The Exchanges

Q I was going to say, do you think that's going to be one of the main reasons of the success or failure of this is how it can be integrated in all of the other pieces that Apple puts out in the world?

A Well, I sort of tweeted halfway through the presentation. I feel very sorry for Meta because they spent, you know, however many tens of billions of dollars on their product that, you know, no one's really, only a couple of million people are really using and doesn't have traction. And after this, why on earth would you develop for that? Yes, like that's three, 400, the Quest three is 400 bucks and this is Two and a half grand, but like, even so. Um, but kind of the reason I mentioned Meta is Apple has all this other stuff to leverage. They've got the developer community, they've got the tools, they've got the Mac and the iPad and the iPhone and the chip team and everything else. And Meta sort of had to try and desperately try and build all of that from the outside and kind of bootstrap all of it from zero. I mean, God only knows how much Apple has spent on this, you know, probably more, much more than Meta, I would guess. But they're doing it in the context of having all of these other things to plug into it, which, which, which Meta doesn't have.

AI assessment note: “Apple has all this other stuff to leverage. They've got the developer community”

Q of what's possible, is that also the meeting in which you are asking the very tactical questions and operational questions of, okay, that's a great idea, but how does it work? Break it down. Or is that not the meeting where that happens, and you're like, no, this is a fucking phenomenal idea. We'll figure it out when we need to figure it out, and we'll ask the tactical questions.

A Well, so, so it depends. So I think we've done other podcasts about this. Um, you don't expect people to have every single detail worked out. Quite often, the point is that you're, that you're, you're betting on the entrepreneur's ability to work out how this is going to work. But, um, the metaphor I always used to use is tractors and rocket ships. So there's some kind of companies that have very high margins, like Facebook, and your job is to keep it pointed roughly upwards and not argue about the thrust to weight ratio. Are we going up very, very fast? Yes. Don't argue about what the thrust to weight ratio is. The answer is lots. And we'll work out the revenue later. And if you looked at Facebook in 2005 and said this thing isn't making any money, I'm not investing, you would have been in more.

AI assessment note: “you don't expect people to have every single detail worked out”

Q against one political party or another. They're here to make money. Why do we seem to forget that? So they're obviously going to build the best platform that actually enables more money to be made. So to your point about, well, what if we stopped sharing on Facebook? My mind went to, would that help with their business model? Would that hinder that? And why are we not doing that?

A Well, so you taught my, all my friends at Facebook would basically say, we wish we didn't have news at all. We don't make any money from it. We actually don't make any money from news, although the amount of money is trivial. It's a huge amount of, of, of, of pain and aggravation and, You know, whether you believe the social harm is real or not, there's a huge amount of people shouting at us. The PR damage is enormous. We should, we would love really, you know, we should just block all news domains entirely. Like, all news content. And like, that's not what Zuck says, but like, from a financial point of view, they are not making money out of this. They're just really like, oh, please make it go away and get everybody back to sharing pictures of puppies. They would much rather just be the puppy sharing picture network.

AI assessment note: “from a financial point of view, they are not making money out of this”

Q sounds like, and again, I didn't watch the Facebook connect event live. I watched the re-recording and you don't have the streams of comment and the re-recording, which sounds like that's actually a blessing. Um, but it sounds like the Facebook event felt much more like a live event. And the Apple, which obviously pre-recorded, you don't have that. Am I correct in thinking there wasn't any sort of live?

A Yeah, it felt like you're kind of, yeah, some of it was live, some of it was recorded, but it felt much more scrappy and free flowing, and it was, it felt like you were supposed to feel like this was recorded in this person's living room, and that was recorded in that person's office, and they cut to 20 or 30 different games companies, and You know, one of them, it was literally like it was in his cubicle and he said, like, I really suck at doing these videos, but I'm going to try and tell you about what a great game we've made. I love it. So it was just a much more kind of a grassroots kind of a feeling. Whereas Apple, it was here is our billion dollar office building. And here is, you know, we spent twenty million dollars on cameras and compositing and rendering. It was a much, it was a movie production.

AI assessment note: “some of it was live, some of it was recorded, but it felt much more scrappy”

Q Well, that's the whole network effect, isn't it? Like, it's pointless if you're just hanging there alone, hanging out there alone.

A And one can go down, you know, questions of, well, how does the wallet model change that? So there's all these sort of senses of, like, these different concepts of what you can do with this, and how this might work, that allow you to build a different kind of product. So, you know, very tangible example. Supposing you build an Instagram on this, and I have Lots of followers, and I'm an influencer, and I sell a sponsored post. Well, that sponsored, the way that would work is the sponsor would buy tokens in the system, place them in a smart contract. When their ad appears between me and them, when their ad appears, then mechanistically the money is released to me. Mechanistically, some portion of that gets sent to early investors or early participants or to infrastructure providers. Some portion of it gets to me. Maybe I've automatically installed some, some other application that automatically allocates some to charity and some to tax. And it's all happening mechanistically. It's not that I have to send them a PDF of the Instagram metrics dashboard so that they can see, so they had to see the ad.

