RadixJournal - The AI Bubble Pops？

00:00:00.000 So, what is remarkable about DeepSeek are two things.

00:00:24.040 It's open source. So, the code for this AI and large language model and reasoning models, is what they said, are there to be inspected and looked at.

00:00:37.980 There's also some indications that it's basically forking OpenAI back when OpenAI was, in fact, open.

00:00:49.680 It's no longer open. It's a closed source for-profit company.

00:00:53.580 It started out as, I guess, a charity, a 501c3 or something like this.

00:00:58.440 And it's had massive investment from Microsoft, primarily.

00:01:01.480 I think Elon Musk was an early investor, and now he's at war with Sam Altman, etc.

00:01:07.940 Another thing that's remarkable about it is that it's Chinese.

00:01:11.120 So, you ask yourself, is this political in some way?

00:01:14.800 Like, is this a kind of sabotage?

00:01:19.200 And in this very Chinese way, there's this shallow mimicry of Silicon Valley ideology. 0.93

00:01:30.100 So, the company describes itself as, we're moving towards AGI and long-termism is possible through curiosity.

00:01:40.820 So, it's almost like they cut up a bunch of buzzwords, put them into a bag, shook it, and spilled it out on the table.

00:01:47.240 And they came up with long-termism, AGI, and curiosity.

00:01:51.620 So, it's a sort of dubious, murky company, to say the least.

00:02:01.180 What is happening here?

00:02:02.920 And I've seen a number of people react to this, and they say, ah, you know, this is the Sputnik moment for AGI.

00:02:12.520 So, this is like the Ruskies launching a satellite into space.

00:02:17.600 Sputnik, they've launched a dog into space, no less.

00:02:21.380 And we've got to get back to basics, we've got to teach young people science, we have to make long-term plans, a moonshot, if you will, a going to the moon as a goal for the decade of the 1960s.

00:02:40.120 We've got to get our act together because they are ahead of us.

00:02:43.240 I've also heard the other side of this coin, which is that they stole OpenAI.

00:02:50.360 There's actually a moment with DeepSeek where if you say, like, who are you as a program, it will say, I'm ChatGPT 2.3 or something like this.

00:03:01.460 So, there is a high likelihood, maybe certainty, that they were using some of that initial code that they then developed themselves.

00:03:11.660 But I think it's something different than that.

00:03:18.180 What I would relate it to is the Quartz Crisis of the 1970s. 0.78

00:03:26.660 Now, what is the Quartz Crisis?

00:03:30.680 I'm talking about watches here.

00:03:33.540 Why am I talking about watches?

00:03:35.200 It's one of the most famous monumental examples of disruption that has ever occurred in economic history.

00:03:45.080 So, beginning in the 1950s, it was discovered that a quartz crystal could vibrate at something like 32,000 times a second or some unfathomable number.

00:04:06.000 And it would do it with such a regularity that you could do many things with it, including keep time.

00:04:13.720 And could you, with the use of a battery, send an electrical signal through a quartz crystal and keep accurate time?

00:04:23.280 And you would have that ticking of the quartz watch that we're so used to, as opposed to the mechanical sweep of a watch that is wound and is using other sort of mechanisms.

00:04:36.780 And it was 1969, I believe, Seiko introduced a quartz watch with a battery.

00:04:46.000 I'm forgetting the name of it at the moment.

00:04:48.540 We could go find it if we want to.

00:04:50.660 There was other attempts at this.

00:04:53.340 The Accutron, I believe, was a famous one from Bulova.

00:04:57.760 And the one that is maybe most iconic, at least for me, is Hamilton's Pulsar.

00:05:07.920 So, if anyone's seen the James Bond movie Live and Let Die, Roger Moore, in the opening, the pre-title sequence, he actually looks at his watch and then he presses a button on his watch and a digital readout of the time is revealed.

00:05:27.020 And it's sort of clunky when you see that now.

00:05:31.540 And, of course, there's like an Italian secret agent in his bed and he's checking the time.

00:05:35.920 But, you know, it reminds us of a decade earlier where James Bond, played by Sean Connery in Goldfinger, comes out of the water.

