This week’s podcast is about the generative AI foundation models of OpenAI and Google. Will they have competitive defenses? Especially against open source.
Here is the link to the TechMoat Consulting.
Here is the link to the China Tech Tour.
Here are the two articles mentioned:
Here is a16z’s tech architecture.
My (working) conclusions are:
- Technology does not have moats. Business models have moats.
- We are looking for competitive defensibility, not moats.
- Adoption by dominant players with big moats that are not being impacted will be the biggest beneficiaries.
- Foundation models will likely advance from early mover plus digital marathon to defensible moats in platform business models.
- Innovation platforms. These are likely. And offer a model for coexistence with open source models.
- Learning platform. It is unclear how these will evolve. They will be different than search,
- AutoGPT and Other Tech I Am Super Excited About (Tech Strategy – Podcast 162)
- The Winners and Losers in ChatGPT (Tech Strategy – Daily Article)
- Why ChatGPT and Generative AI Are a Mortal Threat to Disney, Netflix and Most Hollywood Studios (Tech Strategy – Podcast 150)
From the Concept Library, concepts for this article are:
- GPT and Generative AI
- Innovation Platforms
- Learning Platforms
From the Company Library, companies for this article are:
- OpenAI / GPT / ChatGPT
- Google / Bard
Tue, May 09, 2023 10:26PM • 46:26
moats, ai, business models, model, companies, level, app, proprietary, put, gpt, open source, technology, platform, competitive, run, digital, google, spellcheck, build, generative
Jeffrey Towson 00:00
Welcome, welcome everybody. My name is Jeff Towson, and this is the tech strategy podcast where we analyze the best digital businesses of the US, China and Asia. And the title for today, yes, open AI and Google will likely have moats in generative AI. Now, that’s a very specific title. Because there’s been floating around not know the tech world, the last week or two are fairly well known article that basically in theory was leaked from Google, at least from some Google engineers or something, saying that they don’t think they’re gonna have a moat in AI, and neither will open AI. And a lot of people have been weighing in on this. So I’m gonna weigh in and say, I think that’s actually not true. It looks like it’s gonna happen. And as I sort of sit at the intersection of tech and competitive strategy, I thought this was kind of my type of question. So I’ll give you my take on this and sort of know what some other people think as well, that will be the topic for today. Let’s see did a standard disclaimer, nothing in this podcast or in my writing or website is investment advice and numbers and information for me and any guests may be incorrect, the views and opinions expressed may no longer be relevant or accurate. Overall, investing is risky, this is not investment, legal or tax advice, do your own research. And with that, let me get into the topic.
Jeffrey Towson 01:31
Okay, as always, we’ll start with a couple of concepts, really just, I know, two to three today. The first one is not really a concept, but which is called that a topic GPT. And the idea of generative AI was just sort of its own huge thing. I guess that’s not a concept. But important. The two digital concepts for today are innovation platforms and learning platforms. These are in the concept library on the webpage. Basically, it’s two of the five digital platform types that I always talk about. I’ll go through them specifically when we start looking at the future of these sort of foundation model companies like open AI and Google bar and such. But those are kind of yeah, those are important business models, not only the topic I won’t go through now we’ll go through when I get to, but you can always find those in the concept library. Okay, so I’ll put a link in the show notes. To the article. Everybody’s talking. Well, a lot of people are talking about, titled Google, we have no moat and neither does open AI. Now that was from in theory, dear Dylan Patel off zaal. Ahmed, it was it was sort of in a public Discord server or something like that. And then it showed up on our website semi analysis, I’ll put the link. You know, the idea is, hey, this was leaked from Google. I don’t know. But here, here’s sort of a summary of the article, in my opinion, the argument is basically talking about foundation models, the generative AI foundation models. So for, let’s say, large language models, text based, that would be GPT. Right? That’s the one everyone’s talking about. If we look at sort of image generation models, which will be another type of model, well, that could be Dali or others. But basically, these massive AI models that everyone else is building on the large ones, not the small ones, there’s a lot of small ones floating around. But the large ones, of which the two most well known are open API’s. GPT does now GPT four, and then Google’s Bard. And then we also have them coming out of China with Baidu and some others, but those are the two big ones, right? So the argument is that this has basically been a very expensive arms race between these two leading companies to build and to rapidly advance these large language models, these MLMs which is true, tons of money, huge advancement. Point number two, but in really the last three to