The Dawning of a New Era

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 day Wow, it's January the 20th, Friday 2023, the dawning of a new era, at least the dawning of a new year, 2023, happy new year, Keith, you've been away for what, three or four years, it seems like. It does seem like that, doesn't it? But you know what, looking back over the last, I think it's four weeks, we haven't done the show. I'm not sure the world, you know, missed us too much. I'm not sure if the world missed us, but we missed the world. I mean, ever since you've been gone, chat GPT has taken over everything. Was it in South Africa? I know you spent Christmas there. Were they all doing open AI down there? It was my dinner party show off piece. I had it on my phone, and I would give it to like, I had dinner with one of the lawyers who drafted the South African Bill of Rights, and I gave it to him, and it asked him questions about the South African constitution, and he was blown away that he got it right. Every question he asked, he gave the right answer. And then he asked what the role of himself was in it, and it praised him for having a very prominent role, so he was super happy. But I used it a lot with people who are non-technical, just to see their reaction, and the reaction is huge. So Keith, the title of this week's newsletter is Dawning of a New Era. Are you serious, or is that ironic? Especially given how the lead piece is by Gary Marcus, who suggests that there is no dawning of a new era. Well, I think this is one of those cases where it's super important to be nuanced.

Speaker

Certainly, the larger the claim, the less likely it is to be true. But there are some claims that are true, and I think the claim that's true that is the most important is that these chat GPT-like architectures known as large language models that run lots of different actual models are capable of quickly retrieving information from huge data sets in a somewhat clever, I wouldn't say intelligent yet, but in a somewhat useful for anyone who needs to retrieve information or have information at their fingertips, in the same way we used to think about Google search. So it will get things wrong often. And so you have to be cautious, but it gets them wrong less often than you'd think. And when it's right, it saves you a ton of time. So is it, and I hate this word, I can't think of a better one on a Friday afternoon. Is it a new paradigm? It is a new paradigm because it represents an interface to information that is completely different to anything we've so far had. It's an interface to information and it's, you know, a baby and it's often not good, which is Gary's point. However, that interface is so much better than the previous search page results interface, and certainly better than Siri or Google Assistant or Alexa for that matter. So it really is the dawning of a new era, but in a narrow band of tech that will have large implications for all of us because we'll all start using it sooner rather than later. And what about the business side, Keith? This technology is publicly owned. I mean, obviously, it's been a huge amount of news about OpenAI and Microsoft's investment. I think it's a $10 billion investment. But OpenAI doesn't own this technology. Anyone can use it.

Speaker

Anyone can use it, although they have introduced a professional tier that you pay for in beta. And they are about... No pun intended. Professional tiers. Exactly. And they are about to upgrade from the large language model they currently have called GPT-3 to one that's 10 times bigger called GPT-4. And they anticipate much better performance across a whole range of things. But my sister-in-law on vacation said, yeah, but it's a copycat. It's copy and pasting. It's a mimic. And there's some truth in that because it only has access to the data it has access to. However, it can make up its own answers to questions based on access to that data. So it's more than a mimic. And it's very good at selecting which data from all the data it has access to is relevant to the thing you're asking it for. So it's a lot more than a mimic, but it isn't yet a brain. And in contrast with Google, lots and lots of pieces today's news, Google laying off 10 or 12,000 people and the alarm bells going, senior execs at Google about the challenge to search, Sergey and Larry coming back to head up a reinvention of their search. In contrast with Google, when they invented their search engine, it was entirely proprietary. They downloaded the web and then they owned search. Open AI isn't doing that. Yeah, well, it doesn't download the web as the web at all. So, for example, it doesn't have any attribution. It's closer to a human being. If I say to you, Andrew, what's the capital city of Australia? You'll tell me that it's Melbourne, because you don't know the answer properly. I know it's Canberra. And I won't say to you, what's your attribution for that? Human beings don't seek attribution from each other. But what I don't understand is where, so let's say the Melbourne or the Melbourne or Canberra thing, where is it getting the data from that it knows either I've made an error or I got it correct? We don't know. There's lots of, well, they haven't been trolling the web, I mean, open AI trolling the web for all this data. If it is, it's crawling the web and then getting rid of all the links.