00:05:48.900 There's a swan on his head and he's wearing a wetsuit or, I guess, a dry suit.

00:05:53.600 And he opens it up and he's wearing this dinner jacket, tuxedo, basically.

00:05:57.880 Totally awesome.

00:05:58.740 And before the explosives go off that he's planted, he checks his time and what is he wearing?

00:06:05.820 He's wearing a Rolex Submariner watch.

00:06:08.760 Now, today, the Rolex Submariner is, in effect, a highly expensive luxury timepiece.

00:06:17.480 In 1964, for Goldfinger, Goldfinger was 1964, yeah.

00:06:28.200 That, the vibe of it was different.

00:06:31.300 The idea of wearing a Rolex Submariner with a dinner jacket was outrageous in a way.

00:06:38.280 It would be like wearing a tuxedo with maybe like a big G-Shock or like dive computer or like a knife belt, you know, on your tuxedo.

00:06:51.320 The whole point of that scene was the incongruity.

00:06:56.580 James Bond is a cultured thug.

00:06:59.480 He's a blunt instrument in a tuxedo. 0.54

00:07:02.880 He's a total badass with this sport watch, but he's also a gentleman spy.

00:07:11.660 And that's now lost in the sense that, you know, your broker who lives in Wall Street is wearing a Submariner.

00:07:19.660 And it's no longer really a sports watch.

00:07:22.080 It's a timepiece, although still cool.

00:07:26.240 That was also what happened with quartz watches at the beginning.

00:07:30.780 So when James Bond, played by Roger Moore, looks at his Hamilton Pulsar, and he has to press a button to see the time, and it's a digital, red digital display, that was super cool, in fact.

00:07:44.600 And I believe President Ford wore that Hamilton Pulsar quartz watch, and it was highly expensive and kind of incredible.

00:07:56.720 It was like the Apple Watch, but maybe even more radical or innovative, because the idea of wearing something that's computerized, the battery-powered on your wrist, it's a totally new thing no one even thought about.

00:08:12.840 Well, very shortly after this high-end quartz watch came onto the scene, quartz became democratized to a point that it blew away the market for wristwatches in Switzerland.

00:08:33.740 People started making watches in Switzerland because it was cheaper, in fact, many, many centuries ago, and because farmers had nothing to do over the winter, and they could make clocks and movements, etc.

00:08:48.580 But certainly by the 20th century, Swiss watchmaking had become what it more or less is today, high-quality, fashionable, artisan craftsmanship, the equivalent of buying a German sports car Porsche or BMW or maybe an Italian sports car, an Italian suit.

00:09:12.620 That was the vibe that the Swiss had acquired in watchmaking by that time.

00:09:19.200 Well, that's all fine and good, but when you democratize the quartz watch, it starts to go low-end in a radical way that undercoats cuts everything.

00:09:30.180 Why would you pay $250 in that time for a Rolex Submariner when you could pay $25 for some new quartz watch coming from Japan in Seiko?

00:09:45.520 And it was just a radical undercoats, and it was just a radical undercutting of the entire industry.

00:09:52.780 I've read, because I was reading up on this last night, that there were actually 1,600 Swiss watch companies in 1969, and by 1989, 1,000 of them had gone bankrupt, disappeared, and were no more.

00:10:11.880 So the kind of devastation to the industry just can't even be put into words.

00:10:19.680 I don't know if we've seen anything like that in other industries.

00:10:23.840 Maybe someone could come up with an example.

00:10:26.420 So what did the Swiss watch market do in the face of democratized watches from Japan?

00:10:35.020 And also, not only was it much cheaper, but it was actually much better, at least from a certain point of view.

00:10:43.640 So a Swiss watch that's well-made might be off two seconds a day, forward and backward.

00:10:51.080 It might be off five, ten seconds a day.

00:10:53.640 Now, five or ten seconds a day, who cares?

00:10:56.440 I mean, it's still keeping very good time, and you could always reset it once a month.

00:11:01.980 You're totally fine.

00:11:03.140 It's not like it's a terrible instrument or anything like that.

00:11:07.820 The quartz watch, it's like one second a year.

