May 6, 2023 ยท 2023 #16. Read the transcript grouped by speaker, inspect word-level timecodes, and optionally turn subtitles on for direct video playback
Edit labels for this show, save them in this browser, or download a JSON override for the production folder.
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker
I've got sunshine on a cloudy day When it's cold outside, I've got the month of May. Everybody say, I guess you'd say, What can make me feel this way? It's my girl, my girl, I'm talking about my girl, my girl, I've got sunshine on a cloudy Hello everybody, it is mid-afternoon on Friday, May the 5th, 2023, new month. But an old show, an old theme. We're back with That Was The Week with my old friend Keith Tyr, summarising the week in tech. Hey Andrew. And it's been a very familiar week. Keith's newsletter this week, which goes out live by the way now with the show, so this will be the first time anyone's seen it. Robots and Humans. We've had this one before, Keith. It's been showing many times and it will continue to show many times. What news about Robots and Humans got made this week? A lot happened this week, Andrew, in robot land. The first thing is the White House invited all the CEOs involved in the space to come to a conversation, kind of headlined by we need to keep a close eye on this and regulate it or at least have a plan to regulate it. And so everyone showed up yesterday, including Sam Altman, who was about 10 minutes late. It was very funny watching him go in late. At least he showed up. Was he wearing a fleece? He was wearing a suit. A uniform. A very smart suit. I'm guessing that I haven't seen the photos, actually, but I'm guessing the photos, I know the CEOs of Microsoft and Google were there as well. I'm guessing it was a much less awkward photograph than the Trump one when all the tech people showed up to have to deal with him. And they weren't sure if they should be there. Exactly. And they were all dressed up. I remember even Tim Cook was in a shirt and suit looking very uncomfortable. The funny thing was Lena Kahn, as I understand it, was not there, but she retaliated by a New York Times opinion piece. I know you're a big admirer of Lena Kahn, Keith. You couldn't wait, could you? I was just about... So what's your friend Lena been up to this week? She wasn't invited. Do you think she's on the outs on the White House front? I'm not sure she's done a great job. Even I, I'm not as critical of her as you. Nobody is as critical of her as you, but I wonder why she wasn't there. Maybe they don't want to deal with her anymore. Well, when you're the chairman of the Federal Trade Commission and you're not invited to a discussion about how to regulate a trade, you turn to the New York Times and she wrote a long opinion piece called How to Regulate. We must regulate AI. Here's how. And immediately in my mind, I thought... She only wrote it to annoy you. You know that. Exactly. But I also... The funny thing is she doesn't answer the question. And she adopts what I can only think of as chat GPT like hallucinations, as in she's kind of imagining what she might do, but not really describing anything specific. It got written by chat GPT. You quote in the editorial this week, there's the hallucination. This is on the Khan Op-Ed. OpenAI is a startup barely a few years old and challenging Google, Facebook, Salesforce and others, hardly locking in the market dominance of incumbents. What is she smoking? Exactly. What is she smoking, Keith Leena Khan? I have no idea. I mean, she only has one tune, which is monopolies need to be regulated. So she has to invent a monopoly in the AI space. And the funny thing is, as the leaked document from Google shows this week, there isn't one. There's no monopoly to regulate. I mean, not only is there no monopoly, according to the Google document, there's no moat. So nobody, nobody's secure. Not Google, not OpenAI, not Microsoft, nobody. Yeah, the leaked Google document predicts that open source AI platforms will be better than those created by companies. Who leaked it, by the way? I should ask my wife. She's the head of police at Google. Yeah, I have no idea who leaked it. And what kind of document was it? Was it an official white paper or just a thought piece? It was an internal recommendation piece. So it doesn't have the status of an official Google opinion. But whoever wrote it clearly has insight and influence. So I'm sure it reflects. But in all seriousness with the Khan piece, I think you're right. She only has one. She only has one gear. And it's always about regulation and monopoly. But as you said, it's absurd. There's no way. I mean, it's not even an industry. So how can it be a monopoly? Right. The closest risk is the one that Gary Marcus flags over and over again, which is closer to the risk that agreeing to a drug being released before it's been tested, which is the impact on humanity. And Hinton, who resigned from Google this week. I'm not sure if you connect with any of the pieces about that. But that's really been the dominant story of the week when it comes to AI. Yeah. And Hinton's resignation is, no pun intended, hinted at in the leaked document. The leaked document talks about a fight between different parts of Google. And you remember last week they merged the DeepMind part of Google with the other AI part of Google