You’re panicking wrong about AI (with Casey Newton)

On the bad side, we have copyright infringement, exploitation, and misinformation; on the good side, we have its uses for brainstorming, editing work, and combating loneliness. On both sides, everyone is freaking out about AI — but are we freaking out about it for the right reasons? Casey Newton joins us to make the pitch that whatever you think, AI is not “crypto 2.0.” Everyone, in every generation, needs to take it seriously. Our guest Casey Newton is a tech journalist who founded and writes for the Platformer newsletter (https://www.platformer.news/) and co-hosts the Hard Fork podcast (https://www.nytimes.com/column/hard-fork/). Want even more Panic World content? Like ad-free episodes, bonus episodes, and access to the Garbage Day Discord? Sign up for a membership at: https://www.patreon.com/PanicWorld. Sponsors Audio Maverick, a new nine-part documentary about one of the most visionary figures in radio, Himan Brown. Out now wherever you listen to podcasts. Want to sponsor Panic World? Ad sales & marketing support by Multitude, hit them up here: ⁠http://multitude.productions⁠. Credits - Host: Ryan Broderick - Producer: Grant Irving - Engineer: Rebecca Seidel - Researcher: Adam Bumas - Business Manager: Josh Fjelstad Learn more about your ad choices. Visit megaphone.fm/adchoices

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

• 0:01Speaker 00

  Are you ever shopping online and suddenly a discount that helps you save pops up? It makes your day, sometimes even your week. Want that feeling month after month? Well, new customers who save by switching their home and auto insurance to Progressive save hundreds. Cue your happy dance. Get your quote today at progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states.

• 0:28Ryan Broderick?

  The first question for you I have today. How much of Platformer is being written by AI? Like 70, 80% at this point? Like, what are you, what's your workflow?

• 0:39Speaker 02

  Platformer is written 100% by human beings. That's important to me. People are paying me presumably to read my reporting and analysis as opposed to the median solution that was written by all of humanity and then like regurgitated by AI. So I don't use it to write, but I have started to use it for research. I've started to use it for editing and yeah, just in general, I explore other ways that AI might be useful in my life.

• 1:10Ryan Broderick?

  I fed a bunch of garbage day issues into ChatGPT and then I asked it to write subject lines for a couple of issues as me and it was so fucking bad.

• 1:38Ryan Broderick?

  I'm Ryan Broderick and with me as always is our producer Grant Irving, who you'll hear pop in throughout the show. This is Panic World, a podcast about the viral freakouts and moral panics bubbling up out of the darkest corners of the internet. And today we're talking about AI, which of course everyone is freaking out about. But the question is, are we freaking out about AI for the right reasons? There's plenty to worry about, obviously copyright, misinformation, exploitation. But is all of this anti-AI hysteria looking in the right direction? Today's guest, Casey Newton, the co-host of the Hard Fork Podcast and And writer of the platformer newsletter recently picked a fight with a whole bunch of people who say that generative AI is a shell game, this massive scam that's all going to go bust. And they're mad at him because he wants us to get serious and appreciate AI for what it's

• 2:23Ryan Broderick?

  doing right now and where it could go in the future. So to kick things off, Casey, you're, you know, you're a big AI shill. What's your favorite, most optimistic take on AI right now?

• 2:34Speaker 02

  Sure. And at some point, I would also like to sort of share the critiques of AI that I agree with because like I do agree with many of those critiques.

• 2:42Ryan Broderick?

  We're not going to do that. No, we're just going to make you chill for an hour. Yeah.

• 2:45Speaker 02

  Perfect. Perfect. That's great. I see AI as what sometimes called a dual use technology. It can be used for good. It can be used for bad. Fire is the classic dual use technology. You can use it to burn down a house or a forest. That's bad. Or you could use it to heat your home and cook food. That is good. As I have used AI, it has seemed evident to me that it is in that same category of things. It actually can edit your work and find mistakes. If you are a writer staring at a blank page or a marketer or a product manager and you have a particular problem and you want to take it to the AI and say, hey, here's where I'm at in my life. And

• 3:27Speaker 02

  you can kind of sample the collective unconsciousness to get a sense of some possible paths forward. That seems really useful to me. And indeed, it is useful to me on a lot of days when I do it. So I think that's worth paying attention to. But I think more broadly around like my job is a tech reporter. Like my job is to go out and tell people what is happening and what I think is going to be more important tomorrow than it is today. I don't think there's any intellectually honest way of looking at the AI field and not believing that it's going to be more important tomorrow than it is today. And so if you accept that that's the case, then that means that I do have to go out there and use these tools and try to see what they're good at, what they're bad at. And the more that I've used them, the more that I've seen, they're just getting better

• 4:11Speaker 02

  at an exponential rate. And that has a lot of huge implications, some of which are good and some of which are really bad. But unfortunately, we just Live in a really polarized world and on Blue Sky in particular, if you have an opinion about AI that isn't, it sucks, nobody wants this, and every time you use ChatGPT, a forest burns down, you get labeled as a booster. And I just think that's super intellectually dishonest.

• 4:35Ryan Broderick?

  I think it's just interesting that the majority of Blue Sky users aren't taking planes. You know, because they're so concerned.

• 4:42Speaker 02

  Right, or they don't own cars, they don't live in the suburbs. Yeah, they're doing it right.

• 4:46Ryan Broderick?

  Yeah. Honestly, I've been skeptical because so many tech innovations have been rammed down our throats and failed spectacularly. But I'm coming to terms with the fact that the world is different. AI has already changed work and the internet. And that's all very scary when you look at how much things have changed over such a short amount of time. But this is a show where we like to really scare our listeners by laying out what has happened in that short amount of time. So let's, let's look at our timeline, huh? In December of 2015, a startup shows up called OpenAI. You've heard of them?

• 5:19Speaker 02

  Yeah.

• 5:20Ryan Broderick?

  Yeah, you've heard, you've heard of OpenAI. Okay, cool. So in their introductory blog post, they laid out their mission, which I think is really interesting to look back on now. So they write, "Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return.

• 5:37Ryan Broderick?

  It's hard to predict when human-level AI might come within reach. When it does, it'll be important to have a leading research institution, which can prioritize a good outcome for all its own self-interest." Feels like that holds up pretty good, huh?

• 5:51Speaker 02

  Yeah. Yeah, they didn't deviate from that one bit. It's amazing.

• 5:54Ryan Broderick?

  For people who don't know this, OpenAI did technically start as a nonprofit. They changed from a nonprofit status in 2019, and they finally offered their own definition of a thing that we're going to talk about, I think, quite often this, this week on the episode, which is artificial general intelligence. Can, can you sort of give us your definition before I read OpenAI's here?

• 6:12Speaker 02

  Yeah, I mean, it's famously slippery and the definition seems to be changing all the time, but it is basically computer software that can do many generalized tasks, multi-step tasks, and kind of scaling all the way up to a digital coworker who can basically do anything you would assign to a remote worker. AGI would be able to do.

• 6:34Ryan Broderick?

  That's right. And I, our researcher Adam pulled out this really interesting sort of definition that OpenAI provides in 2019 because it is quite different from how they were talking, you know, only a couple of years before. So they describe AGI as highly autonomous systems that outperform humans at the most economically valuable work. So it's interesting how this nonprofit stops being a nonprofit and they're like, we're going to take all your jobs. Yeah. In 2023, as we learned from leaked documents just a couple of months ago, OpenAI changed its definition yet again to an AI system that can generate at least $100 billion in profits. That rule's so hard. That's so grim. Do you have any sort of context on the,

• 7:17Ryan Broderick?

  like, the AI industry side of things during this time period? The sort of, let's call it, like, the pre-Dolly era, where they're, they're trying to define what it is they're even trying to build?

• 7:28Speaker 02

  Yeah, well, they're trying to build machines that can think, that can reason, that can

• 7:36Ryan Broderick?

  Have sex with you.

• 7:37Speaker 02

  That can have sex with you, that can recursively improve themselves so they get better and better over time, and eventually can, like, solve novel problems in math and science and physics. And so in 2015, they have some ideas about how to do this, but they don't know exactly how. When you get to 2019, they actually have ideas about how they're going to do it. And so things started to adjust as they got a better sense of it.

