Loading summary
A
Hello, Search Engine listeners. Before we start the show this week, some news. We're doing something new and I need your help. So basically, I need you, our listeners, to submit questions that you think I would be unqualified to answer. So, like stumpers, don't ask me how to make a podcast or how we find stories for Search Engine or where to get a sandwich in Brooklyn. Federoffs ask me stuff that I probably could not answer. What's the meaning of life? How do batteries actually work? Technologically speaking? How do you raise your child who I've never met? Everybody wants to look smart online, but in an act of daring and bravery, I'm going to completely and valiantly whiff for an hour and reveal the shallow limits of my own intellect. Send your questions, if you have them, to pjvote85mail.com this is going to be a live event online for our Incognito Mode listeners. We're going to do it on October 17th. As always, you can sign up for Incognito Mode at Search Engine show, but everybody can send questions. All right, the show after these ads. This episode of Search Engine is brought to you in part by mubi, a curated streaming service dedicated to elevating great cinema from around the globe. From iconic directors to emerging auteurs, every film on MUBI is hand selected so you can explore the very best of cinema streaming anytime, anywhere. And right now, to mark the 35th anniversary of Twin Peaks, you can stream the complete original series from 1990 and 1991, created by Mark Frost and David Lynch. Plus Twin Peaks, a limited event series from 2017, available for the first time on MUBI. It is so good and it is so scary. Who killed Laura Palmer? That question shook television history. Twin Peaks rewrote the rules of TV. Twin Peaks remains both wonderful and strange 35 years later. From the visionary minds of Mark Frost and David lynch, this is a chance to revisit or discover the show that influenced everything from true detective to severance. To stream the best of cinema, you can try MUBI free for 30 days at MUBI.com searchengine that's M U B I.com searchengine for a whole month of great cinema Free. This episode is brought to you in part by Square, the business tool trusted by over 4 million sellers worldwide. Whether you're running your business business online, in person or both, Square makes it refreshingly easy to sell, manage and grow all from one place. I always notice that at my local coffee shop they are using Square. It makes things very smooth and very easy to check out and the staff gets to pretend to know my name. That little touch of personalization that is Square Working quietly in the background, Square is built for the real life chaos of running a business. Whether it's a lunchtime rush, a last minute staff callout or a line out the door, their sleek hardware and simple software keeps everything moving and behind the scenes. Square gives sellers real time sales data, inventory that syncs automatically, and even marketing tools to turn one time customers into regulars. Square keeps up so you don't have to slow down. Get everything you need to run and grow your business without any long term commitments. And why wait? Right now you can get $200 off square hardware@square.com goengineering that's sq U-A-R-E.com geoengine run your business smarter with Square. Get started last week OpenAI launched a very strange new app called Sora. Sora is actually Sora too. There was a Sora before this that fewer people used and which lived inside the paid version of ChatGPT. But Sora too lives inside a standalone iPhone Apple called Sora. The short version is that it's an app that lets you make very realistic AI videos using your own face. Plug in a prompt and a little bit later it'll spit out a very realistic looking video clip. There's a little uncanny Valley effect where you can tell, particularly if it's a video of someone you know, that it's been AI generated, but it's pretty close to being indiscernible from real. I think even someone relatively practiced at spotting fake videos on the Internet should expect going forward to be fooled pretty often. Sora launched with some safeguards. Nobody can make a video of you without your permission, but if you want to, you can choose to give the app permission for you to use your own face and videos for your friends to use your face, or if you're nuts for strangers to use your face. I got on and I did the thing you do with these AI toys where I started making videos for myself of myself. A very, very fat PJ shook hands with a very, very skinny P. Whoa, you're me. Just thinner and you're me.
B
Plus a few pounds.
A
Good to finally meet. Likewise, Manosphere bro podcaster PJ talked about his nutritional supplements from behind his no neck, huge muscled bod.
B
What I'm running right now. Five grams creatine, monohydrate every day, fish oil, vitamin D and a little Ashwagandha. That's my baseline.
A
I would tell you more but there's this rule of AI, which is that the things we make are like our dreams, very interesting to us, very boring to everybody else. Anyway, there was a moment this week where I felt a little nausea about the future of the Internet. AI slop content made by bots that floods the places where humans are trying to talk to each other has been rising across social platforms. And convincingly fake video seems like it'll make the Internet worse, not better, at least for me. So I wanted to understand why this is happening. I wanted to both know how we got here, how a company claiming to want to create machine intelligence is detouring into the slop business, but also where we might be going. And that's why I had called Kasey Casey Newton. Welcome back to the Tarot.
B
Hi pj. Nice to be here.
A
So can you just explain in a way my dad would understand, like, how do you explain what Sora2 is?
B
Sora2 is a text to video generator. If you compare it to something like ChatGPT, that's text to text, right? There's a box and I type, hey, do my high school English homework for me. And I click enter and it does it perfectly. Sora lets you say, hey, make a video of me doing my high school English homework using my face, and then it does that instead. And it has audio as well, so it might include a little bit of dialogue in what it produces.
A
And the way that it gets the audio is that when it captures your face, when you sort of let it capture your face, you just speak a couple numbers, and from that it is able to kind of create a facsimile of your voice.
B
Yes. Although as I'm sure you've noticed, it's a pretty bad facsimile. You're saying at most four or five words to this thing when you're getting started, which at this point is still not enough data for it to be really good.
