Data First Venture is Coming

Speaker

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

Speaker

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

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month

Speaker

of July, I've got the month of July, I've got the month of July, I've got the month So here's the problem. How good is a company at the seed round? You don't have a lot of information. Firstly, there isn't a lot of data, but there's even less information about a company at the seed stage. Maybe you know who the founders are. Maybe you know who the team is. Those are not really actionable insights that you can learn from, although some people claim they are, they really aren't. You get to the A round, it's roughly the same. The So data can't really play a role in early-stage investing in a meaningful way. At the seed and the A round, it's mostly human. It's really smart people who know what a good one looks like, and that's the combination of what the company does, the technology opportunity, the risks, the market size, the timing, all those things that you take into account early days, that's all very human. But by the time you get to a B round, which may be three years into the life of the company, that company's history tells a story made up of all kinds of variables, who its investors are, how much money they invested, how many times they've invested, what the valuation growth has been. And you've got real data that you can learn from. And if you take several years of that, you can train a model in what the patterns are that lead to a good outcome. And that's really what SignalRank does. And that's more and more happening in venture capital to the point where I think data will be 100% able to allocate to Series B, C and D rounds. Right now. Right now. But that's your pitch for SignalRank. But now it's growing beyond SignalRank. There's a lot more people saying the same thing. We all have very different approaches to the meaning of the word signal and what the predictive outcomes are. For example, we predict the multiple of invested capital at the five-year mark. That's our target. So in data science, you need to have a target. The machine has to have a target. We say, OK, tell us what you think the multiple of the money invested at the Series B will be five years from now. And we need that to be at least six times. OK, so how does this play out in other areas? So, for example, why couldn't you have a SignalRank style company for HR where everyone's work, everyone's jobs become datafied so that when an employee, an employer decides whether or not they want to hire someone, they can run the numbers? Well, I think where you're thinking about human beings, hiring individuals, and it's the same for individual investments, by the way, you can't really use data if you want it to be accurate for a single event. Even at SignalRank, unless we make six investments, it's pretty random what the outcome will be. So when you're hiring, you're hiring one person for one job. It's almost impossible to use data and expect to be... So in a way, you're replicating the VC model where they know that if they make 10 investments, they'll do well if two of them do a home run or one of them's a home run and a couple of them are doubles. A little bit different because most VCs do way worse than one in 10. But let's assume one in 10 is normal. What we do is we're kind of cutting across the VC funds and we're looking for the one or two out of the 10. So we're already scoring into that top 5 to 10 percent of their investments. So when we make an investment, we're avoiding the nine that don't make it. But ultimately, you're still predicating the value of what you're doing on recognising smart investors. I mean, later we're going to talk about Sequoia. There are certain VCs we're going to talk about Fred Wilson. There are certain VCs that just for one reason or other pick out winners endlessly. Yeah. Well, that is what we learn from. So if you look at the table I put in the editorial, I only put the top 20 seed investors just to protect our data. But we basically learn from the best investors. And I mean that in the plural. We don't learn from an individual investor. We learn from thousands of investors who are doing C to A round investing and we score them based on co-investing with each other and the outcomes that that produces. That's something we call round score. And we can replicate round score having learned what it looks like from decisions that we take today that have yet to pan out. But ultimately, your algorithm is also your version, your view of what to value. You could quite easily have competitors to SignalRant doing exactly the same thing, who just build different kinds of algorithms. Yes. So that's part of what I say in the editorial is in the future, allocators are going to pick the best algorithms to invest in, not the best funds. And then there may be a correlation. But if you're allocating to, let's say, Series B, C and D investing, your question won't be how good an individual investor is doing, because increasingly that won't be relevant. It will be how good is the model they're using to select the companies that are their targets? And then secondly, can they get into them? And that's the human bit, by the way. Data can't help you with getting into them, but it can help you with selection. You could have meta versions of SignalRant where you have an algorithm analyzing companies like SignalRant and determining which is better than the other. And of course, appropriately enough, you create a machine learning image. We joked earlier, we were talking earlier before we went live. You think these kinds of AI created images will undermine graphic designers. But I actually, for people who are listening, it's an image of a fake person, a fake young woman in dark glasses with all sorts of electronics in her hair and her head. I wonder whether actually this revalues the graphic designer, because the more images like this we see, the less valuable they become. That's true. Yeah. Great, great design will always be great design. And you can't, you can't. And won't the same also, Keith, become true for venture capital as everything gets more and