• 8:03Ryan Broderick?

  Have you ever read the book Homo Deus? It's the sequel to Sapiens. It's written by Yuval Noah Harari.

• 8:09Speaker 02

  I have not read, I've read excerpts of it, I guess I should say.

• 8:12Ryan Broderick?

  I read it on an overnight ferry in Greece

• 8:15Speaker 02

  That's awesome.

• 8:15Ryan Broderick?

  one summer. It was kind of lovely. And I read the book and I was like, wow. It came out in 2015 and it describes a version of AGI that I read and I was like, this is totally right. This guy clocked it. This is the most horrific thing I've ever read. Turns out he was pro all of the things in that book. I read it being like, you got the, the, but he sort of described this idea of an AI that is assigned to you at birth and it grows with you and it's like your personal assistant for life. And I was like, that's the darkest thing I've ever heard, but that sounds right. And like, I do think that's kind of in 2015, how people are thinking about these things, which is still very utopian. It isn't quantified yet. Does that sound right?

• 8:55Speaker 02

  Yes, I think in particular because if in 2015, you thought that within a decade, we would have something resembling AGI, like you were a lunatic. Like that was an extreme fringe view. Nobody took that seriously at all.

• 9:08Ryan Broderick?

  So at the same time in 2015, OpenAI launches, it's at a press conference co-chaired by a little guy named Elon Musk with also Sam Altman.

• 9:19Speaker 02

  Whatever happened to him, by the way? You never hear about him.

• 9:22Ryan Broderick?

  From what I gather, Sam Altman and Elon Musk are still close friends to this day. And everything is fine.

• 9:27Speaker 02

  Great.

• 9:27Ryan Broderick?

  And so, yeah, they explained that they're trying to advance digital intelligence. And the only sort of interview we could find from this event was in a Medium post. So here's Altman talking to Elon. Okay. Wait, do you want to read this with me?

• 9:42Speaker 02

  I would love to read it with you.

• 9:43Ryan Broderick?

  Do you want to be Elon or Altman?

• 9:45Speaker 02

  I'll be Elon.

• 9:46Ryan Broderick?

  You'll be Elon? Okay. The moderator says here, what's a bad example of AI? And Altman says, well, there's a, all the science fiction stuff. One thing that I do think is going to be a challenge, although not what I consider bad AI, is just the massive automation and job elimination. It's going to happen. The moderator says, what would be an example of the kind of data that would be shared? Altman says, so many things. All of the Reddit data would be very useful training set. And then the moderator throws it to Elon, and this is your part here. So he says, Elon, you are the CEO of two companies and the chair of a third. One wouldn't think you have a lot of spare time to devote to a new project. And what does Elon say here?

• 10:22Speaker 02

  Yeah, that's true, but AI safety has been preying on my mind for quite some time. So I think I'll take the trade off in peace of mind.

• 10:29Ryan Broderick?

  Oi, groynes! I can, I can only do one, I can only do one thing in a South African accent. I learned this many years ago. Would you like your coffee, blick or wet? Blick or wet?

• 10:40Speaker 02

  Wow, that's very good.

• 10:42Ryan Broderick?

  Pretty good, right? Yeah. He doesn't really have that, but yeah. So, no. So Elon obviously just wanted a friend and realized this was the only way. But I think it's safe to assume that Mr. Musk was fantasizing about another big sci fi idea, which was of course world domination. But what about Sam Altman? Casey, you've talked to Altman. Is he a true believer? Like, is he truly worried about humanity this way?

• 11:04Speaker 02

  So I knew Sam as the head of Y Combinator, right? So as a tech reporter, I write about startups. So I think the first time I met him in person, I was down at YC's Demo Day, where all the startups were showing off their latest stuff. I interviewed him, or I said hi to him about that. Later, the ACLU joined Y Combinator and I interviewed him about that. So yeah, all my early interactions with Sam had to do with him overseeing this vast empire of startups.

• 11:33Ryan Broderick?

  Did you like get a good vibe from him?

• 11:37Speaker 02

  Well, I mean, basically, yes. Like, Sam is very friendly. He's very personal. He's gay. I'm gay. Like, there aren't a lot of gay leaders in Silicon Valley. So I don't know. I sort of always thought that was like interesting about him. So yeah, and you know, if you talk to most people who've talked to Sam Altman, they'll actually tell you that like, one of his faults is that he will just make you think that he agrees with you about everything, which of course, in the moment, does not seem like a fault. You think, oh, I like this guy. He's like, he seems really smart. He like seems to see the world the way that I see it.

• 12:08Ryan Broderick?

  So you're saying it's like talking to ChatGPT?

• 12:11Speaker 02

  Yes, exactly.

• 12:12Ryan Broderick?

  It's just like, oh, this works. Yeah, okay, yeah, this is a person. So about a year later, OpenAI releases its first product, Gym, and it's named that because it's training algorithms. And we found a Wired article from the time that sort of goes into how OpenAI was operating. So it reads, "There are certainly some competing objectives. It's a nonprofit, but then there is a close link with Y Combinator, as you said, and people are paid as if they are working in the industry. The real power of the project is that it can indeed provide a check for the likes of Google and Facebook. It can reduce the probability that superintelligence would be monopolized."

• 12:51Speaker 02

  Yeah, so the whole like early history of OpenAI was rooted in this fear that the tech giants of the time would be the first to create superintelligence and that it would forever be controlled by a private company. And so the early like utopian vision of OpenAI was, we will build it first and we will democratize it. We'll make it very accessible to everyone. We will focus on the benefits to humanity rather than corporate profits.

• 13:23Ryan Broderick?

  So I think that would be surprising for people listening that like there was a sense all the way back in like, let's say 2016, that Google and Facebook were working on something like this. And is that true? Like, so it's not like they rushed to product because of ChatGPT. Like, they were already on this.

• 13:41Speaker 02

  They were, but the systems looked different at the time, right? So back in those days, they were less likely to talk about AI, except in a sort of like far future sense. They were more likely to talk about machine learning, pattern recognition, right? Google Photos launches and you can type dog into it and it will find pictures of dogs. That was the AI at the time, but you know, even then, folks who were working on OpenAI and other AI startups, were looking at that thinking, hmm, we think there's actually a path forward here to build something much more powerful.

• 14:10Ryan Broderick?

  And all of that culminates in OpenAI's first big victory in 2017, when they built a model that could beat the world's best Dota 2 player. Do you know what Dota 2 is?

• 14:24Speaker 02

  I assume you're talking about Defense of the Ancients 2.

• 14:27Ryan Broderick?

  Uh, is that what Dota stands for?

• 14:29Speaker 02

  Wow, you don't even know what Dota stands for.

• 14:32Ryan Broderick?

  Yes.

• 14:32Speaker 02

  Fake nerd. You're a fake nerd.

• 14:34Ryan Broderick?

  Yeah, so Dota is sort of like League of Legends. We can do a whole other episode about how those things interact. But yeah, so an OpenAI model beats Dendi, gets big headlines at the time. And a couple of months after that, in the early 2018, Musk leaves OpenAI. I assume because it's a better video game player than him.

• 14:55Ryan Broderick?

  And he states that the reason is that he doesn't want any conflict between Tesla's AI. But it also comes out later that he was most likely to try to take over the company and he couldn't do it.

• 15:06Speaker 02

  Yeah, yeah. He wanted to be the CEO and they wouldn't let him and he left in a huff.

• 15:10Ryan Broderick?

  You know, I can't imagine Elon Musk doing such a thing. I can't wait till he does it to our own government.

• 15:17Ryan Broderick?

  I feel like the age of chat GPT is we understand it starts in earnest in 2019. That's when GPT two launches. They don't share a full version of the program because they're afraid of like malicious use. It even becomes like a meme where, you know, AI researchers are saying, we've done this incredible thing. You can't see it, but it's incredible. It's too dangerous. A little outlet called The Verge covered this at the time. I think you're familiar with them.

• 15:43Speaker 02

  I am.

• 15:44Ryan Broderick?

  Yeah. And so they, uh, The Verge writes, OpenAI has researchers collected their training data by using Reddit as a filter. In some sense, all the work was done by people on Reddit upvoting posts. One researcher for OpenAI said, OpenAI Director Amon Di adds that at least they didn't use a more toxic source like 4chan, which I think is probably good. Yes.

