Oct 19, 2024 ยท 2024 #37. 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.
Transcript Playback
AI. Where to Invest?
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker 2
Hello everybody, it is Saturday, October the 19th, 2024. Time for our That Was The Week Tech Roundup with my old friend Keith here. Keith and I actually bumped into each other at the Silicon Valley DLD brunch yesterday in Palo Alto when we were propositioned, so to speak, by a software engineer whose name we'll I'm not going to bring his name up, who has a new product, an AI platform or algorithm, I don't know what you would call it, that takes foreign English languages and automates them into more mainstream English languages. As I joked to the fellow, I'm not sure he understood my joke, as most people don't, it might be valuable for Keith so that people will understand his Yorkshire English. But in all seriousness, it actually touches on The core issue in this week's That Was the Week, Keith's editorial is entitled AI, Where to Invest, Consumer Cloud Enterprise. The people we bumped into at the DLD brunch yesterday were in the consumer space. Lots of startups there. Keith, do we need an AI algorithm that takes weird-sounding... english dialects like yours and translates them into mainstream sounds well the the
would need one if i talked in me normal accent andrew but again but uh therein lies the story because my accent is weirdly not very yorkshire anymore due to living here for many decades but yes you would need it for a normal yorkshire person you need it even more for scott who are very hard to understand, even for me. So there probably is something there, but I do believe that in this... I hope we don't have any Scottish viewers, Keith, because if we have, we've already lost them. Well, I was about to say, in this age of identity being central to everything, I'm not sure people with accents would appreciate being told that they need to modify their accents. They're probably proud of them.
Well... The person who was hawking this product, so to speak, had a very strong foreign accent, so maybe it was more autobiographical. But as I said, in all seriousness, this touches on the core theme in the newsletter this week, where to invest in AI, consumer cloud enterprise. Is this a question of the stack, Keith, or is it a more kind of structural or restructural issue when it comes to finding the value in AI?
Well, what is interesting this week is there's about six essays that circle around this question. And the question itself is novel because normally when technology reaches a kind of a reinvention stage, as it did, for example, with the Internet and then again with mobile and then again with the cloud and Web3, normally the assumption is that there are layers, if you will, where different companies can make a lot of money. Right.
Correct. And that's the normal assumption. So normally you wouldn't ask the question where to invest. You'd say, OK, well, you know, Google's done the search engine. So let's build, you know, an e-commerce store on top of it. and lots of money, in fact, most money was made by building in the higher layers that face off to either businesses or consumers. We talked last week about how with AI, there's at least a question mark over whether the same patterns will happen due to the fact that AI is so good at directly interacting with human beings.
So the issue with last week, we talked about the idea of the foundation companies who used to be companies like Cisco now owning the whole stack. So rather than open AI, just owning the plumbing of this new AI world, they don't everything. Yes. Is that because they're so well funded or because something structurally has changed in this new economy, not just in the economy, but in, the technology underpinning it?
It's somewhat structural. The closer OpenAI gets to what they call AGI, artificial general intelligence, the more individual tasks it will be the best at, not just good at, but the best at. Teaching is one we've talked about. This week, Sequoia Capital, who should be very well thought of, they made the point that in the past, they thought that the old patterns would repeat themselves. But, and I'll quote here, we're no longer so sure. Cognitive architectures mean there's an enormous amount of engineering required to turn the raw capabilities of a model into a compelling end-to-end business solution. What if we're just dramatically underestimating what all of that means?
And what that really means then is that the... and I know you're bullish on open AI, that its $150 billion valuation might actually be an underestimation. You were on a show this week where you were articulating this, the core of enterprise, and you were quoted as saying, agents are really mini AIs with compute engines. Is this, so to speak, Keith, the core of the argument here?
