The Web’s ‘Existential’ AI Threat, OpenAI’s Microsoft Office Competitor, Tesla Robotaxi Launch
Channel: Alex Kantrowitz
Published at: 2025-06-30
YouTube video id: Le1_D-zp6os
Source: https://www.youtube.com/watch?v=Le1_D-zp6os
Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional coolheaded and nuanced format. We have so much to speak with you about this week in another big week of tech and AI news. We are going to cover um the tiny amount of traffic that AI has been sending to websites and whether that is a sign of a greater collapse to come. Uh we'll also talk about the latest in the OpenAI Microsoft wars. some interesting news coming from Anthropic that we can drop uh today and talk to you about. Uh we also have some news about what Mera Marotti is up to at Thinking Machine Labs. Of course, Tesla uh has gotten its robo taxi initiative underway and Jeff Bezos is getting married. So, joining us as always, not as always, joining us as as a special guest here to speak with us about this is our regular uh guest Reed Alberg, the technology editor at Sound For. Reed, great to see you. Thanks so much for coming on. Yeah, absolutely. I just got back from the wedding. It was great. No, I'm just kidding. I was not there. Wasn't at the Bezos wedding. This is my Easter egg for listeners. I have a list of celebrities that will be at the Bezos wedding, including one who was listed as uh ready to go and now apparently is not going. All right, I'll we'll get to all that at the end. Fine. Fine. I I'll just give the spoiler. It's Katy Perry. Katy Perry's was on the list. Apparently, she's not on the list now. We'll get to the bottom of this. We do journalism here. We'll figure it out. Um, but let's talk about where we do journalism and where a lot of the happenings on the internet occur, which is websites. So, a couple months ago, we talked about how um the CEO of Cloudflare, Matthew Prince, uh had gone basically public and said, "Listen, uh compared to uh a year ago, Generative AI saying sending way fewer visits uh to websites as let's say Google search." Um so, 6 months ago, this is these are the numbers. He said Google the ratio was 6:1. OpenAI 250 pages is pages crawled. So, six pages crawled, one visit. uh from Google, 250 pages crawled, one visit from OpenAI. Uh 6,000 pages crawled, one visit from Anthropic. That was 6 months ago. Now for Google, it's 18 pages crawled, one visit. OpenAI, it's 1,500 pages crawled, one visit. Uh Anthropic, it's was it's now gone up from 6,000 to 60,000 pages crawled one visit. Prince says, "People aren't following the footnotes." Is this time to panic? I mean, I sort of I'm going to title this uh this conversation or the the beginning of this podcast episode, the web's existential AI threat. Am I getting over my skis, Reed? Am I is this hyperbole? I think it's always a great time to panic in the media business, to be honest. Um and I'm and that's not just a joke. I mean I think it's it's like I remember you know being in uh you know studying journalism in college and talking about you know panicking uh in the media business and at that time it was you know just just the web was uh was completely destroying it. I think what's interesting and Matthew Prince has been on this you know he's been talking about this now for a while. I've talked to him about it. I think it's I think it's admirable what he's doing, but I also think that, you know, traffic was never a good metric to judge whether, you know, a story is valuable, a a media article is valuable. So, if that metric goes away, then yeah, I mean, it's going to hurt the it's going to hurt the current industry, but I think it actually is is maybe in the end a positive thing. Um, and I think there's there's two things going on. I mean, one is the traffic and the other is like, you know, that they're scraping all these websites and and pulling in the information and that part. If you if you get rid of traffic as a metric, you don't need uh you don't need to allow these things to scrape your websites, right? I mean, it's possible to to put a hard payw wall in place and to essentially stop that scraping. So, but then you don't get then you don't get the traffic. So, you get rid of that business model. I think in the end it will be a healthy thing for the media. Maybe not for current media companies though. So Matthew is going to be in New York. I'm in the middle of pitching his team to try to get him uh to come on and talk a little bit more about what he means here. We should note I think he's selling something to help publishers stop the scraping. But I don't think that's really going to change very much because ultimately uh bots like Chat Chipt are already so useful. So, I don't think that you're going to have like this this wall that publishers put up and all of a sudden people are going to go to websites. I think that's probably um because it just takes a few websites to get that traffic or to get that usefulness into these bots. Um, but Rita, I I'll just take the counter argument here, which is it's, you know, traffic didn't matter as long as you had some, right? Like, you needed to have a little bit of an audience. And you could say to advertisers, we have a high uh a very valuable audience. We have uh the top executives in the field. Um, therefore, you should do this event with us or this engagement with us. Um, when you have zero traffic, that changes. And by the way, it's not just publishers, right? We're talking about the web which also has I would say um uh entertainment sites, booking sites. I mean obviously Netflix isn't going to go away. Uh but there was a publisher I spoke with World History Encyclopedia. It's just you know a site where people go and learn about world history. Uh and that's taken a big hit from AI. So if all these um different aspects of the web start to go away, that does sound like maybe it could be a crisis. What's your take? Well, I mean, look, you make you make really good points, but I think in the end, you know, if if what you have is like all the highquality journalism is essentially non-traffic based, the business model is not is not maybe you have advertising, but it's not the targeted advertising that we see, you know, in in the traffic business, then let's see what these AI models get from this these few websites that are still traffic-based. I mean, it's going to be complete crap. And so I I think in the end they'll have to figure out some way that they'll either e either have to pay publications to sort of republish on their platforms and that could be that could be a business that maybe supports media other models will will emerge right I mean like you know at some you do it to events right there there there are lots of ideas to fund journalism subscriptions right before traffic even existed the media business was pretty healthy right people paid for newspapers and you know yeah they got they had the classifides business and all that stuff but but it's not just journalism though it's not just journalism that's the point I'm trying to make is that yeah we you know because we're journalists we like to talk about the journalism websites but um again like this is this is going to be everything yeah I mean look if if what you're talking about is like these passion projects that are sort of like the bread and butter of the web if people don't go to these websites anymore and they get just like ingested into AI models and and disappear. I mean, again, like then the AI models won't like they won't be as valuable. Like they'll there won't be anything to ingest anymore. And so they'll something will have to shake out is what I'm saying. Like it's not I I don't think it's like the Armageddon when it comes to just like the long tale of of web content. I I don't think it's Armageddon for that. I think something will, you know, I don't know what it is, but like that will work itself out. I think what I really care about is that is that journalism is it survives, right? That's what to me the most important thing and I think and I think that will survive too. Maybe I just sound like a like a just a total optimist, a techno optimist or whatever now. But I do think that Yeah. No, I mean it's definitely you have of all the people I speak to about this, you have some of the more optimistic perspective uh