A
So we're talking Thursday, October 2nd. How long have you been messing with it?
B
So I was able to get access to Sora earlier this week on the day that it launched, and I have been making videos with it every day since.
A
So can you show me one of your stupid videos?
B
So I thought it'd be fun to, like, imagine myself trapped in a video game and unable to get out of it. And so I wanted to see if it would put me in the original more Combat Arcade game. That, of course is also like copyrighted material. So I also kind of wanted to test, like, what are the boundaries as far as copyright? And I Found that Sora was quite happy to put me into Mortal Kombat. Ow.
A
Wait, wait, wait, wait. Okay, wait. We're watching it now. Oh, it is a really good recreation. It's like you're fighting Raiden.
B
Oh, my God. Someone's throwing lightning at me. This isn't fun anymore. Please don't finish me.
A
It really looks like you've been put into Mortal Kombat. Like, I'm sure that if I compare this to an actual Mortal Kombat screen, there's plenty of details they got wrong, but relative to my Memory of Immortal Kombat screen, they did a good job.
B
Yeah, I'm definitely more impressed with what they got right here than what they got wrong. Like, you might also notice that both Raiden and me have a health bar, as you would normally see in a fighting game. And my health bar does just say Casey on it. So, like, the amount of details that it was able to conjure over the, let's say two to three minutes it took to generate this video did impress me.
A
So those are the sort of videos that Sora can quickly output from a prompt. I wanted Kasey to also describe the Sora app itself. The Sora app is different from the existing ChatGPT app. It's a whole separate thing to download. And once you install it, it's hard not to notice that it looks a lot like a clone of an existing, very popular platform.
B
So the actual app looks like TikTok full stop. In their livestream announcing this OpenAI employee said, We've designed this to look familiar, which is a nice way of saying we have copied something. So when you open Sora, you see a feed of videos that are all about 10 seconds long. You swipe up to move to the next video. You can double tap something to like it. You can add a comment, you can share it. And the one sort of interesting thing about Sora is you can also remix it. So if I see something that is funny, I. For example, I saw somebody took a popular meme character and put it into the world of the video game Death Stranding. And some people hit the remix button, and they said, well, put him in this video game instead. Or instead of this meme character, actually make him Jesus. And you do that just by typing out a few words in text. And Sora will then generate a video which then appears next to the original video. So if you see a video on Sora you really like, you can kind of swipe to your left and videos appear at the right and show you all the remixes.
A
That people have made. So, okay, so that's the tool that's you playing with for a couple days. Like, talk to me about just how Sora 2 was announced to the Internet.
B
It was announced via livestream, which is how OpenAI likes to announce things. They'll get members of the team who worked on it. They'll go live on YouTube and they'll kind of walk through. Here is the research that we've done. Here is the new tool. And then they demonstrate the tool. Welcome back to reality. I'm Bill. I'm the head of Sora. I'm Rohan. I lead the Sora product team. I'm Thomas. I lead Sora engineering. Back in February 2024, we introduced Sora 1. We really view that internally the OpenAI livestreams tend to be somewhat casual. I would say no one is in a suit. Everyone is in their sort of San Francisco, slightly preppy, hoodie, forward uniforms.
A
When you see the feed, you're going.
B
To see all the fun we've been having on the Sora team. Some memes have been emerging as we play with the product. There's one about ketchup, drinking ketchup for some reason, and everyone is in their 30s, it seems, and they just kind of casually talk through the research they've done, the thing that they made, and it's very much presented as well. We made something. We think it's kind of cool. We hope you like it. We're training a lot of models that we think the world can have a great deal of fun with and can bring a lot of joy. So we're really excited to see what you'll ultimately create on this app. We'll see you guys. On Sora, they tend not to hype things up too much. Sometimes they will say, like, you know, we think this is a big leap forward in this way or that way. But these aren't sort of iPhone announcements where the CEO is on stage, like, constantly telling you how amazing something is.
A
The way I, as a sort of, like, when I pay attention to it, kind of tech journalist. I had not seen the announcement, but what they successfully did that caught my attention was that later in the day where they announced it, they were just posting lots of stuff on Twitter. And There was one OpenAI employee named Gabriel Peters who was posting all these sort of funny and goofy videos he'd made of Sam Altman. And Sam Altman had set it so that you could use his face as much as you wanted to, which is such a weird sentence to find myself saying. But the Thing that caught me was there was one point where Gabriel Peters has this post and he's like, I have the most liked video in Sora 2 right now. I will be enjoying this short moment while it lasts. And it was fake security cam footage of Sam Altman stealing GPUs at Target. Please, I really need this for Sora inference.
B
This video is too good.
A
And then being caught and, like, cuffed by a security guard. And I think he was trying to say, look at the funny thing I made. And I. And I saw a lot of journalists having the same reaction, was like, whoa.
B
Whoa, whoa, whoa, whoa, whoa, whoa.
A
This seems like not good technology.
B
Yeah. Making realistic videos of people committing crimes that they did not commit can lead to negative outcomes.
A
What a neutral way to say that.
B
What did you think when you saw that one? I was struck by the fact that Sam had let anyone remix his likeness into anything he would later post on X. That seeing himself so much in this Sora feed was less strange than he expected it to be. But I was thinking a lot about, wow, we have another billionaire who has made himself the main character of a social app. Why does this keep happening to us?
A
Why do you think it keeps happening to us?