AI assessment note: “how does the wallet model change that? So there's all these sort of senses”

Q Are you saying that all of them don't have a network effect, or do you think that there's a few of them that have a potential of a network effect that's starting?

A So the thing people are kicking around at the moment is a memory where the product remembers, not the model we should distinguish, kind of the product Has some history of everything you've asked in the past, and it uses that to inform its answers. Um, but that stickiness, I don't think that's a network effect per se, or that you'd have to do something else for that to become a network effect, and it doesn't seem that hard that you could just ask ChatGPG, tell me everything you know about me, and then go and tell Claude, hey, know this about me. Uh, maybe, maybe not, but it's, you know, it's stickiness, but it's not, it doesn't seem to be network effect, and there doesn't seem to be network effect in the models themselves, because the models aren't being continuously Retrained on everybody's input. And if you were, it's not clear how that would work. Um, so in contradistinction do like smartphone operating systems, say, or indeed Google or say from networks, there aren't network effects you've got. So you've got the products of the space. It is saying the models are basically saying the usage isn't and how would that change? And I think it was kind of an interesting starting point. And then I went and looked at this Publication from the UK regulator, um, competition regulator, the CMA, which is doing an investigation into Google, and so they went and pulled a bunch of primary da…

AI assessment note: “it's stickiness, but it's not, it doesn't seem to be network effect”

Q yeah, obviously, I haven't put, I haven't plucked this one out of Finem. My memory does not work like that. Um, but you say that ML bias will cause problems in roughly the same kinds of ways as problems in the past and will be resolvable and discoverable or not to roughly the same degree as they were in the past. Is that still, do you still agree with that?

A Well, I mean, one of the things, this is the reason I kind of talked about databases. The You have these systems that produce what appear to be a correct answer, but it might be wrong. And there's a people, there's this kind of claim that somehow databases were deterministic, and you understood how they worked, and they always produced the right answer, unlike machine learning. And that's true in a sort of narrow pedantic sort of way, but it's actually not true. You know, databases are full of bugs and mistakes, and there's stuff in the data that isn't correct. And so databases, in fact, can say stuff that isn't true. And they can say something and you don't know why it's, why it produced that result. And so on the one hand, you need to be conscious of when you are building and using these systems to try to not do that. But you also have to have the institutional capability to understand that maybe the data computer is wrong, which is, and I think there's a sort of a slightly uncomfortable and people use these things to ruin people's lives as they have in the past. And you need to have sort of institutional processes to try and stop that happening in the first place. And secondly, to recognize that it has happened and try and fix it. And I think the same thing sort of gets repeated here again.

AI assessment note: “And I think the same thing sort of gets repeated here again.”

Q but mostly to aggregate data and numbers around, you know, what we're seeing around a certain series of sports, but it's interesting to see how it's being picked up. But I think for most people, we're still trying to figure out how, how is this being used and how are we using it on a day-to-day basis? And maybe we're not yet now. It doesn't mean that it's not great.

A Hmm. Yeah. I mean, the, the analogy I used, and again, Lots of analogies here. The analogy I used to use for the last wave of machine learning was to say, this gives you infinite interns. So you want to listen to every call coming into the call center and tell me when the customer's angry. You want to look at every credit card transaction and tell me the weird transactions. You want to look at every single x-ray of your wind turbine blades and tell me if there's anything unusual. Um, doesn't take, you know, someone with early years of experience. It's just an intern. Um, maybe not the last one, but You know, it's, but it, but you could never automate that before because it kind of required a mammal brain. I think someone that I used to work with said that AI will probably do anything that you could train a dog to do, which kind of captures that as well. You, you could train a dog to tell me whenever somebody's angry. That would not be difficult. Dogs can do that.

AI assessment note: “The analogy I used to use for the last wave of machine learning was to say, this gives you infinite interns.”

Q Um, what prompted this conversation? What was this for, for you? Was there anything in particular that you were looking at?

