Dec 15, 2023 ยท 2023 #42. Read the transcript grouped by speaker, inspect word-level timecodes, and optionally turn subtitles on for direct video playback
Edit labels for this show, save them in this browser, or download a JSON override for the production folder.
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker
Hello, everybody. It is Sunday, December the 10th, 2023. We haven't done a That Was The Week, a summary of tech news for a couple of weeks. Keith and I were both on the road. And now Keith has or had COVID. And the focus of his newsletter this week is No Runway. And in some ways, Keith, it's a return to some of our more perpetual themes of the year about venture money rather than all the dramas of open AI. Things return to normal in tech over the last couple of weeks? I don't think return to normal, but a huge reminder to people of what moment we're living through in tech. Maybe let me rephrase it return to abnormal. Yeah, it's hard, isn't it? Figuring out what is normal, what is abnormal is a bit of a Hegelian or maybe even going back to Greek mythology times. It's a difficult task. What's specific, what is happening? A venture fund closed down this week, a very big venture fund book with about two and a half billion dollars under management and a billion of it's still active, closed down. And very large unicorn backed by Andreessen Horowitz also closed down. And there was news from Mrs. D2IQ. Yeah. It had a different name when it started. I forget what it was now. It's something like Merisphere. And then Carter, which is the company that hosts most of the shareholder tables of most startups, announced that the rate of closures is massively accelerated. So it's a week of more and more evidence that Silicon Valley is at the beginning of a very big shakeout. And is the reason for this, Keith, in terms of the articles you select for this week, is it about the kind of capital investment required for startups these days? It wasn't like in the days back in the 90s, we could all do dotcom startups on almost nothing, just off our credit cards these days to do an AI startup, for example, like the ones, Inflection AI, for example, you've got to raise hundreds, if not billions of dollars. That is true. But that isn't the cause of what's happening now. The cause of what's happening now is all the money invested in 2020 and 2021 at very high valuations in more companies than ever before, where most of those companies are stranded, you know, on a high mountain, and the tide's gone out, and they're stuck. And so they're kind of dropping like flies. That's really the cause of this. Now, there is a separate narrative, the one you just referred to, which is, what does 2024 look like? And that, you know, there is a bunch of evidence that the startups that are attractive in 2024 are going to require more capital, because they're what's commonly known as deep tech, AI being one of the instances of that, and have very large capital expenditures required to even exist. So the whole lean startup era, where the cloud drove everything to zero, it is claimed is not true. I'm a skeptic on that narrative. I'm not actually going to agree that that is a correct narrative. I think there is a subset of startups for whom that's true, but there's going to be plenty of startups that become very big with almost no capital. Some people like, what's his name, Jessica Lessing's husband, Sam, at Slow Ventures, actually claims that you may have startups that become very large, that require zero capital, and are solopreneurs, single entrepreneurs, due to the leverage you get from AI. So I think it will be too simplistic to go along with the evidence. So we go through these boom-bust cycles all the time, ever since you and I showed up in Silicon Valley in the 90s. Why is this one any different? So a lot of startups that were overvalued will go bust, and then there'll be another wave, a new generation of startups, as you say. Some people like Sam Lessing suggest that you can get to a billion dollar valuations now without any investment. What's different? This one's a bit more systemic because it's bigger than anyone that happened previously. Bigger as in the number of dollars that are going to be burned, bigger in the number of jobs they're going to go, bigger in the number of VC funds that want return capital to their investors, and bigger in terms of the number of limited partners in venture funds that won't be able to keep doing what they've been doing for various structural reasons. So it's a lot more systemic, and it leads to the question, if you look at the whole value chain of startups, from seed all the way through to an IPO, what does the future look like is a super hard question to answer. It's always been a hard one. No one's ever known the future, and when they think they've always got it wrong. Well, structurally, you've been able to explain it. I think you can't even explain it structurally now. Will there be IPOs next year or the year after or the year after that? You've never been able to know that either. You've never been able to predict the IPO market, have you? Not with any precision, but with general direction, I think you have, yeah. But once the river of money gets broken and not returned to its source, everything's