OpenAI's Flirty New Assistant + Google Guts the Web + We Play HatGPT

It’s been a week of A. I. vertigo - the feeling that we’ve been dragged five years into the future and aren’t sure what to do about it.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner, Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 0:34Kevin Roose?

  My biggest news of the week is that I am in a war with a car wash.

• 0:38Casey Newton?

  Really?

• 0:39Kevin Roose?

  Yes, last week or I guess two weeks ago now. I took my car into a car wash to get it washed. And you know, it's one of these places where they will not only wash the outside of your car, but also vacuum out the inside and clean the floor mats and get all the like crusty food.

• 0:52Casey Newton?

  A full-service car wash.

• 0:53Kevin Roose?

  We love a full-service car wash. This one unfortunately, so I dropped my car off. I go get a coffee. I come back half an hour later and they have lost my key. And so I can't drive my car away because it's the key is gone. And I have to call my wife who brings the spare. And at first they tried to blame it on me. They're like, well, was the key here when you dropped it off? I'm like, yes, because you had to drive it through the car wash and to the place where they, you know, vacuum it out. So clearly they had just lost it. They looked in the vacuum cleaner and couldn't find it there. And so now I am out of key. And Casey, I want you to guess how much it costs to replace a key on a Toyota SUV.

• 1:35Casey Newton?

  I know too much about this because of the following story. My friend got a new apartment and he was like, I want you to have a spare key. And I said, of course, I would love to be able to do that for you, dear friend. And he gave me the key and I lost it the very same night. I went home and I reached into my pocket and it was gone. He's never trusted me again. The same friend got a car and he wanted to again, to consider letting me have a spare key. And so he went to get a key for his car, which is like a 20 year old car. And the replacement key was $1,000.

• 2:07Casey Newton?

  And so the key exists, but it has an air tag on it. Because he doesn't even trust himself with it. And certainly I wouldn't either.

• 2:13Kevin Roose?

  Yeah. So mine was not quite that expensive, but I did get quoted $575.

• 2:19Casey Newton?

  What are they putting in these keys? Does this key have a large language model running inside it?

• 2:25Kevin Roose?

  Yes, it is actually connected to a neural interface in my brain. No, it is just a key, but because it's proprietary and they have to program it and it has to be specially matched to your vehicle identification number. And, you know, it, they're able to jack up the price.

• 2:39Casey Newton?

  We need to bring back hotwiring.

• 2:41Kevin Roose?

  Yes.

• 2:41Casey Newton?

  You know, you need to learn how to just be able to reach right into that steering wheel and.

• 2:45Kevin Roose?

  See, I was looking for regulatory solutions and I would, I would like for the federal government to take this on. Maybe the FTC could take a look at some FT keys.

• 2:57Kevin Roose?

  You like that one?

• 2:57Casey Newton?

  No, absolutely not. We're moving on.

• 3:05Casey Newton?

  I'm Kevin Lewis, a tech columnist at The New York Times. I'm Casey Newton from Platformer.

• 3:09Kevin Roose?

  And this is Hard Fork.

• 3:10Casey Newton?

  This week, OpenAI has a brand new AI model and it won't stop flirting. Then I go to Google I/O and report back on how the company's plans could end the web as we know it. And finally, pass the hat. It's time once again for Hat GPT.

• 3:42Kevin Roose?

  So Casey, it has been a big week in AI news. So much happening. OpenAI made some announcements. Google made some announcements. We're going to talk about all of it. But I would just start by saying today that this has been a big week for what you have called, and I have shamelessly stolen and, and used as my own term, AI vertigo.

• 3:59Casey Newton?

  Yes. AI vertigo, of course, is that sort of feeling of wooziness when you see something that makes you feel like you've sort of just been dragged like five years into the future instantly. And you're not exactly sure what we're all going to do about it.

• 4:11Kevin Roose?

  Yes. So I've had AI vertigo at least twice this week. We'll talk about that. One of them more. Maybe optimistic and one of them may be more pessimistic. But let's just start with OpenAI because on Monday, OpenAI announced their newest model, which is called GPT-4O. The O stands for Omni. This is a new model that has native multimodal support. It can work with images and video and audio. It also appears to be better at a number of reasoning and coding tasks. But the part of the announcement that got all of the attention was the voice. This is a new voice mode for ChatGPT that is going to be rolling out in the coming weeks. And it is uncanny. I'll just say it. It's a lot.

• 4:56Kevin Roose?

  It's a lot to take in.

• 4:58Casey Newton?

  Yeah, it is uncanny. I would say at the top that like we have not used this. I have asked to use it and have not gotten it yet. And so everything that we're going to talk about is a tech demo. And in the demos that we saw, or at least in the big event, they had the phone plugged into Internet to ensure that they had a very fast connection. So I just point those things out at the beginning because, you know, we have just sort of learned in our years as tech reporters, you never want to trust the demo 100%.

• 5:27Kevin Roose?

  Totally. So OpenAI sort of framed this as being like the real life Samantha from Her. Her is the 2013 Spike Jonze movie, of course, in which a lonely man named Theodore falls for his lifelike AI assistant. The AI assistant is voiced by Scarlett Johansson, Joaquin Phoenix plays Theodore. Great movie. And they basically said, yeah, we built that. Sam Altman himself even tweeted her on Monday. Another OpenAI employee posted, you are all gonna fall in love with it. So let's just play a little clip of this voice because I think listeners should just, if they haven't heard the demo, they should just hear what this thing sounds like.

• 6:07Casey Newton?

  Yes.

• 6:08Speaker 03

  Hey, how's it going?

• 6:10Speaker 04

  Hey, Rocky. I'm doing great. How about you?

• 6:13Speaker 03

  I'm awesome. Listen, I got some huge news.

• 6:17Speaker 04

  Oh, do tell. I'm all ears.

• 6:19Speaker 03

  Well, in a few minutes, I'm going to be interviewing at OpenAI. Have you heard of them?

• 6:24Speaker 04

  OpenAI? Huh? Sounds vaguely familiar. Kidding, of course. That's incredible, Rocky. What kind of interview?

• 6:32Speaker 03

  Well, it's for a software engineering role, but I just need to know, do I look presentable, professional?

• 6:39Speaker 04

  Well, Rocky, you definitely have the "I've been coding all night" look down, which could actually work in your favor. Maybe just run a hand through your hair or lean into the bad genie's vibe. Your enthusiasm is what's really gonna shine through.

• 6:55Speaker 03

  I don't have a lot of time, so I'm just gonna throw this on. What do you think?

• 7:03Speaker 04

  Rocky, that's quite a statement piece. I mean, you'll definitely stand out, though maybe not in the way you're hoping for an interview.

• 7:13Speaker 03

  Okay, I got it.

• 7:14Kevin Roose?

  So that was an OpenAI employee, Rocky Smith, doing a demo of the new ChatGPT voice mode. And there at the end, he put on like a bucket hat. And because GPT-4O can see through the camera on your phone, it was able to like see the hat and, you know, tell him that it looks silly.

• 7:35Casey Newton?

  Yeah, and so, look, you know, we have played a fair number of clips of voices on the show that have been made with AI audio. Very little of what we've shown you so far has been like generated in real time in this way, right?

• 7:50Kevin Roose?

  Yeah, so just last week on this show, you had a conversation with my AI friend Turing, in which you guys talked about lots of things, orphanages, fundraisers for my neighbors. And that experience, I would say, was a far cry from what we saw at the OpenAI demo. What do you, how do you think those things compare?

• 8:12Casey Newton?

  Well, I mean, last week's conversation really could not have teed up OpenAI's announcement better in the sense that the conversation with Turing last week was the state of the art last week and pales in comparison to what we just heard, right? The conversation that we had last week was with something that could not process audio and video and text at the same time. And that meant that the model had a really difficult time understanding emotion, understanding sarcasm, responding appropriately to the way that I was interacting with it.

• 8:45Kevin Roose?

  Telling the difference between your voice and my voice.

• 8:47Casey Newton?

  And so then you fast forward to this demo that we see this week. And it's like, okay, here is the leap forward. Here is something that can understand tonality, that can respond with something that approximates a human emotion. And assuming all of this stuff works just as well when we get it into our hands, I think that's very powerful.

