AI不均衡到来、TikTok可能离场、Xiaohongshu与文化交流的乐趣

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

I feel like I'm just seeing a lot of Andrew Sharp. We recorded a day late this week. I'm already back on with you.

Do you need more time?

We're apparently recording—

We can circle back 24 hours from now.

On the holiday on Monday. Jeez.

That's right. I'll happily go back on vacation for 10 days, and you can just kick your feet up for a little while here. But I'm excited. This half of the podcast is excited to be back in the trenches with you—the take trenches.

All right. Let's do it, then.

We've got a lot to wrap our arms around on this episode, so buckle up. TikTok may or may not be on the brink of a full-scale ban in the United States.

1. AI Arrives Unevenly

But before we get there, we'll begin with AI and an article that you published on Stratechery Monday morning. It's funny because that article had me thinking back to our first show after the holidays, where we got a question about AI adoption among non-tech firms and what sort of tech companies may benefit from helping those companies incorporate AI solutions into their business. With your article on Monday, I felt like you presented a theory of the case with respect to AI's impact on the economic landscape, at least as far as it goes in 2025.

The title of that piece was “AI's Uneven Arrival,” and we can begin with your conclusion there. You wrote:

“The most important AI customers will primarily be new companies. Traditional companies, meanwhile, will struggle to incorporate AI outside of wholesale job replacement, à la the mainframe. The true AI takeover of enterprises that retain real-world differentiation will likely take years. None of this is to diminish what's coming with AI. Rather, as the saying goes, the future may arrive but be unevenly distributed.”

“And contrary to what you might think, the larger and more successful a company is, the less they may benefit in the short term. Everything that makes a company work today is about harnessing people, and the entire SaaS ecosystem is predicated on monetizing this reality. The entities that will truly leverage AI, however, will not be the ones that replace them, but start without them.”

Ben, I'll let you drive here. Do you want to expound on that conclusion and explain to people how you got there?

Well, just to reference one thing in there about wholesale job replacement, that was a link to my article last year about enterprises and analogizing AI to the arrival—

Mm-hmm.

—of the mainframe. That sort of wiped out rear-end back offices. I think, to really leverage this, you need top-down decision-making. You need significant integration projects to make this work.

I'm not skeptical about AI helping the individual worker. I think that's happening right now. In fact, the individual worker today is like the newspaper companies in the '90s.

Right.

It's like, “Wow, we have our core business, and we get all these internet customers for free, too.”

Just more revenue. What's the problem?

That's right.

Yeah.

There's this real arbitrage opportunity that continues to be taken advantage of. If you're an employee who has the wherewithal to go and use AI, you're tremendously more productive, and/or you can be tremendously more lazy while doing the same amount of work—because AI is augmenting you.

Yeah.

I think there's a point of frustration amongst corporations, to a certain extent, where there are large gains being realized, but they're accruing to individual employees and not to the enterprise as a whole. That in and of itself is a real motivation to get this done and figured out. But I'm skeptical of the top-down, “Okay, everyone has an assistant now. Go and use it” sort of bit.

Mm-hmm.

Just like not everyone was going to go and use a computer. The people who benefited from computers, even fast-forwarding to the '80s, were the people who wanted computers and wanted to use them, and then had the wherewithal to figure them out and figure out the use cases for them.

So this idea that Microsoft is going to sell a bunch of Copilot products, or Google just announced—which, by the way, was very clever—they're taking away, for Google Workspace, their sort of Microsoft Office equivalent, the Gemini add-on and bundling it with the whole thing, then raising the price of the whole thing.

Mm-hmm.

That's going to have a much larger impact on Google's business. Price increases that apply to everyone, and you get to sell them as, “You get this new feature”—that's going to be great for their business. I'm skeptical that it's going to—

Be great for companies.

I think it's—no, I mean, they're going to pay more for it. They're locked in. They're not going to go anywhere.

Well, right, but I think the thesis is that you're not going to be transforming the productivity of a company based on Google's offerings there.

Yeah. It might trickle in through people using this stuff more and more, realizing what's effective and what works, but I am a little skeptical of this idea that suddenly every company is going to be a gazillion times more productive next year because now they have access to AI.

