Changing the World - One GPT at a Time.

Speaker

I've got sunshine on a cloudy day When it's cold outside, I've got the month of

Speaker

November the 10th, 2023. Not too many Fridays left in the year, but it's Friday, November the 10th, 2023. But if it's Friday, it must be that was the week. My regular weekly show with Keith Teer from Palo Alto on events in tech. This week, it didn't seem to me a particularly dramatic week, but Keith seems to think in the newsletter that it represents this week, and I'm quoting him here, world changing technologies, Keith. What's happening this week? What's so dramatic about the week of November 10th, 2023? Why would we look back at this and think that's when the world changed? I think broadly speaking, this will be the day, similar to the day the iPhone came out, this week will be the day that you can mark the change of software user interfaces from keyboards typed into webpages to what ChatGPT launched this week, which is conversational AI on any topic you can imagine, leading to the possibility of intelligence at a very granular level for everything that the human race knows about. Wow. I'm not usually impressed, but explain. So what exactly did ChatGPT bring out? What did OpenAI bring out this week to trigger this change? Well, the first thing is what they didn't do. They didn't upgrade their ChatGPT4 model. It's the same model. What they did is they launched the ability to do two things. Firstly, they integrated images, text, and computational power into a single chat interface, whereas before you had to do either images, or you had to do text, or you had to run programs. Now you can do all of the above in a single interface, and it will figure out what it needs to use to deliver against whatever your goals are. So it's a much more fully formed equivalent of a human being. It can deal with all media and all computational power in a single interface. How do you call this thing? Say it again? What is it? ChatGPT4 or 5? It's ChatGPT4, but with a new level of capability. And then the second and probably more significant thing is they've made it possible for you and me to create our own GPTs based on that functionality for any topic that we are able to train it in. So for example, this morning I made a SignalRank AI version. I gave it some data from SignalRank, trained it, and it can now answer questions about SignalRank and the data in venture capital without me having to be present. And it's cleverer than I am at pulling together the appropriate information for whatever the questioner is asking, including charts and tables. I've never heard you acknowledge anything's cleverer than you are, Keith. Yeah, my wife is way cleverer than me. You have to say that. Jeanne is under the table there. So you say in the newsletter that, and I'm quoting in your news, how you're describing a newly scalable architecture that opens up the possibility for anybody to have their own GPT. Developers will be able to both build enterprise and consumer GPT applications, often with no code required. So is this just basically offering the OpenAI API to anybody or am I thinking about it wrong? Indirectly you're right, but there's no need to understand what an API is. They give you a chat interface for building your own GPT. So you'll go into the chat and you'll just use English and you'll say, build me a GPT for Mediterranean recipes. And it will start figuring out how to do that, pulling in everything it knows. And then it'll say, you know, specialize in Italian based meals from Milan. And it will then do that. And you'll go through whatever questions and answers you want. And at the end it'll say, okay, I now have a GPT just for that topic. What do you want to call it? You'll say Andrew's Italian cookbook. So why I take, it's an interesting idea, but why is that better than me just going to the generic chat GPT and asking it about Italian cookbooks? From the point of view of the end user, it may not be better. I suspect it will be a little bit better because it'd be specialized, but let's assume for a second, it's just the same. What is different is the chat GPT has now recruited millions of developers and non-developers to build specialisms based on their preoccupations that the chat GPT employees could never scale to build. So you're going to go from, I don't know how many employees they have. Let's assume they have about a thousand employees. You're going to go from a thousand employees and what they can do