• 9:07Kevin Roose?

  Yeah, I mean, the biggest difference for me between the Turing demo last week and the OpenAI demo this week is the latency, because when we were talking with Turing, there was sort of like a three to five second lag between when you would say something and finish talking and when

• 9:24Casey Newton?

  Every single time.

• 9:25Kevin Roose?

  Turing would process and respond. And we edited those gaps down because we didn't want to like give listeners, you know, the experience of having to like sit there and wait.

• 9:35Casey Newton?

  We didn't want your experience to be as bad as ours was actually doing it.

• 9:38Kevin Roose?

  Right. But we should say like the OpenAI ChatGPT voice assistant at least appears to be very fast, very low latency. You barely have to wait at all between when you stop speaking and when it starts. And in part, I think that's because it is not doing what Turing did. It is not transcribing the audio from you or me and then feeding the text into a model, getting a response back, and then resynthesizing the voice. It is natively accepting the audio input and working with it directly as audio without that sort of middle step.

• 10:12Casey Newton?

  One thing I would say stands out is the way that they have tried to make it feel more human by, for example, making it seem sort of halting and uncertain, by varying the pace of the speech, by sort of having the voice talk at, you know, various parts of a register. Like, there's a moment where she's like kind of squealing like, oh, Rocky, you know, sort of like getting way up in the voice. So we are very far away from your sort of, you know, monotone himbo friend.

• 10:38Kevin Roose?

  You know, in defense of my monotone himbo friend, there was a little variation in Turing's voice, but I would say that the ChatGPT voice assistant that we saw this week, it's just much more dynamic. They even showed off a demo where it can sing, which is something that other AI voice assistants have not been able to do. And I just, I'm, I was sort of amazed by like how much variety there was. And with also the things that OpenAI did to make it feel more human, like it will insert little filler words like um or let's see or good question. Like, it's, it sort of feels like more like you're talking with a real person and less Like you're getting answers from an AI model.

• 11:19Casey Newton?

  Well, I'm glad that now the AI will start getting emails about all the filler words it uses because previously only podcasters have had that experience.

• 11:27Casey Newton?

  But, you know, I think it's important to say that everything you just described is true, and it does not make ChatGPT more useful as an assistant, right? Filler words aren't making it more useful as an assistant. The dynamic range of its voice is not making it more useful. What it is doing, it is making it feel more emotional. It is feeling you more drawn to it. This is an engagement hack. This is designed to get you to use this product because there is now something that is tricking you into making you believe that it likes you.

• 11:51Kevin Roose?

  I actually, I think it could be an engagement hack, but I do think it opens up new use cases as well because part of why we haven't seen a lot of AI voice assistants doing things like customer service is that they're sort of in the uncanny valley, right? When you're talking with an AI assistant, it, you know, you're talking with a robot. It's like it's got a little delay on it and it just is sort of an off putting experience.

• 12:14Casey Newton?

  Yeah. So assuming that this stuff works, Kevin, and that you get it into your hands soon, how do you imagine that you are going to be interacting with it? And does it change things for you?

• 12:26Kevin Roose?

  I mean, the first thought was that this is what Siri should be. And in fact, there's, you know, some reports out there that OpenAI and Apple are in talks to build OpenAI's technology into the iPhone. There's also been some reporting by my colleagues at The Times that Apple is doing a sort of overhaul of Siri to sort of make it more like these generative AI assistants. But that was sort of like my first immediate thought was like, as soon as this becomes available, I am going to program the action button on my iPhone so that it goes immediately to this voice assistant instead of to Siri, because this is the kind of assistant that I want on my phone.

• 13:06Casey Newton?

  Yeah. And what do you want to do with that kind of assistant?

• 13:09Kevin Roose?

  I mean, you can imagine lots of queries that you might want to do in the course of a day. You know, can I, you know, how long can something stay in the fridge? Or, or, you know.

• 13:21Casey Newton?

  Because one thing about Kevin is he's always pushing it with those expiration dates.

• 13:24Kevin Roose?

  I really am. I've become sort of a truther about expiration dates on things in the fridge. I think they're, they're designed by big food to just like make you shuffle your inventory more frequently.

• 13:34Casey Newton?

  Remind me to order a takeout the next time I come over for dinner at your place.

• 13:39Kevin Roose?

  But that's the kind of thing like, or just any number of daily queries that you might want to ask to chat GPT. I also think it's going to be very useful for things like tutoring. There was an interesting demo, sort of included in OpenAI's announcement, where Sal Khan, who runs the Khan Academy, was essentially using this new voice assistant to help his son with a math problem. And we should say, like, it is not just through the phone that this thing works. They also announced that chat GPT has a native Mac app for desktop. And so you can do things like have it analyze what's on the screen and then talk to you about it.

• 14:17Casey Newton?

  So, yeah, I think all of that sounds pretty useful. But I think the most important thing about this story, Kevin, is the way that it goes beyond utility. This company has said, we want to make something that is emotional, that you feel emotions about. And I think this stands in really sharp contrast to the way that Google has been talking about its AI, its assistant. It is very clearly a computer. It is not pretending to be a person. It does not have a persona. This is a clear effort by them to have you not think that you are talking to some kind of sentient being, right? And I think they think that that is really important, because they've had cases in the past where, you know, they

• 15:00Casey Newton?

  had this one engineer who thought that he was talking to something sentient, and it created quite a stir when that happened a couple of years back. So they have backed all the way off of that. OpenAI is now saying, well, what if we leaned all the way in? And what if we made something that was friendly, that laughs, and that frankly flirts?

• 15:19Kevin Roose?

  Yes.

• 15:20Casey Newton?

  In a way that made me very uncomfortable. So I'm curious what you make of this very emotional AI that they're now just openly building.

• 15:29Kevin Roose?

  I mean, my first question for OpenAI is

• 15:35Kevin Roose?

  Did you finish the movie Her? Because I did finish the movie Her, great movie. But it is not a utopian story about a man who falls for an AI assistant and they live happily ever after. In fact, spoiler alert, but the movie's like 11 years old now, so if you haven't seen it yet, that's kind of on you. It ends with the AI assistant sort of breaking the man's heart and revealing that she has, you know, hundreds of other companions, that he is not in any way special, leaving him to sort of go off into the cosmos with the other AIs and sort of leaving him dejected and alone.

• 16:14Casey Newton?

  It was the first poly sci-fi story that I'm aware of among the big movies.

• 16:18Kevin Roose?

  The first of many. So I just wonder, like, as these companies are building things that are modeled on these sort of sci-fi depictions of AI in the future, whether they have fully thought through all of the risks. Because people will get attached to this thing. You even saw it in the demo, and that was sort of the most striking thing to me about this series of demos on Monday by OpenAI was just the OpenAI employees themselves are talking to this thing as if it were their friend, as if it were a human. They are saying things like, hey, ChatGPT, how's it going? Before they launch into questions, they're laughing with it. One of them even like wrote, I heart ChatGPT on a little piece of paper and like showed it to the app.

• 16:59Kevin Roose?

  And you could say, oh, that's just a demo, but it really does speak to the sort of fact that even these very seasoned AI experts who who understand how these things work, who are some of the people who are building ChatGPT, who know that it is not sentient, that it is, does not actually have feelings and emotions and the ability to form emotional bonds with humans, even they are tricked into sort of treating it like a human. And so if they are treating it like a human, I believe millions of other people will too.

• 17:26Casey Newton?

  Well, I mean, I don't know that they're tricked, but I do think that this is how they want us to talk to the thing that they are building. And the reason that it's striking to me is that it was less than a year ago that when we were interviewing Sam Altman and I said, how do you feel about AI girlfriends and boyfriends? He was very dismissive of the idea. He said, we do not want to build that. That is for other people. And while nothing that was showed off this week is really an AI girlfriend or boyfriend, it is absolutely flirting. Like when you listen to that clip we played of this engineer Rocky, the way that the assistant is giggling and saying, oh, Rocky, and you know, is sort of commenting on his appearance and trying to make him look handsome for his interview. I mean, this really does just feel like a generic straight guy fantasy of what a personal assistant

• 18:10Casey Newton?

  is like, right? And so I'm curious what changed inside that company that led them go in this direction. And, you know, I would have loved to ask them a question about that this week, but they did not take my questions.