perspectives that I've heard about it. Um I don't fault you for that. I think that we tend to have things um end up growing together uh but not all the time and it is interesting that you'll end up getting a tremendous amount of power concentrated in the hands of these bots in the hands of let's say openai or anthropics claude although they don't really care too much about cloud anymore um but you'll have this power concentration maybe that's why we see all these billions of dollars uh being invested if you get the master chatbot where everybody accesses all the content, the news, sports scores, world history. Maybe they're instead of going to like booking or kayak, they're doing that within uh the bot. Then, you know, these numbers that Matthew Prince is putting out highlights kind of like the big question we've been talking about on the show for a long time, which is what is the valuation uh of of these b of these chat bots and these foundational model companies and when is it going to be justified? And I guess if you gobble up the whole web, you're okay. Well, you know, like you mentioned sports scores. That's an interesting one. Like people used to used to pay for newspapers so they could open it up and see the sports score, right? And then, you know, online publications started putting the scores up. And now I think the way it works is if you if you Google a sports score, you just see the score right there. And I I believe they have a deal. Well, maybe it's with ESPN, but like eventually it's it's going to I think they just get those scores directly from the leagues and they're just going to be paying the league for that information. This is not really a it's not that's not really like a valuable service that somebody that some publication is providing. So, I think if you're if you're developing if you're bu you're putting out some really valuable thing, I think it'll find it'll find a way to, you know, to to be valued, right? Like and and you talk about these bots, like I don't even know if that's how this is going to shake out. Like I think what you might have is sort of everyone will have their own personal bot that sort of handles their information diet and maybe that bot has subscriptions to things and it reads, you know, publications for you, but you're at least paying for it. I mean, that's that's one way this could go. I think like we're we're looking at this right now through the lens of what we already know, which is like the current web, and that is all going to change. everything is just going to be completely different. So I think people often like forget that like the this is such a fun this AI stuff is going to be such a fundamental change that like all these assumptions that we have or you I think you just have to discount all of them. Okay. And it's by the way it's not just publishers not just web publishers also the infrastructure of the web is going to change. And you recently had a conversation with Liz Reed who runs search at Google and you had you asked her about hey well how how's advertising going to work and you actually have this like idea or this thought that advertising is going to be quite different as we move forward. So what is going to happen to the Google stack? Because the I think and you put this in your article, right? They have so many great AI bets. They have Deep Mind and Gemini. They have Whimo. They have Isomorphic Labs. But the reason why the stock is the cheapest of I think of all Magnificent 7 stocks is because we don't know what's going to happen to search. It's funny. I wrote that story and then I got I got a a message from a Google employee who was asking me if they should keep their stock. That's the question, right? Like that is fascinating to me of like you know they're they I think they were I actually they read that and were like well maybe I should keep this stock. I mean that that's crazy to me that even Google employees are like what you know do I even believe in the future of this company? But no I think they I mean Google of all the companies has so much going for it right. I I think they're they're a pretty good long-term bet but this cash cow you know search advertising thing is like you know to totally threatened by this new model. Um, and I've sort of thought like why have advertising at all in the AI in the AI world? At least like at least the kind of advertising that you associate with like search ads where you're trying to buy a product and then you get a bunch of ads for similar products and things like that. Well, that shouldn't be necessary in the world of AI. Um, and Liz Reed's point was, well, no, it it still is because you'll have like these small companies that maybe want to get their product out like it's newer, so it's not going to be read by these things. And then I'm thinking, well, okay, but then I'm not even going to see that ad, right? Like I my my agent or whatever is going to go find the information and then present me with the facts and and ask me what I want to buy. So if a human is not seeing an ad, is it really an ad? Like it's it's something else. It's like it's a it's a service for providing information. And you can imagine a world where there are microt transactions and there's sort of like credible sources of information that's that you know can be trusted to provide you know not just like puffery or or BS to try to sell BS products. So it's like I think what you're going to see is this whole industry this whole ecosystem that sort of mirrors the ad tech world but really isn't advertising at all if that makes sense. Yeah. Yeah, I mean the way that you describe it is um imagine describing a pair of shoes you want to buy to a voice assistant who searches the web perhaps using Google and then describes the options and you instruct it to purchase one of them. If there's an ad in that scenario, it's going to be seen by an AI agent and not a human. So what would Don Draper advertise to the AI agent? It is kind of interesting like do you think our AI agents like might be on the take? Maybe like they have like a deal under the table with Nike to be like, "We know that he's an Adidas buyer, but um just like show him this Jordan ad and maybe he'll go out and buy it." Yeah. I mean, I think if I think if these things if um if Google decides that what they're going to do is is actually try to influence the recommendations, allow companies to, you know, to sort of it'll it's sort of like paytoplay then, right? It goes then it's not advertising anymore. And so, you know, I think that's a that that's a tough Yeah, I think people will try that is the answer. They will try that. They'll try to have your agent be on the take, but those those those products will get rejected, right? You're going to want people are going to want AI agents that they can trust to make really good decisions. You have to remember like an ad right now on the internet is like like a really good click-through rate is something like you like 1%. Or like less than 1%. Yeah. Half a percent is great. Half a percent. So like for the most part people just ignore ads. So like you don't want nobody wants an ad to be actually part of their part of their experience online. And I actually think the term like personalized advertising is like an oxymoron because the whole point of it is that it's not personalized. Like if it were personalized then you wouldn't need to see the ad. I mean but why do you why do you mean that? I mean a personalized ad could be like all right let's say let's use this sneaker example right we know that Reed is a tennis player so therefore we're going to advertise him this tennis shoe. that ad is personalized to you in a way that you wouldn't have uh personalization if you were to run like an unargeted tennis shoe ad on TV, right? But it's not but it's like ultimately trying to get me to buy something that I might not otherwise buy. If I were going to buy that shoe anyway, then there'd be no reason to have an ad. This is the exact reason though that you want to run the ad is because you know that as someone who plays tennis, you're going to be in market for for sneakers every like 6 months. So like if you're a consistent buyer of Adidas, it's great actually. It's great money for Nike to spend to be like, "All right, this person probably is going