关于LLM训练的重大裁决、科技行业NBA式薪酬的盛夏来信、2025年从零开始，以及更多

Hello, and welcome back to another episode of Sharp Tech. I'm Andrew Sharp, and on the other line, Ben Thompson. Ben, how are you doing?

Frazzled, Andrew. I'm frazzled. This is our last episode before the summer break, which we do every year.

The end is near.

As usual. The end is near.

We need the summer break this year. Absolutely.

But I am traveling—

I can hear it in your voice.

—for the summer break. It requires a lot of preparation, this particular trip, and I'm just ensuring—doubling down and making sure I go on this trip and feel, “Man, I needed this vacation.” So that's what I'm doing right now.

There you go. Well, I'm happy for you, and this is why I said earlier in the week that we were going to keep it light, keep it summery, have some fun questions peppered in. The audience responded, and we're not even going to be able to get to all the fun summer questions on this episode, but we'll keep those rolling all summer long.

We have news at the top here. So, as we said earlier—

Unfortunately, we actually have a real topic to get to quickly. I'll try to keep it short.

I know—I feel guilty inflicting fair-use discussions on the audience, but it is pretty big news. So I'll read from The Washington Post:

“A federal judge this week ruled that artificial intelligence company Anthropic did not break the law when it used copyrighted books to train its chatbot Claude without the consent of the texts' authors or publishers. But he ordered the company to go to trial for allegedly using pirated versions of the books. The decision, made Monday by Judge William Alsup of the U.S. District Court for the Northern District of California, represents a win for AI companies, which have battled copyright lawsuits from writers and news organizations for using their work to train AI systems.”

So, Ben, the floor is yours. What do you think of this ruling? I can probably guess based on some of our previous conversations, but break it down.

1. Fair Use Favors AI Training

I think it's great. It's fantastic. It's good news. I mean, it's always tricky with anything legal, right?

Mm-hmm.

You have to separate “Is this a legally correct ruling?” from “What do I want to happen from a policy perspective?” Now, conveniently enough—and one might argue that it biases the discussion—

Some confirmation bias in play.

I think this is both. I've been writing about fair use for a long time. I've written about it in other contexts, but certainly in the context of AI. I thought this was a very gratifying decision because I feel like the judge's decision was a perfect repeat of the arguments that I've made a couple of years ago.

He gets it.

No, absolutely. Absolutely.

It was also a well-written opinion, a very well-written and well-reasoned opinion.

I feel federal judges, by and large, have well-written opinions. Court cases are generally good reading. It's the lawyers and their briefs that are the problem.

Oh, boy.

Not that I ever saw yours, but yeah. So you have fair use. The really big point is that there are these different standards. Fair use is tough because, actually, even zooming out, every fair-use decision is case by case.

Mm-hmm.

That is kind of maddening, but I think it's actually super important to the broader point, which is that intellectual property law is not a God-given right or the natural order of the universe. It is definitionally a trade-off.

It's a government-granted monopoly. The government does that because it sees the benefits of monopoly, but you, as our resident anti-monopolist, should be intrinsically uncomfortable with intellectual property law because it is monopoly, right?

Mm.

And all the downsides that come with that. I think it's really important to ground any discussion in these points because I feel a lot of advocates in favor of stronger copyright laws or stronger patent laws, or whatever it might be, forget that it is implicitly a trade-off. Therefore, it's appropriate to consider the upsides and downsides, the benefits, and everything else that's going on, because that's implicit in the concept of intellectual property rights as a whole.

Yeah, I think that's fair. Intellectual property rights are ultimately just a policy to incentivize the creation of great art, music, literature, and whatever else it may be.

That's right.

We're producing copyrighted content. We have an interest in intellectual property law. We sell—

Yeah.

We sell content for a living. If anything, our biases should be in favor of copyright. But you do see a lot of discussion about this where people just take it for granted that obviously copyright exists. It's like, no, it's not obvious at all.

Well, it—

It's a policy choice.

Yeah, it doesn't necessarily exist in other countries the way it does here. But I think, in general, intellectual property protection is a good thing, and so it is important to weigh that against the upside of AI.

Well, generally, that's why fair use is a case-by-case balancing test.

Mm-hmm.

A balancing test implies that trade-offs are involved, so it's appropriate that it is a balancing test and not a hard-and-fast rule. You have these 4 prongs of the test: What is the purpose and character of the use? Is it transformative? The nature of the copyrighted work, the amount and substantiality of the portion to be used, and the effect upon the potential market for the copyrighted work.

For example, fair use is less likely to apply to a commercial copyrighted work and more likely to apply to something like an instruction manual.

An academic use.

Yeah.

Yeah.

Well—or no. Academic use is one side, but the nature of the original work is another, right? I'm looking at a photocopier here, or a printer. If someone hosts that printer manual in a database of printer manuals, it's almost certainly going to be allowed under fair use because it's a printer manual.

Right.

No one cares, right? That's different from hosting a novel, which is assumed to have more protection.

Mm-hmm.

Again, that seems fuzzy and odd. Why can't it be clear? But that's because this entire area involves trade-offs. Of course, we want to make different trade-offs for a printer manual than we want to make for a novel. So that's number 2.

Number 3 is the amount and substantiality of the portion to be used. Number 4 is the effect upon the potential market for the copyrighted work. The most important is the first one. Well, actually, let's start with the third one.

Okay.

Number 3: the amount and substantiality of the portion to be used. One of the challenges with LLMs and copyrighted material is that they use all of it, right?

I think about fair use in the context of Stratechery, where I quote stuff all the time. I always use The Wall Street Journal or Bloomberg because of the way they write. They have a classic “everything you need to know is in the first 3 paragraphs, and everything else is filling it in” approach. It's very easy for me to grab it, put it at the beginning of an update, and write about it.

Yeah.

I take a portion, and I believe I'm entitled to do so under fair use. I'm not copying the whole article, right? There have been times I've reached out to authors and said, “Can I reproduce this whole thing because I want to talk about the whole thing?” I've gotten their permission just via email or whatever, saying, “Yeah, that's fine. You can do that.”

Mm-hmm.

But I'm careful because I'm taking the whole thing.

Right.

If I'm just taking a paragraph, I think I'm entitled to do that. There's precedent for that.

Well, and also, most of what you're taking in those paragraphs is just raw information that The Wall Street Journal or Bloomberg is conveying.

That's right. I could have gotten it from a gazillion other sources. It's just out there.

Yeah.

That's right. But in this case, LLMs are taking everything.

Mm-hmm.

That's an initial reason to think, “Oh, this must be a big deal.” I thought the judge's understanding of LLMs was very strong. Basically, the point is that it's like Google and the web. Yes, any one book needs to be used in its entirety, but so much material is needed that any one book actually isn't that material, relatively speaking.

Mm-hmm.

You could exclude it just as easily. I think that's the correct interpretation: No one book in an LLM's corpus is critical to the LLM's functioning. It actually is completely immaterial. Any one book could be taken away with no impact whatsoever.

It's sort of like looking at the substantiality factor and flipping it on its head. The book is substantially unimportant, so it doesn't really matter if it's all in there.

Right.

And then that gets to number 1: the purpose and character of the use. Is it transformative? I think, undoubtedly, yes.

An LLM is producing completely new content. Judge Alsup came back again and again to this point, and we should talk about it.

Mm-hmm.