A Well, so I was looking at, I looked at Metaverse, sorry, at Fediverse, and saw, this looks like an open source claim of Twitter, and open source claims of commercial products never work. It's the only people who use them are people who care that it's open source, not people who care about the product, for the product. And then I thought, no, there's a bit more to it than that. And it's more, this is much, this is also a sort of an attempt at an entire web distribution system. Now it's also been around for a decade and there's a bunch of acronyms and a bunch of ideas and a bunch of very passionate, a very, very small number of very, very passionate people who will say, no, you're an idiot. This is all working really well. Um, but I think this sort of the underlying challenge here is interesting. I mean, one of the things I was thinking about writing was a sort of like a comparison to like the laws of thermodynamics in that, like, there's always this urge to make a open distributed content Recommendation system that's flat and doesn't have gatekeepers and will not downrank things. But if you do that, then now I've got a two, three, 4000 things in my feed. And how are they sorted? Well, they're sorted by when they were published. And that's just an algorithm as well. That's just a randomization algorithm. So you can't, you know, put another way, like you, if you, if you, um, if yo…

AI assessment note: “Well, so I was looking at, I looked at Metaverse, sorry, at Fediverse”

Q That is a perfect example of how it isn't a tool that's built with a specific audience in mind. We don't know who the audience is right now. Do we? Like who's the user base, you know, for this?

A Yeah. I mean, it says, well, in a very different sort of context, one of the ways I kind of talk about HEI is to say, I may have said this on this podcast before, is to say, we don't, we had, when we were trying to go to the moon, we had a theory of orbital mechanics. And you can, we had a theory of what rockets were doing, and you could get the spec for the Apollo rocket and show it to Isaac Newton, and he could do the maths and tell you whether it would get to the moon or not. This much thrust, this much weight, this much fuel for this melange. This is how far away the moon is. We've measured it. Will it get there? Yes or no? Whereas with AGI, we have no theoretical model for what AGI actually is, and we also actually don't have a theoretical model for how LLMs work. Other than at a very, very high, high level. So we don't actually know what will happen if we give it three X more data and four X more compute, which is what gets you to this phrase emergent capability. Like it starts doubly starts working in unpredictable ways. And then maybe, well, maybe it doesn't, and we don't know. So we know the point is we don't really know what it is. We just, we keep making these bigger and bigger rockets and they keep going higher and the moon is roughly in that direction. So we think maybe it will get to the moon. Um, and then someone kind of chokes about this thing and says, well, ca…

AI assessment note: “So we know the point is we don't really know what it is.”

Q Yeah, they definitely, they, they kept it more from an entertainment perspective, work and entertainment perspective, didn't they?

A Yeah, so there is this sense of, here are all these different things it could be, but kind of going up a level in abstraction, it is this sense that this is this sort of infinite display. So, and almost that you don't have a display. So you put this thing on, and it's as though you put on a pair of ski goggles, and you see the world around you. And there's a screen on the outside that just It shows some expression or shows your eyes so that people around you don't get the sense that you're completely blanked off. And so the ad they show is like the kid can run up to you and talk to you and you can still see them and interact with you, which I have to say I'm kind of skeptical about. It seems to me that's okay, that's just enough for you to see why I should take my headset off, not that I'm going to wear it all day. They don't show people walking down the street wearing it, which I think is interesting. They show people, they show people sitting down Or in an office or at home. They, and you know, walking around a desk, they do not show people going for a run or walking down a street with it. So there is a sense of there are limitations on where you could use this. Um, but the core of it is you put ski goggles on and it's so the ski goggles are like a magic window. Like they suddenly, like you see the world, you see your room as it was, but then stuff can appear. And you've got …

AI assessment note: “They show people, they show people sitting down Or in an office or at home.”

Q No, I agree. Where and how do you, here's a question. Do we look at NFT, metaverse, crypto as all part of a singular topic and they're all components of it? Or do you think they warrant three separate conversations on three separate paths?

A So I think part of the challenge for both of these subjects is that, or all of these subjects, is that these words sort of escaped out into the wild and got used to mean almost anything, particularly metaverse, which sort of metastised, if that's the right word, From a relatively narrow sense of saying, well, what would happen if everyone using, is using VR and AR as their main device, and how will that change the internet, into meaning more or less anything that anybody happened to be thinking of at the moment, to the point that you actually don't know what anyone means when they say metaverse, because they could mean anything. And so I think the moment that people started saying that NFT is a part of the metaverse, for me was the point And people started saying, you'll be using the metaverse on your smartphone. It was kind of to the point that I just said, I kind of give up. This word just doesn't mean anything anymore. It became anything cool that might happen in tech in 10 years.

AI assessment note: “the moment that people started saying that NFT is a part of the metaverse”

Q And is this the start contracts with AWS, which requires investment in property, equipment, everything that it takes to operate cloud business?