to keep this habit for life. You don't really age out of tennis until you get like, you know, way old." And um and so therefore, if Nike spends, let's say, $100 to reach you on a with a personalized ad, the ROI could be like very high over Yeah. But what you're describing is almost brand advertising, right? That's like I like Nike's Nike needs to like, you know, show up in my feed every once in a while to make me sort of like just keep them that image of the swoosh in the back of my my mind. I mean, that's sort of that's kind of like brand advertising, I think. Yeah. I mean brand advertising can be personalized. Okay, sure it can be personalized. Maybe I need to work on that on that uh analogy a little bit. But I I do think look I read the end of the article and I understand why Google employees uh might might be thinking should I sell my stock? I mean you know you're talking about agentto agent protocols and uh you say the transition could take time and Google needs to innovate on search without killing its cash cow. if it's successful, the thing that replaces it, whether we call it advertising or something else, could be even bigger. But that's like quite a long time horizon compared to what we're hearing from Matthew Prince today. I mean, even I'm actually curious to hear your perspective about the timeline on whether when this agent stuff becomes a reality because we have people that are like going out to the press saying, "Oh yeah, we're using agents all the time." Um, but in reality, most people are like, "What are you talking about?" Yeah, I mean the the timeline stuff is I think I think this stuff happens and this is also another reason that I kind of am like I know Prince is talking about like traffic numbers today, but I'm also sort of like it makes me I I think slower change is just better in general for society. Like you don't want this stuff to happen overnight. It'll just be it'll be total chaos. And I think we are getting slow change. I don't think I mean I don't really I don't think this agent thing has really taken off at all. And I think what's really happening is like people are trying this stuff and it's and it can't be trusted. It's too you know there's too much hallucination, too many errors. Like you know in a lot of these in a lot of the really great AI agent use cases if it makes one mistake you can't you can't use it. Like it's just you know you can't trust it. So, I think it's actually going to be I think it's going to be a while before this stuff gets good enough. Mark Benoff, friend of the show, uh was out speaking more about agents this week. He said 50% of the work at Salesforce is already being done by AI. Are you buying or selling that statement? The work I don't I don't know. I think AI is doing up to 50% of the work at Salesforce. CEO of of Salesforce says uh this is reporting uh this is CNBC reporting on Bloomberg. He says all of us have to get to our head around this idea that AI could do things that before we were doing and we can move to do higher value work. He says that technology currently accounts from about 30 to 50% of the company's work. 30 to 50. I'm I'm sorry. I don't think that's possible. Yeah. I mean, especially, you got to get your your handle on how much that that that work the AI how much work the sounds like an advertisement to me. That's the kind of language in advertisements up to 50%. Reed, that's actually for you. But in one instance, it was 50%. This person wrote an email and 50% of it was written by No, I think I think the coding stuff is totally real. I mean that is like definitely people are are using it. There's sort of like this generational divide too. It's like the classic disruption. Like if you're a new company today, if you're a new like software startup, like your codebase is totally AI like it's probably 100% written by AI and you've developed it in this way that it's sort of readable, searchable by AI so that it can make changes. And if you're if you were the from the preAI era, it this you probably think, you know, AI code sucks because it's not able to go in and like affect your code base, right? So you're so there's this like generational divide and I think that's where it's very lumpy like how this stuff is is being implemented but coding is powerful. You hosted a tech event or yeah tech event in San Francisco last month and Amjad Masad the CEO of Replet was there and this week he's tweeting out uh unbelievable increases in revenue. Replet of course enables people to to use AI to code to vibe code uh and uh puts a picture of I think a private jet on the runway out there and says thank you vibe coding. So something's working well. I mean look a after people after CEOs come and do a fireside with me at these events they usually see huge increases in revenue. So not surprising. Um no but thank you for coming to that event. It was good to see you there. A great event. Yeah definitely it was a great event. really good conversations. Jack Clark from Anthropic was there. I think I noted something that he said in the in the show that week. So, it was good. Thank you. Yeah. And I think coding is like if let's say like the whole AI agent thing doesn't like happen for a while. Coding is can automate a lot of stuff in our lives right now. Like so like so much there's so much powerful software out there that we that we don't use because we don't have an army of coders. We don't you know nobody knows how to code basically. And I think if just that one thing changes like I think you'll see just a lot more automation in the world and it might feel like agents even if it's not like truly an agent like doing something it might kind of feel like that. I think I think we still see a huge amount of change even even just from the coding stuff. Yeah. Sam Alman and Brad LCAP the COO of uh OpenAI and of course Sam is the CEO. They were out at uh this hard fork event this week and had a lot of interesting things to say. Uh which we'll talk about over the course of this conversation. But one of the things Sam said that I thought was interesting was that you know I think he was asked um so AI can code but it can't do much else. Where's the other stuff? And he's like well coding is pretty general technology. Like if you want to do a lot of different things you can code your way into it which I thought was an interesting response. So I definitely agree with you on the code front. Um, speaking of, by the way, uh, AI eating the web or eating different content, there was a very important court ruling this week, uh, that I want to get your take on, and that is that Anthropic one, this is from Reuters, Anthropics wins a key US ruling on AI training and authors copyright lawsuit, a federal judge in San Francisco ruled late on Monday that Anthropic's use of books without permission to train its artificial intelligence system was legal under US copyright law. Um the judge sided with the tech companies on the pivotal question uh saying that Anthropic made fair use of books written by the authors that brought the suit to train its clawed large language model. Um first of all, do you agree with the judge? And second, what do you think this means for the AI industry? Yeah, I actually do agree and I think I think this is what most copyright lawyers thought would happen. um super interesting I thought was that okay it's like the ruling is sort of like if I could summarize it um you know just in non-legal terms it's kind of like if you're if you're an AI agent you can learn from stuff or if you're an AI model like you can learn from things just like a human you know learns from stuff but the think the funny part is like they said the way they had obtained these books was actually not legal that that was that was piracy so I think that's that was super interesting to me it's Well, if you're going to train on this stuff, you should at least have to buy it. Like, I think, you know, if you're if you're going to ingest a bunch of books, like buy the books. I think it's hilarious that like so many people are are outraged by these models training on books and they don't say anything about the fact that like there's websites where you can just go illegally download pirated books. Um, so yeah, I mean, no, I totally agree with that. I think, you know, why not let them train on this