The point was that Claude does not reproduce copyrighted works, and the authors were not arguing that it did.

Right. The OpenAI case with The New York Times is one where The New York Times was able to force ChatGPT—and this is also hotly contested in that case—but, in some cases that The New York Times cited, ChatGPT was spitting back full New York Times articles.

That's right. And that's at stake in a Meta case about this as well.

Mm-hmm.

Let's park that and come back to it in a moment.

Okay.

That's something to follow up on. But I think, by and large, LLMs are clearly transformational. They're generating completely new content, and there's a bit here where the judge analogizes to humans, right? I read a lot of stuff, and I generate new content.

Mm-hmm.

Is that stuff going into it? Yes. Does that mean they can restrict what I write? Of course not. We've talked on this podcast: Is it right to analogize humans to computers doing the same thing at scale?

That was the one piece of his opinion that I didn't really love because I agree that, ultimately, the output from LLMs—he used the word “orthogonal”—is orthogonal to what you're getting from a full book from an LLM. But the scale of what's being done by LLMs when they train on these books, I do think that makes it a little bit different from humans reading a bunch of books, internalizing that information, and then producing their own original works over the long term.

Right. But it's not reproducing the work. It's unquestionably generating original work. And I think this gets to number 4, which is the effect on the potential market. I think it's this: Are LLMs going to be challenging for authors going forward because there's more stuff to read? Is it going to be a challenge for graphic designers? Something I do care a lot about personally.

Mm-hmm.

Is it going to be a challenge for them going forward because anyone can just generate an image?

Yes.

Yes, for sure. That's also not part of the argument here. Copyright law didn't exist so that you'd have less competition from other authors. It exists so people can't copy your work. And at the end of the day, he distinguished between what's going into the LLM and what's coming out, and as long as what's coming out is not a facsimile of the content, it is a fair use case. Again, I get the concern and the fear, but under what copyright is supposed to be, and stepping back to the broader point about it being a trade-off, I think this is the right determination.

Yeah. Well, as far as the scale argument, you've read a lot of books, but you haven't read 7 million books. That's what I was alluding to earlier.

Nope.

What this technology makes possible is essentially copyright infringement on a massive scale.

It's not copyright infringement.

Yeah.

You know, that's the point. No, this is like the antitrust thing we talked about.

Ingesting all this material—

That's the antitrust thing we talked about before. If you think that's a problem, then make a law about it. Don't try to retrofit copyright, which is about copying.

Mm-hmm.

I'd be like, like if co—

Well, that's fair. Look, my main takeaway from all this, reading the opinion, was that it was a well-reasoned opinion. But my main takeaway was to double back to what I was saying earlier in the week about longer-term AI concerns. I would be more encouraged if the federal government were even attempting to reckon with any of these issues, and it's alarming to me that, generally speaking, it seems like all of the most consequential questions in tech are consistently answered by random district courts around the country.

If the federal government wanted to take a lead here and set a policy and reckon with these trade-offs, I think that would probably be a good thing. But confined to the legal arguments, I do think that the judge ended up in the right place on this issue.

Yeah, but I think that, for this specific point, we don't need a new law. I think copyright law and the fair use test are very appropriate, have been appropriately applied, and came to the correct outcome. I think the analogy to antitrust law is useful. There are some cases where antitrust law works fine, and there are other ones, with the aggregators in particular, where we do need new laws.

Yeah.

Now, in this case, I do think there's room for Congress, and that gets to what we just tabled: the discussion about this question of forcing ChatGPT, back in the day—which, by the way, I think is limited now—to reproduce The New York Times.

2. Congress Must Set AI Rules

Well, before we get there, can I just say that the reason I would say it might make sense for there to be some rules of the road here, and might make sense for Congress to intervene, is because you're right. Legally speaking, it's not the court's job to reckon with the future market for copyrighted work, whether it's books, movies, or music. Because if you apply the logic in this case to training on all music created or all video created, you could eventually find a proliferation of AI-generated audio and video content—

Oh, that's already happening.

That's already here. Look at TikTok.

Yeah.

That absolutely competes in the same general market as the copyrighted material, the same way Netflix is ultimately competing with YouTube or TikTok. So the net effect of allowing training on those works is to depress the long-term market for human-generated material that's ostensibly protected by copyright, and I don't know whether that's a good thing. I would hope that people would reckon with it, but I don't know that copyright necessarily solves the problem if you're bringing cases.

Yeah. Well, we can and have had debates about this. My argument about the OpenAI and New York Times case from a philosophical perspective is this: The AI explosion is good for The New York Times because having trusted, branded content that is a totem—as we discussed with CNET strategy or Sharp Tech content—is actually beneficial if you have—

If you're differentiating—

There'll be more of a market for that.

Yeah.

Do we want to handicap our AI industry to protect undifferentiated content? At the end of the day, that's sort of the argument that's implied here.

Mm-hmm.

But I do want to get to that OpenAI–New York Times bit. The way to force ChatGPT—and this was ChatGPT-3, I think, back in the day—to copy The New York Times was that you had to paste the first 2 paragraphs of a New York Times article, and then it would spit out the rest because—

Yeah.

It got locked into the predictive set of tokens that it was trained on. And whose fault is that? I could take a New York Times article, put it on a photocopier, press a button, get 1,000 copies, and walk around and hand them out. Is that the photocopier's fault?

Mm-hmm.

There has, in fact, been a Supreme Court case about this, which is the Sony v. Universal case, the Betamax case, where Universal Studios sued Sony because you could record a TV show and then watch it later. The Supreme Court ruled, number 1, that it was fair use to time-shift on your own, and number 2, because it was fair use, the Betamax had lots of legal uses, not just illegal ones.

Yeah.

So it wasn't their fault, right? I think this distinction between the toolmaker and the user making a decision is a really important one in general. The counter was the Napster case, where basically the decision was, “Look, it's used for nothing but illegality.” But that is also the problem with the Napster case in some respects, and with some of the BitTorrent cases and things along those lines. I do think the distinction, by and large, is an important one, and I think it falls heavily in favor of the LLMs.

Mm-hmm.

LLMs are quite obviously used for far, far, far more than trying to reproduce a New York Times article and should not be held liable. The user who does that can be held liable because they're doing an illegal action, but not the toolmaker. And so, even in this case, the implication of what Judge Alsup was saying is that because Anthropic, in their product—the Claude product—had filters for copyrighted content, that meant they were okay.

Right.

My view is actually that it shouldn't even matter. Filter or no filter—and this is a big problem for open source and just model weights and LLMs being available generally—you shouldn't have to wrap an LLM in a filter for it to be legal.

Mm-hmm.

Hold users liable by all means. The tools themselves should not be held liable.

Right. I mean, that was an element of The New York Times case that rang pretty hollow to me. We’re talking about extreme edge cases that were clearly generated for the purpose of bringing a lawsuit, and so holding—

Right.

—OpenAI liable for that behavior seems unjust.

But there are other ones. Some of the Llama 1 examples—and I linked that in there; Timothy Lee wrote about this—show that it’s actually somewhat trivial to generate a fair bit of content.

Oh, I hadn’t seen that. Okay. So you mentioned in your piece—I didn’t read the Tim Lee portion of what you wrote in your update because it published about 15 minutes before we came on to record the podcast.

Yeah, my fault.

But—

You’re excused.

You said this bears on the future of open-source models. In what respect does it bear on open source versus some of the big model players?

