Latest / DEEP INSIGHT / Why Anthropic's plan to go public could be a milestone for humankind

Transcript

Why Anthropic's plan to go public could be a milestone for humankind · DEEP INSIGHT

• 0:01
  Speaker 1: We're on the cusp of the largest stock market debut
• 0:04
  Speaker 1: in history, and the company behind it is warning its
• 0:08
  Speaker 1: technology could pose, quote, existential risks to humanity. Welcome to
• 0:14
  Speaker 1: the brave new world of Anthropic and its AI. Business-wise,
• 0:18
  Speaker 1: it could be a boom, or for humanity at least, doom.
• 0:23
  Speaker 1: We'll chat with an expert about that in a moment,
• 0:24
  Speaker 1: but first, let's get into what happened on Monday. So Anthropic,
• 0:30
  Speaker 1: the firm behind the clawed chatbot, filed 355 pages with
• 0:35
  Speaker 1: the Securities and Exchange Commission ahead of its listing on
• 0:38
  Speaker 1: the stock market. It's what companies do before stock market debuts.
• 0:42
  Speaker 1: Think of them as a brochure of sorts for investors.
• 0:45
  Speaker 1: But of the 355 pages in Anthropic's filing, 93 were
• 0:50
  Speaker 1: devoted to so-called risk factors. Among them, this warning on
• 0:56
  Speaker 1: page 46. Advanced AI systems could, if not developed, deployed
• 1:01
  Speaker 1: and operated with appropriate safeguards, oversight and security measures, pose
• 1:06
  Speaker 1: catastrophic or existential risks to humanity. But it isn't all
• 1:12
  Speaker 1: just about risk. But there's also this on page five,
• 1:15
  Speaker 1: where it talks about the promise of AI. I quote,
• 1:17
  Speaker 1: we believe this abundance of AI will rapidly transform every
• 1:21
  Speaker 1: sector of the global economy in ways greater than the
• 1:24
  Speaker 1: impact of industrialization, electricity, semiconductors, and the internet, unquote. So
• 1:34
  Speaker 1: is this a boom versus doom scenario? Is it even
• 1:37
  Speaker 1: fair to call it that? Let's ask cybersecurity expert, Dennis
• 1:42
  Speaker 1: Kenji Kipka, who's also scientific director and founder of the
• 1:45
  Speaker 1: Cyber Intelligence Institute in Frankfurt and an advisor to the
• 1:48
  Speaker 1: German government and the FBI, among other organizations. Welcome, Professor Kipka. So,
• 1:54
  Speaker 1: anthropics warnings, anthropics filings warn of many risks and opportunities
• 2:00
  Speaker 1: with AI. What is your take? Is it a boom
• 2:03
  Speaker 1: versus doom scenario or is that being a bit too simplistic?
• 2:08
  Speaker 2: I think it's a bit too simplistic. So what, in
• 2:11
  Speaker 2: my opinion, clearly stood out is that these biggest risks
• 2:15
  Speaker 2: aren't buried in some kind of small print. So we
• 2:17
  Speaker 2: have been talking about AI catastrophe, politics, dependence on others,
• 2:21
  Speaker 2: and they are right part of the center of the IPO.
• 2:26
  Speaker 2: But I would say the length is really interesting because
• 2:28
  Speaker 2: we have 93 pages and there are a lot of
• 2:31
  Speaker 2: different aspects covered, such as insurance. So the company says,
• 2:34
  Speaker 2: for example, a truly catastrophic AI accident could be so
• 2:38
  Speaker 2: big that there is no insurance coverage. And we have politics,
• 2:42
  Speaker 2: these issues around the February Pentagon decision. And we also
• 2:46
  Speaker 2: have four big dangers that are concretely named regarding AI.
• 2:49
  Speaker 2: So misuse, for example, bioweapons, cyber attacks, humans losing control.
• 2:55
  Speaker 2: And I would say this is relevant because lawyers write
• 2:59
  Speaker 2: these sections as exhaustively as possible to protect the company
• 3:03
  Speaker 2: from being sued after the IPO. But none of this
• 3:07
  Speaker 2: is new, right?