into a single unit run by the DeepMind guy. Then Hinton resigns. So it doesn't take too much imagination to realize there's been a big turf war fight at Google over AI, which is the same fight that Gary Marcus is fighting. Are you saying that Hinton was being disingenuous in terms of his reasons for leaving? No, I think he's telling the truth. But I think what he's saying is we lost the fight. Therefore, I can't be here anymore because I don't believe in that path. Well, he's probably, is he in a separate camp from the DeepMind guy? Yes. So the camps are really, there's some things in common. The things they have in common are neural networks and deep learning models. Those are technologies. What they don't have in common is large language models as a path to artificial general intelligence versus deep learning and neural networks using other approaches. So what I say in the editorial is that it's a little bit like the life of Brian when everyone is a Christian, but they disagree on some small element and therefore become enemies. That's kind of what's happening here in the AI space. The Google leak document seems to confirm that. Gary Marcus on Twitter this morning was asking, can anyone think of other examples where, to paraphrase him, small sets in a common space carve each other to pieces over small differences? And I think that is roughly what's going on. I think the interesting thing about Hinton, actually, is how it's been picked up. I mean, no one had ever heard of him before. I know him because he's the brother of a friend of mine's wife. And I think the real story is that he's a big Spurs fan. So he's probably in a horrible mood anyway. And it's just an excuse to quit his job and complain even more. Well, after Manchester United's 95th minute penalty against them. It was a meaningless game. You're still going to qualify for Europe. I hope so. Can't blame Geoffrey Hinton for that. No. Are you going to blame Geoffrey Hinton or Google or Lina Khan? No, I was just going to join you in the depressed football fan part of the world. But you're claiming I shouldn't be depressed because we're going to be okay. And you're in the FA Cup final, I mean. So back to robots and humans. Has anything actually happened this week? I know you linked to an interesting speech that Greg Brockman of OpenAI made at TED. Another interesting one on AI by one I'd never actually heard of. Yejin Choi at TED as well. What are they saying? And has anything actually happened this week rather than just all talk? Yeah. Well, Greg Brockman demoed some of the next things that ChatGPT is going to be able to do. Which is basically carrier actions on your behalf. Using plugins that interface with, for example, spreadsheets or travel booking systems and the like. And it's really worth it. It's a 30-minute video of which 10 minutes is Chris Anderson asking the questions. But it is a great insight into where ChatGPT is at and what's coming next and how. So it's more than just language. Yeah. It's actually having a real assistant as opposed to the failed assistants that Apple and Google and Amazon have. Exactly right. It's not even close. Siri and Alexa and the Google Assistant are not in the same ballpark even as ChatGPT. And this next iteration that he shows off, which includes web browsing, by the way, and being able to summarize. It must have chilled Google because that's a real threat to their business. Yeah. I mean, we're doing this on Chrome. I'm guessing we won't be needed in future, Keith, will we? You and I. We can be replaced by robots. What about Yejin Choi? Who is she and what was she saying? She's a professor, Chinese background, but American professor, who basically tries to explain why ChatGPT is both super smart and super dumb at the same time. And she's really in the same camp as Jeffrey Hinton, but she's quite concerned in how she lays out the landscape and the case for ChatGPT not being the mainline path and how the reason for its flaws are structural. And I think she gives a reasonably good rendition, not the same as Gary Marcus, but within the same ballpark. He'd probably agree with a lot of what she said. And I think it helps us understand all the noise that's going on between the professionals in this space is really fighting for position almost in an ideological way, as opposed to being scientists talking about technology. They've already figured out what they think the endgame looks like and how to get there. And they want to throw as much mud as possible at everything else that looks promising. You didn't include a link to the Biden initiative. He's putting several hundred million dollars into state funded foundations for investigating AI. Is that a waste of time, Keith, in your view? Are they going to lag behind? Will there be any point to it? As long as I'm guessing they don't appoint Lina Khan to run any of them. No, even in my own company, Andrew, when I sit down and talk to the data science team, and I'm not entirely as uneducated in this as the White House is, I struggle to keep up and understand what they're saying and doing, because this is expert led science. It's particularly dangerous for the White House to do that when there's a turf war between different camps, because if they give the money to the other camp, by definition, they'll fuel the fire on that side of the discussion.