Well, Judge Alsup said again and again that Anthropic’s output is not violating copyright.

Right.

The reason its output is not violating copyright is because Anthropic has filters applied, which are really annoying, right? Anthropic is the one that chides you the most or refuses to give answers most often. They’re the most overbearing about that. In this case, it pays off because their product is very good about not letting you generate—

Copyrighted material.

—or reproduce copyrighted material.

Yeah.

Right? The less you have of those filters, if you just train an LM without any of the RL sort of stuff on it or the filters on it, or whatever the product does to stop that copyrighted material from being regenerated when the user is asking it to regenerate copyrighted material, the more likely you’re going to get it, right?

This is the trade-off. We should want there to be more less-restricted, less-censored, broadly available models that anyone can use to do what they want.

Right.

Implicit in that is those models could be used to generate copyrighted content, and my view is: hold the user liable for that. Don’t hold the tool liable. If we hold the tool liable, we’re de facto restricting the market to closed products like Claude or ChatGPT.

Because you’re imposing massive administrative costs in order to prevent—

Massive liability risk.

—that sort of liability.

Yeah.

Yeah. Okay.

Yeah. So, number 3: There’s one more. The other thing about this is that this is just a motion about whether the case could proceed or not.

Mm-hmm.

The case can proceed on the pirated-content issue, and Meta is in trouble for this too. There are archives of millions of books that have fallen off a truck on the internet.

I saw that. I was not familiar with those archives, but I’m glad—

Oh, yeah.

—these model makers are availing themselves of them.

Oh, yeah.

Anthropic’s like, “Well, this is fair use,” and the judge is like, “Well, just because your usage of it is fair use doesn’t mean your acquisition is now suddenly legal,” as in getting all this content.

Mm-hmm.

What was ruled fair use was Anthropic buying a bunch of used books, tearing off the binders, scanning them, and then throwing the physical books away and keeping the digital copies.

Right.

That was fair use, and that’s directly downstream from the Google Books case previously. I think that’s correctly decided. I also think it’s correct that, look, Anthropic pirated a bunch of stuff, and you don’t get a free pass for doing that.

Right.

But this also doesn’t make sense in a way. Anthropic bought mostly used books, so it’s not like the authors are being deprived of revenue. They went through all the hassle of scanning them in and having access to them. At the end of the day, the effect is no different than having just pirated them, beyond the fact that there’s a legal hurdle here.

Fig leaf.

Right.

Yeah.

No, exactly. I understand Judge Alsup’s decision here. I also think if we’re talking about Congress acting, this is where Congress could make a big difference. If the requirement to get books into your system is to scour the earth for used books, cut the binders off, and set up a scanning operation, we’re cutting off startups. We’re cutting off new companies.

Mm-hmm.

Anyone who’s not willing to follow the law or is in another country and doesn’t care about U.S. IP law, like China, has a huge advantage. This is where I think Congress should recognize AI is really important. I believe there are patents in the U.S. Constitution, but by and large, copyright law is all downstream from congressional acts.

Right.

We should tell the Library of Congress—and Congress has to do this; the Library of Congress can’t do this unilaterally—to digitize all your content and make it available to U.S. AI companies to train on, which is fair use, and anyone can come and get it. Not only does that level the playing field between startups and incumbents, it actually advantages U.S. companies specifically by giving them more opportunities to train models.

I get the controversy here because you’re—

You’re prioritizing AI at the expense of the works in the Library of Congress and—

My point is this is happening regardless. Who do we want it to be easy for? Do we want it to be easy for U.S. companies trying to follow the law, or do we want it to be easy for non-U.S. companies and bad actors?

Mm.

In this case, I think the trade-off is worth making. AI is so important.

Yeah. Your argument makes sense to me, and earlier on, when I was talking about Congress getting involved, I didn’t know where Congress should fall. Honestly, you’ve convinced me to some degree that you have to—

Wow.

—prioritize AI.

It only took 20 minutes. Great.

But it would be encouraging if I had a sense that Congress was paying attention to any of this, and that’s where I come back to: Do we have the ability to pass laws and reckon with any of these questions at this point?

Right now, it’s a case-by-case basis, and literally a case-by-case basis because some of the courts elsewhere in the country may reach different answers on these questions. Hopefully, one way or another, we get some sort of black-and-white rules that people can anchor around as we move forward.

No, that’s right. This is the first of many. Most of the decisions have gone broadly in favor of the AI companies. I think this is the clearest. It directly takes on the fair-use question, and I think the arguments are very compelling.

We’ll put a link to the case in the show notes. I think it’s worth a read. Obviously, I just finished writing about it for Wednesday’s daily update. All in all, this is great news.

Mm-hmm.

So—

We’ll see.

Good note to enter the summer on.

There you go. Celebrate this fair-use decision on your Fourth of July next week.

And everyone who’s angry about it and emailing us—sorry, we’re not going to address it for at least a week and a half.

Yeah, exactly. We’ll double back. I’m sure we’ll talk more fair use as the summer unfolds here. But speaking of mail, let’s move to mail. Eric says—

Okay. I’m limiting all my answers to 2 minutes.

Yeah, I’ll believe that when I see it.

3. AI Researchers Command Athlete Money

Eric says, “Regarding Tuesday’s update on athlete money for AI researchers, one of the aspects of tech versus sports is how impossible it is to objectively judge someone’s impact in the moment. I feel like by the time someone is notable enough to deserve $100 million, it’s likely they’re past their productive usefulness to actually deserve $100 million.

“For example, it’s totally true that John Carmack has created a ton of value, but I don’t think that if you hired him today, he would provide that additional value commensurate with what he could command.”

I thought this was an interesting take on the frothy market for big names in AI. It’s also something that I wondered about in the wake of the Jony Ive news. What do you think of that comment from Eric?

Well, I think this is the same in sports, right? Most guys get paid when they make the most money, particularly because of the NBA structure. I wrote about this in the context of the NBA.

It’s so biased toward seniority that you’re generally overpaying old guys and underpaying young guys. There is a bit where you’re always getting paid for—

Past performance, yeah.

There’s a mismatch, a timing mismatch. Yeah. I mean, I think what’s interesting about the AI market—this is the point I made in Tuesday’s update—is that a lot of times impact is really company-specific and it’s a big deal, but that sort of limits the bargaining power of the employee.

And so the way to get your true worth was to start your own company.

Mm-hmm.

In that case, you're directly generating the value that you can harvest later on in an IPO or whatever it might be. What's interesting about AI is, because it is fairly commoditized—everyone's kind of doing the same thing—that is actually a huge benefit for AI researchers. What they do at Google is what they're doing at OpenAI, what they're doing at Anthropic, and what they're doing at Meta.

Mm.

Because what they do is very specialized, but the overall job is commoditized. It's not specific to Meta, and it's not specific to Google.

30 NBA teams. One of my favorite quotes from an NBA player was somebody asking him where he was going to go in free agency, and he said, “I don't know. There's a bank in every city.” There's a bank in Menlo Park. There's a bank in all sorts of places for these researchers.

That's right. There's a bank in Mountain View. There's a bank in San Francisco.

Yep.

No, that's exactly right. I think that's kind of a unique moment. There's always been a case for this, even for infrastructure, right? Google has its own infrastructure, and Facebook has a more open one. But I don't know if it's ever been the case to such a degree where there's such a small number of people who are readily identifiable as being good at this.

