episode, and I'll say it again. One of these people was fired for lying. So, yeah, that's what's coming out by the end of the year. It's, of course, subject to change. And then, more interestingly, in December, Google Alex Heath, my colleague Alex Heath, wrote this in command line. He wrote that Google is planning to release its next big Gemini model in December.

So Open AI loves to front run Google. So that's something to watch every time. If Google's going to have a big announcement, you can almost certainly expect Open AI is going to fire somebody or release a big new feature or something. Yeah, I do think, I mean, to your point about, like, who's going to be a new thing. really going to win here. OpenAI certainly thinks Google is its main competition. And Anthropic is just kind of over on the side doing a different thing. And that might turn out to be true. It's certainly true right now in terms of like shine and resources. But I think Anthropic is closer behind than it gets credit for in a lot of ways. Totally. I agree. And okay, have you ever watched Mad Men?

Yeah. Do you know the elevator scene where he's like, I feel bad for you? And he's like, I don't think about you at all. That's very much Google and Open AIs for you. relationship, I think. Oh, 100%. Yeah. A hundred percent. Google is, yeah, Google is not super worried about it. Like, they're just after winning Nobel Prizes. Like, Google's going to be fine. Right. Demisis will eventually write an even longer blog post and we'll be back doing this again, I'm sure. Oh my gosh. I can't wait. And part of that, I forgot to mention that Alex wrote that Demis is not super thrilled about the capabilities of its Gemini model right now, the next one that it's supposed to be releasing. So that'll be interesting. I think I would love a way that's more

clear than benchmarks of how to weigh these against each other. I don't think we're there yet, and I just don't trust self-evaluated benchmarks in a lot of cases. But yeah, I'm excited to see how these all stack up in December. It'll be fascinating because we've been at this now for, I don't know, a year in change where Google, OpenAI, Anthropic, and Meta have all just continued to release better and better models. And we're at the point where like every two weeks somebody is releasing a slightly better model. And I've been asking, asking people forever when it's going to stop being so easy to improve the state of the art. And everybody I've talked to at least is kind of like, I don't know.

It would be fascinating if it started to be these. If somebody comes out already behind, that might change things. You know, that's funny because the AI people I've talked to just last week where, like, I think that we're hitting that point. We are finally hitting that point where it is getting harder to scale these or to make them better than before. because for a minute, it was really easy to just like leapfrog, leapfrog, leapfrog. But we're starting to slow down in that sense. But I guess we'll see it in December.

Yeah, it might be a messy holiday season in the AIO world. There might be a lot of Googlers working over Christmas. Typical. All right, Kylie, thank you as always. Thank you for having me. All right, we've got to take a break, and then we are going to come back and we're going to talk about earthquakes. We'll be right back.

Support for the show comes from Framer. Framer is an enterprise-grade, no-code website builder, used by teams at companies like Perplexity and Muro to move faster. With real-time collaboration and a robust CMS, with everything you need for great SEO, not to mention advanced analytics that include integrated A-B testing, your designers and marketers are empowered to build and maximize your dot-com from day one. So whether you want to launch a new site,

test a few landing pages, or migrate your full.com, Framer has programs for startups, scale-ups, and large enterprises to make going from idea to live site as easy and fast as possible. Learn how you can get more out of your dot com from a Framer specialist or get started building for free today at Framer.com slash Verge for 30% off of Framer pro annual plan. That's Framer.com slash verge for 30% off. Framer.com slash verge. Rules and restrictions may apply. Support for the show comes from Upwork.

The days of doing it all, all by yourself, are over. There's no romance in burning out while you're trying to scale. Instead, you can check out Upwork. Upwork helps grow your business by giving you fast access to specialized talent across more than 125 categories. So you can fill skill gaps, launch projects faster, and scale without committing to full-time headcount. And finding the right talent is easy. You can browse profiles, review past work, and get help scoping the role so you can get started quickly. Seriously, you could connect with the right freelancer in just a few hours, especially when you sign up with Business Plus.

Their AI powered shortlisting pairs you with the top 1% of talent in under six hours. No endless searcher required. You can visit upwork.com right now to post your job for free. That's Upwork.com to connect with top talent ready to help your business, That's upw-w-R-K-com. Upwork.com. Support for the show comes from LinkedIn. If you're a small business owner, you know that every hire counts,

but time and resources are limited. Finding, connecting with, and screening the right candidates takes up valuable time you could be giving to your customers. That's where LinkedIn Hiring Pro comes in. It's built to be your hiring partner, helping you find the right candidates faster. That way you can hire with confidence without turning it into another full-time job.

Hiring Pro streamlines the entire process from drafting your job to short-listing candidates and conducting AI-powered interviews for initial screenings. Its updated conversational interface lets you describe what you need in plain language. Nearly 60% of hirers find a candidate to interview within a week. With Hiring Pro, you spend less time searching, and more time connecting with the right talent.

And instead of getting buried in resumes, you get a focus shortlist that actually moves your hiring forward. Join the 2.7 million small businesses using LinkedIn to hire. Get started by posting your job for free at LinkedIn.com slash track. Terms and conditions apply. Welcome back. Will Por is here. Will. Hello. Hello. What are we doing here, Will. I know earthquakes. That's all I've got.

What are we doing here? Yes. So to start us off, I have a terrible, terrible sound that I want to play. Our favorite. The best way to start a podcast, I'm told, is to make people take off their headphones as quickly as possible. So with that in mind, that's awful. Okay, that's enough. Okay, can I tell you where my brain immediately just went?

Yes, please. A mix of Jackhammer on the streets of New York City. Yeah. And like the Sarlac pit in Star Wars opening up. Like the really creaky metal door opening up plus jackhammer. That's not fun. I didn't like that. I mean, it's spooky season as we record this.

So that all kind of works. So what is this? What am I listening to? That is the sound of my house getting seismically retrofit, which is basically like earthquake-proofing. I wanted to get my house earthquake-proof because I love. live in Seattle, which is earthquake country. I am also very terrified of earthquakes. And so I was told by a lot of experts that this is a thing that I should do. And so a few weeks ago, a bunch of people came to my house and crawled under my crawl space and drilled a whole bunch of bolts into

concrete right beneath my feet. And I thought it was a good idea to try to work from home while this was happening. Sure. So, which lasted about five minutes. But, This is me trying to work from home. They are crawling around in the crawl space and drilling very large bolts into the concrete foundation of the house, which the whole house is vibrating. Like I can feel this work in my teeth. And I need to get out of here. It was a bad day. I had a bad day.