A It's a whole other thing. Um, whereas, whereas selling ads on the existing website where you need people to do that, but you don't need like a whole new set of people to run all the infrastructure because most of the infrastructures are there and you don't need to build like a hundred billion dollars and more servers to run that because you already got more. And so you're sort of free riding to some extent on the existing, all the existing infrastructure and cost base. Um, and so, forty eight billion dollars. Um, Alpha Amazon, sorry, Google's operating profit, if you remove TAC, which is what they pay Apple to be the default search engine, and if you remove, um, technology R&D, then you get like a 60% operating margin, so if you apply that to 40, forty billion dollars, you get whatever it is, 20, over twenty billion dollars, um, which is roughly the same as AWS, so roughly equivalent operating profit as AWS, And without AWS's 10 or twenty billion dollars of capex attached to it. So, like, giant new profitable business. Haha. Okay, that's the first way of looking at it. However, the more interesting thing is to say that it's not really a standalone business, and the dumb thing is to say, yes, not a standalone business because you couldn't have it without the rest of Amazon. The more important point, I think, is to say, um, okay, could you do a P&L for Prime? So this is more reve…

AI assessment note: “without AWS's 10 or twenty billion dollars of capex attached to it.”

Q I think, briefly spoken about, and that has been mentioned time and time again, which is this idea That there potentially may be new gatekeepers when we open up a new wave of technology, and there's always been gatekeepers in tech. How are you thinking about, and I'm putting it in inverted commas, like, new gatekeepers in this space, and is that even something that we should be thinking about?

A So I think there's two very different places that one could take that conversation, because one conversation is to go very specifically and look at, you know, Apple, Taking 25% off the top of Facebook. It's actually Apple and TikTok between them. Well, at any rate, Facebook said they missed their numbers, and the stock went down 25%, and Facebook said it's because of A, Apple, and B, TikTok, and C, we're old and boring, except they didn't say that part. And, you know, Apple, despite not really having a significant ad business, has sort of somehow become a very significant power player in the ad business. Amazon announced This quarter, what had previously been sort of implicit in their results, so they, they have this line in the back of the account, it's called a breakdown of their revenue, and they had this line called other, which they say is predominantly advertising, and this quarter they basically broke out advertising specifically, and it was all advertising, and so Amazon needed a thirty one billion dollars of ad revenue in 2021, which is bigger than, bigger than YouTube and bigger than the global newspaper industry's advertising revenue. Um, And so here from nothing five years ago, here's this thirty one billion dollar ad business. Then you have Shopify whose numbers, Q four numbers aren't out yet, but in the 12 months to Q three, they did a hundred and sixty five billi…

AI assessment note: “there's a certain kind of person that wants to say, you've got these giant monopolies”

Q Um, so what's the point with that? Ask better questions or follow up?

A The point with that is, it's like, it's very easy to make your list of frameworks for how innovation is going to change the industry, um, in hindsight. It's, and sometimes it's completely obvious. I mean, I do remember a quote from one of the co-founders of BlackBerrys looking at the iPhone and saying, if this thing works, we're competing with a Mac. Even as in public, they were laughing at it. Um, But, you know, the history of the last 25 years of software companies being created is it's really easy to say that it was obvious after it's happened. Not nearly as easy to say that it's to, where to predict where all this is going to happen in advance. And the same thing now with this stuff. I mean, yes, if this, again, if the AI maximalism view is right, then like all bets are off and all software is dead and everything changes, and this is what gets people to talking about like structural changes in long-term GDP growth. Um, because this is like the new steam engine. If on the other hand, no, this is not the new steam engine, this is, this is like the new internal combustion engine, this is, you know, this is the gas turbine, it's just more, if this is a new PC, this is a new web, this is a new smartphone, then it's a, it's a continuation of a process that's been going for 200 years, then, then, then the world looks different, and it's, in a sense, it's easier. It's both easier a…

AI assessment note: “The point with that is, it's like, it's very easy to make your list”

Q Because there are so many or because there aren't any?

A I had lunch with somebody today at Ponte Latour, um, which is a Conran restaurant by, by Tower Bridge, and I remember a time when a new restaurant was something that happened like once every month or two, and it was a thing, and everyone would want to go and eat there at the Sydney restaurant. And which is what San Francisco is like now, of course. Have you been to the new restaurant? There's a restaurant. Have you been there? We're San Francisco. We've got restaurants. Um, whereas now in London, like a new, obviously like lockdown, it's changed everything, but you know, last year, if you'd said, have you been to, have you been to the new restaurant? It would be like saying, have you read the new book? What's the new book this month? Like, what are you talking about? The book. Um, and it's kind of the same thing with startups. You know, there was a, there was a time when like, there was a startup this year. Have you heard about the startup? Um, what's the startup this month? Now, I don't think you could possibly keep track of what all the startups are. You know, there are, it's not, there are not thousands in the way there are in the Valley, but there, you know, maybe, you know, there are not as many as there are in the Valley, but the sort of the volume and the normality of it has completely changed.