Mm-hmm.

Everyone's trying to do the same thing, and outputs are somewhat measurable, so you know who's good and who isn't. It's much more sports-like for top AI researchers than the market for tech talent has probably ever been, which is why they're starting to get paid like NBA superstars.

But is there some truth to the idea that an engineer or a researcher has a prime, and their prime is in their late 20s, let's say, as opposed to their early 40s?

Probably.

Mm.

We're both past our prime, let's be honest. So who gets here next?

Long past our primes.

Indeed.

But the checks keep cashing. That's all that matters.

The other thing that I wanted to mention, because it occurred to me reading your update on Tuesday and it made me laugh—well, it's actually two competing thoughts I had reading your update. On one hand, I was reading it and thinking about our conversations about athlete money for all these tech employees, and it occurred to me: $6 billion for Jony Ive, or $6 billion in stock; $14 billion for Alexandr Wang; maybe $1 billion for Nat Friedman and Daniel Gross. Those numbers are, in fact, way bigger than athlete salaries across all the major sports, even soccer. The soccer guys get paid crazy amounts of money, but not AI money.

And then a second later, I thought about the percentage of overall value at OpenAI and Meta. In that context, $14 billion for Alexandr Wang or $1 billion for Nat Friedman is actually far less than what the best players in the NBA are getting. Just to underscore your long-standing point about all the numbers in tech being absolutely ridiculous, it amused me as I was reading on Tuesday.

It's absolutely ridiculous. It really is. You do appreciate the extent to which, almost definitionally, tech employees have always been underpaid.

Yeah.

But to the point, it's been hard to measure. Also, tech companies have been found appropriately guilty of colluding to keep that down. So there you go.

That's right. I wasn't familiar with that case until you surfaced it on Tuesday, but that was amusing as well.

Oh, yeah.

I had seen the Steve Jobs email to his Adobe counterpart, but I didn't know that was downstream from an antitrust case.

Yeah.

NBA guys are getting close to parity, but a lot of these players are going to make $500 million or $600 million by the end of their career. So they're compensated fairly, and maybe that's where tech employees are headed. At least the very best are headed there over the next couple of decades.

4. AI Automates Medicaid Billing

Following up on my request for cool LLM use cases, here is Alex from Ohio. This is in 2 parts. Part 1:

“Hey, guys. My wife is a former special education teacher who recently started a day program for adults with disabilities. She bills Medicaid, and it sucks. You pretty much need to submit TXT files with incredibly specific spaces and characters to separate different data fields. It's wild. She used to do this manually, quite literally opening TextEdit and counting out as she hit the space key.”

God, that sounds miserable.

“A few months ago, as she added new members to her program, she spun up ChatGPT and decided to ask if it had a better way to do it. ChatGPT created a Google Apps Script that allowed her to use a Google Sheet format she created and then translate it to the Medicaid format. That was great for a while. This week she spun up Google Gemini, our new favorite, and it created a custom script where she just had to hit a checkbox on a spreadsheet, and it spit out the right billing information. It's a huge time-saver, and she was really ecstatic after a few hours of back-and-forth with Gemini.”

He continues from there, but I'll stop. What comes to mind when you hear about that utility?

No, I think it's awesome. The other thing that's interesting is that it speaks to a certain agency point that we keep coming back to. He mentioned that she was really ecstatic after a few hours of back-and-forth with Gemini. Without question, she's going to save way more than a few hours.

Mm-hmm.

It still takes the volition and executive function to devote a few hours up front—

And patience.

—to pay it off down the line. But no, it's a great example.

Very cool. Part 2 from Alex:

“I don't know what this is or where it falls on the vibe-coding spectrum, but it's enabling nontechnical people to create custom-made software for highly specific tasks. I think she should work with Gemini to build an application wrapper and sell this software to other day programs, but that's a topic for another day. This was all super powerful, and what really got me excited was imagining a future where some AI was watching her submit this billing on her MacBook for a week or so, and then it said, ‘Hey, I notice you doing this very manual task. I built you a piece of software to automate that.’ That's what will enable software to truly eat the world. There are so many industries that haven't been disrupted yet because somebody with software knowledge didn't know the problems that existed.”

In Alex's vision for the future, you don't actually have to be high-agency or have tremendous executive function, because the AI will notice you just hammering away at the space bar and say, “There's a smarter way to do it.” This might be the vibe-solution future you were advocating for earlier in the week.

I think to do this, it's going to be really important to do it in a nonthreatening way. Use something familiar, some sort of interaction model. You could maybe make a character that's like a paper clip. You could call it Clippy.

Clippy.

Clippy's final redemption.

Here we go.

There is a long—

Way to bring me down.

There is a long history in tech of ideas that were right, just way, way, way too early, right? Every dot-com company that you make fun of is now a thriving company today, just by another name, and they started it 15 years later.

Oh, yeah.

That will probably be the case with Clippy and all this sort of stuff. Is that viable in the long run? Sure. Do we have a long way to go to get there? Absolutely. Is it incredible how far we've come and what can be done, while most people are not even remotely taking advantage of it? Yep, also true.

Yeah. We'll see how long it takes to get to that utopian vision from Alex. Honestly, the timelines are getting tighter and tighter here, so maybe that's only a couple of years away.

5. ChatGPT Enters Childhood Computing

Pfizer Frank, whose daughter hosted the fashion show judged by ChatGPT, which we referenced on Tuesday's episode:

“Glad to make my return to the Sharp Tech Mailbag. One thing your conversation made me wonder about: My 7-year-old daughter now uses ChatGPT, but it'll probably be another 7 or 8 years before I'll let her get Instagram or whatever social media accounts are relevant in 2032, when she's 14 or 15 years old. Maybe I'll let her get a smartphone a couple of years before that. Who knows?

“But that got me thinking. ChatGPT, along with Roblox and Minecraft, will likely be at the core of her computing experiences before she ever has to make decisions about which phone she wants, what social media apps to use, et cetera. While AI use cases feel very ancillary to my iPhone experience, I could certainly see an LLM-based experience or assistant being way more central to the expectations of what young people will want from a phone or iPad moving forward.”

So, Ben, this was a lightbulb moment for me as a parent, because if I project forward, I'm going to be a lot more comfortable letting my kids get on ChatGPT early than I will be letting them get on any sort of social media—or maybe even the absolute wasteland that the general internet has turned into.

ChatGPT feels safer for some reason, and maybe that means that ChatGPT is going to be the primary computing experience for the youth. I don't know whether you have any thoughts on Pfizer Frank's projections there and how it may bear on the device-maker market, but it's just crazy to think about.

Yeah. No, it's a really good insight. This is why tech companies always fought tooth and nail to get into schools and gave stuff away for free, because once you get kids hooked, you have them for life—from Philip Morris to Google and Apple.

It speaks to what we talked about. The emailer a couple episodes ago was talking about how we're understating the risk to Meta, Facebook, and the social apps. They're going to try to get kids early as well, but especially nowadays, parents are more and more reluctant to let that happen. And, hey, I already have lots of imaginary friends or interactions or whatever.

Right.

That's good enough. Is there a bit where ChatGPT and other apps should be wary of getting into anything social for this reason? You don't want to get stuck in that bucket. I don't know, but it is a very good observation from Frank.