more automated and data driven, what you call data first. The real value of intuitive fellows like Fred Wilson, who you have a piece from this week, will only be compounded. More and more people will want to know Fred Wilson's wisdom, which can't be replicated by an algorithm. What does Fred say this week? You quote him on how this ends part three on the VC downturn. What's his argument? He's arguing that we may have reached the bottom in the in the public markets. And he quite rightly understands that investing is a continuum. The public markets before the public markets, there's late stage investing. And before late stage investing, there's mid stage investing and seed investing. And he makes the point that there's a delay always that starts with the public markets. When it goes down, a year later, venture has gone down. When it stops going down and year to date, tech is very up in the public markets. A year later, venture is back. And the reason it's up is because all the big companies are doing well. You have some interesting pieces on all of them. Apple is developing its own chat GPT style bot. Microsoft is developing what's called vector search and voice cloning. It seems as like this AI boom is slightly different from previous ones, where it's the big companies that are leading the pack. Yeah, absolutely. Because it's so expensive to do. I mean, it sometimes takes billions of dollars of computational power to train a model on something complex like voice cloning. And so startups can't do it. What startups can do is open source stuff. And there's a lot of that happening like this week. Facebook open source is called Lambda 2, which is its large language model as a way of giving it to the community so people can play with it. So Facebook, let's be frank, Facebook never gives anything away. There's no altruism there. I'm sure that they have some reason for doing it. Yeah, there's lots of theories. The best theory I've read is that they want higher engineers that learn how to use Lambda. Do you think that eventually there'll be startups providing these services, Amazon Web Services style products where this stuff will become more accessible and affordable to startups? I think what's going to happen is the ability to train on smaller models that require less computational power will emerge. And secondly, once the models are trained, they can run on smartphones. You don't have to run them in the cloud. And so intelligence will increasingly be pushed to the edge. For example, if I want to code in Python, I can now do it on my iPhone as long as I can type into it and so on. So I think you're going to have competing trends and they're all going to exist. There's going to be huge models that need a lot of money. There's going to be smaller models that are very specialist that need less money. And then the trained models are going to get embedded in applications at the edge of the network where anyone can use them. I forgot to forward you my keynote interview with Hillary Mason, who's a top AI entrepreneur and technologist in New York. She has some interesting ideas. We did the interview earlier this week. Viewers and listeners should check that out about the possibilities of open source. She thinks that this can enable startups like herself. What about ChatGPT, which has its cake and eats it? It's simultaneously a startup and part of Microsoft. The news this week is that ChatGPT allows it now to remember who you are and what you like. It sounds like Google. Sounds like a search engine. Well, it's personalizing. I've actually gone and done it. And to be honest, it's fairly limited right now. It asks for your hobbies and what you work at and what your primary goals are. Nothing very usable, I would imagine. And very lame. Yeah. And it also asks you, do you want formal or informal conversational style? It isn't really a big deal, I don't think. But the idea of personalizing an agent to yourself, there is a big idea there. This week I used it a lot to run queries against data. And I had to train it every time on the same data because it doesn't remember between sessions. I still think it seems, and I talked about this with Hilary, we're still at a pre-Netscape moment. What was fantastic about Netscape, and I remember it very well in the early 90s, is before Netscape, it was really only geeks like you could use the Internet. Because it wasn't easy to do. You could do it, but it took a lot of effort. And the same is true today of AI. But we still haven't come, maybe we never will, to that Netscape moment where accessing all this new technology becomes idiot-proof, essentially. Yeah, no, I agree with that. In fact, my first big success in business was building EasyNet, which included a browser and an installation disk that let people learn what the Internet was without needing to know about it. And if what you just said wasn't true, EasyNet would never have become a billion-dollar company, which it did in 1999. So the opportunity to make money really is in those gaps between what's possible and what's being delivered to normal people. And so today with AI, I can geek out playing with it. I spent a lot of this week doing that. For an endgame, which is trying to have it help me build a model. But the future really is when you unbundle specific things like venture investing or diagnosing cancer or reading DNA. When you unbundle specific things and make them usable either by businesses or by consumers or both, that's where the real value is. So the job really is to jump ahead of where we are today and imagine products that will be helping humans do hard things that are of enormous value. Yeah, we know that the alarm bells are going off in all the big companies from Amazon, certainly to Google, to Apple. Do you think, I mean, it's hard to know, but is Apple's generative AI chatbot, is it likely to be another Me Too product? Or will Apple really work on it as they did with the iPhone to revolutionize this business too? I predict failure for Apple. I mean, Siri looks like a bit of a hobby now. So does Alexa. And Alexa. So I think what's happened to Apple is they've become the Blackberry of AI. We've seen the movie Blackberry. Except that they don't care. It's not