00:11:10.620 The watch that you buy in Walmart for $9.99 is far more accurate than the watch that some rich guy, the Patek Philippe that a rich guy buys for $100,000.

00:11:28.000 It's not even close.

00:11:30.420 That Walmart product is, on one objective level, infinitely better.

00:11:35.720 So, how did the Swiss handle this problem?

00:11:42.360 A couple of ways.

00:11:45.020 During the 70s, a lot of these other companies were making quartz watches.

00:11:49.820 So you have, like, Rolex made some quartz watches and things like that.

00:11:53.280 They were trying to get on the trend.

00:11:55.520 But they soon realized that that was a bad idea.

00:11:59.200 And by, I don't know when they stopped, but certainly by the mid-'80s, Rolex would never make a non-mechanical watch at all.

00:12:08.060 And they started promoting the idea of, like, the sweep of the hand and things like that, as opposed to the tick of the quartz watch.

00:12:15.120 This is better, more elegant, or something.

00:12:17.880 And so, there was a huge consolidation of these companies in what became known as the Swatch Group.

00:12:29.520 There was an interesting man named Nicholas Hayek, who's actually really a brilliant man.

00:12:34.140 He's a scientist and engineer and also a businessman and watch aficionado from Switzerland.

00:12:42.140 He's a fascinating guy.

00:12:43.060 And so, he combined all these companies where they were, you know, achieving economies of scale.

00:12:50.680 They wouldn't all go out of business.

00:12:52.540 They would do mutual marketing, you know, share movements, etc.

00:12:56.840 You know, basically, consolidation is far more efficient than competition with 1,600 different watchmakers.

00:13:05.640 They also, they both embraced quartz, and then they leaned well away from it.

00:13:12.720 So, they created the Swatch notion.

00:13:16.420 So, Swatch is the, it's not Swiss watch.

00:13:19.620 It's second watch, Swatch.

00:13:21.920 So, we're going to sell you a watch.

00:13:24.160 And I don't quite remember how much these sold for in the 80s, but, like, let's just say 30 bucks or 20 bucks or something like that.

00:13:30.280 It was very affordable.

00:13:31.960 You could buy 10 of them.

00:13:34.080 And they were cool.

00:13:36.180 They were fashionable, very 80s.

00:13:38.900 I might even, I'm sometimes tempted to go buy, like, a super 1980s Swatch from eBay or whatever that just looks like a Duran Duran album or something.

00:13:52.080 They're fun and cool, and they still exist today.

00:13:54.500 So, they embraced quartz, and they embraced cheap, high profit margin, or maybe low profit margin, but definitely cheap, high volume, I should say, watches.

00:14:05.320 And that certainly brought in revenue to the overall conglomeration of Swiss watchmakers.

00:14:11.040 They also leaned in to the luxury.

00:14:13.620 So, Rolex was making cool sports tool watches at one point.

00:14:20.760 They started to make a watch that was a luxury item.

00:14:25.300 You know, what is the Patek Philippe line?

00:14:27.640 Like, you know, you don't buy a Patek Philippe.

00:14:31.480 You pass it on to your grandchildren.

00:14:34.140 It was something like that.

00:14:35.520 You buy it for generations.

00:14:37.860 They weren't saying that previously.

00:14:40.060 They now sort of had to say it, and they went up market.

00:14:44.480 So, quartz started up market at the very beginning, but quickly went way down market.

00:14:51.180 And that is the idea of classic disruption theory of a down market solution that is good enough or, weirdly, in quartz's case, better than the mid or high market solution, and it just upends the industry.

00:15:07.420 Quartz crisis definitely was that.

00:15:09.020 So, Rolex, Omega, Patek Philippe, Audemars Piguet, et cetera, et cetera, they did new designs, but they leaned into the mechanical watch.

00:15:20.840 To wear a quartz watch is sort of cringe or declassé.

00:15:29.000 You know, they leaned into the snobbery of, I inherited this Patek Philippe from my father, or I am successful enough to buy a Rolex, or a speed master, the watch that went to the moon, quote unquote.

00:15:47.340 That kind of stuff that they have basically been doing for decades and decades.