four weeks, it’s happened very, very quickly. We have seen a series of smaller LLM projects, so small, large language model projects, either ones that people have sort of built themselves or more likely, based on an open source version, as opposed to a proprietary model. So the open AI ones and the Google ones are proprietary, but there’s an open source version of this floating around which leaked basically from Facebook about three to four weeks ago, and people have been using that open source one to really replicate what open AI and Google can do, I mean, it’s not 100%. But it is shocking how fast one dude with a laptop, the open source code, and one day can replicate the performance of parts of GPD. And that’s freaked people out that these large giant proprietary MLMs may be very exposed to lots of sort of independent low level innovation based on open source. Okay. And the author’s argued that open source MLMs will end up beating proprietary MLMs. Or if not, they’ll pummel the economics where they basically won’t have a moat. Hence the argument No, that we don’t have a moat. Okay, that’s a very interesting question. And it’s not a question we haven’t seen before. If we were talking about an operating system for your smartphone, we could point to a proprietary operating system like iOS, and we could point to an open source one like Android, which then has a licensed, you know, complement to it. We could do the same thing for Windows proprietary operating system, but we could also look at Red Hat. And really, if you look at operating systems going back 3040 years, it has always been a mix of proprietary and open source. Most of the stuff that runs on enterprise servers forever has been a mixed of enterprise and open source and you go to any GitHub community and companies have been building there forever. So there’s always been a bit of a mix here. But basically, the argument is this is happening as well in proprietary MLMs. Okay, I mean, that’s an interesting point. Now, I’ll give you a couple of quotes out of the article who here, quote, open source models are faster, more customizable, more private, and pound per pound more capable. They’re doing things with $100 and 13 billion pounds, not totally sure a parents are some model number that we struggle with that 10,000,500 and 40 billion perms. And they’re doing so in weeks, not months. So the argument is, this combination of open source and independent appears to be closing the gap in terms of capabilities, quote, astonishingly quickly, unquote. Okay, and then they sort of put their toe in the water of competitive strategy. They say this happened just weeks ago. And independent developers have been making astonishing progress using it. The barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person an evening and a beefy laptop, unquote. Now, the argument, Okay, interesting. And you can also point to, and they do that pretty much the same thing happened within image generation. Okay, so LLM is our one type of foundation model for AI. The other is image generation. And you know, the leader in image generation was open AI is proprietary model Dolly, which came out about a year ago. Okay, but then stable diffusion, which is an open source version, you know, it popped out very quickly. And, you know, I think it’s actually better than Dali. I mean, it’s at least as good, it’s probably better. So the argument is that, quote, many are calling this the stable diffusion moment for LLM. Okay, last point, quote, large models aren’t more capable in the long run, if we can iterate faster on small models. So they say basically a losing proposition. So one, the capabilities aren’t superior by being proprietary versus open source, and to the iteration rate is much better, much faster. That’s the argument. And they go on that this means we have no secret sauce, we have no barrier to entry. I think their terminology is strange. Okay. Interesting. These are obviously engineer types who understand the technology deeply, but I think the competitive thinking is, is not clear, which is fine. Okay, let me give you another article. And I’ll put the link in there as well. Those of you who are subscribers, I sent you part one about this sort of summarizing these two articles. I’ll send you part two in the next day. And I’m basically going to lay out what I think is going to happen, which is a guess, but I think this is my opinion. So there’ll be two fairly long articles on this subject. You’ve already got one, the others common Okay, so Andreessen Horowitz puts out a lot of good writing and they have something I’m very, very jealous of. They have access to data for early stage companies, which most people don’t including myself. So I tend to stay looking at public companies because I can get a lot of data. But when we’re looking at generative AI, the very few people see the real numbers for what’s happening. adoption, economics, what’s working cause? Well, they see the numbers because they’re investors in most of these. And the guy I keep an eye on is Martin casado, who is Andreessen Horowitz. Really good analysis, he’s written a lot about AI over the last year, really good stuff. And he, I think he gets to see everything. So him and some other people at Andreessen wrote an article called who owns the generative AI platform, I’ll put the link in the show notes.