Speaker

It's looking for information, not for references. Google counts references. ChatGPT tries to decide what's the best based on some kind of relevance algorithm. Whereas what Google does is it just counts inbound links based on keywords. Google was originally called PageRank, which was, again, a pun on Larry Page's name. So in an odd way, it's not an advance, it's a retreat. It's a back to pre-Google, pre-links. Well, I think it is an advance because I think the future looks more like the ability to ask questions and get answers in a free flow. Like with ChatGPT, I put it in my editorial. I asked it to summarize one of the articles and then after it had done that, I asked it to critique the article and it remembered the article. It's more of a conversation with accumulated knowledge through the conversation that you can assume is understood. It's super close to a conversation. So it's almost premised on Lockean principles, on experience. It amasses data like we humans, according to Locke, amassed experience. And as a consequence, it knows the world. But it's limited by what it knows in the same way as you and I are limited by our experience. Yeah, but obviously it knows. It's almost like it was given the Library of Congress and then asked to have access to that in real time when somebody asked a question where there is stuff that it can leverage to answer and it knows what to answer. It ignores things that are not relevant, mostly. So it's a very large step forward. Now, that said, apparently it's costing three million dollars a day to run. That's not very much if you've just got an investment of ten billion dollars. Yeah, and my guess is it runs. I mean, just the hardware costs is ten billion. Plus software, the computing costs. Yeah. Which will go down. And presumably most of that's supplied by Microsoft. I mean, it all runs on Azure. So is that the cost to AI or to Microsoft, to open AI? That's the cost to AI. We don't know if they, you know, it's interesting that the deal Microsoft struck with them has a remarkable similarity to the one they struck with me when I ran real names and that didn't turn out well. You may have given them some advice as well, but it's a different Microsoft, less aggressive. Well, I still think there's a guy there called Zig Serafin who was in the middle of my deal and I think he's in the middle of this deal. So even you, Keith, would not claim that your thing was of as much importance as open AI and chat GPT. Of course, my thing was a small stepping stone beyond search. My mantra was don't search, just go there. But it was still a page based navigation engine. Well, let's let's go to the newsletter then. Gary Marcus popped a hole in the AI bubble, as only Gary Marcus can do. He revealed a lot of the fraudulent claims. He reminded everyone, for example, that the original Tesla self-driving car was based on a lie. So what's Gary been saying that's really important rather than just short term sensationalism? He is a bit tabloidish in the way he talks about it, because the thing about Tesla staging their demo is pretty normal in tech that you early on in the life of the technology, you stage demos more in order to convey marketing intent and to start pre-selling your product. So it's not as shocking as he makes it out. He then says that chat GPT is using sweatshops in, I think, in Kenya, paying two dollars an hour for training. Yeah, there's a thing in AI called labeling and supervised learning. Those are the two key things. And they need human beings to curate labels and to say yes or no a lot when careers are asked and the wrong thing comes back. So, you know, I don't know that that's a major shock. Every major company in the world uses labor. Although, yeah, and it's just a sort of continuation of whilst open AI is making billions and valuing itself in the tens of billions, they're paying people in Africa minimum wage or probably less than the minimum wage, certainly less than the minimum wage in the US to build their product. But that's another issue. He's right. But he's an AI guy. What does he say on the AI front? I mean, I've had him on my keynote show.