up for grabs because money doesn't grow on trees. Now, there is one bright spot this week, which is CalPERS, the California Public Employees Pension Fund, is massively increasing the amount of money. $4.5 billion venture bet. So what they're investing in, which funds in particular? They haven't announced everything, so I don't know the answer to that, but it'll be the well-known brands. But remember, $4.5 billion is about 1%. It's still a significant amount of money. I mean, because you're the founder and CEO of SignalRank, it always seems to me you have an interest one way or the other, self-interest perhaps, in predicting the end of the traditional venture business. In five or 10 years, are we still going to have all these venture companies, Keith? Yeah, I wouldn't predict the end of the venture business at all. In fact, I think it could be the beginning of it thriving in a whole new way, but it's not going to look like the business in the past. Just as after the bubble burst in 2000, and at that time there was no such thing as seed funds or even seed investments, it led to the rise of accelerators, incubators, and seed funds to fill the gap left by the death of the 2000 generation. And now we live in Silicon Valley with thousands of seed funds and angels. That didn't used to be true. Are they in? Who's in the worst situation now? Is it the traditional venture capital firms with their massive investments? Is it the accelerators, the early stage people? Or is it the later stage investors? I think everyone's in some trouble, but the most is the later stage. They literally have to stop investing. And that leads to a backward set of consequences for the middle and early stages. And it's all driven, of course, by the IPO market being closed, which means there's no liquidity. So the whole thing turns venture capital into the worst version of itself, where it needs a liquidity, which is, you know... It seems to me, and this comes to one of your news stories of the week about Gemini, Google's release of Gemini, I suppose it, at least according to them, a chat GPT killer, is that all the innovation is now coming from big companies. It's not coming from startups. And even OpenAI is, in its own weird way, a large tech company, or it's certainly supported by Microsoft. It's almost a division within Microsoft. Well, it started life as a big company due to who its founders were. You know, there were several billionaires there at the founding, so it could have as much capital as required. And that isn't normally the case for a startup. So OpenAI came to life as a big company, and in that sense is what you might think of as a synthetic Unicorn, by definition. So yeah, big companies, because they have deep pockets, can innovate at scale in AI in a way that startups can't. But that doesn't stop AI being available to startups to build new businesses as infrastructure mainly. And that is happening and will continue to happen. When you look at some of the competitors of OpenAI, like Anthropic, although it's raised a lot of money, it didn't need a lot of money to get there. And even more true of some of the open source, especially the image-based AI companies, they bootstrapped by getting people to pay revenues for early versions, and are now making a lot of money and have never raised any capital. So I do think it's a varied scene, and I would expect there to be a lot of very large companies that don't require a lot of venture capital. Keith, in your lack of willingness to make serious commitments or forecasts, you're beginning to sound like an AI yourself. Are you sure you're for real? I've had COVID this week, so I don't feel very real. I'm still testing positive, as it is today. Well, we are speaking with Keith Teer, my own friend, CEO of SignalRank and the author of an excellent That Was The Week newsletter. Everyone should sign up for it, a substack-based newsletter, which Keith puts together all the most interesting links. Going back to this Gemini story, Keith, did it strengthen or undermine Google? It was hard to tell. When they released Gemini earlier this week, the Google stock price got a good jump, but then all these stories about how the release was cooked, what do you make of this release? Is it interesting in its own way? Well, a bit of context. Firstly, they have a much more powerful version that they're only going to release in January called Gemini Ultra, and they have a micro version called Gemini Nano, which will run on an Android phone. The one they released this week is the middle version, which by their own definition is underpowered compared to where they're going. That said, they showed evidence that it's better than chat GPT on various standard tests. It's clearly a player, as in it's competitive. They made a huge mistake with the marketing, and anyone that pays attention will already know this, but they released a video that was essentially a fake because it was edited highlights tied together as if it was real-time of what Gemini can do. Very impressive if it was able to do it at all. Less impressive when you realize they had to cheat to do the video, and so they undermined themselves. They shot themselves in the foot, and now the narrative out there is Google is over-marketing and over-hyping, when