First of all, incredible podcaster instincts to say this horrible noise is happening. I must record it. The people must know. I applaud you for this. Also, why are you doing this? Why did you do this to your house? So this is sort of the culmination of just like a long journey between me and earthquakes. I lived in San Francisco, the Bay Area for like 10 years. And that's where I developed my fear of earthquakes. I had reason to. I felt them a bunch. And I was told constantly to prepare for the Bay Area big one, which in the case, the San Francisco would be a repeat of that famous 1906 quake along the San Andreas fault that burned down a lot of the city, among other things. So I spent a lot of time thinking about that in San Francisco and then moved to Seattle, which seismically is more quiet day to day. Like I've lived here for four and a half years. I haven't felt anything, but it's also very much earthquake country and not feeling anything is a real false sense of security.

Did you ever read, there was a New Yorker article that came out about 10 years ago. Oh my God, the really big one. The really big one. Yes. Yes. This piece scared me more than maybe any magazine story I've ever read. It was very good and terrifying. What do you remember from that story?

Wait, okay. Can I go find a line for you in this story that I read that I still occasionally think about? To see the full scale of the devastation when that tsunami recedes, you would need to be in the international space station. That's just a sentence I think about a lot. lot. And this isn't the New Yorker. This is not a thing that is like prone to hyperbole and it tends to, can I read you another sentence? Now I'm looking at this piece and it's scaring me all over again. This is fun for me. I'm having a great time here in Virginia. Yeah, here in geologically ancient Virginia. It says Ospac estimates that in the I-5 corridor, it will take between one and three

months after the earthquake to restore electricity, a month to a year to restore drinking water and sewer service, six months to a year to restore major highways, and 18 months to restore health care facilities. On the coast, those numbers go up. What? Yeah, there's a quote, there's a quote somewhere in there that someone said effectively everything of everything west of I-5 is going to be toast. I think those were the words, I live west of I-5. Yeah. And again, this is a like sober magazine article that is essentially like, we're on borrowed time. So I can understand how you would become somewhere between rationally and irrationally terrified of what's happening here. Yes. So let me give you all the other, like the full rational cause to be concerned.

Okay. So at some point, probably, roughly maybe in the next 500 years, but also maybe tomorrow, because that's how earthquakes work, there will be this mega thrust quake in the Pacific Northwest. It's going to be felt from northern California, all the way up to British Columbia. It's on a fault that's called the Cascadius Subduction Zone that runs north-south all up the west coast, just off the coast. It could hit Category 9 on the. Richter scale, it'll produce maybe five minutes of shaking in Seattle, which is bananas. It's going to do a ton of

damage all on its own. And like you remember, it's going to produce this even more devastating tsunami all across the Pacific Northwest. Right. The earthquake is the first thing, but then it's what it does to the water that causes the worst of it all. Absolutely. Okay. So really, really very bad news. And you live there on purpose. And I moved there intentionally after I read this article. And that I don't know what that says. So that's the very, very bad news. The good news is I, not long after I moved here, I heard about this earthquake early warning system called Shake Alert that had just gone fully live all up and down the West Coast.

So there's no way to predict earthquakes. That's not a thing. That's not a science right now. But scientists can detect them right as their beginning. And the internet moves a whole lot faster than shaking from an earthquake. So if there is a big quake today, and I happen to be far enough away from the epicenter, like if it's off the coast, I'll get an alert on my phone that says there's an earthquake coming. And it might be a few seconds or it might be a minute or two before the shaking reaches me,

which is awesome for me and my fear of earthquakes. But the more I've looked into this, the more I realized that making a detection and warning system that's both reliable and fast enough to be effective turns into this, massive, massive challenge. And in a lot of ways, detecting the earthquake is the easy part. The hard part is using existing technology to notify a ton of people all at once about this disaster that might only be seconds away. So I figured the best way to understand all this earthquake early warning stuff is just to follow the chain of technology from point to point. Basically, how an earthquake is detected, how an alert message gets generated from that,

and how it gets all the way to my phone, which turned into this fun goose chase all around Seattle, and it started with this small kind of creepy shed next to a wastewater treatment plan right here in town. So let's pop out. I'll show you everything, and then we can do it. Okay. Let's try to find a light switch.

Is this going to want to stay open? Oh, that's right. Oh, that flickering is not fun? Deluxe. Good thing it's not a video. Yeah, all right? You're hearing Doug Gibbons. He's a field technician at the University of Washington, or the U-dub, as we locals like to call it.

He offered to meet me here at this shed and show me some earthquake surveillance hardware. Yeah, so what we have here, this is a tightened strong motion accelerometer that is bolted to the foundation of the building here. The device is this plain green box, like the size of a loaf of bread. So if the ground starts to shake, electronics inside this thing will capture a three-dimensional snapshot of the movement. It's oriented north to south, and inside of there, there are actually three different seismometers. There's one placed in the north and south direction, measuring, shaking on that axis, one placed on the east and west direction, and one placed in the vertical direction. Doug plugged a laptop in, and he showed me the raw output of the seismometers. It's basically three different graphs that show the amplitude of motion at this one spot.

It's basically like a digital version of those old earthquake readouts you see in movies. the needle that scribbles back and forth on that rolling piece of paper. Oh, yeah, the one that kind of looks like a lie detector test. Yeah, yeah. It's small, and then it's huge, and then it's small again. Yes, yeah, okay. Bigger scribbles, bigger earthquake.

Three of them for three axes of motion. Got it. Then it's so sensitive, it can even just detect me just bending my knees up and down. Literally one little shift in weight and two of the graphs went, boo!

So that is, wow, that is incredibly sensitive. That's wild. Yeah, it's fun. So it must just be doing things all the time. It is, and it's constantly feeding that to places that we'll get too soon, but it's taking samples many times a second constantly. So over the past 10 years or so, field techs like Doug have installed maybe 1,500 sensors like this all up and down the West Coast. It's part of this huge effort led by the U.S. Geological Survey to keep tabs on all the seismic activity in this part of the country. And it's sensors like this one that power shake alert, the early.

warning system for people like me. And we now are confident that we've got enough instruments to issue good warnings to people. So here I feel like we should stop and just do a little earthquakes 101 because it'll help a lot with everything we're about to talk about.