like Blackberry. They don't care because it's not their core business. So they're not that damaged. That's an excellent movie. Did you see it? I did see it. I thought it was very good. So, you know, Apple probably has to play in this arena. They have to. Not probably. They have to. Everybody does. They have to. Will they be any good at it? The jury's out because the same with Google, by the way, and the same with Amazon. And Microsoft. And Microsoft, yeah. So ChatGPT is very far ahead and a couple of the other players. And they've got to figure out, you know, how to play with the big guys or whether or not to play with them. That's always a hard decision. And we'll see. But I wouldn't hold my breath for Apple. I think this is going to be a whole bunch of new companies. The hard thing is getting the scientists to collaborate with the business people who will create the best products. Well, that's always the case. You had a really interesting link on the newsletter this week to a LinkedIn post by Kyle Poyer. I never heard of him, who says that, and I'm quoting him here, the AI craze will implode faster than threads if folks can't figure out how to make money from AI. And he notes that ChatGPT's traffic dropped almost 10% in June. What are the business models here? There's no new business models. It's either subscription or sales. Yeah. It isn't advertising because you can't really advertise in the middle of a conversation. So it definitely has to be subscription-based for consumers or enterprise license-based, developers building APIs or STKs and having those embedded and getting paid for that. Ultimately, the money will come from engagement, though. Does this solve a big problem for a lot of people, businesses or individuals? And if it does and the problem is valuable, people will pay for it. I pay for ChatGPT without even blinking because… Yeah, but you're a startup entrepreneur and you're a programmer. It's useful to you. I mean, I would pay for it if, as a more mainstream user, if I could have, and this is what Altman said, I think it was the week before this, that I had a personal assistant. I don't want to pay someone in Minnesota or the Philippines $30 an hour and have to deal with them all the time. But if I can have a personal assistant that knows what I'm doing, that can manage all my bookings on Keynote, send the emails out, do the social media posts and all the rest of it, then it's a no-brainer. Otherwise, why am I going to waste my time? Yeah, you're not wrong. I think the big thing that no one's yet realized, so this can be one of those things where we said it here first. We always say everything first, Keith. That's why we have such a huge audience. Exactly. So last week or the week before, OpenAI released what's called Code Interpreter into ChatGPT. And what Code Interpreter is is basically a brain. Before they had Code Interpreter, the machine learning model knew lots of stuff and it could figure out how to answer questions, sometimes right, sometimes wrong. It couldn't do math, for example, at all. And so the fact that it couldn't do math is because it didn't have a brain. What Code Interpreter is is a fully functioning Python executable code environment. So now when you say to it, you know, what's the square root of 27, it gets it exactly right because Code Interpreter can write code and tell you the answer. And, by the way, it can read URLs, it can do summaries and all kinds of other stuff because it has access to the Internet. So Code Interpreter is really a massive, massive upgrade. We're getting a brain. That's sales talk. I mean, it's still going to take a while. I mean, realistically, leaving aside the concreteness of a Netscape moment, are we talking 2 years, 5 years, 10 years? What do you think? I don't think it's going to be like that. It's going to be closer to the Internet than it is to a single app like Netscape. With the Internet, if you look at the user growth and the number of websites over time and the number of domain names over time, and then think about AI from the point of view of the number of neurons, electronic neurons, being applied to knowledge and data, the growth of the number of neurons put to work and the amount of knowledge that it sits on top of, including infinite knowledge that Code Interpreter gives you because you can make new knowledge up by writing code, I think the real measure is how much of what we see, read, hear, act on is produced by computers is going to go from 0% to eventually probably 70%, 80%. We have the legal, and we've talked about this before, and we will talk many times about it in the future, the legal dimension here, what their rights are. That's a really interesting and important conversation. Speaking of Meta, I noted that they're not the friendliest of organizations. They've unfriended the news industry. I didn't know they were ever friends with it. What has Meta been up to when, in terms of undermining journalism, it's already done enough damage? Well, look, the most interesting thing about this is that Meta has decided based on user feedback that people don't want formal news from formal news organizations. They're much more interested in what their friends' opinions are about things, and so people are saying they want to engage with other people around topics of interest and not have formal news. So that really is defining a social network as being different to a media platform, and I think probably their stats are right, that that is true, because, for example, when I go on Facebook, I don't go there to read the New York Times, but I do read the New York Times, just not on Facebook. And so I think the insight there is social networks are not for news and politics primarily. Now, that's a new thing for Facebook to discover. I know you're acknowledging that because you've always been telling me that that can't work on a social network. You were very critical of threads, and you're actually right, Kee. You warn that threads will fall as quickly as it rose, and the journal headline from July 21 is that it's already losing its allure. Everyone signed up, and no one's actually using it. Do you think threads will go away as quickly as it came? Well, there's a message, by the