B
I think at the end of the day, everyone is the same. They want the most followers and the most likes and the most reposts. That's what I honestly think.
A
And that people will compete for attention, even if it's negative attention. And the people not willing to do that don't end up in the public sphere in 2025.
B
Yes. We live in a world where attention is perceived to be power.
A
Right? I mean, I saw Sam Altman's face in all sorts of. I saw it in a toilet.
B
I'm a CEO in a porcelain spa. Skiba do.
A
Skip it.
B
Yes, yes, yes. That's me running AI from the back. Can't stop, head above water, Flush that tune, everybody bouncing.
A
I saw Sam Altman dressed like a teacup committing acts of musical theater Silver towers breathing snow Fans revolve in measured.
B
Time keeping circuits still alive.
A
I saw Sam Altman playing with Teletubbies.
B
I am gonna touch the purple one.
A
Like, I saw so many Sam Altman's this week. It was weird. It was like he was putting himself into everybody's dreams. It was a very strange thing. Cause even, like the way Elon Musk dominates X.com, it's not the same as just seeing someone in video clip after video clip after video clip after video clip.
B
There is a, like, Nightmare on Elm street. Quality to it where, like, you're flipping through the sorafed of dreams and like Sam Altman walks into every dream.
A
Yes. Watching one of the Most famous tech CEOs appear in realistic video after realistic video of himself in very fake situations. I had that feeling you get when you watch the future become the present. It's happened so fast that it feels almost cringy. Or millennial are those different words to wring my hands about all these deepfakes. They're here. They're going to keep getting better. And while OpenAI has put safeguards on Sora, users will want AI video generation with fewer safeguards and companies will compete to give it to them. We all live here. I don't have to tell you how it works. What I did find remarkable was how fast we've arrived at this moment. How we went from oh, the robots can draw pictures sometimes to I don't trust political video anymore. I asked Casey about it. Can you just kind of help me remember, like, what the steps along the road that have gotten us here, like, like, where do we start and how do we get here? Where did we start?
B
We started with a crazy and dangerous dream to build a machine gun that most of the people involved said to each other this would be a bad idea the whole time, one by one they said to each other, while this is a very bad idea, I do think I can do it in a slightly less bad way than you are going to. And so I'm going to raise $500 billion and I'm gonna do it my way. That was kind of the start of it.
A
Yes. And then from there, the first image thing I remember was Dolly. But was that actually the first popular one? I remember, like, there was a moment where Dolly had first launched and people were making these incredibly primitive. It was like looking at a three year old's drawings, but the three year old's a computer. Like, it wasn't that they were good, but the novelty of it was exciting. And the watching a computer try to make images was exciting.
B
I did find it exciting. I even thought the images were cool. I mean, at least it was cool that I could say, like, make me a picture of a bulldog on the moon and like 30 seconds later I'd be looking at like a really cute bulldog. Yeah. So, yeah, Dall? E was a moment where all of a sudden the idea that you could conjure not just text, but also an image out of thin air became real.
A
So Dall? E's 2022. The other thing I remember, when does video Start. Was it like 2024? Do we start to get video?
B
Yeah. So SORA gets introduced in December of 2024, so last year, and gets added into paid subscriptions for ChatGPT. At that time, it did not have audio, so you could make a very short video clip, but it would not have sound. And so while OpenAI was very interested in exploring what potential applications of this could be, they start working with, like, Hollywood filmma. It can't really make what you and I think of as video, which is video with sound.
A
Right. And at the same time that they're doing this, it's like OpenAI, this sort of pattern for this is OpenAI will release something. They're often ahead. They'll release something that works pretty good and feels like a sketch of what might happen next, and they'll have rules in place to prevent abuse. And then usually there are competitors, like prominent companies, less prominent companies, who will build similar things with fewer guardrails. So, like my memory, and correct me if I'm wrong, but 2024, one of the big ones people used was called stable diffusion. And stable diffusion was like, you could do a very similar thing, but you were much less likely to run into the chatbot on the other side, telling you that it couldn't generate an image you wanted.
B
Yes. Stable diffusion made models that were small enough that you could download them onto a laptop and just sort of use them without essentially any guardrails other than whatever might have been baked into the original model.
A
And what I recall also from that time is like, it's where you see there's this, like. I kind of want to bring it up. The Will Smith video where he's eating spaghetti.
B
Yeah.
A
Oh, that's hot. That's hot. Uncle Phil, come try this fresh pasta of Bel Air. Do you want to describe it, Casey?
B
So this is Will Smith's face eating spaghetti with his bare hands. His chin is kind of bobbing all around, almost like detaching from his face. It's incredibly grotesque. They are playing the Fresh Prince of Bel Air theme song, though, which brings up warm and nostalgic feelings in me.
A
This infamous video, originally from 2023. Before rewatching it with Casey, I'd forgotten how aesthetically disturbing it really is. A version of Will Smith, whose facial proportions I would describe as wrong and dripping, shoves a mop like mass of spaghetti near the general vicinity of his grotesque melting mouth. It was so bad, it went viral. But as new AI video models have appeared, oftentimes, people will showcase the new, better version of AI with new Will Smiths Eating AI spaghetti. It's one of the informal benchmarks for how the technology has moved forward. You can watch these spaghetti videos get realer. Underneath that progress, though, I wonder what's been happening technologically. Casey says probably some of this is from the models being fed more training data, more video to learn from, and they're using more horsepower, more GPUs, more energy consumption. But beyond that, it's a little opaque. What's driving the rapid progress here?