• 16:04Speaker 02

  Wait, did you say OpenAI Director Amade as in Dario Amade, who left to go found Anthropic?

• 16:08Ryan Broderick?

  Oh, yeah, that's true. Yeah.

• 16:10Speaker 02

  Yeah.

• 16:10Ryan Broderick?

  Yeah, good catch. Um, so I still think this is a moment where like normal people don't know any of this is happening, right?

• 16:17Speaker 02

  Yeah. So GPT two is a pivotal moment in the history of this company that nobody ever thinks about. And the reason that it was pivotal was that inside OpenAI, they said, holy shit, we see the road to AGI now and we're going to put out GPT two and the world is going to go insane and everything is going to start changing right now. And what happened instead was they put out GPT two and you'd like give it the first few words of a sentence like, I went to the store to buy. And then you say like, finish this sentence. And it'd be like, I went to the store to buy a kindergarten for the potato. And normal people saw that and they were like, this doesn't seem very good, but it was such a radical leap forward from what had come before that inside OpenAI, they

• 17:01Speaker 02

  said, holy shit, this changes everything.

• 17:05Ryan Broderick?

  And I feel like that is the core difference between OpenAI and the tech giants, which is that they, and, and we're going to get to it right now in, in sort of the, the crazy like scale up era between 2019, 2022, where they're, they're basically releasing unfinished or unpolished products to the public. So this is when Sam Altman leaves Y Combinator to run OpenAI full time. They reclassify themselves as a capped profit company. Have you heard this?

• 17:30Speaker 02

  I, it was, it was new to me when they did it, but yeah, I've heard of it since. Yes.

• 17:34Ryan Broderick?

  So it's a hybrid between a for-profit and a nonprofit. So if you invest a million dollars, you can only get a hundred million in profit. This is the moment when OpenAI. Becomes one of the tech giants that they had spent most of their early years trying not to become. And this is also the moment when OpenAI basically becomes the one AI company, which they were also created to prevent from happening. This is the launch of Dolly Mini, which is sort of like the first public version of the art generator Dolly. You know, that's when you start to see AI art screenshots all over Twitter. People start using this stuff. And I think it's a lot of the the first reaction was, wow, this is new and interesting.

• 18:19Ryan Broderick?

  Quickly followed by people being like, this is awful and it looks ugly. But of course, a lot of people in Silicon Valley looked at that and said, oh, my God, this is the future. And that that proved to be a very pivotal moment in our story that we're telling today. Because from there, you know, OpenAI gets a billion dollars from Microsoft. And then they continue to claim that they need more and more capital than any nonprofit has ever raised. And this is when, you know, all of the investments start pouring in.

• 18:46Speaker 02

  Yeah.

• 18:46Ryan Broderick?

  Like, I mean, were you skeptical that this was gonna go anywhere? Like, looking back, did you think ChatGPT would look like it does now then?

• 18:53Speaker 02

  Certainly not then. You know, my feeling about AI was like, probably eventually they'll figure it out. I mean, I felt the same way about like flying cars, you know, where it's like, well, according to the principle of physics, you'll probably be able to come up with a flying car at some point. Was it gonna be five years, 15 years, 50 years? I don't know. That's sort of how I thought about, you know, AI at the time. It did change, though, when ChatGPT came out. And then all of a sudden, it felt very tangible.

• 19:21Ryan Broderick?

  Yeah, it moved very quickly. I mean, the spring of 2022 is when Dolly 2 hits. And then they sort of enter an arms race with Hugging Face, which is, for people who don't know, they describe themselves as the GitHub of AI. They launch a free version. Then Dolly Mini launches, which is free. So all of a sudden, people, I think for the first time, people could see what this stuff could do visually, which really, I mean, they're not good by today's standards, which is kind of amazing. But I remember being on Twitter at that time and just being like, how is this possible?

• 19:56Speaker 02

  Completely. I mean, as somebody who like, I used to try to draw as a kid and I plateaued very early. Like, I just was not good at it despite spending many, many hours at it. And so the idea that I could just type like, you know, frog in a bathrobe into a box and then I would get a frog in a bathrobe did legitimately impress me. But I thought, well, like, I thought, well, this is something new.

• 20:16Ryan Broderick?

  I am probably not impressed anymore, which is weird. And I feel like we can talk about sort of the emotional trail of all this stuff. But I was like, this felt new. And I think you and I both at the same time wrote takes that were kind of like, it's here. Whatever this is, it's here. This is also the moment where we start to get like, I think the pattern that we've now seen for a couple of years, which is OpenAI does something, and then you get a open source or free version from someone else that is pretty much the same. So Stable Diffusion launches in August of that year, and then Midjourney launches inside of Discord as a Discord bot, which I still use from time to time. I like it better personally. I mean, do you have a favorite image generator now?

• 21:00Speaker 02

  Um, honestly, I was bullied by readers on Blue Sky into not using the image generators anymore. So I don't really use them, except for just like, maybe making, you know, the occasional meme to send to my friends. It does seem like Midjourney does the best work, but I have been so put off by the Discord-based interface, which my understanding is you don't even need to use Discord anymore. But usually for my image generating purposes, I just see whatever Dolly can do. And if that's not good enough, I just give up.

• 21:29Ryan Broderick?

  Yeah, I feel like that's what most people are doing. And in fact, you know, at the end of 2022 is when we get ChatGPT, which is now running on GPT 3.5. Listeners, you're gonna hear like a bunch of models and like numbers. You don't have to care about any of this stuff. It's just like how they organize it. So I want to read your reaction to ChatGPT. So I'm gonna read your work to you. Okay, are you comfortable with that?

• 21:52Speaker 02

  Love it. Okay. I love that.

• 21:54Ryan Broderick?

  I mean, and again, here's where I just go back to the wonder and confusion I feel about the fact that you can just sort of plop code down and this thing, just using this corpus of text that has ingested, give you the right answer by just like predicting the next word. And it just makes you wonder, is it because the answer to that specific question was somewhere else on the internet and it just matched that? Or is there something more arcane happening? Do you feel the same way now?

• 22:22Speaker 02

  I think I have a more sophisticated understanding of how these systems work, and I think the systems themselves have evolved. And I think the important distinction I would draw between like what I think now versus what I thought then is that in most cases, I do not believe that ChatGPT is simply running a search function. Like if you say like, you know, uh, what is the fastest way to beat Legend of Zelda Breath of the Wild? I don't think it is just searching the training data for like the Reddit post that has the walkthrough. I think it is doing something different, which in some ways approximates the way that a person would reason through

• 23:03Speaker 02

  something. And that is really trippy to me.

• 23:08Ryan Broderick?

  I think that's right. I wrote about this recently, but I used it to figure out what was wrong with my cactuses. So like I sent it a photo because like, that's not really something you can do with the internet. And I'm always sort of looking for like genuinely new things and behaviors you can do with it. And so I sent it a photo and it like looked at the photo, it analyzed the photo and it like gave me some advice and the advice seemed to be pretty good. That is not searching the internet for a Reddit post. That is something quite different.

• 23:36Speaker 02

  Right. And, and by the way, like all of the tests that they set up now to benchmark the performance of new AI models include questions that are known not to be in the trading data, right? And over and over again, we find that the leading models are able to solve these problems that are not in the training data, which sort of disproves the it's just a search function theory.

• 23:57Ryan Broderick?

  Interesting. This is the moment in our story when everyone is kind of realizing this for the first time. In 2023, the very beginning of the year is when ChatGPT breaks 100 million monthly active users, which I think is very important because we still to this day don't actually know how popular these services are. Like, I think that's something that like really breaks my brain a bit is like, I don't know how many people out there are using them. Like I found my sister is like writing most of her emails at work with them.

• 24:24Speaker 02

  Well, I'm curious why you say that because you know, ChatGPT has said that 300 million people a week use them. The information reported this month that I think 15.5 million people are paid subscribers to ChatGPT. So I feel like we have pretty good data that hundreds of millions of people are using these products on a weekly basis.

• 24:41Ryan Broderick?