Well, it's part of the argument. Actually, that discussion was acknowledging that large language models are really good with unstructured data, that is to say text mostly, and not very good with structured data, meaning databases. And AI agents kind of fix that. especially if they have the ability to run Python programs, to both create and run Python programs to answer questions, which are more logical questions. So LLMs are like a big statistical engine. Agents can be more like a brain. And when you combine the two together, you get close to being able to solve more or less any problem. And if that happens... There isn't really much of a role for third parties to build any third party that builds a product. A good example is the current flavor of the month notebook LM from Google, which is this week's video of the week. it becomes a feature of AI as opposed to a standalone business because it's so easy for the root models to replicate it and be better even than it at doing it.
Okay, let's take that example. So your video of the week is about Google Alarm. We've talked about this. Even I've been playing around with it. But Google is the closest competitor to open AI. They seem to own... maybe it's the wrong way of putting it now, but the full stack. How would Google, this Google LM fit into the issue of where to invest? You would assume in the old days, Google would build the foundations and then a product like Google LM would be built by a startup, but now it's Google who own everything. Is that the issue here?
Exactly. It reminds me a little bit, Andrew, when I did Real Names, my co-founder was Jean-Marie Hulot. who was Steve Jobs' CTO at Next. And Jean-Marie took me to see Steve Jobs, and we told him what RealNames was. And Jobs said, oh, that's just a feature. And what we had was keywords pointing to URLs, so you could not have to do search. You could just type in where you want to go, and you'd go right there. He said, it's just a feature. And I strongly disagreed with him. I said, no, it's not a feature. It's a whole operating system on top of the internet. Turns out he was right. It was a feature. Now Chrome and all of the browsers kind of do what real names did. We gave them the idea, but unless you own the browser, that feature is going to end up in the browser. And so real names became worthless, actually. That is more likely to happen with AI than any previous innovation due to the fact that the user interface to AI is a single chat window. And, you know, once you own the user interface, like the browser does, and you are clever enough to answer everybody's questions or do the tasks they want to get done, there's no need for them to go anywhere else. And they can subsume almost any clever idea can become... One of the people who was at the DLD branch
I know you talked to was Bill Gross from Idea Labs. I was talking to him. He has an interesting new startup focused on creating or defining the value of
where AI gets its intelligence so it can pay back creators. Isn't that a classic? application where open ai has created a product then you have an innovator like gross who comes along and builds on top of it well he's trying to he's trying to
repeat what he did with the the browser and search which which is to create a middle layer that he can extract money from i think he'll fail impressed no i i'm not impressed because i i think open ai is already cutting those deals with creators um and know insofar as the creators have power it will continue to do that like the new york times is one where it cut a deal i think it cut deals with new york times um
new york times are also suing open ai the name of the company by the way we should get bill gross or i'll get bill gross on keen on is called prorata.ai so smart people i mean he's got 40 or 50 million dollars invested in this thing so smart people are doing stuff
Yeah, they're doing stuff, but are they building businesses? If they're successful, I predict they're building features that will be subsumed. If they're really, really successful, they might get acquired, although I think that's going to be rare. I think more likely their feature will be subsumed and they won't be able to build a business. And that tells you that AI is different. It's, you know, if you're a Star Trek fan, you know, there is only one computer. When Captain Kirk says computer, you know, there's only one answer and it can answer everything. It seems more likely that that's the journey we're on. And it's unlikely the Starship Enterprise is going to switch from OpenAI to Google to Amazon.
Keith, you're... Your day job is at SignalRank. You're the founder and the CEO there. You invest in series B startups. Are you suggesting that the future of companies like yours, like SignalRank and the whole series B idea of going from A to B to C to go in public, that that's history now? In this new economy, in the AI economy, where everything, for better or worse, is really an AI company.
I think the answer to your question lies in another question, which is, is there a moat that prevents you from being subsumed? Now, in SignalRank's case, there is. We have over 200 partnerships with existing investors in the companies we want to invest in. And we invest via those partnerships in their pro rata partnerships.