actually they've got a very competitive technology that's about to get a lot better, and we assume that OpenAI will respond and have a lot better version of OpenAI in the new year. Meanwhile, it was, you know, grist to the mill for our friend Gary Marcus. Yeah, Gary Marcus, who is marketing himself as a great skeptic, and he's an AI intellectual. He's a smart guy, scientist, physicist. He claims that Turing test hasn't been solved, and all this generative AI is actually not very impressive. He suggests, actually, that when you compare it to some of the earlier AIs, it's actually less impressive. Is there some truth to this? It's pretty much a debate. That is, if we were talking about the history of religion, it would be like a debate between two branches of a Christian sect about one verse in the Bible. Basically, this is in-house AI people debating some very important things that most people won't be able to grasp, but in short, I think he's right and he's wrong. You're even, you're sounding like an AI. He's right. No, no, let me explain what I mean. He's right that large language models are basically statistical word guessing machines, incapable of reasoning, that surprisingly get a lot right, I would say, and he would say, shockingly, get a lot wrong. It ends up being a matter of opinion about whether you're super impressed by what they get right or whether you're super unimpressed by what they get wrong. I fall on the side of, wow, it's amazing what this thing can do. I use it and it's really, really good for what I use it for, which is mainly software engineering. Nonetheless, it's very true that it gets a lot wrong. I have to say that I'm no great techno-utopian, but playing around with these AI interfaces, whether it's inflection AI or the anthropic one, Claude, or obviously, ChatGPT or Gemini, the Google one, to me, they do pass the Turing test in the sense that you can converse with them online. Actually, it's hard to distinguish that interaction from an interaction with customer service rep, a real human. I mean, to me, they're not pretending to be Habermas or Heidegger or Hegel. They're not pretending to be geniuses, but they are coherent and they engage in conversation, which isn't that different from most of the online marketing conversation. Yeah. I agree with that a lot. That means that there are many human tasks. Not only are they not distinguishable from a human, they're distinguishable by how much better they are than most humans at that same job. I do think there's millions of jobs, some of which are quite high skills, including teaching and being a specialist in the subject, that AIs are already good enough for. It's just a matter of building the interfaces and the tools to deliver the full version, for example, of a physics professor that is as good as Einstein in terms of knowing everything that Einstein knew. Einstein was exceptional. He wasn't a physics teacher. He was a genius. I'm not talking about Einstein. No one's ever suggested they will be Einstein. Well, if you define Einstein as a kind of a static end game, that is to say the Einstein that died and everything he knew, a chat GPT can be that person. They'll never invent anything. But only after Einstein. They can't make it. Only after. I agree. Only after. But they can do that. Well, that's already amazing that they can do that. Why is Gary Marcus? Is he just marketing himself as the anti-AI guy? What is he up to? I think it's a literal truth that LLMs and artificial general intelligence, in his view, are not the same thing. Other people, by the way, argue that they agree with him, but it doesn't like artificial general intelligence. They're good enough to fake it. Therefore, they are it. I think what you're arguing is more that. Look, if you look at the science of AI, you can argue this is not even AI. It's a statistical word engine. That said, it's so good, it appears to reason, and it gets things right most of the time. By those definitions, it is probably irrelevant. He's the literal purist standing out, waiting for the real thing to come along. That's his key driver. He's not just trying to be a pessimist. He's trying to stand up for purity. I think he'll fail because I think this stuff is good enough that nobody wants to listen to the negatives. Most humans, I know this from my show, most humans are very predictable. They don't say particularly interesting or original, even if they think they do. In many ways,
this slightly inadequate, mechanical, monotonous, predictable algorithm and the monotonous, predictable humans, they're not that different, are they? I think if you think of this as a human, including all the flaws humans have, you won't be far wrong. Including Gary Marcus, who seems to be predictable. AI, we come back to every show. There was one interesting piece for me that you linked to. Everything you know about the podcast industry is a lie. I have to admit, the more podcasting I do, the more this makes sense. What is Amanda Sibling in TechCrunch saying on this that's interesting? I put it in because it aligned to a couple of other stories. One was about Google shutting down its podcasting service. It's not really shutting down, it's shifting from audio to a YouTube video-based podcasting service. Yeah, but