  But we don't know what they're doing with them. If Facebook says we have a billion monthly active users, you can like go on Facebook and you can kind of get a sense that like, yes, there are a billion people using the site in some capacity. What kind of freaks me about AI tools in general is like, we don't know where that energy is going. Are they all just having sex with it? Or are they doing something else, something more insidious? And I think this is when the paranoia about this stuff starts, right around 2023. This is when you start to get people really loudly talking about, you know, not knowing what is and what isn't generated by AI. This is where you get huge public conversations about where the data is coming from that powers

• 25:22Ryan Broderick?

  these systems. It effectively just made everyone on the internet really nervous and really distrustful of what they were looking at.

• 25:29Speaker 02

  That's right. Yeah.

• 25:30Ryan Broderick?

  It's like when Dr. Manhattan got created in Watchmen. Everyone loses their fucking minds. And so there's everyone's paranoid on the internet and also everyone's paranoid inside of OpenAI. So this is when you get all this stuff about Sam Altman being fired by the board and being rehired. Like, all of that's happening because essentially they've created this extremely destabilizing technology and no one knows what to do with it. You had like a great sort of like summary about this where like no one could like really agree on what they were disagreeing about. And it just feels very much like 2023 is the year we start being destabilized by this technology and the company itself is also being destabilized. What is your read on like, looking back now, what is your read on what happened with OpenAI internally there?

• 26:14Speaker 02

  Yeah, the Sparks Notes version is that people very close to Sam Altman stopped trusting him, took those concerns to the board. The concerns echoed concerns that the board had about Sam, and so they fired Sam because they were taking their mission seriously as a nonprofit and they thought, you know, our job is to safely create very powerful AI. And we no longer trust the CEO. So they got rid of him and that the entire world freaked out. And the collective workforce of OpenAI basically decided that Sam was the linchpin to the success of the company. And so they rallied and were able to restore him.

• 26:53Speaker 02

  And all of the people that, I should say, almost all of the people that raised those concerns about Sam quickly left the company. Most of them have since started their own AI companies. And now OpenAI is just the Sam Altman Show.

• 27:06Ryan Broderick?

  So it started as a nonprofit to ensure that if we were ever to create a computer smarter than human beings, not one company would own it. This company has now become a company that is trying to build a computer smarter than humans that they own. That, that, that is, that is the trajectory here. And now he's going to be, you know, in charge of our nuclear security systems or whatever, which is great.

• 27:25Speaker 02

  Maybe, or, or maybe, uh, you know, Elon will have him thrown in a Gulag. We'll see.

• 27:30Ryan Broderick?

  Uh, that'd be great. Uh, I want to spin through the current AI platform landscape before we get to our next, uh, our next section here. 2023 is basically when all the other big tech companies launched their competing AI. So you get Meta, Google, some other spinoffs from OpenAI, and they all start coming out with, you know, new twists on this. And this stuff just becomes unavoidable for anyone with a computer. I mean, it's in your feeds, it's in your email, it's in your docs, it's in your phone. It starts to change how Google search works. It starts to change how social media works. And if the goal of OpenAI was to keep this godlike power away from being owned by just one big corporation, it failed, I mean, spectacularly. It resulted in OpenAI becoming either

• 28:15Ryan Broderick?

  one of the big corporations that own this stuff, or it created a world where all of these companies will go bust and no one controls the stuff, which seems equally dangerous to me. I'm personally a Grok guy. It's got an 18 plus sexy mode now, and I can't stop fucking that thing. And Grok runs on XAI, which many of the people involved with that are now running parts of the federal government, which is

• 28:40Speaker 02

  That's right, yeah.

• 28:41Ryan Broderick?

  What's clear is that the people shaping our world all believe this is here to stay. And in a fundamental shift in the way we use technology. And I think you, Casey, believe this isn't going the way of the metaverse, that you think this is here to stay too.

• 28:59Ryan Broderick?

  And so after the break, we're going to talk about why you don't think people are taking this seriously enough. And we're also going to talk about how you got yourself in a bit of hot water and look at how your little blue sky beef has exposed a major pressure point in the AI debate, namely, you think it's time to start ringing the bells. Everyone else thinks it's time to start cheering when Nvidia's stock price drops like two points. And we're going to talk about that right after some AI-generated ads from our sponsors.

• 29:29Speaker 00

  Progressive's Accident Response can contact emergency services if you're in a major accident. The only thing worse than someone you love being in an accident is not being able to help them. So Progressive is here to help when you can't. So get peace of mind knowing your loved ones are protected with Progressive's Accident Response. Learn more at progressive.com. Progressive Casualty Insurance Company and Affiliates. Limited availability in California. Driver is responsible for emergency and towing charges not covered by the policy. Coverage subject to policy terms and conditions.

• 29:59Speaker 03

  Your sausage McMuffin with egg didn't change, your receipt did. The sausage McMuffin with egg extra value meal includes a hash brown and a small coffee for just $5. Only at McDonald's for a limited time.

• 30:11Speaker 04

  Price and participation may vary. Promotion pricing may be lower than meal pricing.

• 30:16Ryan Broderick?

  The unsettling question at the heart of the debate that's that no one seems to want to ask is, what if AI is in a bubble? What if it does keep getting better? And how do we destroy robot god? So you, Casey, recently described the discourse around AI as falling into two different camps. The first camp, you say, you associate with external critics that hold that AI is fake and sucks. And the second camp, which you associate with internal critics, believes that AI is real and dangerous. So let's start with the external critics. I want to sort of get your take on like, what you think they don't understand. Like, what are they missing? Because it feels like that's where the distinction you're making is that like, the people on the outside don't know what the people on the inside know.

• 31:01Speaker 02

  You know, among the criticisms made by this camp are AI is really bad at doing basic tasks and will probably always remain so. Nobody actually wants to use AI. There is no future for AI. All of these companies are going to go bankrupt pretty soon. The generative AI bubble is about to pop. And while I think it's rarely explicitly said, the takeaway from this line of thinking is, and therefore, you don't have to worry about it. You can relax. It is not going to harm you. If you don't like using ChatGPT, great. You never have to think about it again. This is just the latest Silicon Valley Bro

• 31:44Speaker 02

  Fest, and it is all about to explode. And we will just sit back and look at all of these companies in flames, just like we did when crypto exploded. And I think this betrays just massive amounts of cope and wishful thinking from people who are usually in their mid 30s and are tired of hearing about new technology.

• 32:04Ryan Broderick?

  So I want to stop you right there. I want to stop you right there because I think that last point is the most important. There is effectively an entire group of millennials that are obviously very stressed out that the internet landscape they grew up on and came of age with has shifted. And there is an entire content economy that you can become very, very popular inside of if you just tell them that everything will stay as it was in 2013, and you don't have to worry about anything.

• 32:35Speaker 02

  Hundred percent.

• 32:36Ryan Broderick?

  And yeah, I agree is very short sighted. Also, like, if there is an AI bubble that's going to pop, we know what happens, which is that like, a couple people survive it, and then they take over the entire industry. That's what happened with dot com. That's what happened with crypto. Fun fact, it's still around. I remember, actually, I was out to like a coffee with like this like AI writer guy who was like very excited about like building AI tools on top of ChatGPT or whatever. And I was talking to him for about a half hour, and I finally was like, look, I gotta ask you a question. If AIs is gonna be as good as you think it is, why do I need an app for it? Like, why do I need a wrapper? And then I was like, to go further, like, why do I need an app at all?

• 33:20Ryan Broderick?

  Like, why couldn't I just, like, ask ChatGPT to, like, do calculator stuff? Or do a calendar? Or, like, make a special app I need if I really need it? But, like, ultimately, I could just run inside of it, right? And so if the bubble does pop, that's what pops. It's all of the people who, I guess, it's called like a wrapper. You're putting a wrapper around OpenAI stuff? Yeah.

• 33:41Speaker 02

  Yeah.

• 33:42Ryan Broderick?

  And I think it is very short sighted to think that all of this will just suddenly go away.

• 33:48Speaker 02

  Yeah. I mean, again, just like, look at the rate of change. Look at the difference between GPT-2 to like this O3 model that now powers deep research. It is an extraordinary leap forward that took place in about two years. But in this moment, most people have not even tried deep research, right? So this is, you know, a bit of a digression, but I think an issue in these debates is always that some of the people who most fervently believe this fake and sucks stuff are the people who have used AI the least. And I think it's important to remember that when you're taking their opinions into account.