Gary Marcus tweeted to me, he can't find VCs that will fund his approach. So there's a very strong self-interest in getting money. Yeah, but Gary's startup has got to do with media and truth. I'm not even sure how rooted in AI. I mean, it's certainly not a chat GPT product, is it? No, I was more referring to his belief that there's a better path to artificial generation. Well, everyone thinks they've got the best path. I mean, whether it's Gary Marcus or Geoffrey Hinton. Yeah, but when that's still in the middle of the early part of a fight, I'm not sure you want the White House giving money. Right, and there isn't going to be a single path anyway. There's going to be multiple paths, and each path will have sub paths. Yeah, it's very important that politicians don't side with one camp in a debate that is practical and is still... Right, but I mean, when it comes to tech and politics, one of the interesting things, thinking out loud, it occurs to me, is given the, particularly the Republican Party's preoccupation with culture wars and banning certain books in schools. I wonder whether it's conceivable that AI will become an issue in the next election. I can't imagine Trump talking about it, although he'd probably be anti-Chinese. And Biden doesn't seem particularly interested one way or the other. Yeah, I think that there's no upside to any politician in getting involved at this point. They should leave it to the science and see what happens. I mean, I'd say there's quite a few years to go yet before we really even know anything. But they're going to be exciting years because the capability of these platforms is going to grow. If you believe the Google leak document, this is going to be free and available to everybody quickly from open source providers. And that open AI and Google will not be able to get significant revenue from it because of that. I'm not convinced that's true. I think open AI actually is in with a shot. Although open AI is actually, in truth, closed AI. It is, absolutely. They're not shy about that. Elon Musk insists on that, although that's only from sour grapes because he's no longer involved in it. Exactly. Yeah, this issue is going to run and run. It's certainly not the last we're going to hear of it, even this month. What else is happening, Keith? It's not just AI this week. You've got some interesting other news. AI, well, going on with the open AI, they closed the $300 million share sale at an almost $30 billion valuation. Are they a series C or D company? I don't know which series that was. I can check while we're talking. And just to say, $300 million sounds like a lot, but they lost $550 million last year. I read that he wanted to raise a billion. I guess that the open AI IPO is going to be like the Netscape or the Facebook or the Google. It's going to mark, I don't know when it will happen, but it will mark an important date in the evolution of the AI economy. Yeah, it's going to be a long time coming, I would suspect. It is a series C. Yeah, actually, thinking of AI, one of the pieces I suggested you put in, there was an interesting piece in The New Yorker by the science fiction writer Ted Chiang on AI becoming the new McKinsey. I thought that was actually pretty good. It's only going to compound the inequality and the strength of companies like McKinsey who control the AI and who own the AI. Anyone doing consulting for a body of knowledge that's easily available can't continue to charge very high prices for their time. It just won't be feasible because these systems will do it. They'll have to dress it up and justify it in their consulting gobbledygook. Yeah, exactly. I thought that was an excellent piece by Chiang. I agree. In fact, it was so good I bought one of his books of short stories. I didn't know much about him, but I think he's a very good writer as well. In terms of essays of the week, a couple of other interesting pieces I thought you linked to. One about half of YouTube's US viewership now on being on TV, changing the ad market. Does that basically mean that these two markets have converged and we can't really talk about a television market or an internet market anymore? I think the TV screen is no longer a TV in any meaningful sense. It's a window onto internet-delivered content. The smart apps that are built into all the newest TVs probably dominate what people look at because they're leaving cable and satellite networks in droves. What else could they be doing? They must be loading apps, and YouTube is probably the number one app they load. YouTube TV, by the way, is fantastic. It's a really good real-time television network that replaces any requirement for cable. Hulu's is quite good as well, by the way. You have to tell my wife, who won't get it because she claims that it doesn't cover all of Stanford sports. I think it does. I'll have to tell her because it's massive. Everyone hates their cable company. One of the pieces that you linked to that I thought was really interesting and actually touched on YouTube as well was the Stratetory. Who writes that, Keith? Ben Thompson. It was a really interesting piece on content. He touches on the collapse of BuzzFeed News. I actually had Ben Smith, who has a new book out on BuzzFeed, traffic on the show. What is Ben Thompson saying that's interesting here on this unified content business model? I guess it's advertising versus subscription.