Jeffrey Towson 10:56
And one thing they came up with is a pretty good graphic about the new generative AI tech stack, basically, that there’s a technology stack that’s emerging, it’s new, we’re in the early days. And if you map that out for doing generative AI, which is large language models, image generation and other things, you can basically see three levels and I’ll put levels and I’ll put the graphic in the show notes, it’s really worth looking at. If you’re looking at iTunes, or whatever, I can’t put JPEGs in there. So you have to click over to the website, there’s a link in there, just click over to my website, you’ll see the JPEG. But basically, they say, look, there’s three businesses here that make up the tech stack three levels. There are apps, obviously. Now apps can be, you know, it doesn’t just have to be a mobile app. Right? It could be something that runs on your desktop. So it runs on winter, an application, right? It could be a plugin for a browser, it could be something that plugs into an existing company like Adobe or Microsoft. So I mean, these apps can plug into a lot of places. So when they say app, don’t necessarily think mobile app, there’s more than that. And you can definitely see mobile apps, you can see a lot of stuff on browser. But you can also see GitHub co pilot, which is really impressive. You can see gas, Jasper, I mean, there’s a lot of cool stuff happening. And these can be b2b, which was something we’d see on the enterprise side. So like Microsoft is plugging this into everything on the b2b side. But we can also see b2c content creators, people like that. And the distinction they make with apps is there’s two types. There’s sort of apps that build on existing models. So these are companies that build on models that they’re not building. So this could be building off a closed sourced a proprietary foundation model like GPT. Okay, so you’ve built an app that runs off GPT, you have the app, but you don’t control the model. Similarly, you could build an app that runs off stable diffusion, which is an open source foundation model. Again, you’ve got the app, but you don’t run the model. So you’re not vertically integrated in that sense. Now, that’s one type. The other type would be end to end apps where you control everything from the app all the way through the model. So it’s all internal, vertically integrated, mid journey, which is fantastic. I love mid journey. I use mid journey, like every day just to play around where you generate images and photos and whatever. I think I told this story already. But yeah, like I was just chatting with my mom online and my mom likes to paint. She was a an art major art minor in college and she likes to paint. She’s been painting pictures our whole life. And she had a little dog for a long time. And so we were joking, I kind of asked her what’s your famous artist, and she mentioned some artists I’d never heard of who used to paint Chinese and purses. You know, 200 years ago, like the style of it. So as we’re talking I literally went on mid journey. And I had it start generating pictures of in this style of this famous artists I forget who it was of Chinese and versus holding a small dog that was a copy of her dog. And I started sending her photos as we’re talking not photos but very high quality paintings. And I think it kind of blew her mind and then I then I started like what I this was like a 1617 century Chinese emperors holding a small Shih Tzu dog that looked like her dog. And then I started saying, Hey, here’s a version of that with the same Chinese emphasis on the street using a flame thrower. And then I made the embers with a flame thrower then I put her in a tank, Mike and I just started sending all these photos, right mid journeys, fantastic. So mid journey is an integrated end to end app that contains its own proprietary foundation model, as well as the user interface and all of that. The other one that I pay a lot of attention to his runway, runway ml, which is mid journeys, about images, runways about video, you can take a video with your phone, I don’t know, waving into the camera just loaded up there. And it will take the image as the basic structure of what to do. And then it will apply styles to that. So I could take a picture of video of myself waving my hand and then say, turn it into the Human Torch. And it would basically do that. Because it would use the found sort of the basic image structure, I gave it by me waving and then I could turn it in whatever I want. And it was those. So in apps, you have sort of those two versions, really three versions end to end, vertically integrated with your own model or just an app. But then it can either be built on a proprietary model like GPT, or it can be built on or Google Bart or it can be built on an open source version, like stable diffusion. So three types of apps, you move one layer down in the stack, and you get foundation models, which is what this podcast is about. And this is what everyone’s talking about, right? The open source versus proprietary closed source foundation models. So there’s sort of the