• 34:23Ryan Broderick?

  Well, one person whose opinions you targeted specifically, talking about very retro millennial behavior in your call out post, then wrote a call out post about you. So Gary Marcus writes, "Newton vacillates far too much between AI and generative AI, when in fact, the whole ballgame is whether the $200 billion investment in generative AI is the right path to AI or not. For example, he gives a long list of things AI has done, but says very little about how those systems work. Some are based on LLMs and chatbots, blah, blah, blah, blah, blah. It's really about whether generative AI in particular can be made into AGI just using more data and compute. It confuses matters to conflate all that together. Newton just doesn't seem to get this. His writing on social media strikes me as astute.

• 35:07Ryan Broderick?

  Here, he seems to be missing a central point." Do you feel like you're missing a central point there?

• 35:12Speaker 02

  I mean, maybe I've missed lots of points and I'll continue to miss points in the future. You know, I think the thing about Gary Marcus is that he's been wrong so many times. Like this man has been calling the end of the generative AI boom since it began. And one of the big reasons is that Gary Marcus just thinks that AI should be built differently. He does not like the large language model approach. He has a different approach that he thinks that the industry should take, which is something called symbolic AI, which I'm not even going to bother to describe. But when I read Gary Marcus's writing, I see someone who is butthurt that his idea for how to build AI did not take off, and another version did. And I think it has blinded him to how successful the AI labs have been just at trying other approaches that Gary Marcus didn't come up with.

• 35:56Speaker 02

  So all he is left to do, as he continues to make predictions that turn out to be false, is to just say, well, just you wait, it's gonna fail the next time, it's gonna fail the next time. And it's become sort of a joke in the AI industry, because no matter what the AI industry does, Gary Marcus is there to say, ah, this is real, don't believe your eyes.

• 36:11Ryan Broderick?

  The point you hit about his belief that it should work differently is really important here, because I don't know if the average person understands like that there are whole philosophical schools warring over AI. And these are very old schools of thought. Like, we haven't really talked about like the EAC people, what is it, like the Ethical Accelerationists? The Effective Accelerationists or the Effective Altruists. Like these sort of warring camps of people who believe in a kind of like old school revelation, biblical sense of AI and how they can use it to like sort of enact the future they want. And there's also all kinds of people who believe in like different versions of how AI should be built and function. I don't remember this kind of like handholdy, kind of like navel gazy stuff around social media,

• 36:55Ryan Broderick?

  because I think social media was just seen as like a very basic infrastructure. Yeah. So, but these people have been blogging about this stuff for decades, waiting for this exact moment. And I think many of them feel a little angry that they're like maybe are being left out in the cold or that their philosophy is not being reflected, blah, blah, blah, blah, blah. So if the listener here is looking for like a clear sense of Marcus's like viewpoint, this is one line that I think is like pretty useful, which is that LLMs or large language models are already dangerous despite their limits. Covert racism, deep fakes, propaganda, discrimination in employment, insurance and housing. Many downsides are already here and potential misuse in the military seems imminent. And then another great writer Edward

• 37:38Ryan Broderick?

  Anguissa Jr. summed it up, writing, "AI can be real and fake and suck and dangerous all at the same time or in different configurations." Which I sort of think that's the middle ground that I subscribe to, which is that like, bad AI is as dangerous as good AI.

• 37:55Speaker 02

  Sure. And that last quote that you had from Gary Marcus, I agree with. Like, when I started this whole fight, I also had a long section about the things that I agree with Gary Marcus about. You know, again, maybe it is just worth saying here, some of the critiques that I agree with, like, the chatbots hallucinate, they get things wrong all the time, you cannot trust them. The way that these systems were trained was completely unethical and maybe illegal, right? Corporations are amoral and some of them are going to die and lose many billions of dollars, right? So like, there's just like vast swathes of the AI critique that not only do I subscribe to, but have been saying from the beginning. And I think that's important to point out. The reason that I, you know, Ryan, you've known me for a long time. I do not start a lot of fights on the internet. You should now.

• 38:35Speaker 02

  But this one, I started on purpose because I wanted to make all these people come out of the woodwork and say that AI is dangerous. I wanted them to go back to their audiences and say, no, look, this guy has it completely backwards. It actually, like, it can be both at the same time, right? It can hallucinate, but also it's super dangerous. And it started this whole conversation about the ways in which AI is dangerous, which was the conversation that I wanted to have. So, look, are there certain kinds of AI that are like truly fake, like people pretending to use AI when they're not? Absolutely. But to me, the vastly more important thing is that these systems are getting very good very quickly. There are not guardrails being built around them. There is no effective governance of any of this. And we have line of sight to superintelligence. Like, that

• 39:19Speaker 02

  is the holy fucking shit big deal. But you have all of these liberals and progressives on Blue Sky that want to retreat into this cocoon of it's just crypto 2.0. And I'm trying to shake those people by the lapels and say, wake the fuck up. The singularity is happening.

• 39:36Ryan Broderick?

  Just for people who are listening, Casey's bright red right now. He's completely red.

• 39:42Speaker 02

  But I'm also handsome. Tell him I handsome.

• 39:43Ryan Broderick?

  Yeah, the AI model that's powering Casey's voice is glitching out really bad right now. He had to step in and talk for himself a couple of times there. That was interesting.

• 39:52Speaker 05

  Hey, this is Grant, the producer. I'm just gonna jump in for one second. Can you define what you mean by singularity?

• 39:57Speaker 02

  Yeah.

• 39:57Speaker 05

  I only think of sci fi when I hear that.

• 40:01Speaker 02

  Sure. The singularity is this long, like prophesied point at which computers become smarter than human beings. Like that, that's the idea, right? Like people like Ray Kurzweil have been writing about this for 20 years. And, you know, you look at the benchmarks and we now have this set of tasks where AI is already almost the best in the world. Like can do better on various forms of tests, is at least as good as a PhD. And again, I'm not saying that it's there now, but it's pretty close and there are not that many levels left to go.

• 40:31Ryan Broderick?

  If, if you want a good look at like, I think the most concise summary of the fears around a singularity AI. And what a lot of people I think crib from to this day. There's a short story published in 1967 called I Have No Mouth and I Must Scream. And it's about a super intelligence that takes over the world. And it basically like tortures like five people forever. And the line that is the title is how the AI describes itself. And then it, I won't spoil it, but it finds a way to make humans understand that. And I think most people like when they're talking about the danger of this stuff, that's when they're visioning is like a super intelligent computer.

• 41:12Ryan Broderick?

  If you want an optimistic one in the effort of fairness, there's a short story from 1956 by Isaac Asimov. Kind of an unknown writer. People don't really know him, titled The Last Question. Which is similar. It's about basically people asking an AI a question, the same question over like millennia. And then the AI like gets good enough that it can finally answer the question. And I won't spoil it, but it's quite beautiful what happens. They're kind of like same, same idea, but different takes. Which I think is the point here, which is that like. We don't really know what's gonna happen with this stuff. And even if it is a bubble, even if this stuff is a blip, it is clearly on a roadmap towards something much larger.

• 41:57Ryan Broderick?

  And I feel like denying that is ridiculous at this point.

• 42:00Speaker 02

  Yeah.

• 42:01Ryan Broderick?

  I really do. I also think if you look at the history of even the most recent history of how copyright, let's just focus on the copyright thing. Cause I think for a lot of people, it really bothers them. If you look at something like Wikipedia, you look at something like Photoshop, you look at something like music piracy, which were all big things around my childhood and were all things I was told would ruin the world. I was told that Wikipedia couldn't be used in school. I was in art class. I was a digital artist, like when that was very early and I wasn't allowed to use Photoshop in my art class because it was cheating.

• 42:33Speaker 02

  Wow, wow.

• 42:34Ryan Broderick?

  Yeah.

• 42:35Speaker 02

  That's fascinating.

• 42:36Ryan Broderick?