AI assessment note: “Now, I don't think you could possibly keep track of what all the startups are.”

Q You bring up Slack and Notion a lot as sort of these outliers. I'm curious, what do you think the two have in common that let them be the aggregation that most people aren't tool builders?

A Well, so obviously one person can build the Notion that other people use, and it's a lightweight collaborative database system that doesn't otherwise have kind of good parallels. Slack had a network effect in that, um, One team would use it, and then someone was working with that team, so they would use Slack, and then their team would use it, and then they, so it would kind of spread organically. There were very, very few pieces of enterprise software where that's actually happened, and where that's worked. Um, it's like the Slack, and that's about, there's like one other I forget. Um, and what this is, you know, it's kind of the point I made at the beginning of the essay, was there was this whole story, kind of, 1015 years ago in the Valley, that like, Enterprise software will become grassroots. It'll be bottom up, bottom up enterprise, bottom up sales. You won't have to go through the sales process. You won't go to CIO. You won't have to wait. They go through an 18 month sales cycle. The users will see the thing and buy it. And it turned out that that gets you five percent of the market. And for everyone else, no, actually you need to go and evangelize. You need to explain to people why this is a thing and why they need it and why they need to budget for it and why they need to buy it. And, um, you need to explain to people why, what, that this problem exists, and then you n…

AI assessment note: “one person can build the Notion that other people use... Slack had a network effect”

Q So how do you reduce it? How do you create a fair process? Is that even possible?

A So I, so I think there's two ways to come at this. One of them is to say, um, That, um, you, on the one hand, need to try and solve it, and on the other hand, you need to understand that there will always be mistakes, which is, you know, the UK post office scandal is companies, post office deploys database, database has bugs, post office that shows shortfalls in cash, post office sues hundreds, maybe thousands of people for theft, hundreds of people go to prison. At no point does anyone in the post office admit that there are bugs in the database. Now that's, SQL, you know, this is 30 year old technology. There's a deeper problem here, which is it now appears that people inside the post office actually did understand that there were bugs and tried to cover it up, which at which point you're in a different problem. Um, but the point is like, you know, we all, you know, we all sort of understand in principle, the computer should be wrong. And if there are bugs in the computer that are ruining people's lives, then that's an institutional problem, not an engineering problem. I mean, it's an engineering problem too, but the backstop has to be a willingness to consider that the computer might be wrong. I think that, so this is that, so those are kind of the two parts to it. I think the other side of this is when you're actually making things, um, there's a real ideological puzzle her…

AI assessment note: “the backstop has to be a willingness to consider that the computer might be wrong.”

Q And you think we're in that phase right now with AI? Yeah.

A The other answer is until the internet, most people actually didn't have a reason to have a PC, which is why very few people did. I mean, they were expensive as well, but there actually wasn't a very good reason for most people to have a PC until the webcam launched. Um, but, but you've got like the, you've got this general purpose computer. Um, what do you use it for? If it's, you could do anything with it. Okay. Like what? Well, anything. Yes, but like what? And then when you poke away at those anythings, it works much better at some of them than others. Um, hence if you think about, like, the early, the cases where we really seem to have traction with LLMs right now, um, it's kind of an interesting contrast. On the one side it's code. It's helping you write code. So it's things where every, literally every comma and semicolon matters. And you can't have any mistakes where it's useful because it fills in a bunch of stuff and it's very easy to see the mistake. If you are a software developer, you know, that should be, it's forgot the comma, but it's still really useful for it to fill all of that in for you to give you that first draft that you can fix. And the other side thing that's really useful is brainstorming.

AI assessment note: “hence if you think about, like, the early, the cases where we really seem”

Q Which I feel like is, yeah, it's always the story in tech. What are the, what are the pieces in tech that are useful for us to understand how it works, and what are the pieces where we just don't care how it works? Based on another conversation we had this week of, we don't care how the Wi-Fi gets into our house. It's here. It works. Great.