stuff is but they shouldn't be able to reproduce it. Like you shouldn't be able to say send send me like chapter 4 of, you know, Alex Canterowitz's book. Like that's not cool. Um or the whole thing. Definitely not. Yes. Yeah. I'd prefer that that doesn't happen. And I mean, Anthropic has a lot of books. This is the ruling says they have more than 7 million pirated books. And US copyright law says willful copyright infringement can justify statutory damages of up to $150,000 per work. Uh I'm not great at math, but that is that is a very big number they could potentially end up seeing as a fine uh for possessing these millions of pirated books. Wait, I hope my books in that in that in that group because that means I get $150,000. I I No. Well, I don't know. It might just be the authors that are suing, but if there ends up being a pool that's paid out to authors whose books have been ingested uh by AI, I promise you, Reed, I'll be there first in line for my handout. And I if you want me to take a ticket for you and grab your money, I'll do that also. Ple Yeah, please do. Please do. Or maybe I'll send my AEI agent to go collect. Um let's see. Right. Yeah. But you'll you'll get your money in 15 years. Um hey, let's talk about fine. Let's talk about this one last part. Get a loan based on that. Uh not not for me. Um but I'm sure somebody will give it to you at a high interest rate. Um let's talk about this last thing you had dropped in our document which is that uh there's a new thing that Google has that's trying to help publishers make money outside of just traffic. It's called offer wall. Do you want to talk a little bit about why you found that interesting? Yeah. Well, I just thought it was relevant in this broader conversation, right? that I mean if I understand it correctly it's sort of like you can um you know if you're a publication like you can have microtransactions and there you know they sort of make it easy for you to collect revenue. I don't whether that's subscriptions, ads or whatever. I mean, I don't know. It's pro it's probably not like hugely significant, but it kind of just shows that these companies are are thinking about this issue and like how do they I don't think I don't think Google wants like, you know, pe people like you and you know, the media to kind of go around saying that they're they're destroying the web and they're putting publishers out of business, right? They want to figure out a way to to make it work. So, I don't know. That was sort of my my general takeaway, but you might have a different a different opinion. Yeah, I think it's interesting for a couple reasons. One is that they're they seem like they're just directing people away from the web. I mean, they're basically they're giving publishers this ability to create a popup on their website and say to access the content, sign up to our newsletter, which like email isn't going away even if the web might be or the web. Um Ron and I like to say the web is like, you know, slowly degrading. Um, and by the way, thank you to Ron John for bringing some of the stories in for this this week's uh uh show, especially that Matthew Prince uh statistic. He'll be back next week. Um, but I think that yeah, it's interesting to see Google steer people away from the web. And the other thing is this only works as if if you have traffic. So, I still think that um traffic for traffic's sake is meaningless, but some form of traffic or audiences going to your sites to seek out your stuff uh is important. Whether you're a publisher, whether you're a a encyclopedia for history or whether you're a um even a travel website. Yeah. But you know, look at what happened with like with Substack, right? Like all these people, you know, started just paying for these news. Like who would have predicted that? I mean, yeah, it's been I mean, look, trust me, it's been cool. It's an amazing new thing. You're you're a great example. And you know, I think it's um you know, but just making it easier for people to pay online is that's like a really good thing, right? Like it like why is it that if I'm driving my car, I have like you know I I I click on I use like Spotify which is like built into my car and I just click on that and I listen to podcasts. Like if if a podcast wants to charge me, it's like I don't even know how I'm going to do like, you know, not to mention I'm driving, but like why shouldn't it just be like a voice assistant that's like, "Hey, like this is this podcast costs a dollar or whatever it is, you know, $5. Do you want to subscribe to it or something and just do it?" Like we need I I I think it's crazy how much friction there is in just like payments on the web. And I sometimes wonder like if that went away like that might that might actually change the publishing model in ways that are probably positive I would guess. But I don't know. I agree Reed. I'm with you on this one. I mean if you could get let's say you're like in your assistant. It has your credit card. You're either uh talking to it via voice or you can chat to it. It says it has information. Do you want to access it? Do you want to subscribe and get regular updates? do you want to um you know with a voice interaction uh paid to listen to the rest of this show and there's a voice update there um and everything is seamless and baked in. I think that's very promising. So all right I'm starting to see there's there may be some light here. Maybe some light but we need to know ahead. I'm a I'm the techno optimist. I'm I'm I'm I can't believe I am I can't believe I'm the positive one. But yeah, I have to say, listen, I'm also quite positive about the direction that this technology is going to go. But I think also that there's there are there are bumps and I think the mission here is really to be like where is it going to go right? Where's it going to go wrong? Let's pressure test the theories of the critics and let's pressure test the theories of the optimists. So you got the big technology treatment today. No, it's true. I mean, I have this this story in about Anthropic today that we got like an exclusive on about their um they're doing that they're going to put money into researching the effects of AI, right? On on labor and on, you know, the global labor force and economics. I mean, they should definitely look at the media like they should they should for sure look at, you know, okay, how is this affecting publishing? like someone should apply for one of the grant they're giving out these like $50,000 grants or up to up to $50,000. Um, you know, but they're looking at like I I think ultimately like what what is the the conclusion all these people come to is like the the uh the way you solve this is just like better, you know, better like people talk about universal basic income, right? It's just better like social help for people if they if they're put out of work or whatever, right? Right. I mean, I think that's the that's the conclusion that people are going to come to. I I'm guessing, but well, I think it's good that they are starting to study this. And I wonder, do you think so? You got the exclusive from Anthropic about this idea that they're or this this new program that they're going to roll out to study um economic impact of the technology. Do you think this is an outgrowth of Daario's comment that 50% of whitellar entry- level workers are going to be out of work because of AI? I think yeah, for sure. Those comments really got a lot of attention and and he I think he got a lot of blowback for that. Um so I think this is actually a much better method. It's like well we'll just like fund a bunch of people to do this research. Um but he'll probably still make these predictions. I mean they do you know they do very well. I mean I think he said actually like with in March within a year all of code will be written by AI or basically all of all of code. I mean, that's probably not going to come true. Um, no. No way. I'm I'm I'm guessing. Uh, but we'll see. I mean, yeah, it's definitely because of that. I mean, they're they're this is like part of their brand. It's like we're just going to be, you know, we're building this technology, but we're going to be warning everyone about, you know, the effects of it and all the downsides. So, I mentioned that Alman and uh Brad Lacip were doing