  Yeah, now it's the industry standard for almost all commercial art. And then like for music piracy and its effect on sampling in particular, like you think about like Girl Talk being sued for how many millions of dollars? And now it is just a thing that is in all music. Like these things are clearly moving towards a world where the copyright that we built in the 20th century just doesn't function correctly. Our idea of how we get knowledge doesn't function correctly. Like I also find it fascinating, particularly on Blue Sky, when people are like, Chat GPT is the devil. It is destroying human creativity. We need to make sure that the Internet Archive can have as many free books on there as possible. And it's like, okay, sure. Um, I love the Internet Archive. I also see the contradiction there.

• 43:19Speaker 02

  Yeah, yeah. I don't find there to be much of a coherent worldview, like in the Blue Sky critique of AI. Other than that, people are justifiably concerned that they're gonna lose their jobs and this thing is gonna ruin their life. Like that's the thing.

• 43:32Ryan Broderick?

  Which is justifiable, you're right.

• 43:33Speaker 02

  Yes, and that's the like, that's where I want to build a bridge to the like, Blue Sky leftist is to be like, you're right. Like this could lead to lots of job loss through automation. And you actually might want to think about building a political movement around that. Right to, you know, to enshrine some protections for yourself as a worker. That makes sense. Or maybe you're worried about what happens if, you know, a handful of big corporations have access to a super intelligence that, you know, might be able to help them do things that we can't even imagine. You might want to have a political movement around that. But step one is taking it seriously and not kidding yourself that it's all gonna stop working tomorrow.

• 44:11Ryan Broderick?

  I fully agree. And to take it back to crypto, actually, this was something that I kind of stumbled across as a thinking about this during my reporting on crypto, like during the pandemic, which is that I was like, okay, I don't understand how this works. And I should. And I know that the easiest way to learn how to do that is to talk to people seriously about what they're doing with crypto. And like, I'll even admit, like, I got a Coinbase wallet during the pandemic, and I put a couple thousand, I got, I put a couple thousand dollars in there, and I played with it to see how it worked. I would like, open up Twitter and like a crypto hashtags and try to trade off the back of crypto Twitter activity. And I had like a whole column running. And I, and like, I just spent some time being like, okay, how does this stuff work? I didn't make any money. I didn't lose any, but I didn't

• 44:55Ryan Broderick?

  make any. I use, I, it was fine, whatever. And I had, and it was interesting, like, it was genuinely interesting. And now I feel like I can understand how to politically talk about this stuff. And I do think you're right. Like, you don't have to go buy a subscription to OpenAI tomorrow to like, march against it. But you should understand it, like, how it works, why it works, who's behind it. Take it seriously.

• 45:16Speaker 02

  And crucially, if I can add one more thing, you have to keep trying it. So many people's opinions about ChatGPT were formed the day that it came out, and they don't understand that it continues to get better on an incremental basis. And occasionally, you have these leaps forward, like, in my opinion, we just saw this month with Deep Research, right? So this is a case where you cannot rest on the opinion that you developed of this stuff six months ago, because the industry, it is so dynamic, that you're just gonna have to keep trying it.

• 45:43Ryan Broderick?

  Well, speaking of how dynamic the industry is, after the break, we're gonna be talking about how everything we've talked about up until this point doesn't matter anymore, because China's in the game now.

• 45:56Speaker 00

  Life is all about making decisions. Sometimes you make questionable ones, like buying a new TV that's too big for the wall because the electronics store employee was a little too good at their job. Then, sometimes you make great ones, like protecting your home and auto with Progressive. You get customized coverage options, so you can be sure you have the right coverage. Get your quote today at progressive.com. Progressive Casualty Insurance Company affiliates and other insurers, not available in all states.

• 46:25Speaker 03

  Your sausage McMuffin with egg didn't change. Your receipt did. The sausage McMuffin with egg extra value meal includes a hash brown and a small coffee for just $5. Only at McDonald's for a limited time.

• 46:38Speaker 04

  Price and participation may vary. Promotion pricing may be lower than meal pricing.

• 46:42Ryan Broderick?

  Here's the thing. The way our tech giants are in an arms race against one another to be the company that makes the chatbot that ruins the world or runs the world. We are also in a cold war against China, who a lot of American investors believe is fighting the same battle.

• 47:00Ryan Broderick?

  I do think we hit a dividing line and the dividing line between the previous era of generative AI and this current one is the launch of DeepSeek. For the uninitiated, can you sort of like run us through what DeepSeek R1 is and how it sort of fits in here?

• 47:14Speaker 02

  Sure. DeepSeek R1 is what they call a reasoning model. Reasoning models we distinguish from a more traditional large language model like a GPT 4.0 because of the way that they approach answering your query. And the way they do it is by allocating more time when you enter your query to try to think through things. And one of the really novel things about R1 is you can watch it try to think through what it's trying to do. You ask it for something, it's like, okay, hmm, I should look that up and I should integrate this and oh, I just discovered that. That's kind of interesting. And the result is that when you get something out of R1, you're more likely to trust it because you've actually watched the entire thought process.

• 47:55Speaker 02

  So while reasoning models had existed before, we hadn't really been able to see them think. And also no one had ever heard of DeepSeek. And so people were really surprised when this emerged.

• 48:05Ryan Broderick?

  Yeah. The immediate reaction to seeing the reasoning was fascinating where people were like, huh, I guess Tiananmen Square didn't happen. Everyone went home and had a nice time. That's really interesting. But just a quick for the uninitiated, a quick sort of run through of China AI development here. So 2018, the US government bans all purchase and use of Huawei, which is sort of the biggest telecom company in China and the world. And they're banned from any government purpose in the US. 2019, Trump steps up with an executive order banning all commerce with national security. Security threats, which seems to be specifically targeting Huawei. 2020, reports come out that Trump is trying to block any advanced computers from China, even if they're made by third-party countries. 2022, between the release of Dolly 2 and ChatGPT, the US bans China from importing any

• 48:50Ryan Broderick?

  heavy-duty GPUs. Reuters says the ban is on top AI chips. So the Cold War is sort of in full swing here. 2023, Joe Biden sees the new Mission Impossible movie and gets really fired up about it and decides that AI needs safeguards and signs an executive order, basically promising to deal with AI. And then in January of 2025, Trump says that OpenAI will be the major beneficiary of a $100 to $500 billion investment to build Stargate, the newest torment nexus, if you will. And it's basically going to be like a bunch of data centers in the US. Do you get a sense of why Chinese AI has grown fast enough over like the, let's say, the last five years to inspire this kind of fear and panic from the US?

• 49:36Speaker 02

  Well, it's rooted in this idea that if you have a superintelligence, a superintelligence would be massively beneficial to national security. It could help you build a better army. It could help you build better weapons. It could help you do better espionage. And so if you accept that all those things are true, then you're going to want to invest in AI. And, you know, these large language models, the thing about them is a lot of the early ones were built in the open. So we basically know how to do it. You know, it's kind of like if superintelligence could be achieved by like roasting a chicken. And like, there's a recipe for roast chicken on the internet. And we're not exactly sure like how you need to roast it and like what other like ingredients

• 50:20Speaker 02

  you need to put in the roasting pan. But like, if you can get a chicken and like put it in an oven, you're going to get a decent amount of the way there. So that's the point when some of the China hawks in the government start thinking, okay, what can we do to make it more difficult for China to roast a chicken?

• 50:36Ryan Broderick?

  I was at VidCon for 2019, I think, and I was going to all the panels with Chinese apps because I just wanted to like see what was out there. And I got an interview with an American employee at Baidu, which is like a massive Chinese tech company. And I asked him, I was like, why is the machine, I was calling it machine learning at the time. And I was like, why is machine learning so good in China compared to the US? And he's like, because we have access to like as much data as we need and want. Like, there's just no, there's no real data privacy laws there. So we can just do, we can hoover up the whole thing. And so I do think because of that, though, and because of the censorious environment they're building these products in, they have a very different feel. Like, we know this from like Temu and TikTok and Shein. Like, they just think about technology differently

• 51:18Ryan Broderick?

  than we do. And it turns out like people kind of like that. And they like DeepSeek enough that it climbed to the top of the App Store when it launched in January, and it basically crushed US AI company stock overnight. Do you think that is largely because it was launched as open source? Or do you feel like there's something larger there that like actually kind of caused the hysteria?