foundation model level, you know, large models versus small models, proprietary versus open source, and so on. And then the next level down, you get to his infrastructure, which basically means you have to run all this stuff somewhere. So most of this stuff is run on the cloud. So AWS, Azure, things like that. And then everything’s using, you know, basically Nvidia chips, as far as I can tell, or, or TensorFlow or something. So all the hardware. So that’s sort of the infrastructure level, the foundation level, the app level. And that’s a really good picture of how to think about this. Now, I’ll give you basically, what I think Andreessen his conclusions were based on this model. And I’m paraphrasing here. So this is my my language, not here. Basically, they said, looks, infrastructures appears to be the place where there are clear winners, one, we can see the revenue today, it’s not like you’re getting tons of adoption, but no revenue, we can see the chips being sold, we can see the AWS cloud and stuff like that. And you really have to run that I’ve been, I’ve been playing with auto GPT trying to sort of create digital agents. And I basically just downloaded it to my laptop, so you have to sort of put it up on a site and you have to, you know, get the API from, from open AI and plug it in. And then there’s a bit of Python and stuff of all, but it’s not that hard. But you know, at the infrastructure and toolset level, everybody’s using the same stuff, whether it’s servers or whatever. So they said look infrastructure level, we can see the revenue, it’s going up and these business models like invaded, it hasn’t changed the business model. So if these companies were already making money, they’re probably going to make money. It looks like they have moats and competitive strength. Fine. I’m not really going to talk about the infrastructure level, because it’s not my area. If you move up to the app level, okay, I’ll give you their take, which is different than mine. Basically, the application level. It’s just chaos. I mean, there are companies absolutely everywhere, there are apps everywhere. And everybody is replicating what everyone else does. I mean, I like mid journey. But there are other image generation apps all over the place. And a lot of them are very, very good. And you can’t tell the now I think mid journeys significantly better. But yeah, you I get the point where they say like, look, this may look like a commodity, where everyone can do what everyone else is doing. But right now, it’s just chaos. It’s early stage, there’s companies everywhere, there’s hundreds and hundreds of these apps I’ve been sort of taking a look at, or at least reading summaries of is more accurate. And even the ones I’m using, I’m using about 15 of these things right now. Almost every day I’m learning a new significant one. So yeah, apps everywhere. Tons of usage, not a lot of revenue. For most of them. There’s a couple of cases that have revenue, creating code that use case programmers, okay, people are subscribing, copy editing. True, people are using this companies are using this for their copy editing and so on. Image generation, tons of usage, not as much revenue as you would think. Okay, so at the app level, they basically say, Look, we’re getting usage, we’re getting some revenue growth. But we’re seeing very little retention, very little sort of differentiation. And very small margins, probably because it’s starting to look like these are commodity tech services, because you can do the image pretty much the same on 10 different image apps. Okay, but that’s early days, I’ll give you my take in a minute. This is their day. And then for model providers,
Jeffrey Towson 20:33
they basically say we don’t know yet. We don’t know. Things are changing too fast, loves a lot of uncertainty. Here’s a quote from the article quote, so far, we’ve had a hard time finding structural defensibility, anywhere in the stack outside of traditional moats for incumbents. And I’m going to talk to that point, because that’s one of my conclusions as well, which is, companies that already have significant moats and business models that are adopting this tech are doing real well, as opposed to companies that are trying to build new companies and moats based on this tech, where it’s less clear, because it’s completely new. Okay, that’s more or less? Well, here’s one more quote from them. Quote, based on the available data, it’s just not clear if there will be a long term winner take all dynamic in generative AI. Okay, that’s a good point. It’s not clear if it’s going to be like a Google search where one company gets 90% of the entire global market. Because we’re not seeing that with cloud servers, we’re seeing three to four major companies. That doesn’t mean they have notes. I think Silicon Valley gets to enraptured with a winner take all market, most markets are not winner take all most good markets with good defensibility and strong moats are giants and dwarfs, its three to four major players and then a bunch of small companies. That’s really what you want to go for. So that’s the