Speaker

And he doesn't seem to think it's as much of a leap as you do, but well, I think secretly he acknowledges it. I think he's probably irritated that he wasn't the one who invented it. Well, look, both things can be true at the same time. Architecturally, what it does could already be done before, but it was done with way less data. So it's it doesn't represent a computing leap forward, and it probably isn't that it probably he makes the point that true AI won't need access to lots of data. It'll learn like a child. Well, he hopes it will be the equivalent, then Berners-Lee's World Wide Web, which didn't invent anything, it just put different pieces together. No, he thinks that artificial general intelligence, which really doesn't exist yet, will will be closer to a brain like our brains come on the earth. And by the time you're our age, we know a lot, you know, almost everything we see gets stored and retained and is usable by us. So he thinks it'll be closer to that than giving a whole bunch of data that you can then selectively search through to find relevance to whatever someone's asking, which he thinks is not really AI at all. It's just I don't think he uses this word, but it's basically brute force against large data sets. Well, Gary's always worth reading, and this essay is particularly good. And the fact that Keith added it or headlined it in the newsletter, I think is significant. There is other non-AI news, VR news on the Apple front. What's Apple been up to this week, Keith? Yeah, MG Siegler writes about about that. I think it's in News of the Week. Yes, it is. This is a machine rendering of Apple's mixed reality as opposed to VR. Apparently, the forward looking cameras are so good that you see the outside world inside and you can layer things on top of it, which you can already do in Meta's latest headset, but apparently not as good. MG thinks that this thing will be released this year and that it will blow away the rest of the market and it will not be to do with Metaverse. There'll be more functional use cases for it, typical for Apple. I've always said, and we said it on the show, Apple understands that you have to target technology at human senses. Yeah, I mean, we always believe in Apple, we trust Apple. But the thing that we're looking at on the screen, Keith, it doesn't look like an Apple device. I mean, it doesn't look like my my iPhone or my watch. I mean, are people really going to put this thing on? Well, it depends what it does. I mean, Robert Scoble might, but who else? I definitely don't think I'll be doing it, although I wouldn't rule out that I'd buy one just to see what it's like. Well, you buy everything that comes out from Apple, but. Well, we'll see. It'll be interesting. I mean, certainly it'd be one of the major news stories of the year, whether it appears or doesn't and how good it is. Another major news story which keeps on going is the regulatory one. Interesting headline you link to on the newsletter on the EU and their quest for digital sovereignty. Is there an increase in divorce? Maybe there was never a marriage between America and Europe when it comes to tech. You know, the reason this is so interesting is the world has basically applauded Europe for regulating tech, and this is the first time when a European think tank has come out and said it's strangling innovation and making Europe be in last place compared to America and China. By the way, Norman Lewis, who wrote this, is a personal friend of mine going back many years, so I might be biased, but I link to his report and his report is a fantastic, quite lengthy piece looking at why government is strangling innovation in Europe and thus destroying potential value and guaranteeing that the value is built outside of Europe, not inside. Interesting, we'll come back to that certainly this year, and it wouldn't be a week without some Twitter news. You refer to a news piece about Twitter banning third party apps. Why? What's the reason for that? Yeah, you know, it's interesting, obviously, the reason for everything at Twitter right now is that Elon Musk believes that's the right thing to do for whatever reason. It's interesting, I've historically thought that Twitter were undermining themselves by not releasing the flow of data openly to any app and any website, because given that they put ads in the stream, that would allow them to make a lot more revenue. The argument for this is a bit nuanced, and it made me think maybe he has a point. He's basically saying, look, the Twitter client doesn't need to be reinvented by a third party and that does just the same thing that the Twitter client does. Yes, the data should be available for other purposes, but not to build an exact copy of the Twitter app. And OK, that's kind of fair enough, I can understand that. They want to control the user interface to their content within an app. I still, on the margin, think it's a mistake. I don't think there's a big win here for him, but I kind of understand the rationale. Startup of the week, Keith, an interesting startup. Not really, I mean, you have a very liberal definition of startup, but what's the startup this week? Well, I put this one in because it speaks to the market conditions that are going to dominate this year's conversation. Blackstone, as we all know, is a huge asset manager. It manages other people's money and puts it into things where it can grow. It's decided that secondaries, that is to say, buying shares from existing shareholders in venture backed companies, is a good idea. So they might come to single rank and say, Keith, would you like to sell some of your shares? And if I said yes, they'll pay for them. I'll get the money and whoever and they'll put them in their portfolio. Secondaries typically are bought at a discount. So let's say a share is a dollar. Last time you raised money, a secondary user will pay 80 cents for it. Well, today they're only paying 40 cents for it due to the market conditions. So this is Blackstone being very opportunistic to buy distressed assets at the very moment they're distressed. So it's probably a very smart move. It's a smart move, although why is Blackstone a startup? Well, it's a startup in doing secondaries. It's not a startup, but this is new for it. This particular fund is a first time fund doing and it's 22 billion dollars. It's a huge first time fund, which tells you that all of the money thinks the best bargains will come from buying stuff that's already been funded cheaply as opposed to investing in new stuff. Well, as always in these sharp downturns, there's a huge amount of money now to be if you've got cash, if you've got liquidity, this is the time to make money, isn't it, Keith? Probably, probably it is. You need to lend me a few billion and then I'll make some money. I saw that Peter Thiel, of course, got out of crypto just before it crashed. Some people have exquisite timing. And when it comes to exquisite timing, the final feature of the week, which is always my favorite, is Tweet of the Week. And this week you have a really interesting tweet. You know, all your tweets are good, but this one is really intriguing. I've never really thought of it, but it's obvious. So perhaps you might remind us of the Tweet of the Week. So this is this is Kevin Zhu is actually the tweeter. But I picked up on Bill Gurley. Oh, I thought it was Gurley's. But anyway, Gurley is retweeting, is retweeting it and then adding his own flavor on top, which is that de-globalization is dangerous in a world where our self-inflicted red tape makes the US non-competitive. Well, it's not dangerous for anyone except the US. Dangerous for the US. And just the recognition that globalization is dead is already a big statement from the original tweet to then say the US will be punished due to that when, in fact, the US has been driving it, you know, with its trade war stuff since the Trump era. So I, as you know, I'm a big believer in a single global canvas for trade. I'm a free trader, if you will, and I don't particularly want to support nation state A versus nation state B in that I want a level playing field. So Gurley is basically saying that that's the point of view that most aligns with US self-interest, a level playing field. Yeah, and I think what's interesting is that the US state is, it's not suited for de-globalization because it was never strong in the first place. So it doesn't know how, whereas I don't know, Turkey or China or Russia are much more suited to these, to this economy. Yeah, certainly because the state apparatus is super strong, meaning it's a top down society and the US isn't really a top down society. It's a chaotic, anarchistic, bottoms up society with no real strong desire, even amongst Democrats, for a strong state. And the lowest status people in this country are bureaucrats, state bureaucrats, which is the reverse in Turkey or China. Correct. Exactly right. Well, I mean, my view for what it's worth is de-globalization has been exaggerated. It's not the end of globalization. It's a stutter, but we will come back to it. That was the week for January the 20th, 2023. Much more on OpenAI, Twitter, Europe versus America and de-globalization this year, Keith. We'll see you next week. Bye, Andrew. Bye, everyone.