B
You know, I tried to get a really good answer that would be accessible to a broad audience on, like, why is Sora 2 better than Sora 1? And the answer proved to be just dauntingly technical. Like, one thing that's better about Sora 2 is that it can generate videos faster than Sora 1. And so I know, PJ, you're a smart guy. You're wondering like, well, how can it do that?
A
Yeah, how can it do that?
B
The answer to that question involves step reduced solvers, rectified flow variance, and kernel level optimizations. So hopefully that answers your question.
A
But it's funny because there's been other kinds of technological advancement where there's been jargony explanations for it that I've red U parse. This feels more jargony and less parts of all.
B
And I think the reason is because it's relatively easy for me to understand how a single image gets made. You probably watch this happen. It uses this process called gradient diffusion, where what it's essentially trying to do is find the median in all of its training data. So if you say something like, make me a picture of a bulldog on the moon, the model works. And it's like, we're going to show you the thing that looks the most like a bulldog, and we're going to show you the thing that looks the most like a moon, and we're gonna show you the thing that looks the most like the two of them together. And you can just sort of watch that, like, gradually come into focus. Now, you think about a video, you have to do that across hundreds of frames over 10 seconds. You have to synchronize that with audio. And so there are just a lot more processes that are running in order to make that happen. And they have simply overwhelmed my pathetic human brain in this moment.
A
Do you get the sense that the people building it completely understand it?
B
What I've learned from my time covering the AI industry, pj, is that no one who works there completely understands any of it. So, no, I don't think that they totally understand it. I mean, when you talk to people who work on the models themselves, they talk about them like they are growing orchids right where it's sort of like, well, these ones look really good. This one has a problem. We're going to see if we give it a little more light, if maybe the flowers grow a little better. But we're not exactly sure why it turned out that way.
A
Sora 2 the Sora app was released to a limited set of users on September 30. By October 3, it would be the most popular app in Apple Store. Setting aside obvious, necessary conversations about copyright or fairness, what was immediately so clear is just that a lot of people really enjoy Sora. It's a novelty. The way that Sora was set up at launch, users could generate videos using realistic images of many copyrighted characters and public figures. Copyright owners work was fair game by default, unless they'd explicitly opted out. That meant I saw a Pikachu from Pokemon Grilled and Filleted Pikachu on the grill here.
B
It's already got a beautiful char and it smells like somebody plugged in a chicken. Let's give it a flip. I'm going to carve it into some thick steaks.
A
I saw a lot of Stephen Hawkings on a lot of half pipes, and I watched last weekend in real life as a group of teenagers and 20 somethings played with the tool and just got totally sucked into it for hours, amusing each other, mainly by trying to see how inappropriate they could be watching. I thought, okay, a tech company has created one more tasty brand of junk food. A new reason to look at our phones a little longer. It was what I expect from this Internet, although not what we've typically seen so far from OpenAI. The company talks about superintelligence, about trying to improve the lot of humanity. And now they built a tool that let people use Martin Luther King Jr's likeness to make more Epstein jokes.
B
I have a dream that the Epstein files get released.
A
To me, this seemed like a pivot. But was it? What were OpenAI's motives here, after the ads, some answers. This episode of Search Engine is brought to you in part by Axel. You know that feeling when you finally sit down to book a trip, spend hours hunting for the best flights and hotels, and still aren't sure if you actually got a good deal? That stress of wondering if you overpaid or missed a better price? It's the worst. Axel is here to make sure you never feel like that again. You can Message Axel on WhatsApp to find deals, book travel, and automatically get refunded when prices drop. After you buy, hotels average 50% less than major booking sites, and with automatic price monitoring, you get instant refunds and credits whenever prices fall. No work required, just text what you need and get options in seconds. Plus, axle is 24 7, so you never miss a savings opportunity with real time notifications keeping you informed. It's like having a personal travel agent in your pocket. On average, users save $400 per trip, and 91% report less travel stress. Stop overpaying for travel. Go to helloaxle.com search for a free trial. Just 35 bucks a year. After that, most users save double that on their first booking alone. Sign up now and let Axel handle the deals and planning while you enjoy the trip. This episode of Search Engine is brought to you in part by Bombas. Fall's here, kids are back in school, vacations are over, and it's officially cozy season. Which means it's time to slide into some Bombus. You know Bombus, ridiculously comfortable socks, slippers, tees, and underwear made from premium materials that actually make sense right now. Merino wool that keeps you warm when it's chilly but cool when it's not. Supima cotton. Softer, stronger, more breathable than the regular stuff. And classic ragwall, that thick, durable, heritage y cozy sock you'll want all fall. The best part is that for every item you buy, Bombas donates to someone experiencing homelessness. Over 150 million items so far. And if you're not 100% into what you got, their happiness guarantee makes it right. I wear red Bombus socks to try to deter sock thieving. It does not work. It just means that I'm gifting, without my own permission, red socks to lots of people. But they're nice socks. Head over to bombus.com engine and use code ENGINE for 20% off your first purchase. That's B O M B A S.com engine code engine at checkout. What's up, guys? It's Candice Dillard Bassett, former Real Housewife of Potomac. And I'm Michael Arsenault, author of the New York Times bestseller I Can't Date Jesus. And this is Undomesticated, the podcast where we aren't just saying the quiet parts out loud. We're putting it all on the kitchen table and inviting you to the function. If you're ready for some bold takes and a little bit of chaos, welcome to Undomesticated. Follow and listen to Undomesticated, available wherever you get your Podcasts. Welcome back to the show. So our question what are OpenAI's motives here? We of course reached out to OpenAI directly to ask them what they think they're up to with sora. A spokesperson said in part, on the road to general purpose simulation and AI systems that can function in the physical world, we think people can have a lot of fun with the models we're building along the way. In keeping with OpenAI's mission, it is important that humanity benefits from these models as they are developed. We think SORA is going to bring a lot of joy, creativity and connection to the world. So that's the company line, but this is the we're here to help humanity company unveiling junk food, an AI slop social media feed that is glugging unseemly amounts of electricity. I wanted to ask Casey about this since as far as I could tell, none of this had been a part of OpenAI's lofty mission at its start.