Yes. Keeping it moving here, Raheem says, “Ben and Andrew, I was wondering, if you both had to start all over again in 2025, how would you go about it? Would Stratechery start off as purely TikTok and Instagram accounts? Would you start a parody account on X as a hack to get followers? More generally, how would you best leverage AI in your own workflows?”

So, a big question there. The thought of Ben starting a parody account on X as a hack to get followers for Stratechery is unbelievably depressing. But do you have any thoughts for Raheem?

No, I would not do any of these things. Number one, it's hard to deal in hypotheticals because I'm aware of the situation I exist in today, and it's hard to think outside of that.

My view of the extent to which these other platforms are beneficial speaks to how difficult it is to build your own platform. It's harder and harder, but that doesn't change the necessity of doing so. If you're a TikTok creator or a YouTube creator, you're stuck on those platforms and what they want you to do.

Yeah.

Again, this is almost certainly very rich of me to be saying, “I have to have my own site with my own name on it and control my own destiny.” At the end of the day, getting followers is getting followers.

I was going to say, would anybody find it if you started from scratch in 2025?

That's right.

It's much harder to spread and build awareness and all those sorts of things that we've talked about. I think realistically, if I were to start again, I would still want to own all my old stuff, and my content would be dramatically more expensive.

Because my assumption would be that I wouldn't be able to amass the size of audience that I have today. And so the ones that would find it, I would charge them a lot for it, and accept the fact that my audience is going to be smaller and make economic decisions in line with that.

I think I could charge a lot more, probably even maintain most of my audience. I like being relatively low-price and accessible.

I'm in the privileged position because I started early, I have a large audience, and people do know about me and can still find out about me and I can still be growing today. But if I were starting again, I think that would be the main difference. I actually would be more exclusionary—

Yeah.

Not less.

Well, and it's a more crowded market than it was when you started because you've made this model work, and lots of people have replicated this independent-creator model. So it makes sense that you would approach the whole problem a little bit differently.

This was an eye-opening question for me because I came of age during an era where there was a lot of VC money flying around at various blog companies. I started at Vox and SB Nation with Jim Bankoff, which was a great opportunity. I love Jim to this day, and I did well there, which led me to Grantland, and there was real upward mobility.

For me, there were other steps and twists in my journey, but the first 4 years of my career in media were great, and those 4 years don't seem to be viable.

And this has spoiled the expectations for basically every millennial online-content creator forever.

Exactly. So if I were starting in 2025 at 23 or 24 years old, I didn't need a ton of money, but I did want to make money.

No, basically, millennial content creators are like the boomers that bought a house for $5,000 and walked into Goldman Sachs and said, “Can I get a job?”

Yeah.

And everyone else hates them for it because they're like, “Just pull yourself up by your bootstraps.” And you're like, “Well, I'm a successful podcaster, and kids should just try harder.”

What's the problem?

Wait, I think I was the one making that argument. But yes, it's true. We are the boomers of online-content creation. We got in when the getting was good.

Yeah. And so the most honest answer I have for Raheem is that if I were starting today, I think I would have considered content creation for a year or 2, then probably gotten impatient by the time I turned 24 or 25, gone to law school, and become a lifelong lawyer, which would have been fine.

And to his question about AI, AI would have been extremely helpful in my workflow as a lawyer, and probably—

Right.

—more helpful in law than it would be in content creation.

Yeah. Using AI in content creation, the problem is, if you use it as a tool such that you're developing a unique vibe or whatever, that's fine.

Mm-hmm.

If you're using it as a content generator, you're, by definition, completely nondifferentiated. Those may sound somewhat similar, but they're different. Is it the content generator, or is it a tool you use to generate a unique vision or to create a unique vision? That's still going to be more important than ever.

We mentioned it earlier with the columnist in The New York Times: being unique still has value. It's harder, harder, harder to break through. And to my point, if you do break through, you have to really capitalize on it.

Yeah.

I feel confident that even today I could get a few hundred or a few thousand people who know about me and value me and charge them a lot more money than I do—$1,000 a month or $10,000 a month.

The price is going up in 2025.

No, independent research analysts on Wall Street charge like $100,000 a year. So that's the approach I would take: not spreading myself thin on social media, but really doubling down on people who are willing to pay.

That's really interesting. Speaking of LLMs as tools, Sajal says, “I find myself going back to old Stratechery articles quite often, a habit rooted in my approach of asking, ‘What would Ben think about this?’”

Sajal sounds like somebody who might be willing to pay $1,000 a month for Ben Thompson analysis. Sajal, the price is going up.

“Basically, I do multiple searches and read through several articles before I can find the relevant information from Ben that I am looking for. I was thinking most websites that are behind a paywall could use a chatbot-like feature, which is basically an LLM trained on all past newsletters. That way, I could just talk to an LLM trained on Ben's writing, a.k.a. Ben LLM, and have a conversation on any topic.

“I'm surprised that this feature isn't already widely present and pushed by major blog publishers like WordPress. Currently, I could paste the articles in a personal NotebookLM instance and chat with that, but it's cumbersome, and I have to refresh the articles and link every time a new article is added. I would love Ben's view on this sort of feature.”

Ben, do you have a view on that sort of feature?

Figuring out how it would work—the user experience, the cost structure, and the cost of these queries—is certainly a thing. I've thought about it. I think Automattic has done some stuff with this in terms of WordPress or having a sort of offering.

So, yeah, I think it's something to consider. I should consider it. I've been busy with other stuff.

There's also the fact that what I'm delivering is facts and words. It's also the content itself. I'm delivering essays in my voice, in my style, and it's a bundle.

And do I want to unbundle myself as far as what I'm delivering? Particularly, what I'm selling is not necessarily past content, but the delivery of future content. Maybe that speaks to why I should do this, because the past content is, relatively speaking, not that valuable.

Those are some of the considerations.

I've thought about this a bit. Probably should do it. Stay tuned. We'll see what happens. No promises, but yes.

Yeah. I mean, do you think this is something that will be more common three or four years from now, when the costs to run something like this come down?

Yeah, I would say almost certainly. That's why it's going to be important for authors and publishers to think carefully about what they're selling and what they want to unbundle and make available. Sometimes you do want to make things more inconvenient for your customers, for better or for worse.

Mm-hmm.

6. China Still Needs Apple

Yep. Okay. To keep it moving, Sam says, “Hey, guys. Relatively new subscriber here, but I did my part to bring Apple and China up in the rankings. Great recommendation. So, stepping back now, it seems that the only reason China hasn't leveraged its supply-chain grip over Apple even more is because Huawei was crippled by America's decision to cut off its phones from Android and our newest chips. If Huawei's HarmonyOS and in-house chip capabilities catch up in the same way it caught up to Apple in product design and manufacturing, how would we expect China to respond? Is this ultimately a race to see if Apple can redesign its supply chain faster than Huawei can redesign Android and U.S. chip capabilities? Even if Apple wins that race, is the clock ticking on its access to the Chinese market?” Ben, do you have thoughts?

I mean, I think this just misses the point, which is that China is very happy with Apple. Why does China want to get rid of Apple? Apple is in there innovating and teaching its companies what to do, how to build things. They're accommodating to China's wishes in terms of the App Store and what they limit.

Patrick McGee didn't say this explicitly because it's never been officially confirmed, but people we talk to strongly believe that Apple has handed over encryption keys in China. Why does China want to attack Apple? I think I'm sort of picking on Sam a little bit, but this is the assumption people have had for years: When is China going to crack down on Apple as retribution for what the U.S. did to Huawei, or whatever else it might be?