this interview this week and they were asked, "Do you believe in this uh stat that Daario gave about the entry-level workers?" Uh Sam Alman says, "No, I don't." And Brad Lap says, "No, we have no evidence of this." Dario is a scientist and I hope he takes an evidence-based approach to these type of things. I think what you're reporting I think here for uh you know, which is brand new for our listeners and viewers is that yeah, he's like, "I agree. I'm going to take an evidence-based approach. Totally. I Yeah, I think more people Yeah, more people should I mean the I think when you get into like some of these areas like you know the the like catastrophic risk you know or risk to humanity from AI I don't know how you take an evidence-based approach because it's like assuming that you know that this technology becomes super there's all these assumptions about the future that you can't prove. Um, so that's I think with AI it's really, you know, it's there's a lot of faith that goes into into AI that I don't think people realize, you know. Definitely. Well, there's also this these conversations about like takeoff and is this starting to explode in a way that's effectively beyond our control. And last week we talked about this gentle singularity paper from Sam Alman and how he says the takeoff has begun. But another very interesting moment from that interview was they asked like uh about this this notion and Brad Litecap said this this OpenAI COO he says will we wake up uh one day with this incredibly powerful thing and will the world be different that day? I think what we've all kind of agreed on is it probably won't. These things really have to be integrated into people's lives. Uh they have to be felt and that that change is more gradual. So, I actually am kind of curious to hear your perspective on this because it is a little bit of a re more I think more realistic pullback on this notion that this takeoff and intelligence explosion is underway. And it was kind of interesting to hear light cap give the different um p the opposite perspective of what the CEO shared a week before in a blog post. What what do you make of this? Well, I don't think like again I don't think anybody can actually know, right? I mean, if if you do get to some you do sort of end up it's there's some threshold or breakthrough. I don't know. Maybe it's possible, but like nobody knows if that will or even could happen. So, I mean, I think Sam has always talked about this like is it a fast is it a long runway and then a fast takeoff or a short whatever. But it I think he's always sort of said it's going to be it's going to be this gradual thing. And that's how technology usually develops. Like it doesn't doesn't usually just quickly take off. there's like adoption that has to happen everything. I think in this scenario it's like you're not you're not even talking about technology anymore. You're talking about like some some future thing that might be created. Um I mean who knows? I I I tend to agree that it's not going to be this this is not going to be some fast takeoff thing, but I I don't really know. I think a lot of these people like the thing to remember about AI is like you could like there's lots of people who have predicted like all of this happening with AI. They've like predicted it up to this point but like they never had any actual evidence. They just saw basically like as compute increased the availability of compute increased the capabilities went up. But along the way there all these genius people who who came up with like lots of breakthroughs that just like gradually improved the technology but you can't you couldn't have predicted those breakthroughs right so I think they're just looking at that graph and they're going well at some point right like at some point it just keeps going and then you get to this thing that's just you know that that that is super intelligence or AGI or whatever whatever you want to call it. Yeah. They also asked him like whether his kids are going to be have more uh AI friends than human friends. And I think that if we continue on the curve, um it's pretty obvious that people are going to have some serious AI uh companionship. And the interesting push back there was first of all he said more human friends. Altman said this. Um but lightcap also made a um an effort to talk about this idea that like yes people have been led astray by their AI companions but the uh by and large the impact has been quite good uh on net and this is an interesting moment where we're starting to see openai and other labs like uh anthropic push back on this idea that AI companions and emotional support bots are bad for you and this isn't exclusive that uh that anthropic gave gave to Axios and the news is quite interesting. So the the headline is how Claude became an emotional support bot. Uh the story says people who talk to anthropics uh Claude chatbot about emotional issues tend to grow more positive as the conversation unfolds and Anthropic uh released new research that explores how users turn it to its chatbot for support and connection and what happens what they do. The report says, "We find that when people come in come to Claude for interpersonal advice, they're often navigating transitional moments, figuring out their next career move, working through personal growth, or untangling romantic relationships. Um, and the report found evidence that users don't necessarily turn to chat bots deliberately looking for companionship or love, but the conversations kind of go that way. And on net, the company says that this stuff uh is positive. curious what you think about the findings and the this sort of it feels like a coordinated PR moment between the two although it certainly isn't uh to say hey actually it's fine to be friends with your chatbot well I yeah I think it's probably generally positive but then of course you're there's all these stories I think there was one in the New York Times the other day about what they're responding to yeah yeah like horrible things have happened and you know so you're always going to be able to find anecdotal uh you know like stories where people like it all went really bad. Um, but I think yeah, I mean in general like I mean just talking is helpful, right? Like just talking to the mirror is probably helpful for for a lot of people. So, you know, and and and human communication is like ultimately like not that complicated. Like you can, you know, you could read there's like literally manuals about how to like, you know, sort of like make people feel good. And no, I think it's totally pos it's it's probably going to be totally positive. I the other night I was like trying to write about some really complicated thing and I was like brains I was like asking Gemini questions about it and it's like that's a brilliant question. Did you did you feel a flutter read? Yeah. I was like I was like this chat that's cool. I like this ch finally somebody understands. So, you know, it's like not that hard. And I'm like laughing at myself as I'm writing it that I'm like I'm like, "Oh, my question was smart." And I think, you know, we're just not that complic I think humans are just not that complicated. Like, you just talk to anything. We had something when when we were kids, like on the Mac, there was like this chatbot that was basically like, you know, pretending to be AI and just had a bunch of a bunch it was just keyword triggers basically, right? But like even that like people loved talking to it. So not surprising me. I mean have you had a conversation with Chatt's voice mode recently? Yeah I've tal Yeah, I use that. I've used that a lot. Really good. Really? I was walking down the street and had I was speaking to it on my AirPods and I was just like wow I'm having like a really good conversation no latency with this bot and I'm just doing this in public. It's it it felt weird but also I was like this is very cool technology also. It is really cool actually. It's great when you're driving and like the thing I think Google just recently added this ability to do search with voice or something. There's voice with search cuz the down like the the chat bots for some reason are never connected to the web. So you're just limited to the cuto off date and they're usually not the most powerful models