B
When OpenAI gets formed, the idea is to create artificial general intelligence that benefits all humanity. It's famously set up as a nonprofit. It seems like the last thing that they would want to make is a social app that just sort of sucks away your time. And on Wednesday, October 1st, there's a person on X who posts Sam Altman two weeks ago. We need $7 trillion and 10 gigawatts to cure cancer. Sam Altman Today we are launching AI slot videos marketed as personalized ads. And he's got a point there, right? And Altman actually responds and says I get the vibe here, but we do mostly need the capital for build AI. I assume he meant for building AI that can do science. And for sure we are focused on AGI with almost all of our research effort. It is also nice to show people cool new tech products along the way, make them smile and hopefully make some money given all that compute need. So that was really interesting to me because it was an acknowledgement of the fact that this is a money making enterprise. And in part it is a money making enterprise that is going to be designed to subsidize the absolutely enormous cost of running OpenAI.
A
So talk to me about that. Like the idea is you make one more stupid app to make people look at their stupid phones and if you can do that and it causes people to sign up for the more premium version of ChatGPT, that money is just sort of like revenue that OpenAI somewhat desperately needs.
B
Yes, the idea is find ways to generate revenue now that that help to offset some of your losses. Across the rest of your business. And here's where we can talk about one of the most interesting and under discussed phenomenon that we've seen at one of these companies recently, which is what I think of as the Facebookification of OpenAI.
A
Meaning what?
B
So when OpenAI gets started, it really is mostly a research lab. You have some startup people in there, like Sam Altman, who used to run Y Combinator, was just sort of like deeply enmeshed in the Silicon Valley startup ecosystem. But it's a lot of sort of big brains who are trying to figure out how do you build large language models that can do everything ChatGPT does today. Over time though, OpenAI, in addition to being this brainy research lab, also becomes one of the hottest startups in Silicon Valley, maybe the hottest one, and they start to attract a lot of talent from other big companies, including Meta. Recently, OpenAI hired a woman named Fiji Simo to be what they call their CEO of applications. So in charge of running the various apps that the company has, which now include ChatGPT, BT and Sora. Fiji Simo came from Instacart where she was the CEO, but previously she had spent almost her entire career at what was then called Facebook, and she worked on systems that help people spend a lot of time looking at feeds and buying ads based on what they were seeing in those feeds. And when you look across OpenAI, according to some research that we talked about on my podcast recently, we think There are about 700 people at OpenAI who used to work at Facebook. So that's not to say that all of them were Facebook true believers and they want to make ChatGPT look exactly like Facebook. But a lot of the DNA of this company is now people who worked on social feeds that were monetized by advertising. And as we saw with the launch of Sora, there is part of the company that is now moving full speed ahead in that direction.
A
And for you, like, I think one of the interesting things about trying to understand technology in 2025 is that most of the prominent tech reporters in our cohort generation, or at least maybe those are the ones I pay attention to. They're people who watched the rise of and were disappointed by the social media Internet. And one of the questions has been the AI tech companies, like, how much are they like what came before and how much are they not? Like, how do you see OpenAI as at least representing itself as having a distinct mission? And like, what does it mean for it to become Facebookified?
B
Yeah. So in terms of OpenAI's mission. Over the past few months, as there's been more concern that ChatGPT can lead to delusions or even psychosis in the people who are using it, OpenAI puts up a blog post in which they try to assert their mission. And what they say is, we want to help you get stuff done, and then we want you to stop using ChatGPT. We do not want to trap you in some endless engagement loop where you're refreshing a feed a hundred times a day. ChatGPT is to help you get stuff done and to improve your life.
A
And I just wanted to say that this is very, very different from. That's almost like when I think about why I have positive feelings towards this company. I mean, I've never felt the need to put screen time limits on my ChatGPT use because it's something that I look stuff up on and then I put my phone down.
B
Right. And I have felt the same way. And then the week before Sora comes out, they put out something called Pulse. What is Pulse? Pulse is a feed of stories personalized to you. Some of those are news stories. So if you're like me, read a lot of news, it'll be like, hey, here's what's going on with the TikTok sale. But then it'll also say, hey, you were looking up some stuff about meditation. Here's a quick meditation you can do in between your meetings today. And, you know, we know you got a dinner coming up this weekend. Like, here's a side dish that you could make. And on one hand, this is, like, good and helpful stuff. I basically like the idea of it. But is it also a feed I can come back to and check every day? Will they send me a push notification if I let them, letting me know that my pulse is ready? Is this an obvious spot where they could integrate advertising in the near future? Of course.