it's shutting down what most people think of as podcasts, which are just the audio bit. It's migrating all the content onto YouTube, you're right. Then Spotify laid off a very large percentage of its workforce. The narrative that came with that was that the podcasting bit hasn't paid off for them. She keys into some of that stuff. I think she's more or less poo-pooing the idea that podcasting is a business. You read it as well, Andrew, and you're more into this than me. Tell me what you think. I agree. I don't think it's much of a business. It's full of middlemen who only muddy the water, who are selling the promise. It's built around advertising. It's very hard. I know my podcast Keen On does quite well, but it's very hard to actually sell advertising, especially programmatic advertising. I actually have a feeling that Spotify may be one of these companies that ultimately, it may not be quite a fraud, but I think it's on very thin foundations. I'm not convinced that Spotify is a real company. Are you? No, I think it's a real company. I think it's real in a way that makes it in any way exceptional. I know what its valuation is, but it doesn't strike me as a company that has any really long-term viability. Ultimately, it'll probably be acquired by someone else. It's a typical middleman in the media world in that it doesn't originate content. It has very small margins and can only make money at all at scale, therefore, because each unit of time it sells, it makes such a little amount from. It has to sell a lot of units of time. It's a very, very low margin, but huge business. It's a big business. When it's profitable, it can throw off enough cash.
There are real businesses in this space. Obviously, Apple, Amazon in their own way, even Netflix. I just don't see how Spotify really competes against them. I think they're over-invested in podcasting. It may not be quite fraudulent, but the whole thing is so dodgy on so many fronts. You're mixing words that suggest criminality with words. No, I'm not. I'm not suggesting they're criminal. I'm suggesting that it's one of those platforms that seems to go with the wind and use whatever is fashionable at the time, but never really has a coherent business. Another way of saying what you're saying is that there isn't enough value being produced by that business for it to survive against. Right. I don't know what historical equivalents are of those kinds of companies. They always seem to be in the right place at the right time, but actually are always late arrivers. Almost any aggregator is like that because they're having to pay for content usually, and they're not selling it for very much. By definition, they're on very thin ground. Whereas Apple makes money from the iPhone, Amazon makes money from Amazon, Netflix invests money to make its own content and gets all of the subscription revenue back for itself, not just a piece of it. Those are real businesses with high gross margins. That's the secret of startups. When I teach startup life to founders, unit economics, meaning what do you sell and how much for, and gross margin, meaning how much of that comes to you, those are the keys to all businesses. If the answer is that you make either a lot or a small amount with what you sell, but the gross margins are small, that's a very different answer to the gross margins being 80% or more, which is typical for software businesses. I think at the very core, this is an old story. It's an old story. It's old technology. It's old ideas. It's not a disruptive business. I've got Keith to go human again. He's going short on Spotify. I'm particularly short on them. We'll see how that works out. One thing is for sure, Spotify is not our startup of the week. Your startup of the week, Keith, is a much more revolutionary idea, Animate Anyone, which according to your headline, heralds the approach of full motion deep fakes, which is, of course, part of our new age of AI. Tell me about Animate Anyone. These are Stanford students. Where have we heard that before? Larry and Sergey back in the day. They've suspended their PhD courses having built algorithms that appear from their publicity. They haven't yet shipped anything that we can use to see, but they've shown demos. It looks as if they can animate from a photograph and create basically a personality. When you add to that what's happening with voice synthesization, or is that a word? Anyway, you know what I mean. The day when you and me could be animated versions of photographs of each other, and when we are speaking, it could sound like us, and we could just throw a script into the streaming service instead of having to be there ourselves, and no one would know the difference. It kind of... Yeah, and I think this speaks to the real shift in this age of AI, which is an ontological one, to redefining what exactly we mean by the word reality, or perhaps even undermining the notion of reality, of not knowing what to trust or who to trust, because the products of companies like Animate Anyone suggest that you can create deep fakes which are indistinguishable from what we once thought of as real. So, this is the beginning of something very different, and again, very different from companies like Spotify, which seem to be imprisoned in the old