So when an earthquake happens, when two parts of the earth's crust suddenly slip past each other, there are two kinds of waves that ripple out from the epicenter. They're called the primary and secondary waves, or P waves and S. waves. The primary waves move really fast, like a little more than three and a half miles per

second, but they're too weak to do any damage you probably wouldn't even notice them. The secondary waves are the ones that can really cause harm, but those move about half as fast as the P waves. And that's the big opportunity for early warning. All those really sensitive seismometers like the one I just saw, they can pick up those P waves when the real short, shaking is still miles away. Let me put this all in real world terms because I think that'll help it make more sense. So let's take this really big one, that 9.0 earthquake that I am completely petrified of. That could start anywhere up and down the west coast. Say it starts due west of Seattle. This is like worst case scenario for me. Better to just face your fears head on.

That's about 135 miles west of Seattle off the coast. The Cascadia fault is about 40 miles offshore. So that means that those first P waves have a good 40 miles to get ahead of the actual shaking. So when the P waves ping those coastal sensors, the real earthquake is still about 10 seconds away from the shore, which means it's 40 or 50 seconds away from downtown Seattle. So that chunk of time, let's call it 45 seconds or so, that's the window of time that Shake Alert has to warn this major metropolitan city that this huge, huge natural disaster just appeared out of nowhere. So step by step, here's how they actually do that. When the P-waves hit those first sensors, those little seismometer graph spike, and that data gets forwarded straight on to U-dub over radio or hardwire internet or satellite,

really whatever is available right where the sensors are located. And that part happens really fast. Within a fraction of a second, within a second or less we want this data to head back. back to UW. And that's from shaking in the ground to the sensor recording it, creating a packet, sending it to a communications device, that communications device, either sending it through a radio wave or through the internet, back to UW. It lands at a server on site at the university.

This one. So we just walked into kind of a small-ish room that has a bunch of computer racks in it, including and a whole bunch of computers that just have green, blinky lights. lights. All the data from the field comes in here continuously and then gets processed. We're now in the basement of an Earth and Space Sciences building at UW in Seattle. My tour guide is Renata Hartog. She helped build Shake Alert's algorithms. The servers in here, along with others in Northern and Southern California, they run a pair of algorithms that analyze any incoming quake.

So one algorithm, it just looks for, it's just, it's just, looks at the very beginning of the signal, and then it tries to base on the amplitudes. So it associates it into, oh, this is an earthquake, just has a point on the map and an estimate of what the magnitude of the earthquake was. Basically, it's like a snap judgment of the epicenter and the magnitude of the quake. They have the second algorithm that's slower, but it can model the whole quake beyond just that first point on the map, which is a thing I never knew about earthquakes, incidentally. the epicenter is really just the starting point.

As an earthquake grows, it can get bigger and bigger as more miles of the fault line slip, and that means longer shaking and more people affected. So for example, a large earthquake off the Pacific coast here on the Cascadia seduction zone, if it starts somewhere on the southern end, it will take minutes probably before that whole fault plane has ruptured

all the way from the southern end to the northern end. And finally, based on all of that, Shake Alert estimates the geographic area that will get hit the hardest, and it creates an alert to be delivered to phones. And that part happens super fast, too, by the way. The first alert leaves the server after just a second or two of processing.

And if the quake keeps growing like they do, it can send out additional waves of alerts a couple times a second. Okay, wait, so hold on. Do the math for me right now. where are we in terms of like lag and I'm imagining the P waves are going, the S waves are going, where are the waves and where are the alerts at this moment in time? So this is what I kept trying to do as I was having these conversations.

I was trying to add up all of the lag from the beginning to now. And so I kept asking all of these people about all these steps, how long does this take? How long does this take? And everyone was like, I don't know, a second, less than a second. everything that we've talked about so far is just super fast. The sensor's picking up the P-waves, sending that information over the internet to U-dub, the algorithm running and making decisions. All of that added up is like two-ish seconds. So in our scenario, the earthquake is still like 40 or 45 seconds away from me in Seattle. Okay, that's pretty good. So it's like,

I'm imagining sort of Jaws music as it's like slowly creeping towards us as we get here. But we still, we're still ahead of the game here. We still have a solid chunk of time to, I mean, 45 seconds isn't forever, but it's something. It is something if it actually gives me 45 seconds, because there's this, there's one other link in this chain that we have not talked about yet, and that's the alerts getting from this server to my cell phone, and that's where things kind of get messy. So the deal is, we already do have this really good system in a lot of for getting alerts out to a gazillion cell phones all at once. It's called the Wireless Emergency Alert System, or WIA.

So if you've ever gotten an Amber Alert on your phone or a tornado warning or some other kind of emergency ping, that's probably WIA. I'm sure you've seen these things before. The presidential alerts are versions of this. Is this the one that also failed spectacularly in Hawaii and scared a bunch of people that something like a bomb was incoming? Oh, the ballistic missile. Yeah.

the incoming ballistic missile, the alert that was just like, hug your loved ones. Yeah, basically. And then they were like, sorry, never mind. That was wea? Yeah, yeah, that was wea. Okay. So we do know how to do this. Exactly.

That was developed back in 2012 by the federal government. So this has been around for a while. And it uses this kind of funny technology called cell broadcast. And basically, it uses cell towers like giant bullhorns. They just blare out the simple digital message to end. any device that's programmed to pick it up. So this is great for earthquake alerts because there's no need to opt in or download anything.

If your cell phone is in the danger zone when an earthquake hits, you'll get an alert on your phone as long as your phone and your carrier support WEA, which somewhere between most and all of them do. Our WEA messages that get delivered to cell phones are built for 90 characters because we know that the majority of

even non-smart phones, will be able to get those alerts. That if you go above those 90 characters, people might not be able to get them. So we're in the position of crafting messages so that we can make even people who still use flip phones to be able to access earthquake early warning. That's Bob de Groot. He leads the Shake Alert Operations team out of Pasadena, California. He says that Wea is the single biggest best channel for reaching the 50 million or so people on the West Coast.

