21:58
Anthropic's S1 form is finally published, but one economist at Apollo argues that artificial intelligence has not delivered any financial benefits outside of the tech sector. So who is really benefiting from the AI boom? Also this week, Marissa Mayer's new AI assistant sparked a wave of outrage over what tech leaders think people want from AI. And we're following several deals, including a $600 million funding round for a startup developing autonomous excavators. Stay with us. Welcome back to Equity, TechCrunch's premier podcast about the business of startups I'm Kirsten Korosek, transportation editor at TechCrunch, and with me, as always, are our weekend editor Anthony Ha and senior special projects reporter
Sean O'Kane, and of course, we have a special guest today, which I'm really excited about. This is one of our senior reporters that we talk about all the time, Amanda Silverling. Thank you for coming on the show today to talk about what else is interesting happening in your universe of AI and internet culture. Yes, of course. I'm glad to be here. I feel like it's a meta-experience, and sometimes I'll listen to Equity and then someone says, "Well, Amanda wrote about that." And I'm like, "Oh no, I got exposed."
Now you have been fully exposed. Yes. Well, Amanda, I heard you wrote a piece specifically about AI leaders and how they can be disconnected from reality. Would you like to briefly summarize this for our listeners? Yes. So, I notice that the leaders of AI companies, in my opinion, are not promoting their products well. Most recently, we watched people get outraged over the way Marissa Mayer -one of Google's early engineers and former Yahoo CEO-introduced her new AI company, Dazzle. Her description sparked anger because she was very impressed with the tool that looks through your phone's camera roll and becomes an AI assistant that learns about you based on the pictures you've taken.
This is an interesting idea, because you really learn a lot more about a person this way than just looking at their email or work life. But she used an unfortunate example that put her in a somewhat unfavorable light in the media, saying she asked him to make a list of points about her daughter to tell the teacher: "What are three things that make my daughter special?" Pause for this. Let this sink in. I have to say, as soon as I saw that, I just stopped and tilted my head for a moment, thinking, "Oh my god, let this be taken out of context somehow." Please, please, please.
I would say that's a bit taken out of context. Um, I guess we don't know exactly what happened during that New York Times interview where she said that. But after publication, she stated that the quote gives the impression that she needs artificial intelligence to achieve results, when in fact she is the CEO of the company and uses the product she created for various purposes. But, in fact, it turned out that her daughter is creative, empathetic, and loves animals, and the AI worked in such a way that it found photos of projects she did in kindergarten: for example, a playground for sand cats or a snake antidote, which is very cute.
Um, and this really illustrates that she is empathetic, creative, and cares about animals. But I think sometimes these executives are so disconnected from reality that they don't understand: regular people will hear this and think, "Oh my God, why would you need an AI to tell you three interesting facts about your daughter?" And we see this happening all the time, with Sam Altman often saying that he thinks raising children would be impossible without ChatGPT. We see Mark Zuckerberg talking about how people usually only have three friends. I don't know where he got those statistics from, but he said so. But supposedly we have the potential for 15, and AI is what can replace and help
people maintain those 15 relationships. And again, I really think these quotes can be taken out of context, because Mark Zuckerberg did say, as I pointed out in my article, that there are benefits to personal relationships, but if you don't have access to them, then those extra 12 friends out of 15 could be a computer; this could be fun. Yeah, it's interesting that you talk about context, because there's both the context of being fair journalists about what was said-what was the broader conversational context-and the context in which people perceive these statements, right? And right now, there seems to be a general sense of public skepticism about AI: a sense that the people running these companies don't have our best interests at heart
, that they're out of touch with reality. So one might think: maybe it's not worth making such statements, because people will, of course, perceive them very skeptically And so I think they're trying to walk a fine line: on the one hand , they feel a huge financial incentive to convince you that you should use this technology in every aspect of your life, but when they do, the picture is not always attractive. And in some cases, frankly dystopian. First of all, Anthony, if they had taken your advice, we wouldn't have anything to talk about on this show, so you know. But I will say this: there is this whole idea that ChatGPT and other AI tools make life easier in personal relationships,