Well, the question you're answering is this. Could AI gather and allocate capital to startups that it algorithmically has decided are better than other startups? Could it do that? I think theoretically, yes, it could, theoretically. It's going to be one of the final steps it would get to, and I think by then, I hope we're big enough that...
There's never a final step in technology because it's always going somewhere else. But in a way, your argument, which is very uncharacteristic, Keith, which I think reflects your own openness to realities which don't necessarily suit your worldview...
are worth the rest of the market put together. Because if what you're saying is true, these tiny handful of big tech companies are going to essentially own the AI market and there won't be another Google, there won't be another Amazon.
Well, I do think OpenAI and Anthropic are two companies outside of that core that are the most likely to succeed. Due to having no legacy and only having one thing to go after, There's a lot of essays this week around the difficult relationship between OpenAI and Microsoft. At the core of it is a new discovery, which is the contract between the two allows OpenAI to terminate the contract once it has created AGI.
Whatever that means. Yeah, there's an interesting piece in The Times. You have the TechCrunch article. analysis of it in the newsletter. Microsoft and OpenAI's close partnership shows signs of fraying. That's a euphemism for blowing up and basically my reading of it and i'm guessing you're reading is that open ai have woken up to the fact that they don't really want to be in bed with microsoft so they're trying to figure out any kind of excuse to get out of it and agi is such a nebulous term that they're using that their lawyers at least i don't know some of their lawyers at open ai used to work for my wife at uh google uh they're using it as an excuse to to get out of the microsoft relationship
Well, at least to not be captive to it. I don't think they want to get out of it. They did just get married. Is it rather like having a wife but being able to play the field as well? Well, to that very point, OpenAI just did a huge deal with Oracle for computing power because Microsoft can't supply the quantities that OpenAI needs. And they had to get Microsoft's approval to do that, which they got. So OpenAI is becoming more and more independent now. the fact that it's about to be embedded in Apple's iOS, I think, uh, iOS 18.2 is due soon. Um, and open AI will be built into it with chat GPT. It's already available on Android, obviously as a standalone app, even though Google has its own alternative. I think open AI is going to become an independent company that other people want to do business with for revenue reasons. But, uh, they'll continue to own the interface to to users yeah and that's why we talked
about that last week you mentioned anthropic you put them in the same sentence as open ai as one of the two companies that can emerge out of the fog of this ai revolution Dario Amadai, their CEO, had a very long piece. It's one of the most noteworthy tech pieces of the week, of the month, indeed. Machines of Loving Grace, How AI Could Transform the World for the Better. Anthropic's always been seen as the more moral AI startup. Tell me about this essay, Keith. Is there anything new in it? What does it signify for us, or what should it signify?
It's his attempt to write a piece that shows the potential of AI as opposed to shining a light on the fears. At that lunch we were at yesterday, somebody from Google's AI team spoke and said she was worried about how good it is.
And then somebody replied to her that he doesn't think there's any danger. Now, this piece from Amodi is focusing on the good outcomes that AI can deliver. He looks at biology and health. He looks at neuroscience and therapy, actually. He looks at poverty and economic development. He looks at war and peace and governance. And he looks at work and the meaning of life. And he kind of goes through each in some detail saying how AI can enhance human experience.
I don't think there's any intellectual breakthrough in it. It's just all of the arguments in favor of what is currently called AI are in one place for people to go and read. So in that sense, it's worth reading.
And there's another interesting piece in the newsletter this week, the second $100 billion AI company by Rex Woodbury. Given that Amadai, for better or worse, seems to be trying to position Anthropic as the good AI company, does Woodbury imagine Anthropic as the next $100 billion AI company, the one after OpenAI?
No, actually, he argues, and I actually agree with him, that consumer use cases are going to dominate AI success. Not business to business use cases, but consumer. It makes me think of that time when Instagram introduced filters. And, you know, taking pictures and showing them was already a thing. And Instagram brought in filters and that led to Instagram becoming the dominant photo sharing platform because of a feature. So the other side of me saying things are features, Woodbury's arguing that some features are so compelling they get to become businesses.