But at the same time, he's also really aware of its Achilles heel. When the wireless emergency alert system was developed didn't have Shakler speeds in mind. Operating at the horizon of minutes was sort of, wow, that's fast. But now we're asking a system

to operate in matters of seconds and fractions of a second. Wea is just not as quick as the rest of the shaker. alert system that we stepped through. In fact, when Shake Alert was in development, no one even knew what Wea's lag was. Bob's team actually set up these two tests in California, where they pushed out an alert to a specific area and then measured how long it took to reach their phones.

We took a bunch of cell phones, including mine and including my colleagues. We put numbers on them and then laid them on the table and then took a GoPro, basically, and filmed all those phones. He sent me the video of this test, and it's really funny. It's this long table with a whole row of different kinds of phones spread out, and a lot of people are all gathered around these phones waiting for the test to start. So we waited for the alerts to come through, and the phones lit up when the alerts came through. The very fastest turnaround that they recorded was around four seconds. The median was in the six to 12 second range.

A bunch of phones just never got the alert. and then there were residents within the test area who reported an even wider range of latencies. So the point is, even with that controlled test, wea is still this really frustrating black box. How do we not know better? I mean, this seems like A, solve technology, B, very important,

and C, a thing we've been doing and using for a long time. It seems like this should be a totally known quantity at this point. I had that same feeling, and it took me attempting to map out all of the people involved in WEA to understand that it's just not one thing. It's not this technology that is implemented by one organization that can be optimized throughout this whole chain. There are those are the federal government, there are telecon companies, there are handset manufacturers, and every carrier might handle implementation of this standard differently. They might prioritize it differently. on their networks. All of this is voluntary for them, so it's kind of whatever systems they've set up. And so there seem to be all these differences in delivery. There's different kinds of lag across the 2G and 3G and 4G and 5G networks,

across different styles of phones. Bob said that, like, a burner phone that he got at the 7-Eleven turned out to be the fastest delivery mechanism. So if you, like, really want to be on point with WEA, just go get yourself a burner phone. No, babe, it's not a burner. This is my earthquake phone. If it's my earthquake detector. So it's just a little bit of a mess.

There's no built-in way to measure end-to-end latency. And again, this wasn't built for earthquake early warning. So no one in any part of this process was ever told you have to find ways to shave off every second possible from this system. And again, we've been tossing this 45-second number around in our scenario. And so given that a second here or there might not seem like a big deal. But the thing you got to keep in mind is that that's the amount of time that Shake Alert has to notify me in Seattle about a Cascadia earthquake. First of all, even in that scenario, there are a lot of people that live closer to the coast. So they're not going to have 45 seconds to play with.

And even for me in Seattle, there are other faults to worry about. And good evening, a rather typical late morning in the state of Washington was turned suddenly into a rumbling and rolling panic. An earthquake measuring 6.8 has left million shaken. The damage is widespread. The injured list continues to change, and at least one person has lost their life. That was the 2001 Nisquallyquake. It hit right underneath the Seattle area, and it shook the whole region for about 45 seconds. It ultimately did maybe $2 billion worth of damage. It's one of the costliest quakes in U.S. history.

And the fault that produced it is really busy. Before 2001, it produced similar quakes in 1965, 1949, and 1927. That's basically every 20 to 40 years. Those are our most typical significant earthquakes, I guess you would say. That's a once-in-a-generation event. as opposed to the Cascadia earthquake, which is once in the life of a nation kind of event.

Gabriel Lotto works on engagement with the Shake Alert team here at Udub. And he explained to me that the fault responsible for the Nisqualequake sits a good 40 miles below the Seattle area. So let's say you have a repeat of the Nisquale earthquake today. That earthquake starts deeper down, but it's right below our feet, specifically right below, the city of Olympia. Since the waves are coming

directly up from below us, it means more of the region gets hit all at once. And for this one, we just get a lot less warning time. As the area of shaking spreads out, you could get five, ten, 15 seconds, depending on where you are exactly. Just to bring that all home,

five or ten or fifteen seconds is roughly the amount of lag time that's in the WIA system. So for this quake, the one that I will very likely experience myself in the coming decades, that squishiness in WIA could determine whether or not a lot of people get warnings in time. There's definitely a deeply bleak world in which you can imagine getting this notification like two or three minutes after the earthquake has hit. Yeah. Which sounds both very plausible and really truly awful. And just, yeah, just the biggest knife twist.

So it sounds like if WIA is not the answer, or at least not the perfectly tunable answer, are we just on our own? Is there just not a better system here? Wea is not the whole answer, clearly. But the people building Shake Alert are thinking way past WIA to all of the possible platforms for delivering a message like this. So, for example, a team at UC Berkeley built a smartphone app called My Shake. I have it on my phone, and it works like a normal smartphone app. I share my location with it, and if I'm within the alert radius of an earthquake, it sends me a push notification. And push notifications generally arrive in one to two seconds, much better than 5 to 15.

Okay, so this makes a lot of sense to me, and is one of the things I was going to ask is I feel like, for you as a Seattleite, if somebody was like, hey, download this app, And give it your location and we will tell you if there's an earthquake coming. That is the single best case I've ever heard in history for why you should download an app. Absolutely. But I guess there are challenges there too. But it does make sense to me that you would say, just do this one thing and then we can do something much better. Yes, absolutely. And the shake alert people are shouting that from the rooftops.

the fact that it's opt-in versus opt-out for something like this means that out of the gates you are capturing such a smaller chunk of the population that you are otherwise. There's just no getting around that. The Shake-a-Lor team has been at this for a while, and there are now three and a half million or so registered users of this app, which is awesome and huge progress. But again, there are 50 million people on the West Coast.

So it's getting people signed on to this before a big earthquake becomes this consumer tech problem, it becomes an app development problem, it becomes a social problem. It just gets messier. Now, Google has been thinking about this for a while, and they have done one better. They baked Shake Alert into Android at the OS level, which gives them the power to issue a couple of different kinds of alerts that you can't do just with an app. If it's a smaller earthquake, it just looks like a push notification. If it's a larger earthquake, it's a full screen takeover. It will break through silent mode or anything like that. You have no choice but to look at this thing.