question for me. Okay, that’s a summary of what I think they said. Now we’re going to jump to here’s my conclusions. And the question I basically am trying to answer is a little bit different, which is, let me preface it. To make money I’ve argued you always need three things. You need growth and or a big market. That’s your top line. You need attractive unit economics. That’s your gross profits. and to a lesser degree, your operating margin and you need long term defensibility. You need all three of those things to generate economic wealth. And that kind of syncs up if you read Hamilton Helmer. He cites two factors. I break it into three, but it’s basically the same point he calls two of those power. I think it’s three. Okay, given that framework, I think the cure, the real question is what is the long term competitive defensibility for app businesses, and for foundation model provider businesses, those other two levels? What is their competitive defensibility? Because if you don’t have that none of the other stuffs gonna work. But I don’t use the word moat. Because if you’ve read any of my books, you will realize that I think you can build competitive strength on three or four levels. moats are the best level for sure. But you can also build barriers, no competitive advantages, barriers to entry, soft advantages, digital marathons, hence my books, moats, and marathons. So I think you can build competitive defensibility, particularly in the urgent early stages of an industry without having Moats. moats usually appear later, historically, they don’t show up at the stage, sometimes, but it’s very rare. Usually, you build other types of competitive strength in these early stage and then over time, a more powerful moat emerges. And I’ll explain why that is. So I think that’s the key question. Okay, let me give you my sort of sowhat because I’ve been teeing up a lot of theory, or here’s what I think all of these sorts of writers are not getting right. In my opinion. First of all, is the point I just said, there’s different types of competitive defensibility. And they matter at different points in the lifecycle of technology, and in the lifecycle of business models and products and whatever you can be if you’re an early mover in a business like Netflix was for a long time, that can give you tremendous competitive advantage, strength defensibility versus, let’s say, cable companies, and that’s really what Netflix did for a long time. And it gave them good profits, great growth, all of that. Now as as the market has started to mature in over are the top streaming, then you need to have, you know, competitive advantages and barriers to entry, which I don’t think they really have. So you kind of want to look for competitive defensibility more broadly. And I’ve broken this into six levels. But there’s really three to four levels where I think that plays out and you kind of got to know how that evolves over time as a product or business matures. Now, the second point. Oh, and the levels I look at are, you know, tactics, operating basics, digital marathon’s, barriers to entry competitive advantages. Alright, next point, which I don’t think they’re getting right. Technology doesn’t have Moats. Business models have Moats. Technology is, you know, it just bleeds into everything. It gets adopted by companies, it’s all over the place, you can have a very powerful moat with very basic top technology, or you can have tremendously powerful technology and no moat, it follows from the business model first, and then after that it follows from operating performance. So this whole I think so much of this discussion is like do large language models have Moats? Well, that’s just a tech. No, you have to look at the business model. And we’re not going to really know that until we see the business models of these companies emerge and get clarified, or at least know what they’re going to turn into which I think we can tell at certain points. And so if you ever look at any of my, I use the same sort of six levels graphic all the time, and it’s got sort of six levels on the right, I’ll put an example. And but on the left, I always have types of technology that are emerging, and then how do they impact the levels? I’m always looking at? What is the newest technology or trend? And how does it change the business model at what level? But I don’t really ever ask the question. Does the technology have a moat? No, we think business models, I think that’s kind of weird the way they’re talking about it. Okay, so let me give you my conclusions. And I’m going to give you about three to four conclusions on sort of what I think’s going on. All right, number one. This is about a new technology emerging and developing. It’s chaos. It doesn’t have anything to do with businesses. It doesn’t I mean, this is people in laptops late at night creating stuff on Tuesday that didn’t exist on Monday. And it’s advancing quickly. And that 75% of what we’re seeing is just a new technology emergence as its own phenomenon. That has nothing to do with Moats. It has nothing to do with business models, it