Yeah, although it's hard to imagine Instagram as a $100 billion company. So is Woodbury disagreeing with your idea that there aren't going to be vehicles of massive value in this world.
Implicitly, he's disagreeing, although he doesn't know he is. He's more arguing the case that in the world of consumer, things can get virality unexpectedly and take off and then carve out a niche.
Well, I guess TikTok would be a better example than Instagram. I don't know what TikTok must be worth. I mean, if it was sold as a real company, it'd probably be worth $100 billion, wouldn't it? More, I think, yeah. And that's just a feature. I mean, there's nothing new about it in terms of a social network, is there? Yeah, exactly. But I'm still waiting. I think most people are still waiting for a really easy-to-use AI personal system that will learn everything I need and do everything. most of the dirty work that I have to do every week, every day, every hour online. I don't see why that's so inconceivable. At the moment, we don't have that, right?
Well, they just had a massive reshuffle. Actually, we should do a whole show on Google. I'm obviously biased because I know a lot of stuff that I'm not allowed to talk about. I strongly disagree. I think they're still the grown-up in the room. They have an incredible team of enormously talented people. They don't have the Sam Altman problem. They don't have the Microsoft problem. But that's another issue.
Well, when I say they don't have proper leadership, I'm describing their internal working methods. They're like a PhD university where students have to pitch to write doctorates. And if everyone agrees, they get to go and write the doctorate. Notebook LM. Winners like Google LM. Yeah, well, so notebook LM didn't come out of a top-down decision. It came out of a random team building something cool. And, you know, I think it's probably got leadership backing now, but it originally didn't. It was a kind of a one-off project. And that's how Google works. It's an internal kind of democracy in a way. It's not very capitalist in some ways internally.
I mean, there are two different styles. Obviously, Sam Altman and the Google CEO are very different. Maybe we'll do a whole show on that. But does all this mean, Keith, we're beginning to take for granted that AI is going to indeed transform the world. I mean, people like Gary Marcus, who I know you've debated, been on my show, fairly influential, thinks it's still not really here. You had a piece by Andrew Maynard suggesting that And it's commenting on the Amadai piece, is this how AI will transform the world over the next decade? I mean, do you think there's more and more consensus that it is? We're not entirely sure how it's going to transform the world over the next 10 years, but it is going to for sure.
Yeah, this article kind of embraces and reorganizes some of the thoughts in Amodi's piece. But it basically agrees with the fact that the speed of change is going to accelerate due to AI. And humans who are constrained are going to overcome a lot of those constraints by using AI. And complex things which... are really hard for humans to grasp are going to become trivial. And, you know, he does talk a lot about the need for data. And if AI has a weak spot, it is its inability to deal with data.
Well, that's where people like Bill Gross and his prorata.ai might come in. I mean, they're trying to figure out the business model associated with data. On the other side, of course, it brings out the dystopians. You have a link with Matthew Crawford, an old friend of mine, very talented writer and thinker. He has an essay this week, Bye Bye Reality. Are the Crawfords of the world beginning to realize that everything they cherish is about to be swept away, Keith?
Well, he's focusing on the military use of artificial people. a little bit like the video of the week where there's an artificial person responding to the creator as if they're real and having a conversation. He's pointing out that there are government contracts to try to create millions of fake human beings with personas that participate in social media and act as if they're real in order to pursue whatever policy goals the military has in social media. And he kind of posits what a scary future that is. When you read it, you realize it's highly likely to happen. Yeah,
and I had a biotech show recently where someone was talking about rumors that the Chinese and the Russians, because it's always the Chinese and Russians, are developing biotech technology biotech applications that will allow for the development of humans who don't feel pain so they can put them in their militaries i think robots might get this was a this was part of a show on christmas so i mean these things may be still slightly science fictional but not absurdly so
They're certainly within the realm of scientific possibility now. So the missing piece is policy. And this piece by Crawford is about policies catching up. Right.