• 3:08
  Speaker 1: I mean, these warnings have been out in the public
• 3:10
  Speaker 1: domain from Anthropic itself.
• 3:14
  Speaker 2: Yeah, so the dangers themselves, they aren't new. So what's
• 3:16
  Speaker 2: new is that they are now written into a legal
• 3:19
  Speaker 2: document very concretely that investors can sue over. And Anthropic
• 3:24
  Speaker 2: has been talking about these dangers regarding cybersecurity and so
• 3:28
  Speaker 2: on since 2023 with its own safety rules and a
• 3:32
  Speaker 2: trust that watches over its mission and so on. But
• 3:34
  Speaker 2: what is new that these risks now come with a
• 3:36
  Speaker 2: price tag. So the insurance gap that I mentioned and
• 3:39
  Speaker 2: the open conflict with the US government regarding military use
• 3:43
  Speaker 2: of AI and half a trillion dollars of fixed computing bills,
• 3:48
  Speaker 2: which is also really a lot. And the filing costs
• 3:50
  Speaker 2: um admits um in my opinion that being the safety
• 3:53
  Speaker 2: company which entropic definitely wants to be and being a
• 3:57
  Speaker 2: profit-seeking company can pull in some way different directions and
• 4:01
  Speaker 2: we'll have to figure out where this will go to
• 4:03
  Speaker 2: in the last in the next month and years.
• 4:05
  Speaker 1: But just back to the security warnings that are coming
• 4:07
  Speaker 1: because they uh this is as you said this is
• 4:10
  Speaker 1: also uh not new and open ai for example has
• 4:14
  Speaker 1: also been talking about these risks from ai very publicly
• 4:18
  Speaker 1: and that uh What does that say about the timing
• 4:21
  Speaker 1: of these warnings that they are coming before any IPO
• 4:24
  Speaker 1: from Anthropic, for example, or any other movement on the
• 4:27
  Speaker 1: stock market? Why does this focus on security keep coming
• 4:30
  Speaker 1: from these companies themselves?
• 4:34
  Speaker 2: I guess it's a point of fear. So we currently
• 4:36
  Speaker 2: see that we have very strong AI regulation in the
• 4:38
  Speaker 2: European Union, which is being enforced at the moment, and
• 4:41
  Speaker 2: it will also be enforced in 2027. And as we
• 4:44
  Speaker 2: have read this year, especially since summer, about a lot
• 4:48
  Speaker 2: of dangers regarding AI, cybersecurity issues, bioweapon issues, mistrust in
• 4:54
  Speaker 2: AI systems. These companies also have a fear that there
• 4:57
  Speaker 2: might be some kind of AI bubble because we have
• 5:00
  Speaker 2: billions of dollars of investments into AI. And if these
• 5:03
  Speaker 2: investors see that there might be stricter rules which don't
• 5:07
  Speaker 2: allow the companies to place their products on the market,
• 5:10
  Speaker 2: there might be the case that we have an AI
• 5:13
  Speaker 2: bubble so the companies could not work anymore and that they,
• 5:18
  Speaker 2: for example, also lose their net worth.
• 5:22
  Speaker 1: But there is an AI bubble, isn't it? I mean,
• 5:24
  Speaker 1: with this filing, for example, Anthropic is saying that it
• 5:27
  Speaker 1: could reach a potential valuation of $ 2 trillion today. once
• 5:31
  Speaker 1: it goes public. I mean, what is behind this sort
• 5:35
  Speaker 1: of a valuation for a company, especially that in the
• 5:38
  Speaker 1: first half of this year had about $ 16. 3 billion in revenue,
• 5:43
  Speaker 1: but it lost $ 80. 7 billion in the same period. So
• 5:47
  Speaker 1: you have a company that isn't really making a profit,
• 5:49
  Speaker 1: yet it expects a valuation of $ 2 trillion, all in
• 5:54
  Speaker 1: the midst of this conversation about an AI bubble. Can
• 5:57
  Speaker 1: you make sense of any of this?