A
I see. And so the question, one of the questions you have watching this right now is OpenAI says, and let's just like, give them the benefit of the doubt. OpenAI believes that they have a mission, and the mission is in a real way to improve humanity. And, like, obviously there's lots of potential downsides which we can talk about. There's energy consumption and climate risk and job loss. But they have this optimistic vision, and the optimistic vision requires tons of money. And so one of the ways that they are trying to suck money into that mission is to be a little bit more like Facebook. And one of the questions you have to have is like, how much do you trust them? To be a little bit like Facebook versus in two years, ChatGPT is a super optimized product that gets you to stare at your phone using all the tools the existing social platforms use to get you to stare at your phone, most of which leave you feeling worse off when you're done, right?
B
So unfortunately, I do think the trend is that the OpenAI products are starting to look more like social media apps. And in that they are asking for more of your time, they are becoming more explicitly commercial products. One of the other changes that they've announced the week that we're talking is that there is now a commerce integration into ChatGPT and so you can start buying products directly from the ChatGPT feed. Over time, all of this adds up to something that could very credibly serve not just as a replacement for Google, which is where a lot of the attention has focused so far, but but possibly for the meta family of products as well. So on one hand, there is still kind of the big crazy AI story of, oh gosh, what happens if the machine God does get built and none of us have jobs anymore? But there's this other much more normal and traditional Silicon Valley story, which is hot new startup comes along and starts to eat the lunch of the incumbents. And a lot of what we've seen OpenAI do over the past few months has been much more in that direction.
A
Remember, according to Sam Altman's tweet about all this, his belief is that if OpenAI is going to deliver on its promises to humanity tomorrow, it needs capital today. That's why they've begun to nibble at the lunches of these existing social media companies. OpenAI needs more capital, seemingly always, because the resources that these AI companies are fighting for are very expensive, particularly compute power, which comes from enormous data centers which need oceans of electricity to run. This is the subject actually of a couple of stories we're working on that'll be out in the next months. But that said, for the purposes of this story, all you need to remember is that OpenAI is currently spending enormous amounts of money to try to stay ahead of its rivals, because nobody competing in AI seems particularly interested in second place.
B
There is a belief within AI world that this is maybe a winner take all race, and if not a winner take all race, then probably a winner take most race. That has been the case with many of the big Internet categories. Right. There's more than one search engine, but Google took most of the market share. Unfortunately for your show, we keep trying.
A
To hire, we keep trying to head.
B
On from Google, there's more than one social network, but for long time Facebook had taken most of that market. So the question is, is AI sort of the same thing where one company creates the AI that is used by 80 plus percentage of the population and if that is the case, you want to be that company? And right now the race is still unsettled. OpenAI has done a great job so far of cementing itself as the leader in that category, but there are a lot of challengers nipping at its heels and the big incumbents have started to wake up to the fact that, oh wow, these guys actually could eat my lunch. You know, Mark Zuckerberg is browsing that sort of feed and he's gonna go put the fear of God into his employees because this is something that he thinks about constantly, right? It's like, what is the thing that is finally going to dethrone me? So you're just starting to see all of these very competitive, very ego driven executives taking a look at the future and thinking, it had better be my name that gets written in the history books as the person who builds the machine God, and we are going to spend as much money as fast as we can to get there.
A
And so for this narrow thing this week, Sora 2 launched as basically AI TikTok. Do you think it will work? Asking people to make predictions is kind of unfair, particularly about something measurable like the success of an app. I understand what I'm doing is unkind, but do you think that OpenAI will make the money it needs to off of this thing? Do you think people will use it the way it seems to want people to?
B
I'll tell you the prediction that I have that I am confident about and that I think I might differ from my Peter Zahn. I think this feature that they're calling cameos, which is the ability to take your friends likenesses and your own likeness and put it into whatever situation you can dream up that the content guidelines will allow. I think that's here to stay. I think people are really going to like that. I think it's the sort of thing that enlivens any group chat, which is where most discussion now takes place. And I think you can make a lot of fun and funny stuff that will drive memes and engagement online. Can OpenAI take that basic creative tool and build an entire social network around it? That's a much bigger challenge and it is impossible to judge on the first week whether a social network has any staying power. By default, every social network dies There have only been a handful that have managed to last a decade. And the ones that have done that successfully have essentially found some kind of influential community that had a reason to be there and liked being there. And they just kind of stuck around and they drew more and more people to them over time. And the creative tools evolve, and maybe the people who are creating things there find ways they can make money. These are some of the ways that these networks can take off. And it just remains to be seen whether Sora is going to be one of those things. But we have reporting that OpenAI has thought for a long time not just about creating a TikTok clone, but also maybe wanting to go after X, wanting to replace the Twitter of old with something. Would that.
A
I don't understand. Why would that. A Twitter where nobody's real?
B
Well, I think the idea was, like, there should be a place where you can share and discuss your creations in OpenAI. So, you know, at the time we were thinking, maybe it's the images that you're creating. Maybe, you know, you want to share some of the conversations that you're having with ChatGPT. I realize none of this sounds like fun or interesting to browse, but, like, these were some of the discussions that people were having. Now along comes Sora, and it's like, oh, well, maybe that's actually the core of it is, like, people are sharing these videos. But it would not surprise me at all if this thing starts to gather momentum. And OpenAI says, you know what, you can actually share text posts in the feed now as well, if you like. Right. You could use the Sora feed as the basis to create something larger. So if they feel like they're having some success, they may go for it.