• 51:38Speaker 02

  It's hard for me to say exactly what caused it. I think there were a few factors. Like, one, I think people did not realize that China was going to be able to catch up as quickly as it did. Two, as you point out, it was free. So at the time, you could not actually use OpenAI's reasoning model for free, but all of a sudden, you could use this thing that was like almost as good for free. So that was pretty interesting. And then three, it was this huge media story, right? Like, everybody wrote about this. And so there was just this kind of like panic. Now, two weeks later, you look back on it, and it looks like a bit hysterical, right? Like, the stock prices all recovered,

• 52:18Speaker 02

  and no one is arguing that DeepSeek actually has an advantage over of the US AI labs, just that they were essentially able to reverse engineer it and copy it more quickly than people thought they would otherwise be able to. But, you know, at the same time, the fact that China can quickly copy US AI models becomes important as those models get more powerful, right? So like, it's both a bigger and smaller story than I think people give it credit for.

• 52:43Ryan Broderick?

  Uh, your close friend and colleague Ed Zetron wrote in his newsletter, "The AI bubble was inflated based on the idea that we need bigger models that both are trained and run on bigger and even larger GPUs. A company came along that has undermined the narrative, both weighs both substantial and questionable." And I do think that is kind of a piece here that like,

• 53:04Speaker 02

  It's not.

• 53:05Ryan Broderick?

  You don't think that that kind of flies in the face of Sam Altman saying that we need billions of dollars for years?

• 53:10Speaker 02

  No, not at all. And I think that's why it's so important when you're reading about AI, to read people who actually interview people who work at these companies and understand how the technology works. Because the entire industry has been on this curve where they are trying to find micro innovations that reduce the cost of training the models and to reduce the cost of what they call inference, which is when you actually enter a query into ChatGPT. And if you plotted the curve of how the cost has been falling over time, DeepSeek is on that curve, right? So everything that DeepSeek did, it was expected by the AI labs that someone would be able to do. The novelty was just that a Chinese company did it. So to say that it like upends expectations of how AI would be built

• 53:54Speaker 02

  is just purely false and is the opinion of somebody who does not know what he's talking about.

• 54:02Ryan Broderick?

  I do love, I do love how like, there is no civil way to talk about AI anymore. Uh, I do love like how.

• 54:08Speaker 02

  I thought what I said was very civil. I mean, there's things that are true and there's things that are false. Like you can choose which ones you want to believe.

• 54:14Ryan Broderick?

  So you don't think any long-term damage to like OpenAI, let's say, was caused by DeepSeek? In terms of reputation, reputationally?

• 54:22Speaker 02

  No, no.

• 54:23Ryan Broderick?

  So I guess then that leads to the next question, which is, if DeepSeek is this moment where, you know, a company that no one is, you know, no normal person's ever heard of, comes out from a different country and they're like, "Look, we made a thing that could basically do the last like 10 years worth of developments. It's all here. It's free. You can have it." I guess historically, that would mean that like that's the starting point now. And I feel like a lot of tech critics forget this, that like technology does grow exponentially. So like if you can, I mean, we see this with like Raspberry Pi, right? Like you can go buy a really powerful computing hardware and you can build something that would have taken someone 50 years ago, months and months, if not years to develop. So now like AI functions that way where you can download the open source deep seek model

• 55:07Ryan Broderick?

  and then build your AI on top of it. Is that the right, is that the right way to look at this then?

• 55:12Speaker 02

  Yes, and let me say what concern that actually does create, which is how do these companies have a moat? Right? Like this is a real concern, which is, okay.

• 55:21Ryan Broderick?

  Can you define like the idea of an AI moat?

• 55:24Speaker 02

  Yeah, so let's say that OpenAI is in the lead right now and it has the best models. And it is the case that other AI companies, including Chinese AI companies, can quickly reverse engineer that and like offer that for free. That does become a problem for OpenAI because well, how are you going to sell the same service at the same cost if you're being sort of undercut on all sides? And I do think that that is a challenge for them, but I don't think it is insurmountable. Right? Because what has worked up until this point may not actually work in the next era. What do I mean by that? Well.

• 55:57Ryan Broderick?

  What do you mean by that, Casey?

• 55:59Speaker 02

  So you know, sort of how chat GPT works today? You have a query, you enter it, it does some mojo and that mojo is essentially on the internet and you can read how they do that and you can actually build it yourself in your spare time if you have the right hardware, right?

• 56:11Ryan Broderick?

  Right.

• 56:11Speaker 02

  Then the next era, which people are calling the agent era, is that there are going to be tools that can do multi-step tasks on your behalf. So maybe they can order groceries for you, they can sort through your email inbox, they can plan you a trip somewhere, right? It's not yet clear that whatever secret sauce goes into building that is going to be as easily reverse engineered as the earlier generation of stuff. It may be that it is, but we don't know that that's the case. And I believe that most of the economic value that OpenAI and these other AI labs is going to create lies in that agentic realm. If they're able to actually create. Virtual coworkers and all of a sudden you're the CEO of your company, you know, you're the CEO of Panic World, Ryan.

• 56:55Speaker 02

  You all of a sudden realize that for $200 a month, you can have an entire production team.

• 57:00Ryan Broderick?

  I can replace Grant with an AI for 200 bucks a month. Oh, my God, I can't wait.

• 57:04Speaker 02

  It could happen, right? And so they may have an opportunity there that something like DeepSeek is not able to as easily replicate because they don't actually know how to do that. So that's the idea.

• 57:13Ryan Broderick?

  I got in an argument with my readers, actually, probably a year or two ago, where they're like, you gotta stop writing about AI. And I was like, okay, I understand why you are upset about this, but at the time I said, you can run stable diffusion on your laptop. If the government outlawed OpenAI tomorrow, if no more generative AI ever, that would still exist. And I was like, you do understand that like this technology exists now, it can't go anywhere in the same way that like,

• 57:43Ryan Broderick?

  yeah, like, like the hard drive is full of like pirate, like, uh, what is it like a hacked fire stick? Or a Plex server, like this pirated stuff is out there and you can use it forever now. And I don't think the average person who is really mad about this stuff has really sat with the implications of that. So now the idea is not only is stable diffusion open source, something equivalent to ChatGPT is open source. And I feel like that is so massive that I'm still trying to actually process what that means because, yeah, the moat there is gone, then now you have to build something else if you want to survive. And then that is very accelerationist technologically.

• 58:21Speaker 02

  Absolutely. But man, I am so stuck by that story that your readers don't want you to write about AI because like, what is feeling that? It is truly that people do not want to think about, which again, I am sympathetic because who wants to think about the fact that they might lose their job? But like, you know, you and I are reporters and our job is to not turn our eyes away from this stuff. So it's like, you actually have to keep telling this to people, even though it is going to upset them.

• 58:39Ryan Broderick?

  I agree with that to, to a point, I do understand, I understand particularly from rears of a certain age, audience members of a certain age, this idea that like, the thing that they prided themselves on growing up with, like, how do you rotate PDF? The core millennial experience has like moved on in a way. And I understand why, like, a generation of people who believe that they invented technology and the, and grew up with it, now feel alienated by this new way of thinking about it, which is very. Very opposite to the like, we're the we're the only generation that understands how a file system works. Like, Gen Z doesn't know where files go, and they're never going to, because ChatGPT will handle it for them. And I understand that I understand the anxiety there. I actually do.

• 59:24Speaker 02

  Me too.

• 59:24Ryan Broderick?

  But the thing that I am the most struck by, and I think the thing that your sort of story here about your fights with the leftist blue sky kind of illustrates is the political naivety around this stuff. I wanted to share an anecdote with you, which I feel like has kind of changed my thinking about how politics and technology work together. Because right now, you know, obviously, there's a lot of kind of like leftist ludism that I'm finding very annoying. But a couple of years ago, I was talking to a PhD student in Brazil, and she was working on her thesis that was kind of looking into the way the socialist Chilean government before the coup was trying to build their own version of the internet. So the socialists in power

• 1:00:06Ryan Broderick?

  in Chile, they were trying to get this project off the ground. And basically, the way it would have worked is that they were going to network factories. So every factory was a server, and the Chilean internet would function in a totally different way. And I guess they kind of cribbed it from a Soviet idea of how to build the internet. And what I think is so cool about that is that, you know, as I said, like, most of the leftist discourse we're seeing right now are tired millennials that are like, unplug computer, computer bad. But there was a moment in time when that wasn't that wasn't the way that people were trying to solve problems. They were trying to find ways to embrace technology to support their politics on the left, because the right is definitely doing that.