• 6:00
  Speaker 2: Yeah, I would say, and the demand on one hand
• 6:02
  Speaker 2: is real. So people, of course, they really use AI
• 6:05
  Speaker 2: and Entropix revenue is up 14 fold that has been reported.
• 6:09
  Speaker 2: And Nvidia also reports a demand is bigger than the
• 6:12
  Speaker 2: supply of chips, for example. But we see a large
• 6:15
  Speaker 2: depth here. So some data center companies, pay around 50% interest,
• 6:21
  Speaker 2: nearly half of this year's data center's bonds are junk rated,
• 6:25
  Speaker 2: we also see. And the last week, several data center
• 6:28
  Speaker 2: firms also have postponed their stock market debuts. And I
• 6:32
  Speaker 2: would say that this is really a clear signal. And
• 6:35
  Speaker 2: we also see the risk of circular money. So chip makers,
• 6:38
  Speaker 2: for example, such as NVIDIA, they invest in AI firms.
• 6:42
  Speaker 2: which then buy their chips. So it's a bit like
• 6:46
  Speaker 2: a shop lending customers money to buy its own goods.
• 6:49
  Speaker 2: And as a lot of companies do not have clear revenue,
• 6:51
  Speaker 2: they do not have any clear numbers, this is really
• 6:54
  Speaker 2: an investment into the future that we currently see. So
• 6:56
  Speaker 2: we cannot be sure that all these investors will benefit
• 7:01
  Speaker 2: from the current billion-dollar investments into AI. So there might
• 7:05
  Speaker 2: be the case that all these investments are, concretely speaking,
• 7:08
  Speaker 2: an AI bubble.
• 7:09
  Speaker 1: Yes. So when Anthropic in the middle of all of
• 7:12
  Speaker 1: this says, okay, we're going to be making our debut
• 7:14
  Speaker 1: on the stock market, is that a cause for optimism
• 7:20
  Speaker 1: or is that just, I don't know, having a shot
• 7:24
  Speaker 1: in the dark and hoping it hits something?
• 7:28
  Speaker 2: Yeah, I would say it's all about valuation. So we
• 7:31
  Speaker 2: are talking here about the price of the company of
• 7:33
  Speaker 2: the future, not its current worth. So what investors think
• 7:37
  Speaker 2: a company will earn over many years converted into today's money.
• 7:40
  Speaker 2: So this is what you currently see. It can be
• 7:42
  Speaker 2: compared a little bit to a young orchard. So you
• 7:45
  Speaker 2: don't price it by the years, this year's apples, of course,
• 7:50
  Speaker 2: but how many apples will deliver over decades. And tech
• 7:53
  Speaker 2: firms are mostly seeing the same. So most of the
• 7:55
  Speaker 2: value is that we also seen and also the investments,
• 7:58
  Speaker 2: these billion-dollar investments sits far in the future. This is
• 8:02
  Speaker 2: of course a very sensitive thing because if you would
• 8:07
  Speaker 2: change the growth assumption a little, the price can move
• 8:10
  Speaker 2: really a lot. So I would say the question that
• 8:15
  Speaker 2: should be asked is not, or is asked, how big
• 8:17
  Speaker 2: can AI become? Which share of everything companies spend on
• 8:22
  Speaker 2: software and stuff could it capture? But this is really
• 8:25
  Speaker 2: a look into the future that we currently see. And
• 8:27
  Speaker 2: what does this mean for.
• 8:28
  Speaker 1: I don't know, ordinary people like you and me? Maybe
• 8:30
  Speaker 1: not you, but me, for example. If I were to
• 8:33
  Speaker 1: look at this in the form of the stock market
• 8:35
  Speaker 1: and all this sort of bet on a future valuation
• 8:38
  Speaker 1: and how valuable this company could be, it could all
• 8:41
  Speaker 1: come apart very quickly as well, right?
• 8:44
  Speaker 2: Definitely. So this is clearly a bet on the future.