to millions of people and what they can do. And it starts today. I mean, I built one this morning in 10 minutes. I already had the data. What you have to do is so anyone, a developer or an end user brings their own data to this. So you bring your signal rank data, which they don't have to it to give it more intelligence. Is that right? Yeah. So for example, this morning, I'll just use my example that somebody had already built something called Venture AI. And I used theirs and it was just using Crunchbase and news as a source of intelligence. So it's fairly limited with what it could do. And most of what it told me, I already knew because I can read the news as well. So then I did my own one where I uploaded some Excel files with actual data in and told it to build a Venture AI using this data. And it was already a million times better than the generic one that I used earlier. So yes, it can use web data or you can upload your own specialist data. So let's say you're the Khan Academy and you have math classes in logarithmic algebra. You could teach it from your data, everything to do with logarithmic algebra, and it would within less than a day, you'd have the world's best teacher for that topic. So the implications of that for how human beings learn is massive. I mean, that you would no longer need. Yeah, although we can look at the dark side, it's quite corruptible as well. Is this equivalent to the way in which, for example, Google transformed the entire internet into a vehicle for its own advertising? It switched from a search engine page to being everywhere. Is this conceptually the similar sort of thing that OpenAI is trying to do here? I think it's not. I think it's different. Here's how it's different. Let's take the total amount of money spent on education in the world today, which would be in the trillions of dollars. And let's imagine that you could do a better job of education using AI. The cost of the AI would be massively less than the cost of education. I'll just use my example. I spent $75,000 a year sending my oldest kid to college. I would only have to pay CHAT-GPT a fraction of that for an equally good education. Well, theoretically, but they wouldn't go to college, you wouldn't meet people. I mean, that's an old argument. No, but actually, I'm focused on your question, which is the economics of this. Basically, CHAT-GPT will shrink the amount needed for humans to learn and deliver it for way less money. And they'll get paid from subscriptions, just like a university gets paid from fees, but there'll be way less. So the net plus for the human race, assuming this all plays out, ignoring all of the scary stuff around the outside that you might be concerned about, let's just assume they can do it. It's a huge win for the human race. Well, I'm not so convinced. Firstly, I don't really understand how it works. So let's take your son's school, Syracuse University, where your oldest son went. Why would Syracuse University cannibalize its $75,000 a year students to create this platform? Well, that's the innovator's dilemma, the Christensen thing. The answer is they won't. Somebody else will. Well, we've heard this before with Khan Academy and online learning, and it never works out. Actually, they did both work out massively. But the universities are still even more exclusive, and schmucks like you and I still spend hundreds of thousands of dollars sending our kids to college. Well, I think that is because we live in a status-driven world where we believe that that's still important. But this week marks the beginning of the end of that world. I guarantee it. I mean, there's no doubt. This week marks the beginning of the end of what world? The world in which the label of which college you went to is a form of status. Well, could we say this week marks the beginning of the end of the pre- AI age? The pre-AI age. Alexander in the chat asked here, do you export or embed these separate GPTs? Thanks for the question, Alexander. But the answer is no, it's just a URL. So let me ask the question in terms of open AI. Open AI doesn't know everything about what you know a lot about, which is data associated with startups that you've aggregated at SignalRank. When you build, they're not APIs, when you build your own version of GPT, will they have access to that SignalRank data? In other words, is it arguable that they're cannibalizing their own value? No, they're not. Because a user who uses my GPT is paying them $20 a month.