A
It's funny, you know, I've been a bystander to between an adult and a teenager about social platforms. And, like, there's been an attempt to hold the line against a teenager getting on Instagram. And what's been hard is that for the teenager, they're like, this is where my friends are. Like, plans are being made here. Homework assignments are being shared here, not by the teacher, but, like, it's communication that they legitimately have a need to have. And as an adult, you're like, the platform doesn't care about your relationship with your friends. The platform is holding that relationship hostage so that the platform can have a relationship with you.
B
Yes, correct.
A
And the fear isn't what your friends are gonna say to you on the platform. The fear is that once you get on the platform, what your relationship with it's gonna be. And there's something very confusing to me about right now where if OpenAI wants you to have a relationship with its platform, but its platform isn't even kind of about other people. Like, it wants you to share your funny Sora videos so that you see the funny Sora videos and then more people get online and talk to their chatbots. It's very weird trying to imagine what an AI mediated social platform is gonna look like. And I do kind of have faith in their ability to draw people into it.
B
Yeah, I basically do too. You know, one of the dimensions of this conversation that we haven't discussed yet is that I think a lot of folks who are around our age have a powerful disdain for what they would call AI slop, and they're in disbelief that companies keep creating tools to make and share it. And I'm sympathetic to that view. It's weird for me to go on Facebook and see Shrimp Jesus and an obviously fake woman who says that she just turned 101 today and made a birthday cake, but there's no one here to celebrate. And it has like 800,000 likes and comments from people wishing her a happy birthday. It's weird to me that that's like so much of our reality and it's easy to get from there to, this is all going to burn itself out. This is obviously stupid, but I do think that this is one of the lessons of the Internet over the past year is that you can say, on one hand, I don't like AI slop. It makes me uncomfortable. I can't imagine spending much time looking at it. And then on the other hand, digging into it a bit and finding out this stuff is crazy popular. You know, you read the comments of these things and people seem to have an awareness that, like, their brains are being cooked as they're watching it. Like they're just sort of being hypnotized into a stupor by the series of images. We haven't talked yet about Meta's Vibes app, which Meta had rushed out the week before Sora launched and is a kind of just worse version of it. Your friends and family aren't in it. It's just sort of random AI generated video. And as I'm flipping through this thing, I'm thinking, this is Cocomelon for adults. This is just raw visual stimulation designed to keep me hypnotically scrolling as long as I can. And on one hand, that might not seem very different than TikTok today, but on the other hand, it's like, well, at least that person made a TikTok. Like, at least that girl is dancing, you know? And now it's like, I'm gonna get into my bed, but my bed is made out of cookies and I'm gonna fold the cookies over me and there's no real point to it, but, God, it looks weird and I kind of wanna share it with my friend. And like, this is the sort of cuckoo town direction that everything's going in.
A
Are we just gonna end up looking at colors and shapes?
B
We're so close. We're so close to colors and shapes.
A
Underneath all this conversation. And maybe it's just some anxiety that I think everybody who covers tech has right now. The invention of social media was so expensive for society that tech journalists like Casey Newton are now trained to spend way more time anticipating when new technology shows up, what it might break instead of what problems it could solve. There's this hope that if we can see the danger ahead of time, maybe we can somehow be useful. Is that true at all? I don't know. I remember trying to impress a grade school history teacher, quoting that line about how we have to know history so we're not doomed to repeat it. He asked me, when has knowing history ever stopped anybody from repeating it? Me, I continue to use our decaying Internet. I continue to play with these new AI tools and I try to stay curious, which means for me, I try to ask myself a lot these days, what do I think this technology is doing to me right now and do I like it? And so what do you think? Just, like, as a person who needs to, like, not having high screen time is not really an option for you with your work. Like, you need to look at and make sense of an Internet that is becoming both less sensical and less human driven. How do you prepare for the next season?
B
So to try to inject a little optimism here?
A
Yes, please.
B
I do think that there are these avenues where you see people really hungry for a human connection. I think podcasts are a big part of that. Like, I, I look at these AI generated podcasts that people are now making, and I may have some blinders on here because it would be in my interest for this to be true. But I just think, I don't know, in the next two or three years at least, I can't imagine wanting to listen to an AI podcast, weigh in on events in the same way that I want to listen to all of the podcasts that I listen to, investigate all of the subjects of their interests. I think podcasts give us a connection to real people who we like thinking through difficult problems over long periods of time. And I think that's one reason why podcasts are so popular. Right? I also think that you see this in the rise of newsletters and other kind of niche publications, right? There really has been a boom in blogging that I sort of thought we wouldn't see again. I thought it was like kind of one and done, but now I feel like almost everybody I know has a substack that they send out once or twice a year. And I think it's because people want those authentic connections. They want to be able to tell you what they actually care about and how they they're thinking about things. So I don't think it's as if everyone is giving up on human connection and deep thinking at the same time. I do think there are more and more powerful distractions being injected into the bloodstream of the information economy and they are kind of sniping people one by one. And it is worrisome.
A
I think it's genuinely scary. But I think you're right that it's worth remembering that as some parts of the Internet get more senseless, people express a strong demand for sense making and that, like, sense making is still happening. The solution to all of this is clearly podcasts.
B
The solution is podcasts. I bet your listeners weren't prepared for that point, but we got Casey.