• 8:48
  Speaker 2: And this is the first time that we can look
• 8:51
  Speaker 2: at the real books of one company and one of
• 8:54
  Speaker 2: the top AI labs. And this shows definitely a company
• 8:56
  Speaker 2: growing at breathtaking speed, but also one that has signed
• 9:01
  Speaker 2: up to enormous bills. And Of course, this filing shows
• 9:07
  Speaker 2: huge long-term bills for computing power, reportedly more than half
• 9:10
  Speaker 2: of a trillion dollars, and these risks of warning. But
• 9:13
  Speaker 2: we also see that the trust in these companies is
• 9:17
  Speaker 2: very important at the moment. And they are trying to
• 9:19
  Speaker 2: build up this trust with all these reports about security
• 9:22
  Speaker 2: risk and so on that we can currently see in
• 9:24
  Speaker 2: the IPO report.
• 9:25
  Speaker 1: So we're talking about security risk, trust, and also this
• 9:29
  Speaker 1: aspect of regulation you were talking about from Europe. a
• 9:32
  Speaker 1: short while back. And this is all in reference to
• 9:34
  Speaker 1: just purely American companies, it seems, at the moment. Is
• 9:38
  Speaker 1: there an opportunity here for companies or AI companies or
• 9:42
  Speaker 1: AI development from other parts and other regions, such as Europe,
• 9:46
  Speaker 1: for example? Could Europe step into the breach here and
• 9:50
  Speaker 1: build an AI of trust or build an AI that
• 9:53
  Speaker 1: is reliable financially as well?
• 9:54
  Speaker 2: I would say this is really possible because... we have
• 10:00
  Speaker 2: two different problems here. And I would also say it's
• 10:02
  Speaker 2: a wake-up call and a chance for the European Union.
• 10:05
  Speaker 2: So the first problem is that AI is very good
• 10:07
  Speaker 2: at finding security holes in software. And Antropix Mutus, for example,
• 10:11
  Speaker 2: found thousands of unknown flaws and was kept from the
• 10:14
  Speaker 2: public release a few months ago. And the second problem
• 10:17
  Speaker 2: we currently see is that models are being tricked and
• 10:20
  Speaker 2: misused by cybercriminals And last year, for example, state-backed hackers
• 10:25
  Speaker 2: used cloud code against about 30 organizations. And we also
• 10:28
  Speaker 2: see in the European Union this whole debate on digital sovereignty.
• 10:31
  Speaker 2: So dependence lessens when we had this export order of
• 10:35
  Speaker 2: the US government in June, which led to the situation
• 10:38
  Speaker 2: that frontier models were not usable for about three weeks.
• 10:42
  Speaker 2: and Europe's response is here that we have a clear
• 10:45
  Speaker 2: action plan that has been launched in summer this year.
• 10:47
  Speaker 2: We want to get access to our own models, build
• 10:50
  Speaker 2: our own AI and safety security testing capacities, And also
• 10:56
  Speaker 2: build up a shared test platform. And I would say
• 10:58
  Speaker 2: this is really an opportunity for us as a European
• 11:01
  Speaker 2: Union because AI for regulated sectors is needed. And here
• 11:05
  Speaker 2: we need not just functionality, but also trust, data location,
• 11:09
  Speaker 2: and compliance when we think about banks, hospitals, or businesses.
• 11:13
  Speaker 2: public administration, defense industry, and these are the points where
• 11:16
  Speaker 2: I see that in the near future, new AI models
• 11:20
  Speaker 2: might be interesting for government or private companies or even
• 11:25
  Speaker 2: critical infrastructure sectors.
• 11:27
  Speaker 1: Moving forward in the short term, what do you see
• 11:30
  Speaker 1: happening when it comes to AI? Do you see newer
• 11:33
  Speaker 1: companies coming up with newer models, whether out of the
• 11:36
  Speaker 1: United States or Europe or some other part of the world?