A Yeah, exactly. And as I sort of said, right, there is a sort of counter thesis that says, no, this will be so much bigger and so much more powerful. It will not work like this. I mean, in a sense, all of these sorts of arguments are really, is this going to work like every other tech cycle? That will be verticalized and horizontal. There will be stuff that's done by hyperscalers. There'll be stuff that's done by startups. There will be stuff that disappears. There will be new features. We will change our work around this. And in 10 years time, we'll have forgotten about it like we've forgotten about machine learning, like it's just software. The counter argument is to say, no, this is going to be like a fundamental step change in generalization, and most of what I've been talking about is why, yeah, maybe that might not be as easy as it sounds, like a generalized, completely generalized interface. I mean, you know, we talked about error rates at the beginning. I think you could hypothesize that even if, you know, you were using a chatbot to talk to an AGI, it still would not be the right interface. Like, if you're using, you know, like as a kind of thought experiment, if, if, if, if at the other end of a chatbot was a person, if, if JackGPT was actually not, if it was actually a bunch of real people, then it would always be right, but that still might, might not be the right in…

AI assessment note: “in 10 years time, we'll have forgotten about it like we've forgotten about machine learning”

Q And what does a successful, where I keep going to is, what am I, not me personally, but what are we trying to achieve? What, what's the outcome that would make us all feel that this is the right outcome, and I put right in quotes here, in air quotes, but.

A Yeah, well this is, what's the underlying policy objective? Yes, the artist should be able to live. How many artists? On what basis? What does that, what does that mean? And we kind of went through this with music, um, And we are now back at the point that revenue from the recorded music industry is sort of two thirds of what it was in 2020 adjusted for inflation, sorry, in 2000 adjusted for inflation and is heading back upwards. But then there's a whole argument about how you distribute, um, the money from streaming. Um, cause as, as, as it's not to go down a rabbit hole, but as you know, basically the formula is that, you know, this is the total amount of money that Spotify gets and you divide it by how many plays you got. So if you've got two percent of all the plays, you get two percent of all the money if you're Taylor Swift, and if you get, you know, um, and of course the problem, then you kind of get a problem because you've got, you know, so, well, why does a play at a half hour track count the same as plays as a three minute track, um, which is an issue for classical, um, does it, because there might be three tracks over lasting an hour and a half, so why should that get paid, paid the same as three tracks from Taylor Swift? What about a 32nd noise track? What about all the people pumping Spotify with white noise, um, and a spam? Um, what about somebody who's a super f…

AI assessment note: “what's the underlying policy objective? Yes, the artist should be able to live.”

Q And then there's a fourth question, does it need VC money if it's not a tech company, if it's not tech related?

A What kind of, you know, who is it that invests in those kinds of things? I mean, there's a much more, there's a much more general conversation I've had a bunch of times in the last couple of weird years, just sort of along the lines of a presentation I did at the beginning of last year called The Great Unbundling. Like, there is clearly going to be a wave of new company creation, right? Outside of software. So whether that's apparel, or retailing, or restaurants, all sorts of things, like the whole basis of how you buy stuff online changes. 40% of UK retail is now, non-food retail is now online. All of the Media channel, the way you acquire customers changes. The retail channel changes. The way that you would build a brand changes. Every component of that gets kind of broken up. So there's going to be a whole bunch of new companies created here. Half of them will fail. That's fine. A bunch of new companies will get created. Those companies are probably not 20 or 30 X returns. You know, they're not going to put, you're not going to go from a five million dollar check to a billion dollar exit, but they are somewhere further along that risk reward curve, and they might be, there's a lot of three X exits in there. There's a lot of new Victoria secrets and gaps to be, and all birds to be created in here. There's a bunch of Sephoras to be created in here. There's a bunch more Warrior…

AI assessment note: “Those companies are probably not 20 or 30 X returns.”

Q Yeah. It's, um, can you, I don't know if those numbers break down into luxury goods versus necessities, because I thought that was interesting also. I was looking at an OECD report that was saying that There's been a shift in e-commerce from luxury goods to necessities, which ultimately makes a lot of sense. Um, but just, yeah. Any, anything interesting there in those numbers?

A Well, so yes and no. So what both the US and the UK give you is, and there's sort of several things to say here. So they get, they do a split out of e-commerce by category, but online and it's online owned and it's by retailer category. So they have online only retailers and then a couple of other categories. So they have grocery, they have home goods, they have clothing, and a couple of others. And so the picture you get is that online-only retailers, which means Amazon and anyone else who only has a website, doesn't have a store, physical store, is about 50, 50 to 60% of total e-commerce. And that's been gradually drifting down over the years as more and more e-commerce cities and Amazon, basically, like more and more physical retailers build an e-commerce presence. And that dipped down a few points in lockdown, basically because people had to buy more stuff that was from people who aren't Amazon. Not like a really big change, but it's kind of noticeable in the chart, and I posted that chart as well. Um, there's a little bit of texture within that, if I can channel the way equity analysts talk, can you give us some texture on that call please? A little bit of texture in the US, um, online clothing sales, online sales by clothing retailers. Doubled.