has nothing to do with competitive. That’s the overwhelming phenomenon we’re watching. And we’re gonna have to let the dust settle for a while. Before we know so, you know, that’s the number one thing I’m thinking about. Number two, the first major impact of this new technology is there is a quantum jump happening within operating performance. There’s a I mean, it’s a world changing event. So if you look at my standard models, I put structural advantages and moats on the top, I put operating performance on the bottom, and I symbolize them by Warren Buffett on top, Elon Musk on the bottom, operating performance, who’s faster, who’s quicker, who’s smarter, who can execute who can generative AI, and particularly digital agents, which I talked about last week, they are fundamentally changing the nature of an operating entity, where I’m basically replacing humans, with digital agents that can do things that humans do. If I build a business with 100 Digital agents and three humans. That’s an entirely new entity in the world. And the operating performance we’re seeing across the board is just leaping forward. You know, one graphic artist used to take weeks to do one character. Now they can do 50 In a day. You know, one doctor used to have to go to medical school and then train forever. And generative AI GPT. Four can pass the medical licensing exam at a 90% level. So at the individual level, at the worker level, at the business level, at the project level, we are seeing a total transformation and operating performance, where everything is going to be dramatically more powerful. So like that’s the first step to quantum jump in what we consider the operating performance of anything
Jeffrey Towson 30:00
I mean, I’m doing this myself like, I’m literally because I do a lot of content creation, writing, thinking modeling all that analysis, I’m building out a team of digital agents to do this, I used to be me and several people that I’ve worked with, I’m literally building out an entirely new structure of, I’m using about six to seven digital tools right now that I wasn’t using a month ago. And I’ve literally changing everything about the way I do it. And I’m starting to deploy digital agents that operate on their own everyday without me telling them what to do. And that’s kind of so I’m doing this, even with my own little small operation. So that’s number two. All right, let’s go to the app level, because now We’ll answer that, I’ll try to answer the question. When we look at the app level. I sort of put these into three buckets of what I think to happen, is going to happen. Number one, that within this, I think, dominant, successful companies with big mozz today that are not being disrupted, I think they are going to be the biggest winners of generative AI, I think their adoption of the technology. It’s one thing if you’re using the technology and trying to turn it into a moat, it’s another thing, if you already have a moat, in your core business, and you’re putting in this tech to make your business even better, and it’s not being too disrupted. Those are the winners. So who is that? That’s Adobe, that’s Microsoft. That’s gaming companies, that’s Tencent, they already have powerful business models. They are putting this tech directly into their features already. So they’re going to get the best of both worlds. They’re going to have all these new capabilities, they’re going to be able to pass that value on to their customers, which customers are going to love. And they already have a mode. So if this thing turns into another mode, beautiful, if it doesn’t, that’s fine. Now the caveat there is they can’t have their moat already being disrupted by this tech. In that sense, a company like YouTube might be in trouble. A company like Microsoft is going to just be phenomenal. A company like YouTube has a trouble because, yes, they have an existing model. Yes, they can adopt this technology into their model, which I’ve talked about. But this tech is also going to disrupt to some degree, their core business. So we want the scenario where it’s like, we get the best of all worlds. And I think that’s Adobe, an example of this would be cap cut. I mean, tick tock is a cool business model. Right, powerful business model, great cash flow, great service, network effects, long tail, all of that. They went from there into cap cut, which is just a video editing tool that they started giving to video editors, people who make videos, I use it. It’s great. So that was an additional technology service that complemented their core, but didn’t disrupt it. That’s the scenario we’re looking for. Now, there’s a famous quote by Philip Fisher, who’s kind of the, you know, one of the first technology investors, and he had an old quote that said, quote, it is as powerful to invest in companies adopting technology as those creating it. That would be this strategy, basically. And I think that’s kind of what Andreessen Horowitz said, they kind of alluded to the same thing. So we’ll call that companies that I don’t think are going to win. I call these the spellcheck, companies within apps, all