Speaker

The data is more valuable than $20 a month. So they're not going to know everything about SignalRank. The truth is I get to decide what data I give it, and I'm not giving it my most valuable data. I'm giving it enough data to be the best venture analyst in the world, but not my predictive data, my proprietary deal data, which is where I make my money. So I'm using it to give people enough knowledge and intelligence that they understand about venture capital, but I'm not giving it so much that I put myself out of business. And no one else has that proprietary data. That's my value. And you're particularly sophisticated. But won't this only compound what social media and the internet's generally done, is create a lot of free trash? Because they're bound to come out with free versions where they get your data in return. I don't think they are. I think they're pretty consciously on a mission.

Speaker

They're not going to become an advertising platform. They're just not. So they're not going to have the data in return. They might say to you, well, you pay me $20 a month if you want to keep the data for yourself. But if you give us all your intelligent data from all your series A and B investigations at SignalRank, then we'll give it to you for free. But I think they understand that their reputation relies on them not doing that. You know, there's probably a billion likely payers on the planet at $20 a month. I mean, $20 billion a month would make them the biggest company in the world and then some.

Speaker

Well, perhaps. I have to admit, it's still hard to understand exactly what this is and why it would be better to do this than just use their broad platform.

Speaker

I would say their broad platform will be nowhere near as intelligent. Like, if you go to their broad platform and ask about the kind of questions you can ask the SignalRank AI I built, it wouldn't have good answers and you'd immediately think it was garbage. If you ask the SignalRank one, it's super knowledgeable. And so it's to their credit that they understand that in order to get really good intelligence, you have to allow third parties to build intelligent versions of their knowledge base. And, you know, that's really forward thinking. It's very different to Google, which is highly centralised. This is highly decentralised, but with some central resources being applied to it. And from a personal point of view, presumably it will allow individuals to pour their own personal data into this thing and create really much more personalised, accurately personalised GPT. I'll tell you what I'll do, Andrew, before our next session. Which is in two weeks. I will build a that was the week GPT and give it all of the stuff we've written and recorded. And we can do a live demo of how much it knows about the that was the week universe. So I should do one for Keenon. How would it help if I poured every interview? I've got almost 2000 interviews now on Keenon. It would help you by creating an interface for anyone interested to ask about any of the authors you've ever interviewed and what they wrote about and how they answered various questions. Wow. Well, it may be a historic week. According to Keith, it is this week marks the beginning of the end of the pre AI age. But other stuff happened as well, Keith. What other we've got lots of other headlines. Tech is going to get much bigger. I guess that's connected. That was an interesting essay by Paki McCormick, who you like. It's really the essay of the week, actually. I mean, there are many essays of the week but I put that one first because it stands out. We all use this phrase, big tech. And we think of Apple and Google and so on. Amazon. He is making the point that these are small companies compared to what's going to happen next. And he really is amplifying what we've just been talking about, which is what will be the impact of AI on the economy and what kind of companies are going to be built and how fast and how big. And he's making the point that, you know, it's like that old song. You ain't seen nothing yet. Canadian band. I forgot what their name is now. And my brain is now trying to remember the name of the band. I won't be able to. So I should stop. But yeah, and it's a longish, thoughtful, detailed exposition on why tech is about to get much bigger. And in the middle of a venture collapse. That's the other thing that's interesting because you've been taught, you've been and I have to give you credit. You've been saying this for a while now, that tech is going to get much bigger. A few weeks ago, you talked about OpenAI in five years being the most powerful company in the world, being worth trillions of dollars. You've also, and this is part of what you're doing at SignalRank, you've been charting what one of your articles this week talks about the death of venture capital. How are they connected? On the one hand, we're going to have bigger and bigger tech companies. On the other the end, perhaps, or certainly a radical transformation of the venture capital industry, are these parallel or are they interrelated? They are interrelated and parallel. So think about venture capital. At the end of the day, venture capital is about making a decision to put money into a company. So AI is an aid to decision making. How big an aid is a function of each individual's predilection. So in the case of SignalRank, we use data only. There's no human oversight or override on an investment decision. Whereas Kyle Harrison, who wrote that piece, and he's spectacularly good recently in what he's writing, he would be in that world where individuals are still making a big part of the decisions. Now what is going to happen, and I think most venture capitals don't yet realize this, is that their personal role is going to shrink. It's going to become less and less as data intelligence guides good decision making. In most cases, it won't go to zero because of the personalities of the individuals concerned, but eventually it will. And that's the path where any given discipline, venture capital is only one, is enhanced by the use of data and to make it better. And some people consider that scary. In the case of venture capital, it's almost a crime against religion to even suggest that data might be able to improve an individual venture capital's decision making, because there's a lot of arrogance that they know best. But they're wrong. It will. Well, how will that change the signal? Because your signal, you've got feet on both sides of this. On the one hand, you obviously rely on the data in terms of SignalRank as an AI platform for determining success. But on the other hand, and correct me if I'm wrong, isn't the whole point of SignalRank to determine the best VCs and weight it accordingly? But in this future data utopia or dystopia, however you want to describe it, they're all going to be the same, because they're going to be working off the same data. So what will happen to your own particular way of determining investment reliability? How will you determine other people's investments when the data is doing it for them? Well, I mean, the first thing to say is I think due to the nature of venture capital, we're many years away from that dilemma, because these individuals will cling on for dear life as long as they possibly can. You're suggesting they're all kind of losers anyway. So they're the ones who are going to die, probably not the ones you're going