It makes a sound. It takes over your screen. It buzzes. And it's possible to turn off these alerts, but they are on by default, which I think is a really good thing. When life safety is on the line, yeah, you don't want to mess around with whether or not a user will pay attention to an alert. Apple so far has not built anything like that into iOS. There is a way to speed up via alerts,

but it's this toggle that's buried in the iOS system settings. I will put a link in the show notes about that. I did ask Apple about all of this, but they didn't get me any more information. Anyway, the point is there are these other ways that you can get an alert, but it depends on where you are, and what phone you have

and what you're proactively signed up for. So this is where everything outside of WEA just gets really piecemeal. Right. Yeah, it's an interesting adoption problem because you get to a point where you're like, okay, even if we've solved this problem, you know, the only way to get people reliably

to download an app is to promise them like free, cheap clothing from China. So like, I will pass that along to them. Yeah. But you mentioned three and a half million people have downloaded this thing, the system seems to be sort of fully realized. Has it been tested? Have we been through a shake alert earthquake before? Yes. In California, there have actually been a lot of them that have put all of these different platforms to the test. Actually, just back in September,

there was a pretty decent quake in Malibu. And it all worked. And some people got like 20 seconds of warning, which is enough time to take cover. It's enough time to get away from Windows. You can do a lot in 20 seconds. Overall, the rollout has not been perfect. Back in 2021, there was another quake in Southern California. That Shake Alert ended up misinterpreting as multiple smaller quakes, so a lot of people just never got that warning. In some ways, though, the mistake that experts worry about even more

is a false positive, an erroneous alert. Here's Gabriel again. If we tell people there is an earthquake and there isn't, We lose credibility. And that reduces public safety because then you have the boy who cried wolf syndrome. That did happen once in California back in 2020. USGS had to send a follow-up message, which I kind of love because it read, and I quote, Shake Alert message canceled, investigating.

If you protected yourself, well done. It's almost like passive aggressive. Yeah. It's a little bit like, this was kind of a test. in a way. Yeah. Which is great, but also really not great. Because as we all well know, we have spent years and years and years getting really good at ignoring annoying app notifications.

And if earthquake alerts get that reputation, then the whole system collapses. Then people will turn off their notifications. They will go into their phone settings, dig around, and find out how to turn these off. Right, right. But even without any mistakes, alerting becomes this kind of funny mix of art and science. With any automated system like this, there is always going to be some alerting threshold.

A preset that says above this amount of shaking will warn people, and beneath it, we won't. But finding that threshold is a social problem more than anything else. I actually happened to be in Los Angeles for a couple of back-to-back earthquakes back in the summer of 2019, And back then, shake alert was this experimental app. It was just for people in L.A., so I didn't have it. But for those earthquakes, actually, no one got an alert because the shaking just wasn't heavy enough.

But the response from the public was, we just made this earthquake alarm. Why didn't we use it? They felt shaking. They didn't get an alert. They've been told that there was this awesome earthquake early warning system that was going online. So that turned into a whole thing. And the feedback from the public ended up helping USGS decide to lower some of the alerting thresholds. So that's just one example of one quake.

But the point is there's always going to be these edge cases and they're going to have to keep dialing in these thresholds. And poor Bob de Groot from USGS hears about it every time there's an earthquake. I manage our Shake Alert X social media account at USGS underscore Shake Alert. I always entertain these questions of people saying, I got the alert, I didn't get the alert, why did I not get the alert? I didn't feel very much at all. Why the heck did I get an alert? And of course, insert your favorite word.

And the response, generally most people are good-natured, and I love interacting with people on X because they have really good, genuine questions or very thoughtful. But sometimes people get a little excited. As somebody who lived in San Francisco for a bunch of years, and thus experienced San Francisco earthquake Twitter, I feel Bob's pain. It's both a good time on the internet

and as somebody whose job is the earthquakes, I feel for Bob. Right, yeah. If you're the one all those tweets are going to, that's a different story. Yeah, exactly. But I feel like if I'm thinking about this big picture, you've sort of described a bunch of pieces that we need

that all exist, right? Everybody has a device in their pocket that can, in theory, be alerted. We have the apps. The systems are getting better. You were saying the sensors are all over the place. But there are a lot of different parties required to get this together. Right?

You mentioned the carriers and the operating system companies and the government and so on and so forth. Do we just have a stalemate here where we're going to have a bunch of possibilities, but it would require somebody to just put them all together, and that is harder than it seems like it ought to be. We have this weird bureaucracy, collective action problem with earthquake alerts. Well, so all of this energy has been put into cell phones because that is like the obvious massive delivery mechanism. Right. And so, like, getting into the weeds around cell phones has taken up a lot of time and energy. And there are places, especially around WEA, where it does feel like, okay, it is what it is.

You know what this makes me think of, by the way? This is a total random aside. But remember early... in the pandemic when Apple and Google started to work together on some of the notification systems for people who had been exposed. Yes. And just the sheer amount of work and wrangling it took to get that stuff to work and talk to each other and make sense. And that was, obviously, earthquakes are a big deal. Like, that took a literal global pandemic for these two companies to actually, like, sit down in a room together and be like, okay, how do we make this stuff work?

Yes. And I think clearing that bar with anything short of a global pandemic. even for something as important as a natural disaster, just strikes me as very complicated. Completely. And they were under the gun during the pandemic in a way that everyone involved in earthquakes talks about the fact that until there's a big earthquake that resets the idea of earthquakes and everyone's mind and pushes earthquakes to the forefront of everyone's minds. It's just hard to get people to pay attention sometimes and to identify it as a

potentially really important thing to do. Right. And like you said, with phones, it's without those two platforms playing along, you can only get so far. Right, exactly. So that's been all of the cell phone development. But there's a lot of other things that you can do with these alerts that go way beyond smartphones. And people are working on that too. So you think about smart speakers, you think about home assistants, you think about smart TVs. The people I interviewed for this story were basically like anything with a speaker or a screen should be a delivery mechanism for earthquake early warnings. Like all of those companies, all of those platforms, those could all be distribution channels. Then you think about big public channels like highway signs or PA systems in schools and hospitals.