A
Thank you for being being part of the solution here.
B
My pleasure.
A
Casey Newton. You can find him on his podcast Hard Fork and at his newsletter, Platformer. I recommend both. Sam this episode is brought to you in part by Vuori A New Perspective on Performance Apparel. Perfect if you're sick and tired of traditional old workout gear, Vuori clothes are incredibly versatile and comfortable. Perfect for whatever your day brings. They're designed to look great beyond the gym, whether you're running errands, heading to the office, or meeting up with your friends. One specific Vuori item I would recommend is the core short. This is a short that started it all for Vuori. It is one short for every sport you can do for whatever sports you play. It's ideal for fitness, running and training, but also genuinely stylish and comfortable enough to just wear all day. Vuori is an investment in your happiness. For search engine listeners. They are offering 20% off your your first purchase. Get yourself some of the most comfortable and versatile clothing on the planet@vuori.com PJSearch that's V U-O-R-I.com PJSearch exclusions apply. Visit the website for full terms and conditions. Not only will you receive 20% off your first purchase, but enjoy free shipping on any US orders over 75 bucks and free returns. Go to vuori.com pjserch and discover the versatility of Vuori clothing exclusions apply. Visit the website for full terms and conditions. This episode of Search Engine is brought to you in part by Dell Introducing your new Dell PC. Powered by the Intel Core Ultra processor, it helps you handle a lot even when your holiday to do list gets to be a lot because it's built with all day battery plus powerful AI features that help you do it all with ease. From editing images to drafting emails to summarizing large documents to multitasking so you can organize your holiday shopping and make custom holiday decor in search for great holiday deals and respond to holiday requests and customer questions and customers requesting custom things and plan the perfect holiday dinner for vegans, vegetarians, pescatarians and Uncle Mike's carnivore diet. Luckily you can get a PC that helps you do it all faster so you can get it all done. That's the power of Dell PC with Intel inside backed by Dell's price match guaranteed. Get yours today@dell.com holiday terms and conditions apply. See dell.com for details. This episode of Search Engine is brought to you in part by Perfectly Snug. I am a hot sleeper. I do all the right things and still wake up at 3am wide awake and sweaty. Perfectly Snug is a fix for that. It's a 2 inch mattress topper with whisper quiet fans that actively move heat and humidity away from your body. The sensors actually work. You don't have to fiddle with settings in the middle of the night and if you like me, need quick relief. Burst mode cools things down in about 10 minutes. I recommend their dual zone setup so that one side of the bed can run cooler than the other side of the bed and there's no water to refill and nothing snaking off the bed. It's just a slim topper that sips power. Setup is a few minutes. They offer a 30 night risk free trial with free shipping and returns. If you, like me, are tired of sweating through the night, try Perfectly Snug. Search Engine is a presentation of Odyssey. It was created by me, PJ Vogt and Shruti Pinamani. Garrett Graham is our Senior Producer. This episode was fact checked by Mary Mathis. Theme, original composition and mixing by Armin Bazarian. Our Executive Producer is Leah Reese Dennis and thanks to the rest of the team at Odyssey, Rob Morandi, Craig Cox, Eric Donnelly, Colin Gaynor, Maura Curran, Josefina Francis, Kurt Courtney, and Hilary Shaf. Our agent is Oren Rosenbaum at uta. If you'd like to support our show and get ad free episodes, zero reruns and even extra interviews, please consider signing up for Incognito. My you can join Incognito Mode at Search Engine Show. Follow and listen to Search Engine wherever you get your podcasts. Thank you for listening. We'll see you next week.
B
And Doug, here we have the Limu Emu in its natural habitat helping people customize their car insurance and save hundreds with Liberty Mutual. Fascinating. It's accompanied by his natural ally, Doug.
A
Limu is that guy with the binoculars watching us.
B
Cut the camera. They see us. Only pay for what you need@liberty mutual.com Savings very unwritten by Liberty Mutual Insurance Company affiliates. Excludes Massachusetts.
Host: PJ Vogt
Guest: Casey Newton
Date: October 10, 2025
This episode explores the rapid evolution of AI-generated video apps, focusing on OpenAI’s new "Sora 2"—a tool and social feed for creating hyper-realistic, AI-generated videos starring your own face or public figures. The conversation digs into how and why Sora was built, its social and technological implications, OpenAI’s pivot toward addictive feed-based content, and the broader consequences for Internet culture, authenticity, and human relationships. The hosts also ask: are we just heading toward an Internet of "Cocomelon for adults"?
[05:16]–[10:20]
[10:20]–[15:05]
[16:10]–[24:45]
[29:38]–[33:39]
[31:09]–[36:19]
[37:26]–[41:42]
[43:49]–[46:13]
[47:29]–[49:36]
Sora 2 marks a striking shift from OpenAI’s high-minded AGI rhetoric to the reality of engagement-maximizing, AI-generated content feeds. The episode frames this as both a technological marvel and a source of existential anxiety for Internet culture and human connection: as the lines between “slop” and sense-making blur, the value of authentic, human-driven communities—like podcasts—becomes newly clear. Whether Sora becomes the next TikTok, a flash in the pan, or just another layer of distracting noise remains to be seen.
The solution, for now according to the hosts: More podcasts.
For more in-depth insights on technology, AI, and Internet culture, follow PJ Vogt’s Search Engine and Casey Newton’s Hard Fork and Platformer.