AI assessment note: “they do a split out of e-commerce by category, but online and it's”

Q your iPhone isn't in any of the pictures that we shoot that weekend? Which then becomes a really interesting question of what's going to happen if you close a deal with Apple? Apple is going to, I don't know, it becomes interesting when Apple is selling Products and services, and now getting into the, the media rights space, the exclusivity contracts in Formula One are going to get really interesting.

A Yeah, I mean, I think there's a whole media rights and advertising story there that's, that's, as it were, a TV industry conversation. The Apple conversation is, it's just like, I mean, there was this story a long time ago, um, that I think it was, um, Seagram invested in a company in Hollywood, um, in one of the movie studios for some reason, sometime in the sixties. So way before Edgar Bronfman and, and, and Universal, it was sometime in the sixties and they bought a stake in one of the movie studios and old man Bronfman who'd, Created this thing. It's basically selling booze to bootleggers in the twenties. Bill created this giant brewing company called Canadian Brewing Company called Seagram, which no longer exists. Anyway, so they, they own this stake in, as it might be MGM, and old man Bronchman says, that's a lot of money for young Edgar to meet some actresses. To which young, young Edgar says, it doesn't cost that much to meet actresses. That's not what you, you don't need to spend that much money to meet actresses. And there is a little bit of like, you know, is this just like the vanity trip of like, They like having the shows, and Eddie Q likes being the big deal, having the big checks in Hollywood.

AI assessment note: “is this just like the vanity trip of like... Eddie Q likes being the big deal”

Q Yeah, yeah. Oh, that's an interesting thought. Are there other examples that you have, like Bloomberg, of things that just weren't, to your point, that's never going to be disrupted because people just don't know exists, can't have access to it?

A I don't know. I mean, you know, there are, you know, there are whole worlds of, you know, there's a huge wave of stuff going on in fintech at the moment. Um, and you know, people who know a lot about fintech and use words like rails, which is not just not my field. No, in a sense, I don't know. I don't know that field. Um, but those, that sometimes those arbitrage opportunities are kind of hiding in plain sight. And, you know, as soon as somebody says, hey, you could do this, people go, oh, wow. Um, I mean, that's like, um, you know, one of the companies I talk about a fair amount at the moment is frame.io, which again, you know, As soon as you say one person edits a video and 10 people need to see it, and that's a spreadsheet with time codes, you go, well, yeah, that should be software. Um, but until somebody actually says, hey, here's this opportunity, you know, nobody knew it was that, or nobody, you know, and people, and of course, people who are doing that wouldn't necessarily realize it. But that's, again, this kind of, we kind of circle back to this point, you know, would they get into this industry? And maybe this is actually kind of another kind of useful framing. Would Google get into frame.io's business? And again, of course not. Um, there is this phrase that I think, Larry or Sergei, Larry Page or Sergei Brin co-founders of Google used, which was, um, the toothbrush…

AI assessment note: “one of the companies I talk about a fair amount at the moment is frame.io”

Q a tech company or is it a fashion company? Can we just say it's a brand new fashion company built in an era where tech and digital is first? And so that's a new type of fashion. That's a new type of fashion company. That's not a tech company. It's just built today. Like, cause it's brand new, isn't it? She, and it's, it's, it's a couple of years old.

A So I think you could certainly say, um, it's a company that leverages a new channel, which is my, the point of my kind of Walmart analogy, you know, is Walmart a car company? I don't think that's a very productive conversation. Is it a company that presumes everyone has a car? Instead of a company that presumes everyone walks to a supermarket, what does it mean if everybody has a car? How do you build your supermarket in different ways if everyone has a car? If people are driving away with what they bought instead of walking away with what they bought, what does that mean? Um, and I think, you know, you have to think about how good the car park is. You have to think about what it means that people drive there and sort of, and so I think that is kind of the Shopify point.

AI assessment note: “it's a company that leverages a new channel, which is my, the point”

Q It's interesting, is it also for a small group of people, or is it going to actually happen in our day-to-day lives for all of the normal people using it on a day-to-day basis?

A Yeah, I mean, is it more like the web, or is it more like open source? Um, and of course, you've also got to kind of strip aside all of the, the absurd amount of noise around it, for people who say it's all a Ponzi scheme, and people who say it's going to replace civilization. You kind of have to strip both of those apart, and think, oh, in five to 10 years, five to 10 years time, you might be able to build Facebook on this, and what would that mean, and how would that work? Um, the other extreme might be something like, um, cars, say, where we don't really know what or when autonomy is going to look like and how long that's going to take, and it may be that we never get full autonomy, and full autonomy is a bit like general AI in that, you know, this is a joke in AI that, you know, AI is anything that doesn't work yet, because as soon as it works, people say that's not AI, that's just a database. That's just image recognition. That's just, that's just, you know, pattern recognition. That's not AI. And it may be that, you know, we have a car that can drive itself by the freeway, but you know, you have to take over when you leave the freeway. And so people will say, well, that's not full self-driving. That's just driver. That's just lane keeping. That's just driver assists. That's just automatic braking. And we, we never get full autonomy. So we don't know. It's one of those thi…

AI assessment note: “So we don't know. It's one of those things where like come back”

Q But that's why we wanted to have this conversation of like, what are actually the interesting questions to ask about this topic?