• 18:24Kevin Roose?

  Yeah. So let's talk about a few of the other things that OpenAI showed off. They also showed off the coding abilities of the new GPT-4O. They showed off the translation abilities. There was a sort of a cool demo where they had one employee speak in Italian and ChatGPT translated it immediately into English and speak in English, and then they would translate immediately back into Italian. So that kind of thing is very cool. There was even a demo that I think we should listen to, because it was sort of interesting and kind of wild, where two ChatGPT voice assistants were set up to talk to each other, and one of them was using its phone camera to describe the room to the other. So let's play that.

• 19:08Speaker 04

  Hello there, new AI friend. I'm O. Ready to explore the world through your eyes? What are we looking at today?

• 19:16Speaker 05

  Hi, O. We're looking at a person wearing a black leather jacket and a light-colored shirt. They're in a room with a modern industrial feel, with exposed concrete or plaster on the ceiling and some interesting lighting. There's also a plant in the background, adding a touch of green to the space. Anything specific you'd like to know or explore?

• 19:40Speaker 04

  Hey there. Sounds like quite the stylish scene. I'm already intrigued by the modern industrial vibe. Let's start with the person. Can you tell me more about their style? Are they doing anything interesting like reading, working, or interacting with the space?

• 19:59Speaker 05

  The person has a sleek and stylish look with their black leather jacket and light-colored shirt. Right now, they seem engaged with us, looking directly at the camera. Their expression is attentive, and they seem ready to interact.

• 20:15Casey Newton?

  By the way, this conversation is very funny if you imagine that both of the AIs are assassins.

• 20:22Speaker 05

  They might be getting ready for a conversation or presentation, given their direct engagement with the camera.

• 20:28Kevin Roose?

  So that is GPT-4O, and I think there's a chance that it produces a kind of similar reaction to the original ChatGPT back in November 2022. I think we're a little more desensitized to AI than we were back then, but I do think that the ability to talk and receive responses from an AI voice assistant that sort of sounds like a human and responds in more or less real time is going to make a lot of people sit up straight.

• 20:56Casey Newton?

  I think that's true, although, you know, I've never quite known where these like voice-based assistants are supposed to sit into my life, just because it's always been so much faster to use typing and text. Now, I also think that Alexa, Siri, the voice assistants that we've had so far have just generally been pretty awful. And so this is kind of the question is, well, what happens if you have something that is like really fast and pretty good? Even then, though, I can imagine using this while like walking around town on the rare occasions when I'm like driving a car. But I don't know that I'm going to quickly get to a place where I'm just sort of sitting at my computer saying, you know, "Hey, computer, look this up." But do you think you will?

• 21:38Kevin Roose?

  I think I'll use it. I mean, I'll obviously test it because I want to, you know, spend time with it and see how the product evolves. A thing that caught my eye about this announcement, along with the product itself, is the fact that it's this new model, GPT-4O, is going to be free. So unlike previous changes to their models, which have been sort of rolled out in stages, starting with sort of paid premium users, this new model is going to be made available to free users of ChatGPT, which to me says that not only are they really like betting on this to sort of bring them a new wave of users, but also that they have done something on the backend to make serving these models, to make the inference cheaper. Because if they're giving it away to free users,

• 22:23Kevin Roose?

  you have to imagine that they've figured out some way of making that a little less costly on their side.

• 22:28Casey Newton?

  Yeah, I mean, they've said that this model is just way more efficient. You know, at the same time, I'm sure it is going to continue to be very expensive for them to serve this to the masses. This product has more than 100 million users a week, we think. And so that just is going to cost a lot of money. But I think it is good news for people who've been using the free version of ChatGPT. Because for so long, there's been this real gulf, I think, in understanding between people who were using it on the free plan versus using it on the paid plan. Because the paid plan just tended to be so much better. That if you'd only use the free plan, your sense of how powerful this is, I think, would be really limited. So I'm curious to see what happens now that more people have access to the good stuff.

• 23:08Kevin Roose?

  Yeah, but you have to imagine that if OpenAI is releasing something like this now, a year from now, there will be many versions of an AI voice assistant like this. Some of them will be open source. Some of them will be able to talk to you about basically anything you want. And there will be very few, if any, safety filters on them. So I just think that we are we are careening toward the future as depicted in the movie Her. And I don't know, something about that just makes me a little queasy.

• 23:37Casey Newton?

  Yeah, I mean, although it is important to say that in the fictional world of Her, those assistants really do have sentience, right? They're acting on their own, they're having Independent relationships with other virtual assistants, right? It imagines a world essentially where what is often called artificial general intelligence has already been achieved. We are a long way from that. At the end of the day, these are still predictive models. They are guessing and yes, they are getting uncannily good at it and they really do cast a spell. And I'm sure I will be taken in by these assistants, you know, just as much as anybody else. But at the end of the day, they are fictional and they are not what is actually being represented as true in the movie.

• 24:18Kevin Roose?

  Yes, correct. That is important to say. It is also important to say, like, we have not gotten our hands on this thing yet. And so I will be waiting for OpenAI to open this up so that I can put it through its paces myself. Although honestly, I've spent a lot of time talking to AIs recently. I'm kind of ready for some human conversations. So maybe I'll, I'll, I'll leave that for you.

• 24:40Casey Newton?

  Really? I was gonna invite ChatGPT to co-host the show with us.

• 24:44Kevin Roose?

  We should try it.

• 24:45Casey Newton?

  It, it, you know, it, it, it's so friendly and it, you know, it's, it's just gonna love us so much. You wouldn't want to have that around.

• 24:50Kevin Roose?

  Oh, you go for that. I'll take the week off.

• 24:52Casey Newton?

  All right, sounds good.

• 24:54Kevin Roose?

  One more thing on OpenAI. This was not part of their announcement on Monday, but has broken in the days since. We learned this week that Ilya Sutskever, the chief scientist of OpenAI, who was a central figure in the messy board drama that involved firing Sam Altman last year, is leaving the company. Also leaving OpenAI is Jan Leike, who is part of the leadership of the so-called Super Alignment Team, a major figure in the world of AI safety. And so he is resigning. He announced this week on social media.

• 25:32Casey Newton?

  And here, here's why this is a really important story, Kevin. It was July of 2023 that OpenAI says that we're gonna create this thing called a Super Alignment Team. And Super Alignment to OpenAI was going to be the way that they ensure that as these systems that they are building grow more powerful, they're going to ensure that they always act in the interests of, of human beings, right? And of course, there's a lot of controversy over whether we should be Paying any attention to this at all. Maybe it's all just a pipe dream. There are a lot of really smart people that think that this is something that absolutely needs attention from really smart folks. And so Jan and Ilya were the two people who were leading that effort at OpenAI. There was, I would say, a relatively few other prominent people there who cared about

• 26:16Casey Newton?

  safety. Helen Tonner, who was on the board, left. There was this other guy, Leopold Ashenbrenner, who worked on safety stuff, and he was fired recently. So now their super alignment team has no leaders, and I'll just say it, no known employees. So I would love to hear more about how they're thinking about safety, because my strong sense is that this is an area where they have pulled back investment. And frankly, they just look more and more like a regular tech company and less and less like a nonprofit research lab that they were set up to be.

• 26:45Kevin Roose?

  Yeah, I agree with that. And I want to know more, obviously, about why Ilya and Jan are leaving the company. I want to know, I mean, I think in some ways, it is just sort of the natural fallout of what happened last year with the board drama and the firing of Sam Altman, which is that he sort of came back in this triumphal return. His enemies were vanquished. And obviously, if you're a person like Ilya, who had voted to fire him, your future may not have been very bright at that company. Now, obviously, OpenAI is trying to put sort of a positive spin on Ilya leaving. You know, they posted a photo of Ilya and Sam Altman and Greg Brockman and Mira Maroti

• 27:27Kevin Roose?

  and other company sort of leaders on what appears to have been his last day, making it seem like this was all sort of, you know, very cordial as a departure. But I think we both know that there has been a lot of tension inside this company between the faction that is sort of pushing for more growth and more commercialization and more capabilities, research and more products, and the faction that is very concerned that all this is heading too fast, in a dangerous direction, and that we should pay more attention to safety. So I think that debate is still very alive and well inside OpenAI. But I think it's safe to say that if you were a person who was worried about the safety posture of OpenAI last week, you should be even more worried about it today.