to want to recommend anyway. Well, the venture asset class is exemplified by being full of losers. I mean, the game of venture capital is a game of losing until you win. It's like gambling. The power law that we've talked about on this show, you know, it's interesting, yesterday data came out that only 1% of funds find a power law company. So not only is it true that less than 1% of companies have this power law growth characteristics, it's also true that only 1% of investors ever find one. So by definition, 99% of investors are failures. So that is true. And single rank extracts data that finds the best investors, and within their decision making the best companies, because it doesn't trust, we don't trust any single investor. So we're looking at scoring across investors. And so we basically find the best within the best. And it works. So when is this going to filter down to everything, Keith? I mean, I know it's kind of taking it to this ludicrous extension, but you're suggesting that venture capital, for example, investment will be determined by data. What about startups and entrepreneurs? At what point does everything depend on data, then no one's even going to trust a human being with a startup idea, because it's not rooted in AI. When a smart young kid has an instinct, are people just going to reject that? I think it's going to be harder and harder to disambiguate whether an idea came from a human or an AI or both. I think most smart humans, as they're ideating what to do, you know, they're trying to think about you and me, if we were brainstorming a new startup idea, what do you need? Well, you need historical context, you need a sense of what comes next in any given specialism. You need to know whether the things required to build what comes next already exist or need to be invented. If the latter, then R&D is the first step, not building a startup, and that is less attractive. So timing is a huge factor in what comes next. And then it's people. And all of those things don't go away, they'll all be required. But increasingly, AI will help you get the historical context, get a short list of things that might come next, maybe find appropriate people using data that could be part of a founding team. And even though a human being is driving that, the extent to which the human being is answering the questions will be less and less. So I think the whole idea of a solely human decision will, there will be no such thing. Oh, you're going to get nailed on that one, Keith. Let's see, we will determine this. You cited Paki McCormack's piece about tech going to get much bigger, but it seems as if everyone's need to get much bigger. You have an interesting piece from Variety suggesting that Hollywood needs to create its biggest hit ever. Is this true for every industry, for entertainment, for manufacturing, for venture capital? It's only going to be the survival of the largest. It's increasingly a winner-take-all world. Yeah. Well, this is Jason Killar, who's the CEO of Warner Media. And he's a quite prolific writer. He was one of the founding team at Hulu in the past. And he's making the point that, I mean, this is very specific to streaming. He's making the point that you can't retain the subscription fees from an audience of 200 million, which is what he thinks you need to have a good business. You can't get those people to keep paying a subscription unless you have a lot of really good content coming regularly. And he uses Netflix as the positive example of somebody that has done that. And he says that, you know, Paramount and Peacock, Comcast, and all the others don't have as compelling a product as Netflix. And he postulates that maybe Hulu could be that if they all banded together. And he's saying unless they band together and get to that level of scale and quality and regularity, they can't compete with Netflix and can't make profits the way Netflix does. Apparently, Netflix is super more, the others all lose money. Netflix doesn't. I think Amazon doesn't lose money either on its streaming. So, you know, going back to Christensen and the innovator's dilemma, what you have here is an old media company with some current streaming products admitting that the legacy of the past has created a failing business that needs to learn from Netflix if it wants to succeed. Interesting. You suggest this week marks the beginning of the end of the pre-AI age. This week certainly marked the beginning of the end, if not the end of the end of WeWork. What does it tell us the collapse of WeWork? We've already talked about Sam Bankman-Frieden, a kind of collapse of crypto. When the new begins, the old ends. Is there some historic significance to the death of WeWork, Keith? You know, it really represents the death of the era that started roughly in 2015 and ended in 2021 from the point of view of the preparedness of investors to underwrite large checks on capital intensive businesses in the case of WeWork buildings. But what Alex Wilhelm does in this piece, which is why I published it, is way more interesting because it looks into the question, why did anyone ever believe that WeWork could be a the unlikely scenario that WeWork was a profitable business? So it's really, I think it's a very good piece of work that he wrote. It's not just a headline piece, it's a deep... It's always easy in retrospect. Back then everyone was saying, Oh, WeWork's the future future sharing of work. Today it's AI, this open AI model you suggest can create, turn open AI into the most valuable, powerful company in history. In retrospect, these things are always obvious. Well, one wonders to what extent the virus, the COVID virus was the main catalyst for failure here. Alex argues that the core business model was broken. I think you could make a counter argument. You could say that this was a very expensive business to build and it certainly was going to make a loss for a long time. But ultimately, if people abandoned long-term office leases and replaced them with desk leases, that the owner of the buildings would have created what is effectively a timeshare business at great scale. And it could possibly have done what Uber has done this week. Uber's it's now being described as a cash machine. Although up until this point in history, it was described as a loss-making public company. It's possible that WeWork could have gone through the same trajectory. I don't think that's impossible. I think... Airbnb is going through a crisis this week. I mean, the one variable here, Keith, which we've talked about before is regulation on AI. One of your videos of the week is from your pinup, Lena Khan, the FTC chair on AI regulation. You may poo-poo or dismiss her, but the issue of regulation isn't going to go away, particularly if you're right and this is going to radically transform everything. I think regulation is a little bit like chasing a cat. The cat's faster than you, it can jump higher, and the chances you can catch it unless it wants you to catch it are very low. Unless you're a big cat. You can give it a bowl of milk. Well, if you're a lion or a tiger, you can catch a cat. True, but let's not pretend for a second that the Lena Khan's organization... Well, the big tech, the current big AI companies are actually in some way sympathetic to regulation. We've talked about this before, because in some ways it might actually make them stronger. Yeah, yeah, they are open to regulation, mainly because,