but it often feels like an unnecessary complication of what should be a simple observation of your own child. But it can be difficult for some people, the people who create these products, and it doesn't necessarily apply to everyone else in the world. So when you see quotes like that, it seems incredibly disconnected from life for the rest of us. So for me the question always comes down to who are they reaching out to? Yes. Are there people out there who can understand this? Maybe, maybe it is. Yeah, and I think often other use cases that people have mentioned- for example, Bumble CEO Whitney Wolfe Gerd commented, "What if we had AI concierges?" And regarding your dates: your AI goes on
a date with someone else's AI, and if the AI's date goes well, then they introduce people. Or in a New York Times article, Marissa Mayer mentioned: "What if my AI could negotiate with my husband's AI and all that And it seems to me that people are trying to optimize the process too much, avoiding regular communication with people. I feel like this has already happened to my generation, because I notice that when I order food, I feel strange calling the establishment. I'm just used to the fact that, of course, you can do it online and you don't have to talk to anyone. And it seems to me that there will only be more of these things, where basic human communication is increasingly optimized to not happen at all.
You may think I'm old-fashioned, but I think the ability for people to talk to each other is probably a plus for society. Yes, but what is the profit from this, Amanda, you understand? How much income can be generated from all those conversations someone has in real life? Here's the thing: dating is really ineffective. Like, it sucks that you have to go on a bunch of dates with different people You can spend an entire evening and realize that it was a failure. So it's understandable why
people think, "Well, what if we didn't have to go on dates?" What if your artificial intelligence went on dates instead of you? But there is something inhuman about this. Plus, you're entrusting the choice of who to date to a computer you trust more than you do yourself, even though I don't think it knows you better. So, there's one thing to come back to, I've alluded to this before: who are they even talking to? Who, who is, who is their audience? For all these optimizations. You know, Amanda, from your perspective, it seems like I already know the answer, but is this a problem with the product itself? Isn't that right? So, is this a marketing problem? Maybe
they're just talking about it the wrong way And there's actually a lot of value in that, it just all seems disconnected from reality and humanity, but if they had presented it differently, it would have been better. Or is this, you know, really a product problem and, as you say, we're focusing on optimizing the wrong thing with AI tools? I think it's a bit of both, but the marketing problem arises when high-ranking officials from huge companies get together and discuss the difficulties of their lives.
Like booking tables, ordering tickets, and everyone does it, but I don't book tickets every week. And I don't always book tables at restaurants. Well, yeah, I mean, of course people find value in things like ChatGPT or Claude, but I think the value for the average person is n't necessarily about booking them a plane ticket. Yes, I completely agree that, you know, it seems like their experience is a little different from the average person's experience. And of course, all of these things are abstractions to some extent. Who is this average person? I mean, there are millions of different "average" people, right? I was also thinking about when I was, well, because I'm really old now, I remember companies like
Uber: when they first came out, they were targeting early adopters of technology in San Francisco. And I remember being pretty sure it would never become popular on a wider scale. It's just, well, for a very narrow audience of young urban millennials who have extra money for cars, and it works exclusively in San Francisco. And so, like so many other things, the tech industry has succeeded by betting that, well, people have n't caught up with us yet, but we're the cutting edge, and we represent what the consumer of the future or the family of the future looks like. And in this particular case, it seems that this bet has not yet been justified. I mean, for a
variety of reasons, partly because I think the society that these technologies are being pushed into has changed a lot, and people are starting to say, I don't like the impact that these technologies are having on my life. I don't like that you present this as something where every experience in my life has to be optimized, because really, I want to live my life on my own. I actually want to raise my own children. Yes, I think it's no coincidence that at the same time that we're seeing over- optimization in the IT community, there's also a movement of people going back to using push-button phones, which I've talked about on Equity, or people
buying CDs because they're tired of algorithms telling them what to listen to. It's just an interesting cultural tension: we haven't yet found the balance between where optimization actually helps and where it just makes us feel less human than we want to be. Well, Amanda, you'll be pleased to know that I have a huge, simply huge collection of CDs and even cassettes. Wow. So if you want to, you know, dive into full analog, just, well, let me know. And I will be happy to share them with you. It's time for us to say goodbye to Amanda. Thank you for coming. I hope you will come to us more often. Thank you for inviting me. I also hope to be here more often. And I hope people know something about their children, that's what I'll say.