• 28:13Casey Newton?

  Yeah, and I would say if you were a person at OpenAI worried about safety last week, you might not work there this week.

• 28:20Kevin Roose?

  Right. It's not a safe time to be involved in safety at OpenAI.

• 28:24Casey Newton?

  They need a safe space over there.

• 28:28Casey Newton?

  When we come back, I head to the developer conference of Bing's biggest competitor.

• 28:33Kevin Roose?

  You're talking about Google?

• 28:34Casey Newton?

  That's right.

• 28:52Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner, Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 29:26Speaker 06

  Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? Without systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And Outsystems gives you the platform to make it agentic on your terms with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 30:01Speaker 07

  Today's AI builders need infrastructure that scales with their ambitions. Iron is a vertically integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Iron's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Iron, AI cloud for the next big thing.

• 30:35Kevin Roose?

  All right, Casey. Well, the other big AI news this week happened at Google I/O, the company's annual developer conference, which was held on Tuesday down in Mountain View. I couldn't make it this year, but you went and attended. And tell me what it was like.

• 30:49Casey Newton?

  Well, once again, the heavy burden of doing the actual reporting on this podcast falls to me, but I was only too happy to do it, Kevin, so we'd have some fun things to talk about. But I knew you would be sad to miss it. So I went ahead and I got you the official 2024 Google I/O tote bag.

• 31:05Kevin Roose?

  Oh, that's very nice.

• 31:06Casey Newton?

  Yeah. Now, if you're just listening to the podcast, you should know Kevin just accepted the tote bag in flagrant violation of the New York Times ethics policy, and we do have that captured on video. But listen, Kevin.

• 31:16Kevin Roose?

  I think I can accept gifts as long as they're not worth more than $25. That's definitely the policy.

• 31:21Casey Newton?

  That definitely isn't. It's made out of publications that Google killed last year. So anyways, we had a great time.

• 31:28Kevin Roose?

  Well, thank you for the gift. I will put it in my tote bag of tote bags.

• 31:31Casey Newton?

  Okay, great. Well, listen, we had, we had a good time down there. Now, I will say, I do want to tell this story, and this story makes me look annoying, but I just got to get it off my chest. Anyway, it actually sucked getting into Google I/O this year. I've never had a problem getting into Google I/O before. I've been going for, you know, more than 10 years. And yet, this time, there was something I, now, maybe this has been true in previous years, but I did not notice it until this time. They had Live Nation managing this event. Okay?

• 31:56Kevin Roose?

  Wow.

• 31:57Casey Newton?

  And I think that's just because monopolies love monopolies. And so they want to make sure that they sort of help each other out. And so anyway, there's this one entrance that the press has always been able to use. And so I went to use it this time around. And they were like, basically, scram, kid, go wait in, you know, in a line with everyone else. And I said, okay, fine. And I stand in 15 minutes in this other line, you know, with everyone else. And then finally, a man with little paddle says, oh, by the way, if you're in the press line, you should go to that other entrance, which was the first place that I went.

• 32:24Kevin Roose?

  Oh, thank God. I was worried you were gonna have to wait in a line for 15 minutes with the hoi polloi.

• 32:29Casey Newton?

  I only got through half the line.

• 32:30Kevin Roose?

  Thank God you got your VIP pass that lets you cut the line.

• 32:34Casey Newton?

  So then I get back to the press entrance. I say, hi, I'm the press. They just told me it was the press entrance. The guy goes, no, you can't use this entrance. And Kevin, I had to raise my voice. I said, I've literally just been told to come here. And then finally, you know, the guy throws his hands up. And it's like, well, okay, you know, if you're gonna make a whole thing about it, then just use the entrance.

• 32:52Kevin Roose?

  Wow.

• 32:52Casey Newton?

  So yeah, so I got in there. By that time.

• 32:54Kevin Roose?

  Why, you give this guy a podcast, he starts asking. Like a diva. All of a sudden, it's waiting in this line. Do you know who I am?

• 33:02Casey Newton?

  Listen, all I wanted to do was find out how this company is about to destroy my business. And the very least they could do was let me into.

• 33:07Kevin Roose?

  Did you say Google me?

• 33:09Casey Newton?

  I did not.

• 33:11Kevin Roose?

  I'm shocked. It's very effective around there.

• 33:14Casey Newton?

  All right. So.

• 33:15Kevin Roose?

  So set the scene.

• 33:16Casey Newton?

  It's a developer conference. It's at the Shoreline Amphitheater in Mountain View. It was a balmy day, sort of really California, you know, showing off its best side. And in the morning, they have a big keynote where for two hours, Google says, here are all of the things that we have been working on that we're planning to release in the next year that we can fit into a two hour presentation. And, you know, as you probably guessed, they had a lot to say.

• 33:41Kevin Roose?

  Yeah. So let's go through some of what they announced on Tuesday. They announced a new AI video generation tool called VO, a new AI image generation tool called Imagine 3. They also announced another version of their Gemini flagship AI model called Gemini 1.5 Flash, which is optimized to be fast and cost efficient. And they announced a version of Gemini 1.5 Pro. Is it Gemini Pro 1.5 or Gemini 1.5 Pro?

• 34:13Casey Newton?

  That's none of my business. There was so much word salad at this event, and I'm just going to see if any of it ever shows up in my life ever again. That's my approach with what everything you just said.

• 34:22Kevin Roose?

  Right. So they announced a larger context window that will allow developers to try 2 million tokens in their context window.

• 34:30Casey Newton?

  Now, that's actually a big deal.

• 34:32Kevin Roose?

  Why?

• 34:33Casey Newton?

  Okay, so, you know, if you're not familiar with what a context window is, it's basically how much of your stuff can you bring to the AI to have it look at? Why does this matter? Well, maybe you're a graduate student and you have a bunch of papers that are PDFs and you want an AI to help you summarize them, to ask questions of them. Well, now, all of a sudden, instead of maybe just uploading one PDF, with 2 million tokens, you'll be able to upload, you know, like thousands of pages potentially.

• 35:03Kevin Roose?

  Yeah.

• 35:04Casey Newton?

  So, you know, we should say that the 2 million. Token window that is available only in preview. But if you pay Google 20 bucks a month for their little Gemini advanced thing, you now have up to a million tokens and that is much bigger. So that's important.

• 35:18Kevin Roose?

  Yeah. So that's some of their AI announcements. But I want to spend most of our time right now talking about something else that Google announced, which is that Google is now bringing generative AI answers directly into its core search engine. So, starting this week, hundreds of millions of Google users in the US will start seeing a feature called AI overviews. This is a feature that used to be called search generative experience when Google was sort of testing it out.

• 35:46Casey Newton?

  Another huge win for the Google branding team. Search generative experience.

• 35:49Kevin Roose?

  Right. So this is a feature that basically when you go to Google and you search for something, Google will generate an AI summary that will live in the search results page above the traditional search links. This feature has been rolling out gradually, but they said that as of this week, hundreds of millions of users in the US will see it. And by the end of the year, it will reach over a billion users. So what do these AI overviews actually look like?

• 36:16Casey Newton?

  Well, so there's what they look like today and there's what's coming. What do they look like today? It is a box on top of search. It's, you know, some bullet points, essentially a webpage within a webpage that summarizes the topic for whatever that you've been looking for. But Google's plans are much more ambitious. They showed off a ton of examples. You know, for example, if you move to a city, you can just sort of say that to Google and you'll get an AI overview that says, here's everything that you might be interested in. Now that you move to a new city, we'll help you find a dog walker, we'll help you find a dry cleaner. Maybe you want to plan an anniversary in Dallas. It will say, well, you know, here are some great spots for an anniversary dinner. And based on the time of year, we're going to show you a place with a roof deck because it's warm outside. Right? Or, and you know, Google executives love to talk about travel.

• 36:59Casey Newton?