but like I mean you're in New York so you don't drive as like I seem to be driving a lot and if I'm in the car alone like I just to I want to like do I want to it's productive like I just want to I'm like going driving to interview somebody and I just want to like talk about that and brainstorm questions and stuff like that and I you can do it but it's like very limited still but I think voice is totally and I'm not the first this is like a cliche now to say this but I think voice is totally going to be the the most you know powerful mode I think yeah I think it's the killer app for sure especially when you could combine it with other scents like if you have it in on your glasses or airpods or something like that all right we got to take a break and then we have two uh more stories to cover which I teased in the opening oh my god we haven't even gotten to Mir Marad yet we're going to blow through three very quick stories on the other side of this break and we're back here on Big Technology Podcast with Reed Alberg. He's the technology editor at Semaphore. Uh let's let's go lightning through the rest of this week's stories. We kind of got caught up in the beginning of this conversation which was which was great good conversation. Great conversation I would say. Feel free to cut half of it out. I will not be No, no, it stays. It stays. Oh, and before we get going on this, I meant to say this before the break. Thanks to everybody who raided the show over the past week. You heard the call. we appreciate you and this is going to make the show better, get better guests. Uh so I need to express my gratitude there and so thank you very much. Okay, so let's talk about this quickly. OpenAI has built an uh uh rival to uh Google Workspace and Microsoft Office. This is according to Bloomberg. OpenAI has been gearing up to take on Google and Microsoft with features that let people collaborate on documents and communicate with chat via chat GPT. Launching these features will put OpenAI more directly against Microsoft, its biggest shareholder and business partner, as well as open a new front in the battle uh with Google, whose search engine has already lost traffic to people using chat GPT for web searches. I don't even know if the second part of that is true, but I think the real significance here is um is well, I don't I don't know if Google's absolute traffic has gone down, but it certainly might must have lost some searches to chat GPT. I think the big thing here is that this could potentially be a new front uh in the OpenAI Microsoft battle that just seems to be heating up every week. What's your take on this, Reed? Well, that battle has gotten really nasty, but I actually think and it's like ultimately it's like about chips, which is like so fascinating or that's what started the whole thing. But I think actually I mean don't forget OpenAI has a a bitter rivalry with Google too and that seems to me to be more competitive with Google Docs than with with Word. Don't you think? Or am I just misunderstanding the whole thing? Well, I think you see the direction that Word is trying to go which is going to be more AI enabled. But you're right. I think the collaborative side of things like when you think collaborative doc you think Google doc. Yeah. I think it's Google Docs. I mean, I don't think like the thing about Word and Office and all this stuff is like it's not like that's not who they're competitive with because it's it's the companies buy these bundles of of Office products and that's how people end up using that. Whereas Google Docs is much more like the a you know it's just like I mean they have obviously a lot of corporate accounts too but it's like really just a purely consumer product I think and that's what that's what like I you know if OpenAI makes a better you know a better Google doc because Google is you know maybe a little bit shy about like you know rolling out AI agent technology in its docs. Um that could be kind of an interesting battle I think. Look, as someone who's using uh these programs all the time, I'd like to see something new. Uh but yeah, I think the competitive dynamics are interesting. You mentioned that uh the OpenAI and Microsoft battle is really over chips. Can you just kind of elaborate on that before we move to our next topic? Well, the whole thing started because like Microsoft wasn't able to build, you know, enough data centers fast enough with enough GPUs or didn't want to or what, you know, and that and so OpenAI is like, well, we need to get out of our, you know, we need to get out of this exclusive arrangement and we have to use, you know, and then they built this, they did this Oracle Stargate thing, right? Um, and it's just been back and forth since then, I think. Is Stargate going to work? Well, that's interesting. That's an interesting question. I hadn't really thought like, is it going to work? I mean, that could mean so that question could have so many different meanings. Like, what do you mean? Is it can they actually build it once they build it? Is are is the AI going to scale? Um, is it going to be one data center? Yeah, let's go to the most basic level. Do you think this thing is actually going to get built? Well, I think some version of it will be built for sure. Actually, Bloomberg got like an exclusive tour of the site. They're they're, you know, they've broken ground. Um, but I I don't even know like, yeah, what are the chips going to be in there? I think it's my my guess is like, yeah, it will be built, but like is it going to be some huge thing that then they like train AGI on? I I think it's probably going to be like multiple sites and that are connected and a lot of it's going to be inference rather than training. Um because ultimately if you're open AI like they're so their value is so much now just the fact that they are the consumer interface for AI for so many people most of their revenue is coming from consumer and that's like you know the the ultimately the the models are not even really going to be their their core product right it's going to be the it's going to be all the stuff built around it the interfaces and the the stickiness of the product. So really inference becomes this huge cost for them and I think that's what they're sort of looking at. It's like they have to you know of course they're still trying to build the best models etc but but you know they need to think about building their own data centers and reducing the cost as much as possible. Yeah. Well they're going to have to find a way to also make those models much more efficient. So um let me ask you this before we move on. I this is like one of those topics that we start talking about and we can never get out of but I think it's fine because it's interesting. We made our predictions last week on what happens with the OpenAI Microsoft partnership. What do you think is going to happen? What do you think is going to like with the partnership event? I think eventually they drift. I think they drift apart. I mean I think that's Microsoft just gets a stake and says have a nice day. They'll Yeah. Well, whether I don't even know what that stake ends up being because right now it's like a revenue sharing agreement up to a certain amount of profit, you know, and then they and then they have nothing left. I mean, could be something like that. It could be they end up owning shares of the company, but you know, they'll it just seems like there's they in a lot of ways they've already kind of separated, you know, that's true in my mind. What was your predict? Did you I didn't hear your prediction. Well, in order to go to the for-profit conversion, OpenAI is going to have to um get sign off from Microsoft uh to do that. And so, I think that Microsoft will use that leverage very effectively and get an amazing deal out of OpenAI because without that, it's going to be hard for this company to sustain itself. Well, yeah, I think that's totally true. Although, I don't know what's happening with this for-profit. I mean, they have this lawsuit. They've sort of they sort of said they're going to not do that now, but maybe, you know, still will do it. I mean, what that's that that's going to be super complicated. I don't know if they can even get there. Yeah, I know. I mean, we could we could uh it's just