I think part of the point of Patrick's book is that Apple has been fantastic for China, and China is well aware that Apple has been fantastic for China.

Yeah.

Why would they want to wreck a good thing?

That makes sense. I mean, it does seem like the relationship between Apple and the Chinese government has gotten a little bit more complicated recently. They've banned Apple phones in Chinese government offices and stuff like that, and there's sort of a nationalist streak in the consumer market over there. But from a macro standpoint, your logic makes sense, and that's certainly proven true over the course of time.

Tesla and Apple are always the targets that people mention: “Oh, boy, they're going to crack down on Tesla. They're going to crack down on Apple.” But Tesla and Apple are also companies that the Chinese government points to as evidence of what's possible if you get in deep and really embed in China. I don't think they would want to crack down on those companies unless things devolve to a point of no return, which may be possible.

No, but again, I'm not disagreeing with you. That is true. They can point to them. It's also a fact that Apple employs a million people in China, if not more.

Yeah. I believe Foxconn is the biggest private employer in China.

Yeah, but the point of the book, and I think it's very well made, is that Apple is actively beneficial to China, not just as a marketing scheme. Apple has trained the entire manufacturing industry and the consumer-electronics industry there, and all their dominance in all these different fields is downstream from Apple spending billions of dollars in China to train manufacturers there, and they continue to do so.

Every new innovation Apple develops is dispersed throughout the Chinese ecosystem as they build these things out, and that's priceless.

Well, what Sam is asking, though, is whether there's a point where that relationship outlives its usefulness. If Huawei develops in-house chip capabilities and develops its own OS that's better than Android, at a certain point, does the CCP just turn its back on Apple and the relationship over the last 15 years, which obviously has yielded massive benefits to the manufacturing base in the PRC?

Well, that argument is no different from the longstanding argument, which is, “Oh, Company XYZ is about to start out-innovating Apple. Apple's doomed.” I mean, sure, if your bet is—which, by the way, maybe it's your bet—that Apple can't design or do anything innovative anymore in terms of manufacturing or new devices, then they're expendable.

But that's sort of betting against history. It's part of Apple's entire modus operandi, and why they're so successful, that they actually do actively innovate and create new things, particularly in terms of hardware. As long as that's the case, China is the biggest beneficiary.

Interesting. All right. To keep it moving, Fraser says, “I work in the online-gambling industry, which, due to heavy regulations, provides users with many ways to restrict how much, how often, and what they bet on. This is reasonable given how harmful a gambling addiction can be. Why, then, is there no way for users to restrict themselves from short-form videos? I would love to be able to ‘self-exclude’ from ever seeing a YouTube Short again, but Google will never add this option on its own. Will a powerful lobby ever emerge against this type of content that is addicting all of us and melting our brains, especially our youth, in the same way that lobbies emerged against other vices like drugs, cigarettes, and gambling?” Ben, do you have thoughts?

Well, I think this is more of a sociological question in many respects, but I think the motivation for a lot of these regulations and capabilities is about harm that is done to other people.

Mm-hmm.

If you get drunk and drive drunk, you hit and kill someone. The problem is that you're killing somebody else, right? If you're gambling, it destroys families. People go into debt. There's crime that follows from that, et cetera, et cetera. I think, by and large, if you're melting your brain and wasting tons of time, that's a you problem.

Yep.

I appreciate the sentiment here. I find YouTube Shorts in particular extremely irritating.

Except the YouTube Shorts that we publish on Sharp Tech—

YouTube is such a great resource.

—that we publish on Sharp Tech and our YouTube channel.

I don't know what you're talking about. But because YouTube is such a resource, having it polluted with these—particularly in search results—is really annoying.

Yeah.

So I agree. At the same time, this isn't really where I want to see government intervention in something that doesn't harm other people, that affects people who can't control their own time. That feels like—where do you draw the line? You know, there's lots of stuff people do that's really bad.

Yeah, I mean, it's a perceptive observation.

I think where we generally agree the government steps in is when it starts extending to other people.

Yeah. I hadn't considered that it's really the effect on law-abiding people that pushes things over the edge and forces the government to get involved. I mean, drugs would be a counterexample. Cigarettes would be a counterexample.

When those laws were passed, they were often framed as additional health-care costs—

Secondhand smoke.

—secondhand smoke, drugs, broken families, crime. This is where we put in our monthly boomer complaint about walking around cities that smell like weed.

Legalized marijuana.

Yeah.

It's pretty terrible.

It's terrible.

When I'm walking my 2-year-old son and the block just reeks of weed, it's pretty depressing, but it is what it is. Yeah, I think moral regulation can be underrated.

My main reaction to Fraser's question is that if this option existed, I would check, “Yes, please restrict how many short-form videos I watch,” because I can't think of any short-form video I've watched where I've been like, “Man, I'm really glad for that 90 seconds I just spent watching a short-form video on X.” So if I could experience the internet without those—

Well, yeah, the X ones are terrible. The X ones are terrible. You've got to give some credit to TikTok here. Some of those are pretty good. Let's see how much time we can waste.

Oh, God.

He's got a whole week off from work. He can definitely have some time.

Yeah. Ruin my vacation by besieging me with short-form nonsense.

Fraser, thank you for showing us the change that I personally would like to see in the world. Elizabeth says, “I almost unsubscribed after Ben’s Taylor Swift comment on the podcast last week. Dangerous territory.” This, of course, is in reference to you explaining the situation between Taylor Swift and the record executive.

Mis-explaining, I think, is the word we’re looking for.

Mansplaining. Very, very dangerous territory. Do you have any thoughts for Elizabeth?

Elizabeth, let me explain this to you.

Oh, God.

I think I’ve said enough.

Oh, God, yeah. Well, look, Elizabeth, take heart. Take solace. He went through a Taylor Swift phase over the last several years.

No regrets.

Don’t worry. Ben was a Taylor Swift fan.

No, I remain a Taylor Swift fan, from a business perspective. I think the way she’s approached music in her career is great.

You applaud her exercising her leverage and mobilizing her fan base.

Absolutely.

I mean, you’ve written to that effect.

No, the way she leveraged her fan base and did the Taylor’s Version of her songs to depress the value of her masters so she could reacquire them—and I have my daughter coming to me saying, “Oh, I can finally listen to the good version again.” Very impressive. She got all her fans to listen to inferior versions of her music so that she could buy it back at a cheaper price. How can you not? It’s brilliant. It’s amazing.

It’s unbelievable, yeah.

Yeah. Also, her complaints about someone else owning the masters are still BS, but it’s impressive how she got them back.

Well, and look, deep down, you’re a Swiftie yourself. You’ve been to concerts. You’ve listened to lots of Taylor Swift in cars with your daughter.

Yeah, I have been to a lot of concerts. I know a lot of Taylor Swift songs. So, Elizabeth, get off my case here.

Don’t worry. All right, Greg says, “Andrew, you and Ben need to try the live video function of Gemini Live if you’re looking for AI cooking help. It’s incredible.” Have you tried the live video function on Gemini Live?

I have not. I’ll try to get to that. But I will say, being back—not in my condo in Taiwan, but at a house in America where stuff needs to be fixed and done—it’s actually almost a problem because my use of AI is so out of control. I’ve talked about using AI for interview prep.

Yeah.