  Can I just tell you my quick joke about every tech demo? This is, this is my impression of every tech demo. Okay, ready? You know, I love to visit Antarctica for lunch with my wife, but there's only one problem. Where am I going to get a helicopter? That's every tech demo I've ever seen.

• 37:12Kevin Roose?

  It's so true. It's so true. It's like, you know, when I'm, when I'm looking to furnish my vacation house in Tahoe, but the only place with the, the custom boar's head statues is, is in Zanzibar.

• 37:27Casey Newton?

  That's right.

• 37:27Kevin Roose?

  How can I arrange two day shipping? It's, it's ridiculous.

• 37:31Casey Newton?

  Exactly. So, so they had a travel example, which was basically like, oh, you know, I'm planning this trip. And, you know, with this, this AI tool, it'll say, okay, well, you know, we know, we know what your flight times are because we read it in your Gmail. And based on that, we're going to sort of have you do this dinner, the first night that you arrive, and then we'll plan a walking tour for the next day. And they'll, they'll put together the whole itinerary for you, or they'll plan a whole meal plan for you. So this stuff is starting pretty small and frankly boring, but the vision here is this stuff is going to be doing planning for you. This thing is an AI assistant that lives within Google now that is doing research projects for you and that, you know, pretty soon is probably going to be booking travel for you.

• 38:09Kevin Roose?

  Yeah. And I knew that this had been announced at I O because my phone started blowing up with texts from other journalists, people who work in the media industry that mostly said some variation of, we are so screwed. And I think we should just explain a little bit why the online media business is freaking out about this change, because I don't think it's totally obvious unless you work in our industry.

• 38:34Casey Newton?

  Yeah. Well, so much of the way that the internet is funded depends on people visiting individual webpages. Those webpages have ads on them. A lot of those ads served by Google, by the way. They also, as people visit those webpages, publications, publishers, businesses have a chance to get you to sign up for a newsletter, maybe subscribe to something. So Google is this massive funnel that winds up being the economic engine of the entire web. And the way that it works is people Google things and they see links and they click on the links, right? And this has been the state of affairs for so long now that I think people take it for granted.

• 39:13Casey Newton?

  Once you take an AI and you start to just summarize those links and essentially tell people, You don't have to visit websites anymore. All of a sudden, that thing starts to break down. You know, there was this phrase that they used a couple of times at the keynote, Kevin. They said, let Google do the Googling for you. And it struck me so hard because the implication there is that being on the web is a chore. And the Google of the future is not going to take you to the web. Google is going to give you everything you need within Google itself.

• 39:45Kevin Roose?

  Yeah, that struck me too. And they also talked about sort of taking the legwork out of search. And as they were saying that, I was thinking like that legwork has funded basically the entire digital media. I mean, you know, I don't want to overstate things that are, you know, Google is not everything, but is by far the biggest supplier of traffic to many digital publications. So Google executives have obviously anticipated that publishers might freak out about this change and they had some responses ready. They said, you know, in our tests, we've found that users who saw these AI overviews tended to conduct more searches and visit a more diverse set of websites. They also said that the links that appear in these AI overviews, because there's a

• 40:29Kevin Roose?

  little like section at the bottom of the overview where you can go to click on things to learn more. They said that those links got more clicks than the sort of traditional search links below them. And Liz Reed, the Vice President of Search at Google, said on Tuesday that the company would, quote, "continue to focus on sending valuable traffic to publishers and creators." But I think it's fair to say that publishers are skeptical.

• 40:52Casey Newton?

  Well, and she said, we're going to send valuable traffic. She didn't say we're going to send more traffic because guess what? They're not going to. And when analysts look at this, they're predicting somewhere between 20 to 40% of Google search traffic is at stake here. And I think it's fair to say that most publishers do not have a plan for what happens if they lose 40% of their traffic in a year.

• 41:10Kevin Roose?

  Yeah, clearly the the risk here is that all of this traffic will just evaporate. I also read a good post on Search Engine Land, which is a good website that covers the search industry.

• 41:20Casey Newton?

  And one of my favorite theme parks.

• 41:23Kevin Roose?

  They were basically talking about how the problem for publishers here is not just that they might see their overall traffic decline, but that they can't really do anything about it. With other chatbots, other AI experiences, publishers can say opt out. They can, they can block the crawler for ChatGPT or Claude from accessing their site if they don't want those products to be able to summarize what they find on their websites. But with Google, because their technology also crawls for the search index, there's no way right now to decouple those things. You can't be a publisher and say, I don't want my content to appear in these AI overview summaries. I only want to be listed on regular

• 42:06Kevin Roose?

  search results in Google. You have to kind of say yes or no to the whole thing.

• 42:10Casey Newton?

  Yeah.

• 42:11Kevin Roose?

  But I think overall, if you are a publisher that derives a big portion of your traffic from Google, this is a big deal for you. This is the big one. I would be pulling the fire alarm.

• 42:21Casey Newton?

  Yeah, it is absolutely time. But you know, Kevin, at the same time, I have to say, in some ways, we had to know this was coming. You know, I've been writing about Google for more than 10 years now. And you go back a decade and they were talking about Google as the Star Trek computer. This is their vision. You know, if you've seen Star Trek, Captain Kirk or whoever will just say, computer, and then, you know, is able to ask the question and then they can fire the photon torpedoes or whatever. And that is a world where, you know, the computer was not saying, and actually I got that fact from the New York Times, right? And here's a link.

• 42:55Kevin Roose?

  Here's the Wirecutter's favorite photon torpedo.

• 42:58Casey Newton?

  The sci-fi vision of this has always been, all of the inputs, all of the labor is abstracted away, and you just have this perfect assistant. So they've told us that this is what they were building. I think it just seemed so fantastical for so long that publishers, and by the way, not just publishers, anybody else who depends on Google traffic, helping people discover their business, which is a lot of people, right? It was so easy for those folks to just dismiss that. But the point of what we're saying today is, the moment is here. The Star Trek computer is coming into view, and people really do need to have a new plan.

• 43:33Kevin Roose?

  Yeah. Is there anything that publishers can do at this point to sort of keep their Google traffic from falling off a cliff or to adapt to these new AI overviews?

• 43:43Casey Newton?

  You know, I think the only thing that works, and it's really hard is, but you have to build a direct connection with an audience or with a customer base if you're a different kind of business, right? People have to think of you independently and they have to, you have to have some way of maintaining a connection with them. So, you know, this is why email newsletters have become such a hot commodity in the publisher world, right? Because it allows us to sidestep the need for you to, you know, look up a piece of information in order to find it. Instead, someone will just kind of send it to you. So, you know, I think podcasts are another thing, frankly, right? Because like, we don't rely on Google search really for people to find the podcast. It just gets delivered to them in a feed once they become subscribers. So, these are the kinds of things that I think more publishers need to be thinking about.

• 44:28Casey Newton?

  But even then, Kevin, the scale of the businesses is going to decline because in its heyday, Google really would send millions and millions of visits to websites. And in a world where that's about to decline, the size of the audience that you can build is smaller, which means that the staff that you can afford to pay gets much smaller. So, I just think that we are going to be living in the fallout of this for a really long time. And I gotta say, man, I was, I was emotional yesterday in a way that I was not proud of. And, you know, I mean, not that I was like crying at Google I/O, but I just felt really out of sorts, you know? Within the past 48 hours of this recording, this, I have one friend who got laid off from their media job, and I had two other friends who are currently employed in media jobs who told me that they're looking for other jobs right now.

• 45:13Casey Newton?

  And in every case, the answer is the same. The money just isn't there the way that they all thought it would be, right? And so that's just, you know, my story, past 48 hours, three people looking to leave. We already, you know, we could name the publications that have shut down in the past year. There are more that are going to shut down in this year. And so I understand, you know, maybe this just sounds like two reporters navel gazing and, you know, maybe we're spending too much time on this. But what I'm telling you is, the thing that is happening to the media businesses is going to happen to the other businesses. It turns out that in a world where everything is online, how people Find stuff, how they discover new businesses, that matters a lot. And when all of that is about to be hidden away behind some black box algorithm that summarizes what it thinks is the best thing based on whatever and who knows?

• 45:59Casey Newton?

  I truly do not think people are prepared for this world.