00:15:52.540 They've never stopped it, and I don't think they will stop it, because that's sort of all they've got.

00:15:58.520 But it has been remarkable.

00:16:02.140 So, why have I been talking for 20 minutes about Swiss watches?

00:16:05.120 Well, I'm sort of fascinated by the whole thing, first off, obviously.

00:16:09.320 But there's something else to it.

00:16:13.080 So, I think what is dramatic about Deep Seek is that it's basically the quartz watch of AI.

00:16:25.320 And in this hilarious, you could say, way, it's proving how AI is not all that it's cracked up to be.

00:16:37.360 So, maybe they did fork open AI from a couple of years ago.

00:16:45.540 Maybe they're copying from other companies, and other companies are copying from other companies, et cetera, in some big circle.

00:16:51.880 Regardless, on a fraction of the investment in training, they have basically produced something that is good enough.

00:17:04.220 And I've even seen some evidence out there that it's, in fact, better.

00:17:09.860 And so, you don't, like, open AI, the way I view it, it's almost like the Swiss watch market without any of the romance.

00:17:23.300 So, what I mean by that is that it's not open AI.

00:17:26.460 It's a closed system that has received billions of investment with the assumption that it's going to be highly profitable from groups like Microsoft, et cetera.

00:17:35.320 But why would you pay for this service if you can get something much simpler and easier from the Chinese knockoff company, the Seiko of our time? 0.99

00:17:52.480 The Swiss watch market was able to survive because they leaned into the romance.

00:17:58.880 There's no romance to Sam Altman.

00:18:01.420 There are a couple of things that he's playing on recently.

00:18:06.960 There's the one-day-ism of one day in the future, we're going to have AGI, and it's going to think for us.

00:18:16.980 I saw Mark Andreessen was saying how, you know, we worry about replacing labor with AI, but we might actually replace the C-suite with AI.

00:18:26.260 I don't know what he's playing there.

00:18:27.720 And there's the little tech Silicon Valley patriotism that has been promoted by these people, how we need to dominate the world in AI.

00:18:39.020 We can't let the Chinese do it. 1.00

00:18:41.760 And look, we have this $500 billion investment.

00:18:45.040 It's going to cure cancer one day, maybe once we get there.

00:18:48.760 But keep the money flowing in because we need these massive, massive server farms to train this data and to build it up.

00:18:58.260 I don't think an LLM is ever going to merge into AGI.

00:19:05.920 That is something that has a thanks for itself and in a way has a will of its own.

00:19:11.260 That's just actively solving problems, reproducing itself, maybe destroying the world in order to make more paperclips or whatever.

00:19:19.720 I have never bought into AGI doomerism, but I also don't buy into AGI utopism as well.

00:19:26.540 I use AI on a, if not daily, weekly basis for sort of things, but I'm really not sure it's much of anything than if-than statements on a search engine.

00:19:42.040 It's a kind of development of Google giving you responses that are, in effect, averages.

00:19:50.300 It's giving you the normie response.

00:19:53.920 It's a midwit normie generator in terms of ideas.

00:19:57.860 That can be very useful.

00:19:59.880 Don't you think that, to some degree, it might be cheaper for them to do what they're doing because they don't have to train it on so many, like, on all these racial grievance stuff and all this, like, feminism stuff? 0.84

00:20:16.260 That may be, but, I mean, this is the other problem with these companies like Gemini with Google, and I assume Llama has the same thing, and also OpenAI, is that, you know, with, say, the Swiss watch companies, people loved these companies.

00:20:35.640 With AI, people sort of hate them.

00:20:41.180 You know, like, Google came out of the gate with this image generator that was making George Washington black and gay and female.

00:20:50.700 And that pissed off the conservatives, but I think it sort of pissed off everyone and created a huge credibility crisis.

00:21:01.040 I don't think DeepSeek is better able to do that because there's not the political correctness overhang, but maybe that had an effect.

00:21:08.240 But I think the bigger issue is that it's not, like, the political notion that we can dominate the world in AI and that we can prevent China from catching up because you need all of this massive investment in infrastructure.

00:21:27.540 I just think that's been revealed to be fake.

00:21:33.440 Large language models aren't that great.