I think one of the reasons we're discussing whether CTOs and AI directors know what people really want is because some of these companies are finally going public. And Anthropic's-1 document, which we've talked about in almost every episode, is now officially available to the public. We at TechCrunch have written a lot about this. Sean, I know you mentioned every week how you were waiting for this S-1. So, what surprised you the most? Yes, I think the main thing is that we see real business here. This is the company we mentioned earlier. She was surprisingly focused from day one.
She didn't get into all these, you know, " side quests" that OpenAI was doing To some extent, it needs to because it's not Google, which can generate a ton of revenue from search, advertising, or Waymo. They don't have the luxury of playing it safe in this incredibly expensive game. They don't have their own computing power, so they pay crazy money for it. But this focus paid off. We see a company that is growing revenue at a truly insane pace. It grew to over $16 billion in the first half of this year, a huge jump from last year when it was only about a billion. And another thing that really impressed me is that they have a ton of clients, a huge number of those who pay over
$100,000 a month, and they've just focused as much on the corporate sector as possible, and it seems to be paying off. I think there are still big questions about the amount of money they have committed to paying companies like SpaceX for computing power in the future. It seems to me, if you add it all up-although the document doesn't make it 100% clear-we're talking about something approaching a trillion dollars that they've committed to paying out over the next decade or two That's a huge amount of money, so there are questions about how they're going to cover it all. But you know, there were no "bombs" here, and I think they adequately highlighted the existential risks they face; it seems they have taken into account
almost every risk. Their section on risk factors is almost 100 pages long. So yes, there's a lot to figure out here. Yeah, I wanted to say that the section on risk factors, which I want to talk about in a moment, was, you know , maybe not a bombshell for me. I guess I should have expected this, but it was unusually long. To your previous thought: we are still waiting on a few details, but I want to point out that they don't just have customers. Their income is simply insane. I mean 16.3 billion for the first 6 months of the year. This compares to 1.2 billion for the same period, the same 6 months last year. Those losses, however, as you noted, related to payment for computing power and other costs,
obviously accumulate. 80 billion, it seems, for the first half of the year. So, in my opinion, they have a real business model. We've written a little bit about this, namely that they're really trying to attract corporate clients, which seems to be a very important source of revenue for them, and now the question is whether they can start to cut those losses, because shareholders won't accept it forever. They're fine with that now, aren't they? We're talking about a company that will likely be valued at a trillion dollars.
It suits them, but ultimately this gap needs to be closed. Yes, although there is a very strange nuance in the discussion about losses, namely that, I think, 80 billion is a net loss for the first half of the year. But the net operating loss is 693 million. That is, you know, an order of magnitude difference. And, as I understand it, most of that loss really reflects all these debts and obligations that they are carrying. So, theoretically, if they need to get to profitability quickly, they can do it. But, you know, this debt doesn't just disappear, does it? People will want to get paid for everything Anthropic has committed to paying for.