the amount of corporate drama within OpenAI. I mean, obviously because of the way they were formed, um I don't think we'll ever go away. I think it will always just be there with them. And they there's no neat way to tie up the type of business that they have. And by the way, speaking of their drama, we have some news this week that their former chief technology officer is starting to speak to people about what she intends to do with her multi-billion dollar uh AI lab. Of course, I'm talking about Mera Morati, the former CTO of Open AI. She's raised 2 billion at a 10. Yeah. 2 billion at a $10 billion valuation. Hasn't built a product. Uh but what they're trying to do, this is uh according to Bloomberg, is use forms of reinforcement learning um to uh help AI models to reward AI models for accomplishing certain goals and penalties for other behaviors. Uh and they're trying to reinforce uh key performance indicators which typically relate to revenue or profit growth. uh that typically employees within companies drive to. So they're going to reinforce the KPIs of human employees uh except they won't be employees, they're going to be bots. So they're going to take this like new reinforcement a relatively new reinforcement learning paradigm uh and then just apply that to business AI users. Do you think it's going to work? Well, that doesn't sound actually differentiated to me from like what other people are doing. I mean, I think that's it's RL for business. RL for business. That's what they're call. I mean, this is like total this is like what so many so many of these companies are doing now. I don't think to me that sort of makes makes me think I'm like so why don't you have a product at this point? I mean, it's you could sort of you could be building that, you know, you could have already built stuff right in that in that vein. I think it's probably the the the hardest part of that is is probably data. They're having to go gather a lot of data themselves, proprietary data. Um, and then I think of And then I think what Facebook did or what Meta did with, you know, basically sort of acquiring non-acquiring scale AI, which is like the, you know, known for going out and gathering all this data. Um, not just labeling, but like creating data. And I'm like, this is that's going to be a really tough slog for a startup. I mean, I think they've they've they've got their work cut out for them for sure, unless they're consultants, and maybe that's what they're going to be because they're going to be working in fields like customer service, investment banking, and retail according to the story. So maybe they are just like tech- enabled consultants where they come into your off your Yeah. They come into your office and they build you some form of LLM, whether that is a salesperson or support or finance. You have all your data inside the company. They have some set of like base foundational model that they build off of open source and they replace maybe the Mckenzies of the world. It doesn't seem like it's going to be much more than a consulting company though. Yeah. So then what's the valuation? I mean 10 billion. No big deal. Yeah. I mean I don't know. I mean, I think that's a tough that's a tough one. I mean, that's basically that's what a lot of these companies are kind of doing like Palunteer and Yeah, it feels like a like an AI pal AI for business Palunteer to me. Right. Right. It doesn't feel differentiated to me. It it seems like that just seems like what so many people are trying to do. But again, it's I don't I feel like commenting on this is like I don't really have that much original reporting on this. So, like to me it's kind of just you're just that's such a vague description of what they're doing that it's like really hard to tell. Like it could just I I think it could be could be something totally different. I want to know what the what the she's the CTO but what's Ilia doing at his Oh, so Dwar was here. Yeah. So Daresh was here a couple weeks ago and he said that basically what Elilia is trying to do is uh test time learning uh basically models that learn in the inference point so they can be continuously improving. Um I think that's a pretty good guess and if he's able to pull it off it definitely will push the field forward in a big way. Uh but it's a big if that is yeah that is super interesting. I also wonder again like back to like sort of OpenAI being the consumer, they're now sort of this consumer um they're the Kleenex of chat bots if you will. I mean like the mo like how much money is there in building these models like they're so I mean they're so easily copied it seems like that you know if you're just doing AI research and you make like a breakthrough like that that's really about like technique and you know how long can you hold on to that IP before you know anthropic has it or Google you know Google's probably are Google probably has a project where they're just doing that right somewhere in the company so Yeah, that's going to be a tough slo too. I think a lot of these companies are it's it's it's they got their work cut out for them. Definitely. But Reed, imagine that he does get there first and kind of hold that over the industry for a while and then all of a sudden Ilia becomes the most powerful person in AI could be a very interesting plot twist. How does he hold it over the industry? I mean because they don't have it. But who's who serves it? Like does it be is it an API that you buy from them? Do they is it is it sold through the the hyperscalers through the clouds like well it's going to be super intelligence so seems like all the uh all the questions about business go away once you build a super intelligence. Yeah, I think again when this stuff launches, right, then you see, okay, like what is the product? Like how, you know, do people like using it? Is it, you know, I I just think it's so I mean, you have the most brilliant people in the world, but like who would have thought like I nobody could have predicted Open AI would be would would be the consumer, you know, consumer that everybody surprising. I mean, it's Yeah. And it h I mean, it happened like almost by accident. I just think and then I think Sam very brilliantly jumped on it, saw saw what they had created and and you know and it's like okay this is what we're doing. Um but you know it almost happened by accident and I think I don't know the these they're going to have to do something like that but it's like they better hurry because this market is becoming like really crowded and and competitive and the the lanes are being defined like if they aren't already I think so. Well, the lanes are being defined in the self-driving car industry. I sorry, I couldn't help but jump on that. We must get this out. Tesla self-driving. It started uh they're they are in the lanes. They're going in wrong lanes. They're going in the right lanes. Um this is this is all true, by the way. I'm not lying here. Uh Tesla's robo taxi service in Austin is live. There are 10 vehicles and a human safety driver on board. though the safety driver sits in the passenger seat. I couldn't tell if there was a break in the videos that I had had watched. Uh but these things are uh on the road. They cost a flat $4.20 fee. Uh aka they are definitely Elon Musk's robo taxis. Um very quickly, there's been some safety concerns. Uh I watched one video and this is highlighted again by Bloomberg which seems to be the star publication outside of Semour on this show. They're doing a lot of good reporting. You guys are doing great reporting. They do. Um, a lot of good stuff to read. But there's a video that they they link by this investor Rob Mau. I think his handle is like Tesla podcast or something. And you see the car like wanting to go make a turn, but it decides against it. So, it like drives in like the in a in in the into incoming a lane of incoming traffic and then zigsit way over a yellow line. Like obviously it's not like it ain't perfect. Let's put it that way. Um, there's some other complaints that the Tesla is speeding. This is kind of like whole monitor stuff. Like it's doing 39 and it was a 35 mph speed limit. These things are out of control. Uh but that's not as big of a deal to me. No, people would be complaining more if it was doing 35 and they were