00:21:36.280 At some point, they're going to produce the cheap Seiko that is good enough or weirdly better than the stuff that they have dedicated hundreds of billions of dollars towards.

00:21:50.420 And AGI is never happening.

00:21:55.120 And selling people on the notion that HAL will one day save us all is nonsense.

00:22:02.840 The overwhelming truth is that they have produced an LLM that is just as good as Claude, Gemini, Llama, OpenAI, ChatGPT, et cetera, et cetera, et cetera.

00:22:21.960 And that this notion that they're years behind or we could prevent them from catching up or that there's something altogether special about these LLMs, I think, has just been debunked.

00:22:35.960 It's also of interest is the degree to which little tech, David Sachs, who's the AI and crypto czar, and Andreessen and all these people.

00:22:57.900 They covered AI in this veil of populism or nationalism or conservatism.

00:23:08.140 And it's like, we've got to win this war.

00:23:11.760 You know, and it looks like they've lost.

00:23:18.440 And I'm sure Trump is exceedingly furious at them because he imagined that last week's announcement of, you know, we're going to cure cancer one day was this huge win for him, this way of appealing to a base beyond MAGA.

00:23:37.280 I mean, MAGA probably hates the idea of mRNA vaccines created through AI.

00:23:42.760 They're like the least prone to liking those kinds of things.

00:23:47.500 But he could reach out.

00:23:50.940 He could get a win.

00:23:52.200 He loves the stock market.

00:23:53.800 And I feel like unless OpenAI can advance beyond what they're doing right now, they've sort of already lost.

00:24:03.440 Because AI, or I shouldn't say AI, LLMs are now commodified.

00:24:10.700 Period.

00:24:12.060 Like a quartz watch is kind of commodified.

00:24:15.220 I mean, let's just say, I mean, with a smartphone, you don't even need a watch.

00:24:18.660 But, you know, let's say you're on vacation and you forgot to pack your watch.

00:24:26.220 You just go to some store and just like, yeah, what do you got?

00:24:29.500 Oh, that one.

00:24:30.660 Oh, that's the cheapest.

00:24:31.580 Yeah, this one's $5 more, but I kind of like the color.

00:24:34.960 I'll just buy that.

00:24:35.700 It's just a pure commodity.

00:24:37.940 It's bacon in a way.

00:24:39.940 It doesn't matter what brand of bacon.

00:24:42.180 I mean, maybe it matters to some degree, but you get the point.

00:24:45.580 It's eggs.

00:24:47.180 It's bacon.

00:24:47.720 It's toilet paper.

00:24:48.720 It's just a commodity that can easily be reproduced.

00:24:52.660 And it's not something special.

00:24:54.920 Like, what you add to the AI could be special in your niche market.

00:25:01.600 But the LLM thing, it's been done.

00:25:06.980 We have seen maybe the limits of its potential, in fact.

00:25:11.720 And now the question is, what are you going to do with this commodity that everyone has?

00:25:18.640 What's up on minus one another emoji?

00:25:20.760 Well, that's because you're going to be gerrym怎么了.

00:25:22.380 So like, of course, they might be a big challenge.

00:25:23.800 So like, you're going to get there.

00:25:25.140 If you do that, there'll be somelines like this. 0.98

00:25:25.880 There's chill on the righteousness.

00:25:27.380 On the bottom, there's a decrease in the mouth.

00:25:28.760 It's not a good sign.

00:25:29.540 I find ity6.

00:25:29.980 There's the loud kommunaler. 0.98

00:25:30.500 So like, of course.

00:25:31.760 I'm doing this.

00:25:32.400 Bye.

00:25:32.580 See you next time.

00:25:33.760 If it's wall out.

00:25:36.600 I'll be surprised.

00:25:38.800 Fireflies.

00:25:39.220 It's magic worthwhile.

00:25:40.480 It's practical nine times.

00:25:41.600 I just go a good sign.

00:25:43.380 It's epic.

00:25:44.140 It's obvious.

00:25:44.440 It's like they're here.

RadixJournal - January 29, 2025

The AI Bubble Pops？

Episode Stats

Summary

Transcript