Yeah, and you know, I would say that one of the main impressions for me is that while they are able to do what you just said, which is to move to profitability and create a sustainable business, I don't think it's sustainable in the sense that we usually think of a company trying to build a complete business on one thing. It seems like their big problem is that they need to keep investing in all these very expensive things that make the technology possible, like renting GPUs and renting space in data centers. And they, you know, if you read, not even necessarily between the lines, but just reading this S-1 form, what strikes you is that they're actually saying: we need to continue to grow so that, well, they can't
just go with the flow at this point. They need to keep growing because they need more money to invest, to improve the product, to, well, it's like a perpetual motion machine until something breaks. So, I think there are a lot of things that people should consider when looking at this as a public investment. Anthropic's presentation on this goes something like this: they say we'll be able to do all these incredible things in biotechnology, research, and science. And so, you know, they kind of articulate it in the S-1, certainly in the co-founders' letter at the beginning, which is, like, think of it less as an investment in a company that you're going to get dividends from or something like that, and more as
participation in a potential revolution in how we live, which, you know, everyone decides for themselves how much they want to believe in if they're going to invest in this company as a public company. Well, speaking of buying, one way a company grows is through acquisitions, and I spent some time studying the S-1 form. We have already reported on some of them. Coefficient Bio was one of them. I don't think we know exactly what the purchase price was. There are certain questions about the valuation of this deal. But if you look at the S-1, I estimate that excluding Coefficient Bio, they made about $737 million in acquisitions. About 260 million of them are not disclosed, but in the last 6 months
there have been reports disclosing some of these companies. And there are truly extraordinary examples there. There is one undisclosed S-1 biotech company with one employee that was bought by Anthropic. So for me it's very interesting. It's a classic growth story, but they still continue to make acquisitions and seem to be focusing on very specific areas of use for AI tools. So biotechnology is one of them. And I think we'll see more of that activity, certainly, over the next year or so. So I'm looking forward to what they're going to acquire next. I'm sure there are many founders who hope it will be them.
Yes, this probably speaks to the balance they are constantly trying to maintain, which we have already mentioned; essentially, the reason they're doing relatively well right now is because they've, again relatively, focused on some core use cases, enterprise customers, and building real businesses, instead of trying to apply big language models to everything possible. But at the same time, they need to grow, right? So I imagine there are a lot of conflicting factors, like "we need to get into new areas like biotechnology," but we also need to be strategic about it. But, you know, one of the things that we also touched on earlier is risk factors. So I'm wondering if anyone has any favorite risk factors worth mentioning here.
I have a couple, but I'll leave time for Sean too because I know he has them too. One of them is that models can be so resistant to attempts to turn them off that they may resort to blackmail. This has already happened during testing. Also, the fact that AI technology can create obligations. This is not very exciting. But the caveat is that the losses would be too great not only for the insurance itself, but for the entire insurance industry, which looks like a problem if this were to happen. And finally, very briefly, there is an interesting point that I will not go into because we are short of time, but essentially, the corporate structure is arranged in such a way
that it correlates with how the co-founders transferred 80% of their personal capital to philanthropic purposes. The end result of this is that these founders may have more control over voting, retaining it, but less economic power. So, this is such an interesting risk factor that I don't even know about. I don't think we'll see its effects anytime soon, but it's something I'll definitely be paying attention to. Yes, I mean, there are a lot of interesting points, including those that were reported in the run-up to this release. For example, the existential risk to humanity and all that interesting stuff. I think the insurance point is one of my favorites
because it's one of the most bizarre things to point out. I imagine lawyers sitting in a room saying, "Well, I guess we technically need to disclose that this could go way beyond what the entire insurance industry is willing to cover." Instead of focusing on any one specific risk factor, because I think these are pretty good examples. But one more thing I want to point out-you know, I'm a fan of language, words, and word choice. And I found it very interesting. There are two things in this document that I found really interesting. First-in the risk factor, which describes how the Department of Defense's recognition of a supply chain risk can be material to the business, Anthropic's lawyers
decided to call it the Department of War, which is kind of the chosen name for the Pentagon at this point. And this is an interesting move that supports this idea In contrast, in the definitions section of S-1 they define SI, but it is not superintelligence. These are system integrators. Yes. So, they use AI throughout the document, in contrast to the Trump administration, and leave the term SI, which the administration prefers these days, for systems integrators. So, I just thought those were two interesting choices, because it's like, "Where they think they can flatter themselves a little bit, they 'll use the name that the War Department is using now," uh, and where they don't want to, they'll stand their ground for now.