His end point is why worry about the distinction? Why not offer both? He focuses on Substack and makes the point that no individual Substack author is going to have the gravitas to attract serious advertising. Substack, the company across all of the authors, would. Why doesn't it create an advertising bucket, if you like, that could be distributed across authors that want to have advertising and make a lot of money from it? He's being logical about it. We've talked previously about sponsorship in the same concept that Substack could be the sponsorship intermediary for these things, which is not quite the same as advertising. Of course, Substack is resistant to that because its entire vision is founded on not needing advertising. There's a zealotry around that. I wonder whether the Substack model, which does seem in a way attractive, whether it's going to hit a wall when people simply are not willing to spend money on more subscriptions. They have to be able to bundle these subscriptions. You're right, bundling is the solution. Maybe a hundred of your favorite writers on Substack for whatever it is, $20 a month. Because otherwise, at a certain point, most people are still not spending money. There was an interesting piece, it's not in your newsletter, that I got today from Andrew Sullivan, really attacking Smith and BuzzFeed and Gawker, suggesting that they were the sinners, that they fetishized. Smith kind of acknowledges this. They fetishized traffic and the viral quality of the internet, which has essentially destroyed credible information online and created monsters like BuzzFeed and Gawker. What Sullivan concludes, which is in his interest, is that it's all led to Substack, which he's on, which he's a superstar on Substack. But most of us aren't like Andrew Sullivan. We don't have that kind of following. We don't have 10,000 people willing to spend a few dollars a month to subscribe. It's not viable for most people. All it's creating is a new superstar system, a winner-take-all economy. What's being missed in all of that conversation is the value of an audience. There are things you can do that reduce the value of an audience down to an atomic unit where every person is the same. The same is not very much. It's two cents per thousand clicks or something of views. But there's another way where the value of an audience can be enhanced. For example, if you could get all Spurs fans. There are any left, Keith. It's me and Geoffrey Hinton. Let's just play along with the idea that you could get all, let's say, Premier League fans looking at one site. The value of that site to an advertiser who wants to attract Premier League fans would be between $20 and $60 per thousand views. But if you put those people and spread them across the internet and treat them just as random individuals, it's two cents per thousand views. So what's being completely missed in the ecosystem is the value of an audience that has common interests to an advertiser that wants to target those interests. Glam Media did this back in the day with a fashion website. Yeah, but that was slimy as well, Glam Media. Well, the concept underneath it was not slimy. It is that audiences are more valuable when clustered around interests. And usually that implies demographics. And the Google effect on advertising is to destroy that by atomizing everyone down to just being an individual equal to every other individual. And now that they can't track you using cookies based on which sites you follow, that atomization and therefore reduction value is even worse. Somebody sometime is going to figure out how to pull audiences into common interests where the content is excellent for that interest and the value of the audience is high, which is what cable TV did, by the way, by creating all these channels. I don't understand. I'm seeing Jรฉrรดme Lanier the weekend and he's always been a champion of this. I don't understand why there isn't a service where if I want to click on an article, whether it's on Substack or a newspaper or a blog or anything else, I'll just pay a few cents. In that way, it's just automated and I don't have to think about it. And it's like going across the bridge on a toll. It's just a toll meter. And you can have a maximum if you don't want to go over a certain amount each month. But that would be the simplest. Why hasn't that been tried? It seems so obvious. Go and look again at the News of the Week articles. There's four of them. And look at the last one. Twitter plans to enable publishers to charge per article to bypass paywalls in the app. Oh, you see, you beat me to it. So Twitter's already doing this. Musk is thinking ahead of the crowd again.