way, from one of our listeners, Stuart Soffer, or Soffler, my eyesight's bad, Soffer, I think, saying keep the faith. Keep the faith in what, threads? Well, definitely not in threads. Threads is really an example of Zuckerberg getting it wrong. His instinct is super good, and he often gets it right, and you should be giving credit for that. But when he copies, he almost always gets it wrong. I can't think of a successful copy. Is threaded copy just a copy of TikTok? I think it's trying to be a copy of Twitter, and he has Reels. Is it Reels, whatever that was called, that is now becoming video, and they're changing the watch tab to a video tab. So they're going to have a TikTok on video, and they're going to have a Twitter-like thing on threads, and neither one is going to do very well. Do you think, though, that he's playing with casino money on threads? He doesn't really care. He probably didn't spend a lot of money on it. It's all built off Instagram anyway, so who cares? Well, everything he spends is, in that sense, casino money because Facebook isn't- No, because with the meta idea or the metaverse, he could lose big time. There's no core business there. But with Instagram, all it is is a social network built off Instagram. It is. That's all it is. But what I mean by casino money is Facebook mints billions of dollars every quarter, free cash flow, and its investors will let it spend a sizable amount of that on experimenting with the next growth wave. And so they can almost try anything. But also, we talked about this before, would it also be fair to say that the news, the Wall Street Journal and the Times and all the other journalists, mainstream and other journalists, their obsession with threads has nothing to do with threads. It has to do with Twitter. So the interest, the story here is not the rise of threads. It's the collapse of Twitter and many people's hatred or ambivalence about Musk. Well, I agree with the second part of your sentence, but not the first. I don't think Twitter is collapsing. I didn't say that. I just said that because you're in the Musk camp, but I'm saying the perception or the hope that it collapses. Yeah, yeah. So, yeah, I think hatred of Elon Musk is the key driver. And it's interesting. I mean, why do people hate Musk? They really hate him because he's not prepared to cancel right-wing views. That's the only reason they hate him. I'm cancelling you on this. Too boring. Let's move on to Startup of the Week. Sequoia, of all places, one of the blue-chip venture firms, Keith. What has Sequoia been doing? They're shaking themselves up. The end of Mike Moritz, one of the most legendary of investors, the guy who put money first into Yahoo and Google and many other companies. Yeah, so it's a generational shift driven by some short-term catalysts. I don't think the short-term catalysts really are, you know, enough of a story to explain all this. They've lost two of their crypto investors. That probably is driven by their FTX investment and the catastrophe about that. And Sam, I mean, maybe for next week, Sam Bankman-Fried is – I mean, he always looked like the creepiest of guys, but the news this week makes him sound even creepier. It's astonishing. The one about buying an island. Well, that's his brother buying an island, but now the headlines today is he's undermining this partner he had, a partner-girlfriend. So he really is a nasty piece of work. But that's another week, another story. Well, one of our readers is asking, what's this about? So my advice is to go to thatwastheweek.substack.com and look at a couple of past episodes and then you'll figure it out. We can't really answer that yet. But yeah, no, I think Moritz left a long time ago, honestly. So that is just news catching up with fact in that case. But I do think the primary narrative, which is that Rulof Mutter is stamping his authority on the next version of Sequoia, that's accurate. And he's had some bad press recently. Is he doing a bit better? I know he's a personal friend of yours. I mean, he's a friend in the Silicon Valley sense that we know each other and I certainly respect him and I like him as well as a person, but I wouldn't say we're friends. He's a former chief financial officer who's had a long history now of working in venture that is now the boss. So the bar's raised and his skill set, to be honest, is still growing. He's not a finished product. So this is his chance to shine and we'll only really know five years from now, maybe 10 years from now, how well he did. Any chance in five years, Keith Signal Rank will have acquired Sequoia? No, not at all. But you are in all seriousness, and this comes back to where we began, this data first investing. Are companies like investment VC firms like Sequoia, are they beginning to understand this or recognize at least what you're saying? Not everyone probably believes it, but is this argument beginning to filter down? Permeate? Yeah, no, I think the venture community, especially those who invest late stage, will be the last to believe that data is going to drive investing, certainly not lead investing. So I wouldn't hold my breath for, you know, Vinod Khosla or Mark Hendricks. Is that your way of saying they're dinosaurs? They're not dinosaurs. It's a transitional phase. They're increasingly less relevant late stage, but by the way, Sequoia is a really, really good early stage investor. Very good. One of the best. As is Andreessen. And that doesn't go away. So it's just that the game changes. That human instinct that they all are very good at. So why do you make, I mean, you told me that both, Bolof, whatever his name is, is not really ready yet. Moritz was already out the door. Why do you make them start up at the week? What's the big deal? Well, because it's a startup now. Sequoia, the name feels like continuous, but Sequoia in July 2023 is now a startup with Ruloff at the head. And let's see how it does. You're not confident, no, are you? Doesn't sound it. I'd be somewhat confident. They're smart people. That doesn't sound very confident. What does your data model say? What happens if you ran Bolof through your thing? Ruloff himself doesn't score very high at this stage, but some of his colleagues do.