Then you think about all the other automated things that don't have anything to do with just warning you as a person. You can open fire station doors so that they don't get stuck down. if they're damaged by an earthquake. Water utilities can shut off valves or pumps. You can slow down trains or you can recall elevators. There's all this other stuff that you could do with this information. And in a lot of ways, it's just early days in figuring out how to plug this information into all of these different systems. Also seems like it makes that threshold of when you do and don't trigger an alert become very important.

very important and potentially different for all of those different systems. Totally, fair. Yeah, it's just a mess. So some of that is already happening, but it means working with one organization or agency at a time. It's really slow going. But Shake Alert has only fully been online for a little more than three years now. So this is still really early days of actually plugging it into stuff. Is the sense that Shake Alert is kind of a thing?

sort of a project out of a university that is like interesting research for how to do it, or is it the thing? Oh, it's on the west coast of the United States, it's the thing. Okay. There are other similar systems elsewhere, but there's, thank God there isn't like

a competing earthquake, early warning, like standard. Like, we have to do a whole other story about... You have a folder full of apps. Yeah, all of the rival engineers and scientists trying to make a different one. No, it's, it's, the system is in place All of the state and local governments are aware of it.

All of the big technology providers are aware of it. It's just a matter of making it as useful as it can possibly be. Okay. So do you have the My Shake app? I do. I absolutely do. How many notifications have you gotten from it recently? What do they make you feel?

This is the thing about living in Washington. It's just, it's like, it's quiet. It's too quiet. It's too quiet. And I swear to God, a lot of these folks that I interviewed for this story all told me some version of, I just want there to be like a good medium earthquake in Washington or thereabouts that just reminds everyone about earthquakes.

It doesn't do any damage per se, it doesn't kill anyone. It just puts it back in people's minds. They just want like a nice, fun reminder quake. The earthquake that just knocks like one plate out of a cabinet. Yeah, everyone loses a plate. And it's like maybe everyone's favorite plate. And so they're like, where's that plate? Oh, the earthquake got it.

Yeah, 100%. Yeah, that makes sense. Has all of this made you more or less terrified of earthquakes? I apt, like, strong both. Okay. Talking to the science people about all of this taught me about all the other faults that are around me besides just the Cascadia's abduction zone. There's a fault that runs basically under my house, like very shallow. and it's much less well-known and well-understood than the others.

So that's just like a new abstract fear for me. So that's cool. On the other hand, it's genuinely very heartening to see this really messy constellation of organizations try to figure out early warning. Government agencies, states, local governments, universities, telecom companies, cell phone makers, like they're all doing their best. and the result is something that I can almost certainly say I will benefit from at some point in the future. I just have to wait and see when.

Also, I really want an Android phone now. We're all buying Android burners. That's my takeaway from this, is everybody gets an Android burner, and it's going to save the world. 100%. So, yeah, back to the house. We're done with the noises. How was the process? Was this, are you glad you went through this crazy earthquake-fitting process at your house?

I am. It was only a couple of days, and now it is, like everything else about earthquakes, it's completely abstract until it's not. So I just like, I'm just standing here in my house knowing that the house is now bolted to the foundation. And I just have this like magical thing that I don't see protecting me in theory. And so now I just have to go about my business. But I will say just like, it's like having an amulet. It's like much more real than that, obviously, but just knowing that that happened, I can kind of just like put that part of my mind to bed. Totally. All right.

We got to take a break. But real quick, give people the call to action. If you're on the West Coast and you're worried about this, what should you do about it right now? Yes. So if you're on the West Coast, if you have an iPhone, we'll put in the show notes how to turn on a feature called local awareness. That's the thing that speeds up via messages. Android and iOS, you can download the My Shake app. And you can go to ready.gov for just general earthquake advice for before and during and after.

Love it. All right. We'll put all that in the show notes. Will, thank you as always. Should I play just like a little bit more of that drill sound to play us out instead of the Vergecast music? Yeah, let's really ruin everyone's day on our way out. Let's do it. See, everybody. All right, we've got to take one more break, but we'll stick around and you're going to help us do a question from the Vergecast hotline. We'll be right back. Support for the show comes from Anthropic. Not every question has an easy answer.

And the ones that are really worth asking usually come with a healthy mix of inspiration and backpedaling, aha moments, and quiet meditation. When you're working through one of those problems, you want a partner to bounce ideas off of and figure out where the deeper issue lies. That's where Claude can help. Claude is the AI for minds that don't stop at good enough.

It's the collaborator that actually understands your entire workflow and thinks with you, whether you're debugging code at midnight or strategizing your next business move. Claude extends your thinking to tackle the problems that matter. Plus, Claude's research capabilities go deeper than basic search. It can have comprehensive, reliable analysis with proper citations, turning hours of research into minutes. Ready to tackle bigger problems? Get started with Claude today at cloud.a.ai slash vergecast. That's clod.AI slash vergecast

And check out Claude Pro, which includes access to all the features mentioned in today's episode. Claude.complex and unprecedented, the Spanish authorities are calling it. Passengers who'd been stuck aboard the Hanta or maybe Hanta virus-stricken Dutch cruise ship disembarked in the Canary Islands this weekend, prompting the highest stakes game of where are they now since maybe COVID? Some of the evacuees, American and French, have since tested positive for the virus, and yet public health officials seem remarkably calm. We do have one individual who was taken to the biocontainment unit early, early this morning, and we assess that individual.

They are doing well. Possibly because this is not the one to freak out over. Today, explain, drops every weekday afternoon. Buzzwords like progressive and affordability are thrown around all the time in politics. But what do they actually mean? For me, being a progressive means at least two things. One, being willing to unite lots and lots of people, all of the folks that are getting screwed over

against the powers that be that are making your life worse. And then second, being progressive is essentially a hopeful enterprise. That you think, I think, that the world can be much better, that we don't have to settle for crumbs or settle for the status quo. And is there a difference between what it means to the elected officials and what it means to the people? So money is essentially the root of everything.

I don't care if you're gay. I don't care if you have all that. That's like secondary, third. Like that doesn't, that's not a priority. That's this week on America Actually. Let's begin. All right, we're back. Will Por is still here.

Hi, Will. Hi. I haven't left. Will just lurks in the background of every vertcast, whether you hear him or not. He's always here. I could appear at any moment. Just know. that's always a possibility.

Yeah. So we've been thinking over the months about all the stuff we can do with the Virchcast hotline. We get tons of good questions. Lots of them are weird and are basically just like, is my iPhone bad? And we say like, no, it's fine.