Why don't you go through Twitter for this? Why does Twitter have to put itself in the middle of it? Well, somebody has to be a horizontal media sitting on top of the New York Times, the Wall Street Journal and HBO and all the other things. But then they just become a tollkeeper. Why should I have to pay Twitter? Why can't I pay Google? No, you're not going to pay Twitter. Twitter's going to enable you to pay them. Yeah, but they'll take a cut. No, they're not planning to take a cut. So what's in it for Twitter? Audience. Twitter will get an audience. If you can find every article. So if there's an article, say, in the LA Times and I don't subscribe, if I go to the latimes.com, I won't be able to look at it. But if I access it through Twitter, they will have an arrangement with Twitter and I'll be able to access it and pay for it. Basically, you'll have a wallet on Twitter and it'll deduct a hundredth of a cent. Why not just do it on a browser? It's so much simpler. Well, if you do it on a browser, the browser has to implement some kind of billing and payment system. Well, I'm sure that our friends at Google could figure that one out. They might. They might copy it, but it's the right instinct. But they're lower down. I mean, they're higher up the chain. It's much more foundational. I don't understand why you want to do it through Twitter. Because Twitter is a meeting place for newsreaders. It's many other things as well. No, I don't want to go on Twitter. I can't stand Twitter, but I have to use a browser for better or worse. Yeah, but a browser is not a thing unless you're on a website. Which website would that be? Well, whatever I do, I go through Chrome. So if I access an article, it will come through Chrome and I will pay through Chrome. But Chrome isn't, let's say you're on the New Yorker in Chrome. The New Yorker is going to be pretty upset if you're paying Google and not them. But you won't. I mean, it's the same as Twitter. Chrome will enable you to. Right. So in other words, you're saying Elon's clever and he's doing the right thing. Well, in a way. But in a way, he's believing in his own hype and believing that he's more powerful and that Twitter is not just for journalists, it's for everyone. Yeah. Well, that's his theory. I don't think that's a bad theory. Well, this is his. I mean, I guess this is the beginning of his vision of an X app, which is ubiquitous and for everything and everyone. Well, I think if you look at it from the point of view of the individual, we're going to have some kind of a budget for reading. Right now, it's totally inefficient. It's so annoying where you click on something and you don't really want to get it. And I want to put a screenshot, but I'm not going to pay and I'm not going to subscribe. I agree. I mean, I'm not sure how typical we are. I've got to believe anybody that reads stuff where they click on something and it's either full of ads or a paywall and they have to subscribe to the entire publication to read a single article would cheer at the idea that you can just click on the articles you want, pay for that article and move on. It's a no brainer. Why aren't there a million startup entrepreneurs trying this? Many have tried. I actually tried this in 2005 at Egeo, where we created a digital envelope that could be unlocked by paying. And inside the digital envelope, there could be a white paper, a PDF, an article, music, video, anything like a locked digital artifact. It's an obvious idea and it's not new. That was 15 years, 18 years ago. So you're right. Somebody needs to do this and it needs to succeed. Twitter is as good a choice as anyone as a starting place. And I guess in terms of content, the video model would have been the cable company who figured out a way of doing it. It doesn't make many people happy, but it's worked. Yeah, but forcing a subscription to all you can eat. That's always been the problem with content owners. They don't trust that they have enough good content for you only to pay for what you want. So they force you to pay for all the stuff you don't want as well. And that can't survive. That just doesn't make sense. And it doesn't solve all the problems, but it would also be a shot against advertising. Because if a browser or Twitter or everyone else was offering this to me, and they said, you pay X amount, and you will have this amount of content, and I guarantee you won't have to look at advertising, you won't be tracked. I don't know whether I'm a typical consumer, but I'm guessing there has to be a market for that kind of product. Yeah. Do you use Apple News on your iPhone? Well, Apple News is basically that. If you pay for Apple News as part of your package. I think I do, but I never use it. You have to train it because it learns your taste. I wonder whether, and this was my next question on AI since it's so ubiquitous these days, will AI help this? Because as you say, people have been trying to crack this nut for years. Can AI help? Well, I think that will happen. The Google leaked document talks about personal AI, which is an AI that runs on your phone that knows about you. That probably is the secret to filtering content based on what you want to receive. And I think that will happen. It is already happening. It will happen. Cosmix was the last company that tried to do this. They were acquired by Walmart Labs because they didn't get enough of an audience for reading to make money. But I do think personal AI blended with content in the cloud that's properly labeled