• 46:01Kevin Roose?

  Yeah, and I think Google obviously has, you know, reasons for wanting to, you know, evolve its search engine in this way. I also think it's a very risky move for them for a few reasons. One is, as we know, these AI overviews can get things wrong. They can, they can be untrustworthy, they can be erratic. I saw there was a screenshot going around the other day on social media where someone asked Google, what are some food names that end with um? And Google's experimental AI overview said, here are some fruit names that end with um. Appleum, bananaum, strawberryum, tomatum, and coconut. And like, that's a funny example, but I think there are going to be a lot of examples

• 46:46Kevin Roose?

  of these AI overviews, at least in their early days, just getting things wrong. I do think that will start to erode trust in Google over time.

• 46:54Casey Newton?

  I mean, I think that is possible, but I would not count on it because there are mistakes on all the webpages that people are visiting to. And I'm not sure how much it just matters in the end.

• 47:03Kevin Roose?

  Yeah, I think there's also a business risk here too. I mean, Google, you know, it does make the vast majority of its money from showing ads right there on the search results page. And it remains to be seen whether they can sort of monetize these AI overviews as effectively as their traditional search results. By the way, they also make a lot of money showing ads on third-party websites. Their Google Ad Network is a big part of their business, tens of billions of dollars a year. If all of those webpages suddenly have to close up because they're not getting any traffic from Google anymore, then that part of Google's ad business has to shrink as well.

• 47:37Casey Newton?

  So that is totally true, but keep in mind, Kevin, it is Google that has its fingers on the knobs. Google is the one in control here. It gets to decide how fast or slow it moves in a case like this. If it does something that dramatically changes something in a way it doesn't like, it can always edge that back. The important thing is it is the one in the driver's seat now for what the future of the web looks like. And what they are telling us is Google is just going to keep more and more of the median internet experience to itself within its own walled garden.

• 48:07Kevin Roose?

  Yeah. Do you find any reasons for optimism in what you heard this week at I O?

• 48:12Casey Newton?

  Well, you know, I was so desperate for optimism that I texted a tech executive, you know, who I have some of these back channel conversations with. And this is an executive who I would think, I think it's fair to say, generally thinks that I over-rotate on being, you know, concerned and a worrier, you know, because, you know, you talk to people in tech, they're mostly just optimistic about everything. They think everything is going to be fine for everyone. Of course, they also live in mansions. But anyway, this person said to me that essentially, people still want novel content and experiences on the web. They want to have shared experiences with their friends. They want real relationships, right, with, you know, with publications, with writers. And so if you are the kind of media business, or you're the kind of journalist

• 48:56Casey Newton?

  that does have an authentic audience, that you are doing a really good job for, that you're doing something novel, you're doing something really smart, you might wind up being in a better position coming out of all of this. Because as much as I love the media, it is also true, there's a lot of garbage in our business. And there are a lot of unscrupulous publishers that are publishing lists of, quote, you know, "the best vacuum" that haven't tested a single vacuum in their entire life, but they figure there's some upside in getting some of that affiliate revenue. So if you want an optimistic case, it is that Google will sort of cause all of those bad actors to wash out, and it will leave a few brave, proud publishers that are doing honest work standing to reap more of the benefits.

• 49:38Kevin Roose?

  Yeah. I think my reason for optimism here is that I actually do think that it is in Google's long-term self-interest for there to be a vibrant and healthy internet. I think that if Google starts sort of just aggregating and summarizing the internet and spoon-feeding it to people through these AI overviews, I think that'll work fine for a while. And then I think they will start to see one website close, and 10 websites close, and 100 websites close. And within a couple of years, you're looking at just a very diminished internet. And Google in that world is not going to have, you know, what is its AI going to crawl and summarize if there are no websites left? So I just think that, you know, we will, we will end up in a world where all of the valuable information

• 50:21Kevin Roose?

  on the internet is sort of hidden behind paywalls and subscription publications, maybe newsletters and podcasts. And you will just see sort of an impoverished Google that is out there scraping as hard as it can, trying to summarize what it finds and not coming up with much.

• 50:36Casey Newton?

  Yeah, well, I do think that that is a possibility. I also think Google will just get increasingly good. I was having this discussion with friends last night of like essentially this exact thing, you know, where is Google going to find the information anymore? And, you know, you might think of one example, well, let's say you go to Paris and you want to know like, what's the best bakery? Well, today I would rely on a travel writer who has been there and been there recently, right? Because, you know, the best bakery from in 2024 might not be the best bakery in 2020. And we started talking about it's like, well, you know, Google could actually just kind of look at the foot traffic in all of the bakeries in Paris and say, this one is the most popular. And that is not actually going to require there to be a healthy web feeding at that information, right? These companies collect so much data about us and they are going to come up with so many more novel

• 51:19Casey Newton?

  ways to use it that I really do wonder if maybe we aren't flattering ourselves when we tell ourselves that they need us.

• 51:27Kevin Roose?

  I hope you're wrong.

• 51:29Casey Newton?

  I hope I'm wrong. I've never hoped for anything more.

• 51:32Kevin Roose?

  But I do, I think this is sort of the flip side of AI vertigo, this feeling of like, wow, the future is coming is like, I don't know that our institutions, including our media institutions, understand the gravity of what happened this week at Google, which was that they did essentially declare that we are moving into a very different era of the internet.

• 51:51Casey Newton?

  We had talked on this show for more than a year now about automation, AI, coming for jobs. It is seemed like something on a kind of medium to distant horizon. But I do think that at I O this week, we saw the beginning of an arrival of something like that. And again, it is starting in the media, but it is not going to stop there.

• 52:14Kevin Roose?

  Yeah. Can I just say one more thing before we close the segment? I got a little emotional watching this announcement too, for maybe some similar reasons as you. But my emotion was like, I just feel like I'm being gaslit here. Like publishers, people who work in online media have known that Google has plans to shove generative AI into search results for years. We talked to the CEO of perplexity a few months ago, who basically told us like, yes, if we do start generating these AI summaries of search results. People will click on less news. And I actually admired that he was honest about that. Google is not being honest about this. They are telling people we are still committed to sending traffic to publishers.

• 52:57Kevin Roose?

  People are not going to stop clicking on links just because they have these like AI summaries above their search results. And I I really just wish that someone at Google would come out and say, yes, your traffic is going to fall off a cliff. But here's why we think it's important anyway. But instead, they are trying to signal to users that this is going to be a good and useful thing. They are trying to make publishers feel less scared. And frankly, I it just felt like being it felt like attending a funeral disguised as a wedding.

• 53:27Casey Newton?

  Yeah, and this actually is the reason that we need media to exist in this world is because we will try to tell you the truth about something. We will try to go beyond our own narrow self-interest and tell you what's going on on this planet.

• 53:37Kevin Roose?

  It drives me insane. Just just tell me. We're going to like, you know, basically destroy the foundations of the open internet.

• 53:43Casey Newton?

  No, I felt exactly the same way. All right, Kevin, well, I'm afraid that's all my context window can hold for this conversation. So we're just going to sort of have to purge that and we will have to return to the subject another time.

• 53:53Kevin Roose?

  Yes, I think we will have plenty more opportunities to revisit the subject.

• 53:57Casey Newton?

  And and we'll be you can find us podcasting under an old bridge on the abandoned railroad tracks.

• 54:04Kevin Roose?

  What's your post media plan for when all the Google traffic? Like, are you going to become a farmer if if the digital media dies?

• 54:12Casey Newton?

  I'm getting really into TV and VCR repair.

• 54:15Kevin Roose?

  I think there's a bright future in that.

• 54:21Kevin Roose?

  When we come back, we'll play a round of HATGPT.

• 54:35Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 55:09Speaker 06

  It's time to innovate at the speed of AI without compromising control. That's OutSystems. Take a look at what happened with one global retailer during a period of massive expansion. They used OutSystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. OutSystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 55:44Speaker 08

  Your sales team is at risk of missing quota. Don't panic. Just ask Rippling AI. Built on your real time people and business data, Rippling AI can pull metrics from Rippling and your CRM into a meeting-ready dashboard showing quota attainment, headcount trajectory, and revenue to quota by region. In seconds, you'll see what's behind your quota risk and fix it before it's missed. Head to rippling.ai/hardfork to get the only AI built to give you full visibility and take complex actions across your whole business. That's R-I-P-P-L-I-N-G dot AI slash hard fork. Sign up today.