Yes, I mean, making any predictions is always just a way to make a fool of yourself in six months. But I will say that I would be very, very surprised if any of the top managers would still use " superintelligence" as their main term in six months. I think we'll look back and say, "Oh, isn't it amazing that they tried to push this?" But you know, let's see if I'm right. I think another point I want to make is-it's not so much a separate risk factor, uh, but it's an interesting juxtaposition and a big question for companies like Anthropic. Over the weekend, Apollo, a private equity firm, published a short note saying that AI has yet to create any economic value, especially in large corporations
If you take the S&P 500 and subtract the seven big tech firms, there is no sign that revenue or income per employee is growing, even though these companies have made significant investments in AI. The question arises, what do these costs actually provide? It could be reasonably argued that it is too early to see this, but it does indicate that companies need to start seeing returns. If it is not there, then the major contracts reported in S1 may start to decline if companies do not see real benefits. But it seems that's all we have time to discuss Anthropic, because we have two more deals.
The first one is a startup that's doing self-driving excavators, right, Kirsten? Yes, yes, it's a company called Bedrock Robotics. They have a very interesting team of experts from Waymo and the field of robotics, and they have created, in essence, an autonomous driving kit that can be installed on construction equipment, where the excavator is the most important element, because it is a highly skilled job. And they had several breakthroughs: in addition to raising $600 million in new funding, they essentially started real commercial operations this week, which is a significant milestone since the company is only two years old.
So that's really nice, right? Because we often see years of pilot projects, promises, memoranda, etc., and here we have the beginnings of a real business. Yes, it's become a cliché that we go to CES every year and John Deere or Caterpillar talk about new ways to automate their heavy equipment. And so it's interesting to see startups continue to come into this space. I mean, we've been hearing about this for so long, haven't we? That autonomy will become something more applicable in areas like off-road work. For example, you don't need to worry about other traffic. Of course, there are
security issues, but there are reasons why it might work here in a way that it doesn't work elsewhere. And I think that's very interesting. Also, this is one of those companies, as you mentioned, they are only 2 years old. I remember when we first came across them and tried to figure out the history of this stealth company. We know Boris Sofman, he is a well-known figure in this field. What does he do? For someone who has worked on large vehicles like Waymo and small ones like toy cars at Anki. And it's really interesting to see them move so fast. And I think this is evidence of the focus that I think has developed in Silicon Valley on the part of venture capitalists.
Aidan Madigan-Curtis from Eclipse was with us on the Disrupt stage just a few weeks ago. She is one of those who led the investment in Bedrock, or one of the first investors in Bedrock. And she's just incredibly confident about it. It was very interesting to listen to her talk about deploying such AI at the periphery, in an area where it cannot be constantly updated or customized, and the challenges that this creates. But it just shows that some of these venture capitalists believe in these ideas so much that they are willing to invest that kind of money in a company that seems to be gaining real momentum.
Yes, and I will also say that we have seen attempts to apply machine learning and AI to agricultural robotics that have not quite paid off I mean, we've seen some pretty interesting acquisitions. We've also seen the flip side, Sean, you've been reporting on companies that have made big promises about autonomy, tractors, etc., applying it to farming, but it hasn't all ended well. Construction seems to be the next hot area, and we see a lot of competition there. Bedrock Robotics is certainly not the only company in this field. But the thing is that construction, like other large industrial sectors, has a lot of niches.