A It is. And I mean, I think, you know, the, the, the, The framing that I often kind of come back to is to look at the, you know, to compare this particularly in the way people think about it in the way the state of development to compare it a to the early internet and b to open source in the nineties. The early internet in the sense that again, yes, everyone thought this was bullshit one and two, it was incredible. You needed to be incredibly technical to get anything done. And it was not obvious what you would do with it. And again, whenever I say this on Twitter, people say you're an idiot. There were loads of useful things to do on the internet in the seventies. And I said, Who was on the internet in the seventies? There were useful things for those people to do, but if you got on the internet in 1993, yeah, I could send email. To who? To other internet people. Not to anyone else. And so, you know, and meanwhile, if you actually wanted to do anything, um, it was totally unclear, and I think we talked about this briefly on the last podcast, it's very unclear what the endpoint was going to look like. So not only were there all these layers in the stack that you had to So you had to get a modem, and you had to set up TCP IP on your PC, which, and it didn't come with TCP IP, so you had to know what that was, and then you had to get it from a software shop, because there was no in…

AI assessment note: “the framing that I often kind of come back to is to look at”

Q the weeds and the infrastructure. Like that's the last thing we need to think of. Um, Which becomes fascinating, because I always wonder, like, why are people interested in web three right now? Is it because they want to think of a new way to build that infrastructure, or are they more interested in that projection of what could be and what all of this could build down the line?

A So I think, look, crypto has clearly had these sort of successive waves of enthusiasm, and there was a sort of huge bubble sort of four or five years ago. There was a huge spike in the chart. Now we've kind of, we're in like last 18 months or two years, this is sort of massively elevated price. And of course, you know, it sounds very cynical and unkind, but when somebody, some people are getting very, very rich from something, that does occasionally change how they think about it. It could change how they think what it's going to be. Um, now we've clearly, we've had this sort of, this huge surge within that, we've had this surge of enthusiasm for NFTs. And, you know, it's almost like a kind of a microcosm of my point that by talking about that Web three is an attempt to Shift the discussion from saying this is digital gold and that's all it is, and it's, and it's a cryptocurrency. It's to get away from talking about currency and to say, no, it's a platform, which I think I'm, if I haven't said explicitly, like, yeah, I get that. I believe that. Definitely. Um, but, but the, the end of my, my point is, so if web three is a sort of a rebrand of, of, of crypto and NFT is a sort of a, there's a microcosm of NFT in that because clearly the wave of, of excitement was around these sort of various kind of collectible digital objects. And again, I have no conceptual problem at all with …

AI assessment note: “Web three is an attempt to Shift the discussion... to say, no, it's a platform”

Q oh, well, and that was always one of the tensions between sales and engineering. The sales will be, well, I sold the deal, and they're going to get a discount every month, and then it's going to be, and the engineers would look at them with like deaf eyes of, That is not a thing that our billing system or any billing system can do. What are you thinking about?

A Well, so the loop, the reason that I remember this is that the startup that I was working with, um, they were basically a layer on top. And the idea, the original idea was that this was going to be a platform to allow telcos to sell to MVNOs. So, you know, you're going to have five, anyone not know, MVNO is a mobile virtual network operator. So they buy that capacity on the cellular network at wholesale. And then they resell it on with their own brand. And there's a lot of different degrees of difference. Sometimes they have their own SIM card. Sometimes they don't. So this is what Virgin Mobile was is, for example, is they don't own the network. They're renting capacity on someone else's network. And of course, as a, as a mobile operator, you need a software system to support that. And so the idea was here is a software, it's called an MBNE, Mobile Virtual Network Enabler. Um, at one point, this is all very exciting and interesting now, very much isn't. It didn't really turn out to be a big business, and we could have a whole conversation about why MB&As were not a big deal, except in some places. So in some places they're a huge deal. Let's forget that. Never mind that rabbit hole. Um, the rabbit hole I wanted to go down was that the idea of the realization of this startup was, no, actually what we can do is we can sell this to telcos Not to enable EMVNOs, but as a layer for …

AI assessment note: “now they can actually have a system where they can change the price every day.”