and indexed. Yeah, I do think this will happen. I don't know how many more years Keith will be waiting, but we will be on it on that was the week. What about the state of the private markets? Fantastic. There is an order. Well, no, it's not Jeanette's, but this is her space and your space. What's happening with that this week? So this is just a follow on from last week's shrink shrinking headline. Carter is a company that stores all the documents when startups raise money. So it knows exactly how much investors are paying, how much they're buying of a company for that price at every stage. So they do a state of the private markets roughly three or four times a year. This is the Q1 2023 one. And it is just a goldmine for anyone working in the venture capital or startup space to understand what's going on. The headline news is not anything different than last week. Everything is down. Everything is shrinking, including M&A. That also means that bad behavior is growing. Things like rounds of financing that include unfair caveats for the investors at the expense of the entrepreneurs. The time between rounds is expanding. The amount invested per round is shrinking. It's just a very, very good. And when are we going to start seeing the beginnings of the AI boom in these sorts of numbers? Will it be by the end of the year? Well, it's already there. It's a little bit like saying, when will we see the next Olympic champion in any given sport? Well, they're currently 12 years old and really, really good in high school. And the same for startups. In 2023, there's been a lot of AI companies funded. The future Olympians amongst them are already there. And it's a question of figuring out which ones they are as early as possible, which is, of course, the game that I play. And I think as long as you can take your eye off of the decline of the historical cohort and spend time looking for the new cohort, you'll see it. But some of the old Olympians are still winning gold medals. You made Elon Musk and his Boring Company startup of the week. I know you're a big admirer of Musk. What has he been doing at Boring Company to justify giving him startup of the week this week? Well, it astounded me, actually. What he's done is negotiated an agreement with the city of Las Vegas to build a 69 station underground, connecting every part of the city to every other part of the city. Now, if you think about places in the world that have underground metros, like London, for example, that have taken more than 100 years to build today and have cost huge amounts of money, what Musk is doing with the Boring Company, assuming it executes, is significantly reducing the cost and time to build sophisticated travel and communication systems. Yeah, I mean, why isn't this bigger news? I have to admit, it's the first time I've heard of it. I mean, won't this cost trillions of dollars? You would think so. But I'm sure the answer is no, because I don't think Las Vegas has trillions of dollars to spend. So it must be a lot less than that. And there's probably a private public partnership. I'm sure the city is paying for some of it. The federal government is probably paying for some of it. And I'm pretty sure that Musk, once it's built, will be making all the money from the ticket sales. So once again... So the Las Vegas underground system is going to be owned by Musk's Boring Company? I actually don't know. The details are not clear yet. I'm sure they're not clear. The Boring Company is definitely building it. Because there's more than a single payer, I assume there's more than a single beneficiary of the ticket system. But one of them probably is the Boring Company. And given the fact that it's happening in Las Vegas, I'm sure that the traditional public nature of transportation infrastructure will be different, which will be another story. Finally, you brought Musk in, one of your favourites. Your other favourite, Paul Graham, is our Tweet of the Week. How did he get in? What's Paul Graham been saying this week, Keith? And by the way, when are you going to change it? Not Tweet of the Week, what's this new social media platform? Blue Sky. What do they call it? They haven't allowed me in. No, but they have a term for tweets. What are they? Oh, it's a terrible term. I forgot what it is, but it's a term. You're sticking with Twitter. I'm sticking with Twitter. Why is Paul Graham the Wise Man of the Week on Twitter? I thought it was a good follow-up to the Boring Company one. His original tweet says, electric cars, fake meat, AI, all three have been around a long time, since he was a kid, he claims, and they were never good enough, so we ignored them. But now all three have crossed the line where you can't ignore them. And the question to his audience was, what's next? And there's a long thread of various well-known people speculating on what the next thing that we assume to be a failure, that will eventually be a success is. His prediction is nuclear fusion, which I would love to be true, but I'm still not convinced. What's your prediction? And what's next? I mean, I buy electric cars and AI. I'm not sure about fake meat. Flying cars would be a good bet, because I think we all assume they're not happening, but I think they probably are going to happen. And what won't happen, Keith? What won't happen? Like they talk about the Brazilian economy, always on the horizon, always next, but never real. Super easy, Andrew. What won't happen is Lina Khan won't regulate AI. It's not going to happen.