• 11:38
  Speaker 1: Or do you see there's going to be a consolidation
• 11:41
  Speaker 1: beginning to happen with, I don't know, larger models, larger
• 11:44
  Speaker 1: companies absorbing smaller ones.
• 11:48
  Speaker 2: So generally speaking, I would say AI will probably get faster,
• 11:52
  Speaker 2: more independent and more woven into everyday work. And Entropic
• 11:56
  Speaker 2: itself is warning already about where that could lead to.
• 12:00
  Speaker 2: So we currently see a switch. We started with chatbots.
• 12:03
  Speaker 2: Now we go to coworkers. So AI is moving from
• 12:06
  Speaker 2: answering just some questions or generating images to these agents
• 12:10
  Speaker 2: that work autonomously on their own for hours. So for example,
• 12:14
  Speaker 2: writing pieces of software checking cybersecurity or even running complete
• 12:19
  Speaker 2: business processes. I would say that also the case of
• 12:23
  Speaker 2: self-improving AI is getting more important than ever because the co-founder,
• 12:29
  Speaker 2: Jack Clark, recently said, it is more likely than not
• 12:33
  Speaker 2: that by the end of 2028, there will be an
• 12:35
  Speaker 2: AI that can improve itself without humans. The company calls
• 12:40
  Speaker 2: this recursive self-improvement. So what I currently see is that
• 12:44
  Speaker 2: there are a lot of ideas, but we have a
• 12:45
  Speaker 2: lot of borders as well when it comes to chips, electricity,
• 12:48
  Speaker 2: permits for data center also in the US, which is
• 12:51
  Speaker 2: why these huge computing deals exist that we currently see
• 12:55
  Speaker 2: that are made by Anthropic, for example, together with Microsoft.
• 12:58
  Speaker 2: So regarding this, I would say when it comes to
• 13:00
  Speaker 2: the global AI development, AI will be more integrated into
• 13:04
  Speaker 2: everyday's procedures in the next years.
• 13:07
  Speaker 1: Thanks for all of that context, Dennis. Always a pleasure
• 13:11
  Speaker 1: talking to you, Dennis. Kenji Kipka, thanks so much for
• 13:14
  Speaker 1: making the time.
• 13:16
  Speaker 2: Thank you very much.
• 13:16
  Speaker 1: All right, so that was Dennis with all of that context.
• 13:22
  Speaker 1: But what does Anthropic think? We figured we'll ask its chatbot, Claude,
• 13:29
  Speaker 1: about what it thinks about this upcoming IPO. So I'm
• 13:32
  Speaker 1: going to try and ask Claude and see what Claude thinks.
• 13:36
  Speaker 1: So here's my prompt in a sentence. and I can't
• 13:40
  Speaker 1: type sentence properly in a sentence is Anthropix upcoming IPO
• 13:52
  Speaker 1: boom or doom let's see what Claude says Right. I
• 14:04
  Speaker 1: don't know if you can read that, but I'll just
• 14:05
  Speaker 1: read that quite quickly. I can't call it boom or doom.
• 14:08
  Speaker 1: I'm made by Anthropic and no one can predict a
• 14:11
  Speaker 1: stock's performance. But the case for boom is the reported
• 14:13
  Speaker 1: $ 100 billion raise at a possible $ 2 trillion valuation and $ 11. 5
• 14:19
  Speaker 1: billion in quarterly revenue. While the case for doom is
• 14:22
  Speaker 1: a net loss of almost $ 42 billion last year. And
• 14:26
  Speaker 1: one analyst view that it looks like a risky investment
• 14:29
  Speaker 1: that will almost certainly carry a hefty premium. This isn't
• 14:32
  Speaker 1: financial advice, and the IPO is reportedly now slated for November.
• 14:37
  Speaker 1: From the horse's mouth there, IPO potentially slated for November.
• 14:41
  Speaker 1: But tell us what you think. You can write to
• 14:44
  Speaker 1: us in the comments below. I'm Bidesh Banerjee in Berlin.
• 14:48
  Speaker 1: Thanks so much for watching. Goodbye.