• 56:19Casey Newton?

  Well, Kevin, from time to time on this show, we like to take a quick run through the news in a segment that we call HatGPT.

• 56:26Kevin Roose?

  Yes, let's play it.

• 56:33Casey Newton?

  HatGPT is, of course, a segment on our show where we take headlines from the week and we pull them out of a hat and we riff on them a bit until one of us gets bored with what the other person is saying, at which point we say,

• 56:45Kevin Roose?

  Stop generating.

• 56:46Casey Newton?

  And then we move on to the next headline.

• 56:47Kevin Roose?

  So Casey, should we use the baseball hat or the bucket hat?

• 56:50Casey Newton?

  I think we should use the bucket hat in honor of Rocky, the OpenAI employee who put on a bucket hat this week when demoing the new her assistant.

• 56:58Kevin Roose?

  All right, I like it. Grab it.

• 57:01Kevin Roose?

  Do you want to do, do you want to go first?

• 57:03Casey Newton?

  Uh, why don't you go first?

• 57:05Kevin Roose?

  Okay, well, let's make, will you mix them up for me? And I want to introduce some randomness here.

• 57:09Casey Newton?

  We have to mix up all the headlines in the hat.

• 57:11Kevin Roose?

  Be sure to just like jostle and hit the microphone as many times as possible.

• 57:14Casey Newton?

  Yeah, we're jostling, we're hitting that microphone.

• 57:16Kevin Roose?

  Yeah, just, just.

• 57:17Casey Newton?

  This microphone is in trouble.

• 57:18Kevin Roose?

  Yeah.

• 57:19Casey Newton?

  It's broken now, actually. All right.

• 57:22Kevin Roose?

  Okay, first story. FTX customers poised to recover all funds lost in collapse. This comes from my colleague David Yaffe Bellany at the New York Times. Customers of the failed cryptocurrency exchange FTX are poised to recover all of the money they lost when the firm collapsed in 2022 and receive interest on top of it. The company's bankruptcy lawyers said last week, under a plan filed in federal bankruptcy court, virtually all of FTX's creditors will receive cash payments equivalent to 118% of the assets they had stored on FTX. A judge has to approve the plan and it may still take months for the money to reach customers. This is an amazing story.

• 58:02Casey Newton?

  That's true. Now, others have pointed out that if you had simply taken the money you had in FTX when it all shut down and just like used it to buy Bitcoin, you'd actually be in a way better position than you are today. So it is not right to say that everybody sort of came out smelling like a rose here. But based on what you've said, Kevin, I do have four words that I would like to say to the people of the United States, and they are these, free Sam Bankman Fried. He did nothing wrong!

• 58:27Kevin Roose?

  No, no, no, that is not the moral here. So Sam Bankman Fried has been sentenced, he is currently doing a 25 year prison sentence for his, uh, his role in, uh, this collapse.

• 58:38Casey Newton?

  His alleged role.

• 58:39Kevin Roose?

  No, it's not alleged if you've been convicted and sentenced and you're in prison. No, it's, we can just say it. He did a fraud.

• 58:47Kevin Roose?

  But this is an amazing result in part because while Sam Bankman Fried does appear to have been a giant fraudster, he did actually have a good nose for investments. So one of the things that he put money into Solana, which is a crypto token, um, part of the reason that they're able to return this money to creditors is because Solana has been booming, Bitcoin has also been booming, and some of the other investments that Sam Bankman Fried made, including a stake in Anthropic, the AI company, have become much more valuable since the firm's collapse, which is just an amazing testament to the fact that you can be a giant fraudster and still be pretty good at investing.

• 59:23Casey Newton?

  Yeah, when this guy gets out of prison, I think he's going to make a great partner at Andreessen Horowitz.

• 59:29Casey Newton?

  All right, let's stop generating.

• 59:31Kevin Roose?

  All right, your turn.

• 59:32Casey Newton?

  And I'll do one here.

• 59:35Casey Newton?

  Bumble's founder says your dating AI concierge will soon date hundreds of other people's concierges for you. This is from Fortune at the Bloomberg Tech Summit in San Francisco. Last week, Whitney Wolfe-Hard, the founder and former CEO of Bumble, predicted that singles would one day deploy AI dating concierges to help them find love. Wolfe-Hard told the crowd, "There's a world where your dating concierge could go and date for you with other dating concierges, and then you don't have to talk to 600 people." So Kevin, what do you make of this form of theoretical future dating?

• 1:00:07Kevin Roose?

  So she got a lot of backlash to this comment. I was at the Bloomberg Tech Summit where this happened, and immediately when it went onto the internet, people started saying like, "This is dystopian." I want to defend this idea.

• 1:00:20Casey Newton?

  Okay.

• 1:00:20Kevin Roose?

  Because I think we are, you know, I'm not on dating apps, but my impression is that most people who go on these dating apps are already basically glorified algorithms, right? They are using one of like three preset lines that they've come up with. They are basically trying to quickly assess compatibility, but they're not, you know, they're not having real, deep, authentic conversations, at least not right away. And if AI could save you some time by going out and dating people on your behalf and saying, "This person's a loser, this person, this person's mean."

• 1:00:52Casey Newton?

  Wow!

• 1:00:53Kevin Roose?

  I think this could save people a lot of time. What do you think?

• 1:00:55Casey Newton?

  I do not think it would save people a lot of time, because you know, here is the thing, you know, I'm also not on dating apps anymore, hashtag soft launch, but what I remember from when I was, was that oftentimes you would say to somebody, you know, "Hey, like, how's your week going?" And they would just respond with, "Hey." And then you would think, "Oh, this person isn't super engaged, I'm gonna move on." A world in which the person who responded "hey" is using a bot to do responses on its behalf is doing everyone a disservice, right? Because if this is a person who can't even be bothered to complete a sentence when I say hello to them, the odds are that I'm gonna enjoy my time with him is just sort of like very low. So is there some world where some sort of AI something is helpful in? You know, enhancing the online dating experience.

• 1:01:40Casey Newton?

  I'm very open to that. But if it is essentially just AIs tricking each other into thinking that the other person is like a really good writer and has a huge personality, then they're going to be wasting even more time than people are wasting on these apps already.

• 1:01:52Kevin Roose?

  Yeah, I do think it opens up some interesting possibilities, but I can see how people would find it dystopian. My main thing is like, I just hope that they, um, what am I, what am I saying?

• 1:02:02Casey Newton?

  What are you saying, Kevin? I've always wanted to ask you this. What are you saying?

• 1:02:07Kevin Roose?

  Yeah. In conclusion, I'm very glad I'm not single. All right, stop generating.

• 1:02:16Kevin Roose?

  Sonos says its controversial app redesign took courage. This is from The Verge. Sonos, the connected speaker company, released a major update to its mobile app earlier this month. To quote The Verge, "In the days since, customers have complained about missing features like sleep timers, broken local music library management, and no longer having the ability to edit playlists or the upcoming song queue. Customers were very angry about this. And in response, Sonos's chief product officer told The Verge, quote, "It takes courage to rebuild a brand's core product from the ground up and to do so knowing it may require taking a few steps back to ultimately leap into the future." Casey, what do you make of this?

• 1:02:53Casey Newton?

  You know, I'm not sure that I would say that Sonos executives' move took courage here. I would say it seems more likely that they took mushrooms or some other sort of psychedelic substance that sort of melted their defenses until all they could do was give us the app that they did. Look, I have been in a daily war against my Sonos system for years now. When it works, there's nothing better. The problem is, it doesn't work a lot of the time. And so when I got this new app, I thought, maybe they have finally solved all of my issues. And I swear to the heavens, Kevin, the volume slider in this app, which let's just say is one of the core things I want to do to my Sonos system a lot of the time, is adjust the volume, it is not persistent.

• 1:03:35Casey Newton?

  It only appears when Sonos sort of knows that something is playing. Anyway, the point of the story is, the number of things that this app could do could fill a segment much longer than ChatGPT.