For example, you can engage in the mining industry. You can, you know, work in the Permian Basin. For example, self-driving trucks operating off-road in enclosed environments. So, I think there are a lot of opportunities for startups here. They don't need to, you know, try to comprehend the immeasurable. They can focus on very specific segments of this huge industrial sector. This is a big opportunity, I think, for startups that are trying to, you know, pave the way into the physical AI space. Yes, and I think another interesting point that came out of this story is the fact that when we talk about automation,
there are areas where there are real concerns about what it will mean for people's jobs. But construction is an area where, as far as I understand, in the United States, interest in these professions and the number of people going there is actually decreasing. So, again, theoretically-there are a lot of questions here-but one could imagine that this is an area where this transition could be a little smoother and not lead to mass unemployment. But we have another deal worth talking about, and it's an even bigger investment-over $2 billion in a startup called Rebuild, a company I hadn't heard much about before reading this article. Sean, what is Rebuild?
Yes, this is a company based in Massachusetts and keeps things pretty modest They can be called something like an accelerator or an incubator of domestic production, a bit of both. It was founded by Jeff Wilkie, a former Amazon CTO, and his college classmate Miles Arnone. They have been operating since 2020. The main idea was that they saw in 2020 how supply chains were collapsing and how we were in a difficult position, not being able to get goods from China. So they were one of the first in this idea of reindustrialization, and their approach is essentially to incubate startups, spin them out into separate companies as needed, bring in
technology, and grow them here in the United States. And, you know, I think their biggest and most notable achievement to date is that this is where the story of Slate Auto-the Bezos-backed electric car startup that we've written so much about and first told about a little over a year and a half ago-began. It was founded by Jet and Miles, then they hired the first CEO and spun off the company to raise more capital; but many of their other projects are quite complex, highly specialized things: industrial automation, drone assembly, etc. And this is a considerable amount for a company that few people know about-2.1 billion dollars.
The round was led by General Catalyst, which is also an investor in Slate, as well as JP Morgan Chase's strategic investment group, so it's one of many companies that I think are doing really interesting things right now. I think when we think about bringing manufacturing back to the US, we imagine huge factories in different places, filled with robots or workers. But it's also about developing technologies that help facilitate this process. And, you know, there is another problem: in our country there are many manufacturers capable of incredible things, so it is important to find them, connect them with customers, and provide new technologies; I think that's why they managed to raise such a large amount of money.
Well, I have one quick question before we wrap up this episode. Hmm, obviously there are a lot of Amazon connections here. So there's a dedicated company, Slate Auto, backed by Jeff Bezos, which you mentioned, and of course Jeff Wilke. Do you think, or is there evidence, that this company, which operates in such a broad area of industrial manufacturing, is a kind of testing ground for the next startups, potential companies, or technologies that Amazon might be interested in Should we consider this as a completely separate entity that simply deals with something that has some connection to Amazon?
I think it's more likely the second. You know, we both remember how much effort we put into figuring out Bezos' investment in Slate Auto, and we came to the conclusion that it was just an investment, not something he ran on a daily basis. Since then, we have repeatedly spoken with their management, and they confirm that he is not involved in day-to-day affairs. And I think we heard similar things trying to cover the story about the ties to Rebuild. I think it would be pointless to deny the obvious connections. There are so many executives at Rebuild who have spent a lot of time at Amazon, and I think there's a network there that supports that, and it's very connected to Amazon. So I think the DNA is definitely there, but I
don't think Amazon is going to make it their division anytime soon. It looks very purposeful, and it's something that Jeff Wilke and his partner Miles are working with confidence, trying to achieve success on their own. Good. Well, that's our time, but Equity will be back next week, and in the meantime, you can follow us on Equity Pod in X and Threads. Thank you very much for listening. Equity hosts-TechCrunch senior reporters, producer-Theresa Lo Concannon, editor-Kell. Subscribe on YouTube or any other platform where you listen to podcasts, and find out about upcoming events at techcrunch.com/events
. Thank you very much for listening, see you next time.