The subtitle of the essay is Bye Bye Republic. The military industrial complex gets up to the new tomfoolery. And Crawford's always one of, I think he's one of the more acute critics of our technocentric age meanwhile coming back just down to earth because not everything is purely futuristic it's been a great week for netflix i mean i've always believed in netflix i'm not an investor but if i was one i think i would always buy netflix stock they always seem kind of like amazon to be the kind of company everyone always writes off and they always come through why has this been a good week for
netflix uh because of their earnings they they they're almost at 10 billion dollars a quarter um which is huge and um that means that they can invest lots of money in great content that keeps people wanting to subscribe and the article is framed as a um i was wrong uh the author basically
thought netflix couldn't keep growing but i mean the netflix going to be able to adapt to the i mean they went from the physical to the digital and then from the digital to the cloud, are they going to be able to adapt to the AI age? What's the future of Netflix in an AI age?
it basically predicted what you'd want to watch next. and it was so good that it led to lots and lots of repeat use of Netflix, surfacing things that you probably will find interesting, and it still does that. I can't imagine that there's a big role for AI in improving that, but I do think there's a role for AI in producing content.
Edward Saatchi, one of the sons of the Saatchi brothers, is based up the road from me in San Francisco. He has a new AI movie company. So I would guess companies like Netflix might be interested in, so to speak, moving up the stack key and owning the studios because the new studios will be AI applications.
Yeah, I'm kind of in wait-and-see mode there. I watched this show on Netflix this week. It's called One Day, and I think it's like a 12-episode, 20-minutes-each episode through the life of a couple that meets in Edinburgh University, and it goes through using St. Swithin's Day every year.
Exactly. But the thing about it was, it was a very unstructured on Hollywood storyline, uh, which me and my wife both commented on and most likely AI can be trained to either do patterns, you know, like superhero shows with act one, act two, act three, roughly following a structure, but they could also be trained to break patterns. And it's going to be interesting to see if the content's any good. My current belief is it won't be, but... Well, we're going to have to wait on that one.
You kindly made my interview with Simon Johnson, who won one of the winners of the Nobel Prize for Economics this week. I interviewed him last week. He's the co-author of Power and Progress. Another...
Northern Englishman, Keith, like you, made it in the United States. He has a different view in Power and Progress. He talks about the danger of technology compounding inequality. I know you don't necessarily agree with him, although if your vision of AI comes true, then this new AI economy is only going to make the inequality, the social, particularly the economic inequalities, of today, even more dramatic, aren't they?
He agrees, and what he argues in the 19th century is it's kind of middle-class guys like yourself and Johnson who came along against the establishment, who seized control of industrial revolution in the united kingdom particularly i mean england and scotland northern england he sees northern england as the the engine of the industrial revolution which he thinks is a good thing but he doesn't seem to think that the digital revolution has the same democratic foundation so i think you'd find his his book and his argument very interesting well you know look the
Well, the money flowed to London. And then, you know, banks in particular and stock markets, all were created off the back of the Industrial Revolution. And the financial... going back to the Lear's idea, was where all the money was made. Yeah,
of course, the new middle class tried to buy up into the aristocracy, made marriages for their daughters, and tried to buy into the aristocracy and got sold on the idea of the British upper class, which wasn't probably to anyone's benefit.
But if you go back to his core thesis... I'd say the lowering of the working day to eight hours is probably one of the bigger things that have driven equality from technology. And I don't anticipate that stopping. I don't think eight hours is the end of history when it comes to how little do humans have to work.
Yeah, and next week, and maybe we'll make this the video of the week for next week, I've got Robert Skidowski on the show. He has a new book out called Mindless. He's a very famous English historian. He's the authority on Keynes, and he's written a book about AI in the context of Keynes. Keynes, of course, imagined the shrinking of the working day would be a good thing, although Skidelsky talks about that. Anyway, lots to talk about, Keith, but we've got to get to our final two features. Startup of the week is, I wasn't sure, are you making Pamira, who acquired Squarespace for $7.2 billion? Or is it Squarespace? Who's the startup, Pamira or Squarespace?