these free tools that people are giving out. If if the development curve and advancement of the capability flatlines. They’re going to be commodities. So that’s spellcheck. Someone invented spellcheck. It was great. But the capability didn’t keep advancing. So it flatlined, and then everyone copied it. Now you get spellcheck for free. I think a lot of these apps are in the spellcheck category, however, runway and mid journey, which are apps, I think the development pathway where they can, I think that’s going to keep going so that I would describe as a digital marathon. There’s enough development possibility that if you’re out in front running faster than everyone, you can perpetually stay ahead of everyone because the capability keeps advancing. So those new free tools the spellcheck not good to companies with the sort of long runway for development advancement, like the mid journey, although I’m not sure about mid journey, but I think much more runway am sell better. And that really is the third group, which is new innovative services, which I would call runway that type. If they can develop proven business models with cash flow, if they can get themselves if they can go from being very successful in the digital marathon, to building a business model we’re familiar with that has moats that I think they get there. And that’s kind of how I argue that like these competitive defense abilities happen at different points. The first move for let’s say, runway ml, like mid journey, is they got their first, they were an early mover, fine, nice, then they had a digital marathon they could run that kept them perpetually ahead of their competitors, fine. And then the next stage would be they get to a proven business model that we know has moats, if they can get there. And that would be the next one. So I would put a couple businesses in that category at the at the app level, but I’ll write this in the show notes, my conclusions, their dominant players with big moats that aren’t being disrupted. If they can adopt this, it’s going to be win win. Not so great free tools that don’t have a long development pathway. The spellcheck type capabilities, not good innovative services that have a long development pathway that can move toward proven models, they look pretty good. Since time in China is like this, actually. Companies that I think do this well are when they’re not entirely digital creatures, where it’s all software, because software is very easy to replicate. If it’s a combination of hardware and software, which sense time is in China. That tends to be a stronger position for this. And it was okay. And then we’ll get to the last point, which is, what about the foundation models, which, you know, Google barred open AI GPT. I’m actually fairly optimistic on this. Because I can see one and maybe two business models I already recognize. And the first one, well, let me the one I’m not sure I recognize is a learning platform. concept for today. I’ve written kind of a lot about Baidu, Google search. And I’ve described those as learning platforms, which is, it’s not a great definition for a platform, I struggle with this one. But platform business models, you are in the interactions business first. And you can have different types of interactions, transactions, that means you’re a marketplace. If the point of the interaction is to increase the intelligence of the service, than I call it a learning platform. So Google gets smarter, the more people use it. And that’s not an ancillary benefit. That is the core service. So people do lots of longtail searches. You know, the crawler goes around and looks at everyone’s Search Engine Optimization checks. And we can see a couple businesses like these, Cora Zhu who in China, which is, you know, one group of users is creating answers to questions that would be Quora, and Yahoo. And one group of users on those same platforms are searching for answers. And you’re in the business of connecting suppliers of information with people looking for information. And the more interactions you have, the better the answers get. That’s what I sort of call a learning platform. But that my thinking is a little fuzzy on this one. It’s not as clean and simple as marketplaces. I think that’s what we see with Google. And I think that’s why Google has 90% market share winner take all. I think we see it in Baidu, I think we see it in cacao in South Korea search engine. I think we see it in Yandex in Russia searcher and we see it in a couple other interesting bits. I call these like the aardvarks. Like certain businesses are pretty recognizable like marketplaces, Amazon Taobao Tmall. To me, those are all like big cats, like lions, tigers, jaguars, they’re different, but they’re all big cats. This little bucket of learning learning platforms, I consider these like the aardvarks and the porcupines. They’re just strange, strange animals. I think they’re kind of hard to categorize. They’re one offs almost. So I’m trying to decide whether what open AI is doing with GPT and as well as Google Bard, is going to end up becoming a learning platform.