But occasionally we get questions that like make us really like go deep down rabbit holes to try and figure out what's going on. And so we've been trying to figure out ways to go down more of those rabbit holes. There are also like an alarming number of people in the Slack room now who see all the hotline questions because everybody loves getting these questions. It's really fun. It is really fun. It became part of my job somewhat recently, but I would have

been doing it anyway because it's just a fun way to spend a minute at a time. It is super fun, except when people are mean to us. Please don't use the email to be mean to us. Yes. But when you call, be nice. That's really all you ask. Yeah. But you, I have no idea what you're about to do, but you found a rabbit hole that you went down. Yes. on the hotline. So set this up for us. Well, there's not much setup other than I found a question that I personally wanted the answer to. So I was like, okay, I will go try and find this answer, and that will be a hotline. So that's what we're doing here.

All right, set up the question. Who's the question from? This question comes from a listener who has a kid who is running around all the time, and he wants to take pictures of that kid, and he has a question about that. Hey, it's Peter from Brooklyn. I had an idea of how to show the new features on the iPhone 16. It'd be really cool to do a race to take different types of pictures in different situations with the new phones versus the old phone. I want to know, can you really take the camera out, open it up, and get a shot to that moving kid faster? Can you really change the aperture faster? All those different settings.

Is it really faster with the new controls versus the old ones? That's it. Thanks. Have a good one. Rock and roll. Okay. So I went straight to Allison Johnson for this. because A, she did our iPhone 16 review, and B, I know from listening to this show that she has

lots of experience chasing small kids around with a camera. Hello, Allison. Hello. So I was curious about this question because I'm one of those people that watches every Apple keynote every year just to see what the camera does now. And I feel like even for me this year was a very like camera. E event. Do you feel that way also? Yeah. It wasn't so much

that like anything drastic had happened you know, with the camera pipeline or whatever. But, you know, the camera control was the big thing on both of the iPhone 16, all the iPhone 16 models and kind of the new filter

options you have. Yeah. So, yeah, it's like a camera forward update, I would say. Well, I watched, I saw all of that. And my question was kind of the same as Peters, which is, you know, how much does that button make it feel like you have a point and shoot camera in your pocket? Which to me is just like a muscle memory question. It's not a question that you can answer from watching the keynote or

watching demos or anything, because it's all about the like routines of how you just like pull out and use your camera without thinking about it when there's a toddler running by doing something cute for example. So I'm curious to get your take on what routines you have around your phone and what routines you've had over the years and then whether the 16 and that button meaningfully changes any of that. So, you know, historically before the 16, walk me through your reflexes for taking a photo really quickly. You've got your phone in your pocket. Your kid is doing something cute. Like, what are the steps? Yeah. So I switch between Android, iPhone, like quite a bit throughout the year.

So that changes things up. But on Android, it's actually super easy because I can double-click the power button. And like, I think on basically every Android phone will just open the camera app from wherever you are. So I can do that without looking on an iPhone. Typically, I will take the thing out of my pocket. And I could be clever and map. the camera to a triple press of the home button or something, I just am not that smart, I guess.

And I always just end up, like, swiping from the home screen. Like, you can tap the camera icon and, like, hold it. Right, that's what I do. Yeah, that's, I think, what a sensible person does. I don't know why I have trouble with that. I swipe, I slide my finger across, and that's how I open. it up. So it's like one extra beat where I have to kind of like have the phone in front of me and the screen, you know, wakes up and I swipe and then I'm ready to go. Gotcha. So there is some like

home screen friction. Yeah. Yeah, a little bit. Maybe it's of my own making. I don't know. Yeah. I know the feeling of just like reaching in my phone to quickly take a picture of something. And then I like, the screen doesn't quite want to wake up or the like it doesn't register the little camera icon tap. There is a little, I've like, I've lost pictures because something in there just doesn't like quite happen the first try or the first two tries. And that's super frustrating. And then I'm distracted like trying to swipe on the screen and then my kid has run off and he's doing something else. Yeah. So disaster.

Yeah. So there's an in-between that experience potentially and this new fancy camera button. And it's the action button on the iPhone 15s. Have you played around with that over the course of the past year? And is that meaningfully different? Because I know a lot of people just mapped their camera to that button. Yeah, yeah. Okay.

Yeah, it's totally reasonable to use the action button as that camera shortcut button. And I find the placement of it a little funny for that. It's kind of like on the other side of the phone that I feel like I'm reaching for and it's kind of up high above the volume buttons. So for me, I feel like I have to kind of do a do a little maneuver to get there. But it seems to work for a lot of people, I guess. Yeah, but it's not it's not the like, it doesn't form the muscle memory for you the way that you want it to. Yeah. Gotcha. Okay, so what about the literal camera button that is now on these phones? How is that, has that changed this like quick draw routine to go back to the beginning? Like, kid is doing something cute. Your phone is in your pocket. What do you do? Yeah. So I just had a roller coaster of emotions with the camera button because it's like coded like everything I would like, you know? It's like a real button. It controls the camera. You can do a bunch of cool. cool, like, change exposure and stuff with it. I don't actually like it as much as I thought I would.

Oh, no. So my kid is doing something cute in front of me. I will definitely press it to launch the camera. The screen has to be awake already before it will do that. Like, pushing it just the one time will wake up the screen. So you have to kind of do it again. Oh, interesting. Yeah.

Which I would like start that process when the phone is in my pocket. So it's hitting the button once to turn the screen on and then again to launch the camera app and then again to take a picture? Yes. And for the third thing, for the taking a picture, I don't like it very much because I know. It is, it's kind of stiff. Like it's a little bit recessed. Yeah. With the, from the edge of the phone, which makes sense because you don't want to accidentally.

take a bunch of pictures. But it just feels like that little bit, it's too stiff and, like, I have to push a little bit too hard on it that I feel like I'm shaking the whole phone. And I'm still, you know, maybe the iPhone can cope with that. But I'm so like, no, I don't want any shake, you know, in this photo. It might be messing with your framing. Yeah. It's going to be a little crooked or something.

I don't like that. So I'm just really stuck on this like one, this like scenario and the like individual beats of this kid does a cute thing. One click to turn the thing on. One click to get into the camera. And then you're hitting the software button on the screen. Yeah. Gotcha. That's been my process.

And maybe the screen will wake up when I take it out my pocket. You know, there's variations in there. Right, right. But that's basically it. Yeah. But it's still a little fiddly. Is it faster, meaningfully? Is there anything about this that does speed up the process, even if it's fiddly?