• 1:00:47Speaker 02

  Yeah.

• 1:00:47Ryan Broderick?

  And I don't need I'm not going to say like, if you're if your communist revolution doesn't involve AI, like, it doesn't work. But I do think you cannot

• 1:00:55Speaker 02

  But that might be true.

• 1:00:56Ryan Broderick?

  It might be. This modern idea that somehow a new technology won't influence politics for on either side is very naive.

• 1:01:05Speaker 02

  Let me put it another way. Do you think that the Republicans aren't going to use AI? Do you think that the fascists aren't going to use AI?

• 1:01:12Ryan Broderick?

  They already are. The AFD, the Alternative for Germany Party, they're like a far-right, they're basically just straight-up Nazis. They came in second in the election this month there, and one of their main players did a whole bunch of press bragging about how useful AI was. And he, you know, what he said, which I think is very important, is that AI helps them show their supporters what the future will look like, what the far-right future will look like. And it's already happening here. I mean, Donald Trump, literally this month, shared an AI-generated video of the Trump resort he wants to build in Gaza. The worst people on earth have realized what this stuff can do,

• 1:01:55Ryan Broderick?

  which is render a more realistic vision of their horrible future they want to force upon us.

• 1:02:00Speaker 02

  Yes, they're literally installing it in the servers of the US government illegally, right? So they see the advantage in it. So I think if you're a liberal or a progressive, and you're deciding that you don't like AI for aesthetic reasons, right? Or some other, you know, reasons that you saw in a reply to me on Blue Sky, you're going to be fighting fascism with one arm tied behind your back. And I think that leftists might want to have a more engaged view of AI that at least asks the question, could this be helpful in organizing? Could this be helpful in persuasion, right? Could this be helpful in developing and articulating a media strategy? Because if liberals and progressives just choose forever not to ask that question,

• 1:02:44Speaker 02

  I guarantee you, the other side is going to have no problem asking it.

• 1:02:49Ryan Broderick?

  I think it's actually easier to just assume we can go back to like 2008. So I think like, I think, I think it's just easier to just turn back the clock. We introduced you as more optimistic than Ryan?

• 1:03:03Speaker 05

  I do not leave this conversation seeing any way that this doesn't just make the world suck more.

• 1:03:09Ryan Broderick?

  Oh, that's a good call. Yeah. How do you give us your, once again, give us your most optimistic take of like how AI is actually good and could be good.

• 1:03:18Speaker 02

  Like I said at the beginning, AI is a dual use technology. There are absolutely ways that it can be used for good. And the three that I think are the closest to like being real, some of them are real today, would maybe be these. Number one, it actually is helpful for knowledge work. It's not great for writing your headlines or writing in your style or like generating new insights. But man, if you need to like write a speech, if you need to give a presentation, if you are exploring a new idea and you just want something to bounce some ideas off of, AI is actually really good for that. And one of the hardest things in knowledge work is generally the blank page. AI really will meaningfully help you with that today, and it's only going to get better over time.

• 1:04:00Speaker 02

  Number two, I would say tutoring and translation. Education in this country is pretty not great. There are certain states in the United States where like certain forms of knowledge are just forbidden. Like, you know, God forbid you want to learn something about like the reproductive system in Texas, right? AI can absolutely be a tutor for you. If you are learning English as a second language, AI is amazing for helping you translate documents. For sort of editing your emails before you send them out, so they sound like they're written in like proper corporate English. So I think that that's really powerful. The idea of every child having a personalized tutor growing up that can meet them wherever they're at, I just think is is super cool. And then finally, I would say companionship, and this is a somewhat fraught subject that could be a whole separate podcast, Ryan.

• 1:04:43Speaker 02

  But when you think about how many people in this country or around the world are suffering, who need support that they aren't getting, who are struggling with an issue. And in your pocket, you have an infinitely patient, pretty wise therapist that can listen to you, can mirror what you're saying back to you, can suggest strategies to you. I think it's super promising. Again, this has some downside risk. We could talk about forever. But I think most people would be better off having a little bit of therapy and you can get a little bit of therapy from a chatbot today. So those are just three things that I would point out. There are lots more, but those things are all real and they are helping people today.

• 1:05:19Ryan Broderick?

  So I will say the translation one, I'm 100% on with you. I think there's a net benefit to translation, like just it is good. It works. Use it. It works better than Google Translate. It's great and it's getting better. Also, at the very beginning of sort of the AI run, I interviewed a bunch of therapists and I asked them like, what do you think about chat GPT therapy? And the overwhelming response was, we felt uncomfortable with Zoom therapy until the pandemic. And now we're sort of facing this other thing. And all of the therapists were just like, yeah, use it. Like, it's not a replacement, but like, anything is better than nothing.

• 1:05:51Speaker 02

  Exactly.

• 1:05:52Ryan Broderick?

  And, and I was really surprised by that, but I do think there is something there. That said, I hate AI. It's bad. It has to go away. It's stupid. It's a bubble. Please subscribe to my newsletter, garbage day dot email.

• 1:06:05Speaker 02

  Casey, thank you so much. I do want to say those three points you made about why to be optimistic feels so small compared to everything else we talked about. I'm about to go for a very sad walk of feeling very doomed.

• 1:06:19Ryan Broderick?

  Casey, I want to thank you for coming on. This was a blast. You scared me several times as you were yelling at me, but I enjoyed it. I really did.

• 1:06:29Speaker 02

  First of all, you set this whole thing up to get me to yell. So don't pretend like this wasn't what you wanted, Broderick.

• 1:06:34Ryan Broderick?

  Welcome to the puppet show, my friend. If people want to follow you online, where can they do that?

• 1:06:40Speaker 02

  I wouldn't. Don't follow me. Don't, don't look for me.

• 1:06:42Ryan Broderick?

  If people want to yell at you on Blue Sky, where can they do that?

• 1:06:44Speaker 02

  The listeners of your podcast, I don't want to hear from them. I'm sure they're lovely people. I don't want to hear from anybody. So you can Google me. Look me up in chat GPT, see what that says.

• 1:06:52Ryan Broderick?

  Yeah, perfect, perfect.

• 1:06:55Ryan Broderick?

  Panic World is a Garbage Day production. Subscribe to the newsletter at garbage day dot email. Panic World is written and produced by Grant Irving. It's hosted by me, Ryan Broderick. Our amazing researcher is Adam Bumas. It's engineered by Rebecca Seidel. Our deranged logo was created by Gabby Cash. Please give us $5 at patreon.com slash panic world. Please give us products to sell by contacting multitude at multitude dot production slash ads. For any other way you would like to give us money or work with us or promote us or become financially entangled with us, you can reach out to our fixer, our wonderful bag man, Josh Fjelstad, and you can reach him at [email protected]. And one piece of advice from me to you, chill out, touch grass while you still can.

• 1:07:43Ryan Broderick?

  I'm using ChatGPT every day now, actually.

• 1:07:45Speaker 05

  Me too.

• 1:07:45Ryan Broderick?

  It's a fundamental change in how we expect the internet to give us information. It is not like, oh, OK, I got a weird rash from like going, I wore like some fetish gear to a Matrix themed rave the other day, and I got like a weird rash afterwards. And so I took a photo of the rash, and then it told me what like most likely was wrong with me, which was I had a, it was like a sweat rash.

• 1:08:05Speaker 05

  ChatGPT told you, you're not cool enough to go to a fetish rave.

• 1:08:10Ryan Broderick?

  Yeah, I am not. I'm not.

• 1:08:13Speaker 00

  Life is all about making decisions. Sometimes you make questionable ones, like buying a new TV that's too big for the wall because the electronics store employee was a little too good at their job. Then sometimes you make great ones, like protecting your home and auto with Progressive. You get customized coverage options so you can be sure you have the right coverage. Get your quote today at progressive.com. Progressive Casualty Insurance Company affiliates and other insurers, not available in all states.