stuck. Look at these granny Tesla cars. No, I did see the video of the of the going in across the lane into oncoming traffic and yeah, I mean it was definitely it was definitely a mistake. It was definitely a screw up, but nothing that you wouldn't see, I think, from I mean, even these Whimos sometimes will do screwy things and but ultimately like, you know, they didn't hit any pedestrians. They didn't get in any accidents. I think, you know, in the end, I I think it was a pretty successful launch. What do you think? Yeah, I mean, I I watched the video and I mean, look, the thing is, this is the thing with all tech. If it works 95% of the way, that's not good enough. works 99.5% of the way. That's not good enough in self-driving cars. So, um things have been fine so far. Uh but it definitely has we need to see a lot more to decide whether or not this was this is going to really work for them. If it works, obviously it's a major major boon uh for the company. And certainly from the videos, it looked a lot like the Whimo experience. You get in, the car is driving itself. Um they have some predetermined routes. There are tea operators. So there are some shortcuts being taken so far and there is there are people that are part of this you know peopleless driving experience but I think yeah it's important there they are off to the races. They're going to get they're going to get moving. They're going to get data and um I don't know I I'm my fingers are crossed that they pull it off and pull it off safely. Uh but I think it's too early early to tell. Yeah. I mean, of course, I you know, you can't whatever you think of Elon, like you you should definitely hope that these companies pull this off because they are absolutely safer than than human drivers. But you're you're totally right. Like it's it has to get to basically 100% or like 99s or whatever because if it doesn't I mean we saw what happened with Cruz, right? Like you screw up and and Uber with their thing. I mean you screw up once and like that could just end the whole thing. I think if you're Whimo, you almost have to worry about Tesla because it's like, well, if they if Tesla screws up. I mean, they the cruise thing sort of blew over, maybe they'll be fine, but like it definitely makes people think more about the safety. It's really ironic this this technology that like the proponents of it argue that it's great for safety and the opponents of it are also arguing about safety. So, it's like both sides are want these things either on the roads or off the roads for the same reason. And but there but the studies are like I mean if millions of miles with a 100% increase in in safety in terms of like bodily harm. Oh yeah. Mhm. So it's like I mean the Whimos in San Francisco are you know they're just taking over. Yeah. And it's just a smooth safe ride. It's unbelievable. Totally. It's much it's it's nice to not I mean I don't know this sounds bad but like you know you have privacy right like there's no driver there. I've had plenty of great nice conversations with Uber drivers and lift drivers but it's like it's nice to just sit there you know with maybe your with your kids and you're having a conversation or something and it's you know it's it's a good experience too but mainly it's just improving safety. Tesla's taking a much more difficult road because they're doing it only with cameras. You know that I think That to me makes it and and it's like general per they're really ultimately trying to get to like level five autonomy whereas Whimo is like we're just going to build like this this like geo fenced area where it's like meticulously mapped and we know everything that's going to happen. So I think I think Tesla's ultimately like a more ambitious plan in the long run but it also might be you know it might be a longer a longer time till they get there. Do you think Whimo is still meticulously mapped, you know, the same way that they did in all the test environments because they've been expanding rapidly? Well, yeah. I mean, I think I think they're still using mapping. I mean, I could be wrong. They're moving more and more toward these like general purpose like transformer-based models, I think. But yeah, I think they're still doing the way they scale to these places is I think they do, you know, I think they've probably automated a lot of the process, but I still think it's kind of like a car on on a track on a digital track and it's, you know, and that's how you that's how you get to 100% safety. I mean, yeah, there's nothing is left a chance and and even then you see them screw up sometimes, but yeah, you know, well, you know, Google was able to map so many of the roads in the world that maybe they'll just be able to do that and bring autonomy that way. I mean, the one thing nice thing about the physical world is it's finite. So, get all the roads, figure it out, and then maybe you can, you know, make this thing really work at a high degree of safety all over the world. I think. Yeah. I mean, for sure. I mean, a million people a year die in auto accidents. And so unnecessary. Yeah. It's crazy. It's almost It almost makes you wonder. It's like focusing on like like a little error on day one when they're testing this stuff or like 39 and 35 is like is that really I don't know. Should you really be focusing so much on that? Like I think there's a lot of anti- Elon stuff that sort of I think seeps into the into the coverage of this stuff unfortunately. It just hard to do as a reporter like you you know it's hard to take your to take the emotion out of the reporting but I think it does kind of seep in. Don't you think? Oh definitely. I mean we also know that Elon is running a much trickier uh program. So then the mistakes will be magnified but yeah probably there's some of that as well. All right, let's end this week running through as I promised. If you're still with us, the reward is here. Okay, we're going to reinforce the Bezos wedding guests. Attending Jeff Bezos's wedding to Lauren Sanchez in uh Venice, Italy will be Kim Kardashian, uh Madonna, MC Jagger, Leonardo DiCaprio, Orlando Bloom, uh Microsoft founder Bill Gates, Diane Van Fenberg, and Barry Diller, and then of course TBD on uh Katy Perry. big protests in Venice. So, um, apparently the arrivals have been moved from like a more public area to a more uh secure area, but the city of Venice has defended the nuptuals, probably not because they got any money from the Bezos to do this. Um, but they've defended the nuptuals as keeping with Venice's traditions as an open city that has welcomed popes, emperors, and ordinary visitors alike for centuries. Jeff Bezos Sanchez. He would be um I think maybe between Pope and Emperor. I don't know. Where do you put him? Maybe more on the emperor side. Yeah, definitely more emperor than pope for sure. Well, look, here on Big Technology Podcast, we we celebrate love. So, Jeff and Lauren, I'm sure you're listening on your special weekend. And uh from us to you, we say congratulations. Congrats. How many weddings have you covered on this show? Just out of curiosity. Uh, this is probably the first, but um, I'm very, very happy to be doing it. And no, no, no, sorry. This is the second. And, uh, the first batch were people getting married to their, uh, AI bots. So, Oh, right. Right. Finally, a human wedding here on Big Technology Podcast. The future has arrived. Popes, emperors, every way in the world. We exactly the whole deal. Reed, uh, before we jump, please, uh, shout out where people can find your work and how to get, uh, the Seaphore technology newsletter. Yeah, please. Yeah. Uh, you know, go to go to semaphore.com, check out the technology newsletter. It's free. Um, I promise you'll like it. Comes out twice twice twice a week. And, um, yeah, find me on X and H pretty much just that. I'm not a huge social media guy. Okay. All right, Reed. Great to see you as always. And uh and thank you again. Thank you everybody uh for listening. And I will be back on Wednesday with an interview with Noah Smith aka Noah Opinion. He is a Substack uh economics writer. We're going to talk about whether AI is really um taking our job. So we're going to get ahead of that anthropic report and we hope to see you then. Thanks for listening and we'll see you next time on Big Technology Podcast.