Squarespace, for sure. They just got paid $7.2 billion after, what, 15 years of work? That's not bad. It's good. It's very good. They should be super happy. And Squarespace, by the way, I think sucks. And the fact that Pamira... Well, and Pamira is a British...
Right, and they've got a quote on this from there. We look forward to leveraging our decades-long experience and track record backing internet platforms and technologies that enable SBMs. So it sounds very old-fashioned. Are they Did they just lose $7.2 billion?
And I would have guessed, I mean, I used to use Squarespace, I don't anymore, that for internet, I don't know what you would call them, a platform for building websites, that AI would just blow all that away. You don't need
Yeah. Say that again, Keith. That was a good one. They just bought a horse when the wheel was invented. Why didn't they ask you? I mean, how do companies make such dumb errors? I don't know why they, I guess they're your startup of the week because you think they must be laughing all the way to the bank.
Yeah, well, they got a deal. I wonder if they can unravel that one. Finally, we're no longer calling it X of the week, but it was an X, the post of the week from Michael Seibel, a partner at Y Combinator. Too many startups investors are exploring the B2B opportunities with LLMs and not enough for exploring the consumer opportunities. This is the moment consumer founders have been waiting for. Sybil would disagree with you in terms of your overall structural analysis. He seems to think that we're still back in the layers.
Yeah. Well, I think if there is an opportunity, it is in consumer. So I kind of agree with Sybil and Woodbury. But I suspect that OpenAI knows that as well. And the space left for third parties to build sustainable businesses is smaller than they realize.
But opening is so busy on so many fronts. I mean, Altman's trying to figure out the relationship with Microsoft. He's trying to make sure other people don't leave. He's lost a lot of his senior people. He's trying to rebrand the company. They're not going to have the time and the focus to be able to do what you want. I mean, even, even, um, um, A trillion, double trillion dollar company like Google couldn't do that.
Well, that is correct. And I should make explicit what's implicit in my statement. OpenAI's key asset in this context is its APIs. And third party developers will use OpenAI and pay it, by the way. So its brain, if you want to think of its foundation model as a brain, its brain will power those consumer apps.
Isn't that just a repeat of what's happened historically? No, because in the past there were hundreds, if not thousands, or tens of thousands of APIs that you would orchestrate together to create some experience. In this case, there's probably going to be one or two. And so the center of gravity of the entire ecosystem will come down to one or two companies. And that's different. That means that the value from anything that's successful, it's a bit like Amazon stores. Amazon makes a lot of money from third parties building Amazon stores. Imagine Amazon stores was the only store on earth, then it would make all the money. Well, with OpenAI and Anthropic, I think we've got the two most likely big players that everything else is going to hang off of.
Because Google, unfortunately, Google is implementing AI inside its own universe, like Google Docs, spreadsheets, and slides. And it's not really successfully becoming a platform outside of the Google ecosystem.
But that Google ecosystem is enormous, as is Microsoft. And even Microsoft lived off its own desktop publishing ecosystem, even when it lost in the internet battles of the 1990s.
I agree. But that means they're replaying that playbook from that era and the What is up for grabs now is so much bigger than anything inside their ecosystem that by doing it, they're constraining their impact. Whereas OpenAI and Anthropic are not constraining the impact at all. They can be in everything. Like we talked about, like water seeps into every crack. That is what's happening with OpenAI and Anthropic, I think, to a lesser degree, but also... And I think, by the way, Elon Musk's grok within Twitter is going to be a big player as well, bigger than anyone realises.
Yeah, and of course Musk, for all his warts, understands this as well as anyone. Fascinating show, Keith. We will return to this subject many times in the future. Have a good week. And we will talk again about the week in technology next week. Thank you so much. Thank you. Bye, everyone.