I feel like I do like having just a button to press down that feels to me like I am starting the process. Even if it's not like, you know, bam, I'm going to get a shot just like that. Yeah. One thing it is really useful for, and this is something I've kind of like used more in long-term testing and not exactly in those like quick draw moments is changing. the exposure compensation because when you do that like little half press or the light press on it, that's how you access. You can set it to a bunch of different things. But the one I like is the exposure compensation because I don't know if you've ever tried changing the brightness on a photo that you're taking, but it's such a pain on an iPhone. Is that that, that's where you have to

like tap the screen until the little sun icon shows up and then you have to drag the sun up and down. Yeah, I can never, I can never do that right. It never feels like I'm dragging it enough and then all of a sudden the image is like way too bright and I'm like spending so much time fiddling with that that like whatever I was taking a picture of is long gone. Yeah, this is just like a little little bonus exposure compensation dial, which I love. So that's been my favorite part of the camera control, honestly. Okay. Do you feel like it Can you get to other settings really quickly? I just the vision I have of what Apple wants out of that button

is for people to very intuitively just like wiggle their finger back and forth by a few millimeters and like do 17 things and take a picture. Yeah. Do you feel like you are like psychically connected to this button the way that they want you to be? I do not. And it's a little funny. It's like something you have to kind of get used to because you're like half, pressing this button. It's not, there's not actually a half press. It's, that's the like capacitive,

you know, it gives you a little haptic buzz. Like, you did it. And then you're supposed to keep your finger on there and slide it back and forth. And I end up, like, taking my finger off or like, I'm, I slide it and I need to reposition my finger to like, you know, do the adjustment some more, which feels a little like choppy and not quite right. And then there's, like, no good way to, like, exit the control, which is a weird thing that Nelai and I both noticed in the review is, like, you kind of want to just, like, tap to get out of there, and it's just weird how it works. Huh. How do you get out of there? Do you even know? Have you figured it out yet? Or has that setting been up since you got the phone? Let's see. No, no, no. You tap the live view. Oh, God, the little sun icon is back. I don't want that at all.

Yeah, no, it feels like you should do a light tap again and that would exit. And that's not what happens. That does not work. Yeah, that's the weird behavior. Huh. Okay. Well, I can't say it feels like you're selling this button. I know.

I'm really not trying to, I guess. Yeah, clearly. That is fair and understandable. I guess the like, I guess Peter's question was, is this any faster? And my sort of tack-on question is, is this meaningfully better? And I guess all of that rolls up to, is this a meaningful factor in deciding to buy a phone if you really care about taking pictures quickly? Yeah. The thing I've kind of come around to is like it doesn't hurt.

Like literally nothing was taken away. You can use the camera in any other way that you prefer. and this is like an extra thing that, you know, for opening the camera itself kind of helps. I don't prefer it for taking a photo, but it's not like Apple took away all the other ways to take a photo. Right. And then you get that little extra control that like sometimes does come in handy. So it's like, I don't know if it's a net positive, but it doesn't hurt. That's what you want out of any flag.

feature of a new phone. It does, it's not damaging my experience in any big way. First, do no harm. Something like that. And I think it's just like a lot of other things with phones is like I don't see a lot of new, you know, on iOS or on Android, a feature that I'm like, wow, you should absolutely throw away your old phone and get a new phone this year for this thing. Right. It's sort of like, are you in the right, you know, talking about. in that cycle of, for some people, it's maybe like two to three years or for other people

that they just want to not shop for a phone as long as they can. Yep. It's sort of like, where are you in the discomfort with your current phone and how appealing are the things on the new phone? I think the 16, like the regular 16 in particular, this year is a pretty good value proposition of like you get these new buttons and like maybe Apple Intelligence. will turn out to be something. And it's, but besides that, it's just like the new iPhone and it's pretty good.

So. Yeah. Yeah. That is fair. I guess my last question is that we talk a lot about whether in the future we're all going to keep using our phone for everything or whether the phone is going to get splintered into a bunch of other gadgets. Point and shoot cameras are having a moment. If you care about, you know, whipping out at a moment.

advice in taking a picture really quickly, is there a world where you should just get a point-and-shoot camera and not worry about all of these fiddly buttons and software updates? What do you think? I was really hoping I could shortcut to this could be a cool point-and-shoot camera with the camera button and with the photographic styles. Yeah. Because you get some kind of neat kind of Fuji film-esque, you know. Totally. Film stocks or whatever you want to call them. And it does not feel that way to me. Of course that doesn't. It's a phone. Like, it's still going to be a phone. It's still going to be like pinging you with slack messages when you're out trying to take some pictures. So I think that is true. Like a button is not going to turn this into like, you know, a super great camera in a way that it wasn't before. It's still the camera I use most of the time. But yeah, it's, I don't think, going to sway anyone from. from a nice Fuji film camera or I don't know,

it's the meta ray bands. Like that will take a photo faster than taking a phone out of your pocket. So I don't know. Right. There are other ways. Yeah. Okay. Well, I don't know where Peter goes from here.

I'm sorry, Peter. I just want to like hit this button a bunch of times now. Now I'm curious to see. But thank you, Allison. Yeah. Thanks. All right, that is it for the Vergecast today. Thank you to everybody who came on the show,

and thank you, as always, for listening. There's lots more on everything we talked about at Theverge.com, all of our coverage of what's going on with OpenAI, all of our coverage of the iPhone stuff. We'll link to both blog posts in the show notes. You should really read them. It's a lot of reading, and it gets pretty flowery at some points, but both of these things are good reads,

and I recommend checking them out. As always, if you have thoughts, questions, feelings, or blog posts about AI that you would like to share, You can always email us at Vergecast at theverge.com or call the hotline. 866, Verge11. We love hearing from you. And like I mentioned, there is that Slack room and everyone's in it. And I can't even begin to guess all of the fun places that hotline questions are going.

So thank you so, so much to everybody who reaches out. It's the best. This show is produced by Liam James, Willpore, and Eric Gomez. The Vergecast is Verge production and part of the Vox Media Podcast Network. Nelai and I will be back on Friday to talk about more open AI news, all of Apple's gadget announcements from this week and lots more. We'll see you then. Rock and roll.