• 1:03:46Kevin Roose?

  I'm confused because it seems like the job of The Sonos speaker should be fairly simple, which is play the music that I want when I connect to it.

• 1:03:53Casey Newton?

  You're so right, bestie.

• 1:03:54Kevin Roose?

  I don't own a Sonos speaker, so you tell me, is that your experience?

• 1:03:58Casey Newton?

  No, my experience is I say, "Sonos, please connect to Spotify," which is what I play music from, and Sonos says, "I have no idea who you're talking about. I've never seen this man before, and I want you to leave my house." And I say, "This is the entire job that I've given you." So it's a real problem, and I wish Sonos continued courage as they try to build their first functional app.

• 1:04:17Kevin Roose?

  All right, stop generating.

• 1:04:18Casey Newton?

  All right.

• 1:04:26Casey Newton?

  Oh, this is interesting, Kevin. Inappropriate behavior shuts down the Dublin to New York City portal. Less than a week after two public sculptures featuring a live stream between Dublin, Ireland, and New York City debuted, quote, "Inappropriate behavior and real-time interactions between people in the two cities has prompted a temporary shutdown." The portals, as the sculptures are called, are the brainchild of a Lithuanian artist named Benedictus Gillis, and they were shut down Monday night after videos spread on social media of visitors misbehaving in front of them, including an OnlyFans model in New York flashing the portal and people in Dublin holding up swastikas. So Kevin, what are the odds if you just set up two giant cameras in Dublin and New York that people would troll them? What are the odds?

• 1:05:10Kevin Roose?

  Close to 100%, I would say. No, so I was bummed out about this because I did actually think the portals were a cool idea. This is basically like a giant sort of screen that was placed in kind of a circular enclosure that had a live streaming feed where you could basically go up to this thing in New York City and see people who are staring into the portal in Dublin, and you could kind of have like this wormhole between the two cities.

• 1:05:34Casey Newton?

  And we could recognize our shared humanity.

• 1:05:36Kevin Roose?

  Exactly, but as it turns out, it was more useful for flashing and holding up swastikas. This is why we can't have nice things.

• 1:05:43Casey Newton?

  Well, you know, Kevin, I'm Irish, and I have to wonder if there was just sort of something about our mischievous nature that, you know, led the New Yorkers to misbehave.

• 1:05:50Kevin Roose?

  I do also just like the name "The Portals." You know, it's sort of of a genre with "The Sphere" in Las Vegas. I just think we need more, more things of that nature, things that are sort of the titles of, of Michael Crichton novels made into reality.

• 1:06:04Casey Newton?

  That are just sort of like vaguely ominous.

• 1:06:05Kevin Roose?

  Yes.

• 1:06:06Casey Newton?

  Yeah. All right, stop generating.

• 1:06:11Kevin Roose?

  Oh, this, this one's for you. "After 28 years, someone opened an unopenable door in Super Mario 64." This comes from The Verge. "On April 22nd, user AlexPalax1 posted a video on Discord showing how he got through an unopenable knobless door, which everyone previously thought was impossible. To do it, he used a workaround involving getting a Mother Penguin to push Mario into the door while also doing a turnaround move." It turns out if you use this technique, you can open the door, but it doesn't actually save you any time in playing the game.

• 1:06:42Casey Newton?

  Well, listen, I hate to tell you this, Kevin, but I think Alex Palax has made a terrible mistake. I think some doors are closed for a reason, and I think an ancient evil has been awakened. And if, if hell is unleashed in this country over the next weeks and months, we will truly only have Alex Palax to blame.

• 1:06:58Kevin Roose?

  Yeah.

• 1:06:58Casey Newton?

  I mean, I truly do love the people that never stop playing video games, you know, like the speedrun community for the Mario games or for Tetris. I mean, these people, the lengths that they go to, to shave two seconds off their time to open the unopenable door is truly inspirational.

• 1:07:11Kevin Roose?

  I also just hope that we have a national security plan that involves elevating people like Alex Palax into sort of code breakers, you know, working for the Department of Defense. Like, we could put these skills to real use.

• 1:07:23Casey Newton?

  I also want to say for everyone complaining the younger generation has no attention span, like this person, the attention span on him, I'm super jealous.

• 1:07:31Kevin Roose?

  It's true. All right. Stop generating.

• 1:07:33Casey Newton?

  All right.

• 1:07:34Kevin Roose?

  All right. Casey, last one.

• 1:07:35Casey Newton?

  Last one. Oh, and this is, this is a good one. "Man fools Waymo self-driving cars with stop sign t-shirt." This is from Car Scoops, which is where I get all my scoops about cars. "An Arizona content creator, Jason B. Carr," this cannot be real, "made a t-shirt with..."

• 1:07:54Kevin Roose?

  His name is Jason Carr.

• 1:07:55Casey Newton?

  His name is Jason B. Carr.

• 1:08:00Casey Newton?

  Jason B. Carr is like the Johnny B. Good of the

• 1:08:05Casey Newton?

  automotive vehicle world.

• 1:08:06Kevin Roose?

  Continue.

• 1:08:07Casey Newton?

  And Arizona content creator Jason B. Carr made a T-shirt with a stop sign to see what Waymo's autonomous test vehicles will do. In a series of videos posted to Cars Instagram, he tested whether Waymo's self-driving vehicles actually stopped when they saw him wearing the shirt while standing on the sidewalk. He said he believes that the autonomous vehicle may be confusing him with a construction worker holding a stop sign. What do you think, Kevin?

• 1:08:27Kevin Roose?

  I think we are entering a bold new era of autonomous car hijacks. I think the people are already starting to find so many creative and nefarious ways of messing with these things. And I think that if I were Waymo, I'd be very annoyed. What do you think?

• 1:08:41Casey Newton?

  Yeah, I mean, we see this already in San Francisco. I mean, not as much anymore, but when the AVs just first started rolling around in the streets, people would just walk in front of them to see what they would do. I mean, like, the confidence that people had in these AVs was staggering.

• 1:08:54Kevin Roose?

  But this was different because Jason Carr apparently figured out-

• 1:08:57Casey Newton?

  Jason B. Carr.

• 1:08:57Kevin Roose?

  Jason B. Carr apparently figured out that you don't even need to walk in front of it. If you're just wearing a shirt with a stop sign big enough on it, it will just stop even if you're on the sidewalk.

• 1:09:06Casey Newton?

  Jason B. Carr woke up and went said, "Today, I be stop sign."

• 1:09:13Kevin Roose?

  All right, Casey, that is Hat GPT.

• 1:09:15Casey Newton?

  That's how we play Hat GPT. Time to close up the old hat. Hats off to you, Kevin.

• 1:09:29Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 1:10:03Speaker 06

  Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business. Without Systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And OutSystems gives you the platform to make it agentic on your terms with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 1:10:38Speaker 08

  Your sales team is at risk of missing quota. Don't panic. Just ask Rippling AI. Built on your real-time people and business data, Rippling AI can pull metrics from Rippling and your CRM into a meeting-ready dashboard showing quota attainment, headcount trajectory, and revenue to quota by region. In seconds, you'll see what's behind your quota risk and fix it before it's missed. Head to rippling.ai/hardfork to get the only AI built to give you full visibility and take complex actions across your whole business. That's R-I-P-P-L-I-N-G dot AI slash hardfork. Sign up today.

• 1:11:14Casey Newton?

  Hard Fork is produced by Rachel Cohn and Whitney Jones. We're edited by Jen Poyant. We're fact-checked by Caitlin Love. Today's show was engineered by Isaac Jones. Original music by Alicia Beetup, Rowan Nemesto, and Dan Powell. Our audience editor is Nolga Loughley. Video production by Ryan Manning and Dylan Burgason. Go check us out on YouTube at youtube.com/hardfork. Special thanks to Paula Schuman, Puywing Tam, Kate LaPreste, and Jeffrey Miranda. You can email us at [email protected] with a picture of whatever you did in front of that portal. Please don't. Knock it off.

• 1:12:15Speaker 06

  It's time to innovate at the speed of AI without compromising control. That's OutSystems. Take a look at what happened with one global retailer during a period of massive expansion. They used OutSystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. OutSystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.