And I do a deep research report. Then I go back a few hours later to reference the report, and I have to scroll down to find where it is because I have so many ongoing conversations. It’s out of control.

Well, it seems like Gemini in general is amazing in all sorts of ways, and I just never find myself using Google AI products except for AI Overviews, which I’m never asking for explicitly. So there are probably lots of things that Google is offering that I should try to explore over the next couple of months. I appreciate the tip from Greg, and, Ben, good luck with your ongoing AI addiction there.

A couple more questions. Gabe says, “Can we go ahead and pencil in a crossover episode where Ben Thompson comes on The Greatest of All Talk so we can fix the NBA? Please see the following email I wrote to this effect last year.” Last year, Gabe identified himself as a GOAT who converted to Stratechery after hearing you recap your trip to Davos, which was one of my favorite episodes we’ve ever recorded: 25 minutes on Davos. So, what do you think? Are you willing to guest on The Greatest of All Talk at some point and fix the NBA with me?

If we’re not in partnership with The Greatest of All Talk, I absolutely would, but I would feel like I’m inserting myself under the current scenario.

Oh my God, that’s a ridiculous policy. We’re going to get you on The Greatest of All Talk at some point, because half the time we could fix the NBA on Sharp Tech, but that’s not really what the Sharp Tech audience wants.

All right. When Ben’s on vacation, maybe I’ll jump in. I do have a long catalog of takes that I’m filtering. I’d say they’re getting out through the NBA. I have a few that I think people will be surprised I’m going to claim ownership of, going years back.

Mm-hmm.

So, yeah, maybe I will.

There you go, Gabe. We will get there eventually. It will probably be—

There was one just last week where you wanted to credit somebody else for the idea, and I’m like, “No, I gave that person the idea.”

Ben got mad during Game 7 of the NBA Finals. I was just locked into Game 7 at that point. It was the first half, when it was still close. Ben came off the top rope mad at me for crediting someone else.

But that’s the sort of content you can look forward to when Ben finally joins The Greatest of All Talk at some point as a guest. We’ll do that before we eventually launch an F1 podcast. We got a lot of requests asking for an F1 podcast. We will have a review of the F1 movie, which is in theaters Friday. I might be seeing it on opening night.

I thought we were going to go see it together. That was the plan.

Well, that was one person’s idea. Domin’s idea was for us to see it together, but I’m not going to wait a couple of weeks to see F1. I think I’m going to go see it with Alice. I’ll see it again with you.

Well, the problem is, I might—no, see, I might not see it then. I’ll just wait until it comes out on home video.

That would be the ultimate Ben Thompson move: to wait 2.5 years to see the F1 movie.

7. Mechanical Watches Stay Simple

Hawkins says, “I’m currently on paternity leave, and after having some trouble adjusting to the slower pace, I’ve decided to get into mechanical watches.”

Oh.

“Ben seems to have a well-thought-out answer for most questions, so I’m interested in learning more about his approach to watch collection. I’ve just started learning about the different movements, and I’m trying to figure out what I value in a watch. I’m also wondering how the resident normie, Andrew, approaches all the training metrics available from his Garmin. I don’t look at it personally because it stresses me out.

“P.S. Ben once outlined how the printing press created nation-states. What world will AI create? Will a handful of companies take over?”

So that is an enormously—

A wide array of questions there.

—weighty postscript. Probably the most intense postscript that anyone has ever sent us.

No. Well, I know the answer. I’ve read about this. It’s downstream; it’s a continuation of what’s going to happen to the internet, where you’re going to have these overarching organizational structures, and then increasing devolution into smaller entities—city-states, fiefdoms, sort of like whatever it might be in the physical world.

Again, you go back to before the printing press, where the Catholic Church was the overarching entity, and they had a bunch of city-states in Europe. The printing press undid that, and now we’re sort of going back in time. So there’s your answer to part 2.

Okay.

Part 1: Mechanical watches are like short-form video for very rich men. They’re very addictive and will destroy your life.

Yes. My approach to mechanical watches is to not get too into mechanical watches. That said, I do have a few.

Can you enlighten me? What’s the difference between a mechanical watch and some other kind of watch? Is he talking about a digital watch, like a Garmin?

No. Well, there are smartwatches, like a Garmin or an Apple Watch. You have to charge them, they have a digital display, they might connect to your phone, and you can track your metrics. I hate them.

Mm-hmm.

I hate having to charge them. I hate notifications bugging me all the time. I hate metrics that I feel stressed out about and have to keep track of. Horrible. Very anti-smartwatch. I see the value, and I understand why people like them. I personally do not enjoy wearing a smartwatch.

Yeah.

Then there’s your digital watch, like the old Casios or whatever, right? That shows the time. There are quartz watches or automatic—are they called automatic watches?—but battery-powered watches that keep time normally.

I think when he says mechanical watch, he means a self-winding watch. Actually, it doesn’t even have to be self-winding: a watch that has a movement and a spring that uncoils and actually moves the hand. So it’s all mechanical.

And that can be self-winding, where there’s a disc in there that moves as you move your arm, so you never have to actually wind it. It might be manually wound, where you have to actually turn the spindle, and that gives it its “charge,” which is sort of tightening the spring that slowly untightens and moves the whole thing.

I personally don't really care about movements and stuff like that. I like mechanical things. I appreciate a really nice movement, but I like having an analog thing on my wrist that I don't have to charge, that's always readable, and that generally doesn't stress me out. I specifically have learned that I need to have a date complication. I hate not having the date.

I really like a GMT watch because I live in 2 places. I like a real GMT watch, where the second hand is set independently of the rest of the watch. You have the 24-hour GMT hand that always stays the same. I like the traditional Rolex GMT-Master II, which I don't have. If a dealer wants to hook me up, I wouldn't mind.

Mm-hmm.

The Tudor Black Bay GMT is a great watch. It's relatively inexpensive, and it has the same functionality. I think Tudor's doing an amazing job. Tudor is Rolex's bargain brand, but they're like what Rolex used to be: pretty attainable, well-made, good-looking watches for everyday use.

There you go.

Rolex became a luxury brand over the last 20 years, but that's what Rolexes used to be. That's why our parents or grandparents would have Rolexes. It wasn't a big deal. It was just a nice watch. It was an investment, but it wasn't a collector's item.

That's helpful information, because my dad has a Rolex that he got in the '70s, and I found myself thinking the other day, “Dad, how the hell did you afford a Rolex in 1975?” I guess Rolex wasn't always Rolex.

Yeah, no, a Tudor is around $3,000 or $4,000. So it's a big purchase, but it's not a once-in-a-lifetime sort of thing. If you wear it every single day for years, it's trivially worth it if you calculate that out, right? You pay $1,500 for an iPhone that you use for a year or 2. You could use this watch for a very long time.

The watch I wear almost every day—I've gotten into buying different watches, but part of it was just honing in on what I like. I wear the Tudor Black Bay GMT almost every day. It has a 24-hour hand. I set it to GMT, and then I move the bezel. It has a movable bezel set to the time zone I want to know. I have the bezel right now set for Taiwan time, so I can see the 24-hour hand against the bezel to know what time it is in Taiwan.

The second hand is independent. When I'm flying, the moment I get on the plane, I change my watch. Instead of pulling it out and stopping the second hand and changing the time like you do with normal watches, with a real GMT—there's an actual name for it; I'm just trolling everybody by calling it a real GMT—you pull out the crown a little bit, and the watch keeps running. Then you can move the second hand simultaneously.