Jeffrey Towson 40:01
And I don’t know, because it’s not an Aardvark, it’s a different animal entirely. It’s still in this strange category. But it is definitely different. And I’ve got a question mark on it. That’s what I’m trying to figure out. If it is, then I’ll be able to predict what’s going to happen, because I know how those models play out. But I’m not there yet. So I’ll call that may be one business model that I recognize. If it has that it’ll have network effects. It will have scale advantages, it’ll have a lot of strengths. I think that’s probably true, but I can’t quite get there. Now. The other type of platform, I think we can see, this one’s very clear, I think, open AI, GPT. And Google Bard are clearly innovation platforms. innovation platforms are things that others built upon. So Windows, developers build on Windows, they make all the software app developers built on iOS, we’ve seen this model over and over and over game developers build on, you know, Unreal Engine and other things. Right, we know what an innovation platform is. Both of these proprietary models look like very clear innovation platforms. And the fact that there’s an open source, I don’t think that’s a problem, because we can see that with Android. We know this business model. I think we’ve seen examples, particularly on the enterprise and b2b side of proprietary innovation platforms, and open source capabilities working very well together. And we can see in the last couple of weeks, open AI in particular, has really been moving down this pathway. They’re starting their own sort of plugins. They’re releasing the API’s, and they’re creating some sort of play store that will let others build upon them. So I think we know this model very well. And it looks exactly that looks like what they’re doing. Now, they may not get there. Which is why I’ve kind of said, I think it’s likely, but I’m not 100% Sure. But at least it’s clear what the winning picture looks like. Which is not the case with the learning platform, at least yet for me. But at least with the innovation platform, I know what winning looks like, and I can see their progress towards it. And if they get there, I think there’ll be a Giants and dwarfs scenario. And I think it’ll be quite good. So that is kind of where I end up on all of this. I know that was kind of a lot. I will write that in the show notes. But net net. Yeah, I think Google Bard and I think open AI GPT are gonna become some combination of an innovation platform, almost for sure. And maybe a learning platform. But I struggle with that business model, because I find it very hard to get my brain around. So yeah, and then on the app side, I think the clear winners are the proven existing business models that aren’t being disrupted. Who are the early rapid adopters, Microsoft, Adobe is doing really well. Microsoft, GitHub is absolutely fantastic. So we see those, and then there’s a little bit of a question mark, about some of these standalone app services, like mid journey and runway that might be able to break out and move from digital marathon to competitive advantage. And that’s what I’m kind of keeping an eye out for. Anyways, that’s where I am on all of this. Dojo is going to be a lot of theory today. The emails I’m sending to subscribers, I know it’s pretty, it’s gonna be pretty long. But I don’t know, I feel I feel like I’m 60% of the way there on this one. I think I’ve got some of it figured out. And I’m relatively confident the hard part about this is you can’t really see the data if you’re a guy like me. You know, if you’re, if you’re a venture capitalist, you get a lot more granular look into what actually is happening in these companies. So I’m, I’m guessing a bit here much more than I normally do. Anyway, so that’s where I am the two key concepts for today, innovation platform learning platforms, you go to the concept library. I’ve written a lot about both of those. If you have any thoughts on learning platforms, let me know. I think that’s a really interesting subject since time in China. I’ve thought about this for a long time. Waze is another example where I kind of looking at these, but I’m still kind of fuzzy on it. And that is it for the content for today. As for me, just a good week, running around like always, things are hopping in Asia, which is a lot of fun. I’m thinking about sneaking off to Kunming, which is sort of southwestern China in doing a little Chinese. I got my Chinese is really degraded over the last several years and it really annoys me because I was pretty fluent a while back and now I am now So I’m thinking about sneaking out there. That’s where I actually learned. During the financial crisis. I took a couple months off, and I just went to Khun Ming. And I just studied Chinese for about four to five months. And that’s kind of how I learned. And I’m thinking of going back there. It’s a nice city couldn’t mean the southwest of China’s eunan, which is southwest conveniens, the capital and there’s a lot of famous tourist sites down there at least Jiang shi shuang bond. And it kind of looks a little bit like Thailand, actually, because it’s right near the border. So if you go to see Shuang by now, which is sort of near the border, I mean, there’s elephants and other things. So I’m thinking about trying to sneak away for a couple weeks, maybe sit back there and the other weird thing about Kunming is it’s a big tobacco city, like there’s huge tobacco factories there. So everywhere you go, at least back then everyone was smoking like crazy, which was kind of all the young people were smoking. And, yeah, it’s kind of a strange place, but it’s a nice sort of quiet place. And the air is real nice. So I’m trying to figure out if I can sneak away and do that or not. Anyways, I’m literally mulling this over as I’m sort of planning my trips in China and around Asia in the next couple weeks. We’ll see anyways, random. Okay, that’s it for me. I hope everyone is doing well hope this was helpful. Take care, and I’ll talk to you next week. Bye. Bye.
I write, speak and consult about how to win (and not lose) in digital strategy and transformation.
I am the founder of TechMoat Consulting, a boutique consulting firm that helps retailers, brands, and technology companies exploit digital change to grow faster, innovate better and build digital moats. Get in touch here.
My book series Moats and Marathons is one-of-a-kind framework for building and measuring competitive advantages in digital businesses.
Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.