world, the old pre-AI world. Yeah. By the way, these are students. They're all Chinese students at Stanford, and there's a big... There's more and more evidence that the expertise of Chinese universities and students that come out of China is growing very, very fast, and the idea that there's a wall between China and the US is becoming less and less true. Yeah, I'm guessing that it won't be long before conservatives will jump on this and suggest that the Chinese are in the business of destroying reality, or at least our sense of reality, but the Chinese... And again, generalizing about the Chinese versus the Americans is dangerous, but they are... They understand this, perhaps, or the country understands the future better than the Americans, don't they, Keith? The opportunities, the challenges, and certainly the technology. Yeah, you're asking me to kind of give a binary answer there, but I do think China is an equal in every way, and in some fields is advanced. But by the way, so too is Japan when it comes to animation and computer vision and stuff like that. So America is there too. I think it ends up being about productization more than national origins, and what's happening recently, which is new, is Chinese-sourced technologies are getting productized and becoming popular, TikTok being the obvious example. Some of the TikTok tools to create TikToks are way ahead of anything Facebook has, for example. Well, Facebook seems much more in the Spotify camp. The New York Times talks, and this comes back to the theme of your newsletter, from unicorns to zombies, but it's these new companies like Animate, anyone that's getting the money at them, really convinced that anything is that different. Old technologies. But things just move so fast that what once was new and revolutionary, or at least disruptive, is now conservative. Well, yeah. I mean, think about language translation. We're probably less than a year away from me speaking English and a French listener hearing French in real time, and it better than your Yorkshire English, anyway. Hey. I shouldn't say, hey. Hey. Well, we're on, Keith, to the final feature of this week, which is your X of the Week, and you have a Gary Marcus thing. This is from Gary Marcus too. What's Gary? I'm not sure whether you're critical of him or not, but what's the X of the Week this week? Well, no, the thread is the thread where he was knocking, doing his Turing test, hasn't been proven thing. But then Sam Altman was at the center, and if you put it back on screen, you'll see that somebody replied to Sam Altman with an anti-Semitic jibe, which is in the spirit of current times, where anti-Semitism is on the rise. Gary Marcus, to his great credit, did not accept this person's comment as supportive and said, hey, that's not okay, and came to the defense of Altman. I put it in just because, I don't know about you, Andrew, but my Twitter is full of explicit anti-Semitism. It just blows me away, and in the week where the Senate hearings happened and the head of Penn State University has now had to resign, for reasons we all know. I don't think it's any different, though, from explicit anti-Black or anti-White or anti-Muslim sentiment. And the thing with this tweet is that this guy, and I'm not sure if he's the same one, is probably a fake. He's a bot of some sort, created by someone for some reason, so I'm not sure really of his significance here. Yeah, well, it's significant only in the context. That specific tweet may or may not be significant, but I think the context is real, and they're complex issues. You and me may not see eye to eye on, if we got into it, which we won't, because it's not what our show's about. But I just wanted to put on the record that I, for one, think that anti-Semitism is a growing problem that needs to be addressed. Yeah, I have to admit I don't agree, but that's probably a subject for another show. What I do think, though, and you may be right on all the explicit anti-Semitism and anti-Islam and Whites and Blacks on Twitter or X, whatever you want to call it, is it does speak of the increasing irrelevance of X. For me, that's in the dead pile with Spotify. It's archaic and irrelevant, increasingly. I don't even know why anyone goes on. For me, it's the opposite. It's the increasing relevance of it. I mean, where else is most conversation happening? But it's not even real conversation. It's conversation between you or some people like you and bots. You don't even know if these people are real. They're throwing stupid insults at one another. It's like going back to being four years old. So what? No, I mean, that's an example of that, but I agree. But that isn't what you mainly find on X. There is some rational discussion happening there. But it's also, go to threads or Facebook and try to find anything substantive that's discussing what are seriously real issues. You won't find it, but you will find it on X. Yeah, I think these are the dying days of a pre-AI age, which social media has revealed doesn't really work very well. And we get to the new world of animate anyone, of open AI, of Gemini. We're in a new world where a lot of these issues would just simply go away. They're irrelevant. Well, unless the bad guys produce their own animated avatars spouting