Okay.

If I'm going just to the East Coast, I pull it out, pop the hour, move it 1 hour, and boom, I'm on the right time. My watch never stopped moving. I don't have to reset it to align it to whatever seconds it is.

Then, at least in theory, if I don't want to pull up my phone, I can look down and see Taiwan time. I could turn the bezel and see, “Oh, what's UK time?” The UK is 0, and that one's easy, right? If this place is plus 6 or this is minus 8. The problem, of course, is daylight saving time screws all this up, so you can't just memorize it.

Mm.

My parents live in Arizona, which doesn't have daylight saving time, and that just endlessly confuses me. I could never figure out the time there without looking at my phone.

Yeah.

The world GMT stuff is not as useful as it used to be, because phones are often just easier in a complicated world. But for me, where I generally want one time zone, that's great, and then you have the jumping secondhand.

It took me a while to get to this. I had a watch I really liked, but it didn't have a date function. I had other watches. I like the look of chronographs, but I find I value a GMT function more. I like titanium watches. This is actually a Tudor titanium GMT that I have yet to try, but I think it looks really great because it's really light. When I'm typing, I do feel the weight of a watch.

Hmm.

I really like rubber bands. I got this from the Apple Watch, which actually has an incredible rubber band. With a rubber band, the watch can get wet and do whatever.

Yeah.

I dress casually most of the time, but part of this is just figuring out what you like. For a while, I would get a new one every couple of years. I would say, “Every 10,000 subscribers, I'm going to get a new watch.”

Yeah.

But, uh, but—

Good little reward system. Great.

Yeah. It's a pretty expensive one, and at the end of the day, I found out what I like, and I just stick with it. I went through my phase. I never got crazy about it. I never got into the really extreme stuff. Again, it can get very, very crazy.

Oh, yeah.

If Stratechery gets bought for $100 million, or if Mark Zuckerberg hires me, then I will get more watches. But I think I made it through and got it through my system without going bankrupt, so I'm good now.

Yes. Well, I appreciate the question, because I have been toying with entering an expensive mechanical-watch phase. I do like my Garmin a lot. I don't pay attention to any of the training metrics for the exact same reason that you cite: It stresses me out, and I don't really care about any of it.

I do like the step counter, and the ultimate selling point with the Garmin is that I charge it once every 7 days, as opposed to having to charge a smartwatch every single day. In that scenario, I'm just not going to wear a smartwatch. But I've started to become intrigued with some of the more expensive watches, and I find myself thinking, “Is it time to upgrade?”

To your point on Rolexes, I was actually in a meeting with someone earlier this week who had a really nice Rolex on. It had this green face, and I was like, “That's a pretty cool watch.” I Googled it later, and it was a $20,000 Hulk Rolex, so that sort of stopped me in my tracks. To your point, you need to proceed with caution as you delve into the mechanical-watch world.

Right. But again, that's just because Rolex has sort of become—I think we already did a segment on Sharp Tech about the tiers of luxury.

Oh, yeah. I mean—

Rolex is still fairly low, to be clear, but it's no longer your working man's watch. I think the other advantage I have is that I have very large wrists. I'm big-boned, as they say.

Okay.

Which is great, because I can wear these big, beefy watches. I can wear all kinds of them. And again, get into movements at your own risk. Be careful. That's where you start spending a lot of money, right?

Also, be aware of whether you do or don't have big, beefy wrists like Ben, because there are a lot of guys out there wearing these garish 47-millimeter-face watches, and they look kind of ridiculous. Get something that actually fits your wrist. I feel like that's a good bit of advice as you delve into this.

Yeah. The real watch—you will know that Mark Zuckerberg hired me if I come out with an A. Lange & Söhne. It's actually German. It's not Swiss. Amazing watches. I do look at those, but no, I'm good.

For now, you're good with your Tudor, man.

Yeah, I'm good with my Tudor. There are other ones in that class that are good. I also think some of the luxury companies—Hermès and LVMH, or Louis Vuitton and those sorts of entities—all make watches. They're entry-level and kind of scoffed at. It's the opposite of apparel or bags or whatever it might be. It's like, “Oh, they're not collectible.” They're Swiss-made, but people in the watch world look down on them.

Yeah.

Hermès in particular makes some really great-looking watches. If you don't care and just want a watch that looks really good, that's actually a surprisingly good place to look.

There you go. This is helpful advice for me. I'm not spending $20,000 on a Rolex, but I would like something that looks nice that I can wear with formal wear.

Yeah, that's a good place if you want a more formal watch. Again, just know what you're getting into. You're not buying an Hermès because it's collectible and you want to pass it down. You're buying a quote-unquote disposable watch, but it's a nice mechanical watch that works well and looks really great.

They have this amazing set of square watches that I really like. I can't remember what they're called—the Hermès H08 or—

Okay.

—or something like that. But I like them because I like mechanical watches. I like having a mechanical device on my wrist. I don't like that it's not digital. I like wearing a watch. I hate not having a watch on. So, for me, it's pure function. I want it to look good, I want it to be comfortable, and I don't want to go broke.

There you go. I, too, love having a watch on. I wasn't a watch guy until about 2.5 years ago, and now, if I don't have my watch, I get very frustrated when I look down at my wrist and—

It's the worst.

Yeah.

It just bothers you constantly that you don't have your watch on.

Exactly. On that weird note, we're going to head into the vacation week here, Ben.

Look, don't have me talking about buying watches when I'm about to take a week off. Just be like, “All right, I'm—”

Oh, God.

“This doesn't need to air. This subscription—this guy's out of control.”

This is the Davos Ben self-loathing all over again.

No.

Absolutely, man.

I'm telling you, I'm not wearing a Rolex. I don't have a Patek Philippe or fucking Cas—Vacheron Constantin or whatever.

This is what I'm saying.

Richard Mille. All the F1 watches are around $800,000, so I'm familiar.

Well, this is—Louis, Louis is in your ballpark. IWC has entry-level watches, so maybe we—

Well, we'll continue—

—this conversation offline, you know? You can coach me through my first couple of watch purchases.

Also, Red Bull is, like, TAG Heuer. That's also an entry-level sort of watch.

I was looking at some of that, and I would feel like a dork if I were walking around with a Formula 1-branded watch. So I'm not going to go in the F1 watch direction. But they are nice-looking watches, which is why they caught my eye initially.

I think it's Breitling that people are pretty pleased with. Basically, Tudor, I think, really set the pace.

Yeah.

Rolex is like, “We're not filling the space we used to. We should make really good watches for under $5,000.” I think mine—the one I was talking about—is $4,500 now. I think I bought it for around $3,000, but I've had it for years.

Mm-hmm.

That's where you should look at the various brands, find one you like the look of, and get that. That's a nice place to start. I think Tudors are amazing deals. I think they look great. They are fairly beefy, but that's my recommendation.

Well, thank you, Hakan, and thank you to everyone who wrote in. Again, we got more questions that we're going to roll over. I'm looking forward to the summer. I'm looking forward to the 4th of July, and I hope everyone has a great vacation out there. Ben, you, too, have a great vacation, and—

Thank you. I will see you. I think we're recording in person in—

In person, in an undisclosed location—

—10 to 11 days or so. Yes.

Perhaps we'll have an F1 review by then. But until then, enjoy yourself, and I will talk to you soon.

Talk to you later.