2. Defining AGI And ASI

And this bit about the very structure of AI—even if you get to this world of agents and can give a task to an AI and it will go and accomplish it—there's always this discussion: What is AGI? What is ASI?

I like that. Yeah.

I arrived at the idea that AGI is when you can give the AI a task and the AI will accomplish it. It might have to access different tools and things to do that along the way. That's a step up from the assistant level, or the chatbot level, where it's directly reactive to what you do and it's not a multistep sort of thing.

AGI is like a very conscientious but fairly dumb employee—

Right.

They're not going to figure out what to do, but if you tell them what to do, they'll go do it and do a good job. And by “good job,” it doesn't mean a perfect job, right? You give a human a job—again, to use the self-driving car analogy—

It's not going to be perfect, right?

Self-driving cars are going to get in accidents. The question is—you have to remember, humans get in accidents too, right? You give a job to an employee, and sometimes the employee will screw it up; you give the job to an AI, and sometimes the AI is going to screw it up. But it's going to reach a good-enough level.

What you'll see is this sort of bleeding of human jobs. Maybe humans leave and they're not replaced because the current humans can be more productive, right? That is a tried-and-true method of getting productivity into the workforce: We used to have 10 employees on this team, now we have 8, and you still have the same number of responsibilities, so you're going to have to figure out how to get it done. Suddenly, you reach for AI because you just have more work to do, and that is actually how it gets into the organization. I think that's the way a lot of these things are going to happen.

As an aside, ASI—artificial superintelligence in this framing—is the AI that can decide what tasks to do in the first place.

Right. It can look at the problems you have and come up with different solutions that humans haven't devised themselves.

Or find the problems on its own and just go and fix them, right?

Yeah.

It's interesting because that is, in some respects, very compelling. It's also maybe a little more approachable than this idea of a godlike oracle that is solving the world's problems. But I do think both of these definitions have the benefit of being fairly testable, right? Can it do XYZ, or can it not?

Of course, the actual approach and achievement of them will be on a spectrum, a gradient in terms of whether it achieves them or not.

I was going to say, with artificial superintelligence, for instance, you can give an AI, like an LLM, a list of symptoms, and there are cases where the LLM can identify what's wrong with a patient and a doctor can miss it. I don't know whether that qualifies as artificial superintelligence.

No, to me, that's still just the assistant. It's just returning an answer. I don't want to call it a glorified search engine; that diminishes what it is.

Mm-hmm.

That's the assistant level.

But it's processing a lot of information.

Just speaking in the abstract, it's processing a lot of information and identifying what needs to be done to help a patient heal. So it's sort of like what you're describing on a micro level, but on a macro level, I understand that we're talking about broader solutions and broader abilities than exist now.

An ASI would be more like it working over all the patient data and unilaterally scheduling an appointment with someone and saying, “There are symptoms showing up. You need to come in.” In this case, it's directing humans.

And it says, “Okay, go do this scan on this person,” and then it sort of has a solution and writes the prescription and so on. That is ASI. It's the flip. It's where the AI is starting to tell humans what to do instead of humans telling the AI what to do.

Mm-hmm.

And that is sort of the line. Whereas AGI—and I actually do like it; I think it's been very productive to add ASI as opposed to AGI, because these are 2 different things.

With AGI, the doctor is telling the AI to go and do XYZ, and it has to do a number of different steps to figure out the solution and do XYZ, but it's still under control. The human is still telling the AI what to do.

Anyhow, all of these are going to be fuzzy in implementation, but I do like this framework in terms of giving us a vocabulary to talk about these different steps. So right now we're in the assistant age, and that's just LLMs—what we're calling AI. The next step is the agent age, which I would call AGI, where it will accomplish tasks that you tell it to do.

And then ASI is when it tells humans what to do because it already has it figured out. That's both the more promising and also the more scary leap, for very obvious reasons.

Right. I actually didn't realize that ASI was a recent addition to the lexicon, because I saw people dropping that a couple months ago and I was like, “What the hell is ASI?” So I appreciated your digression in the article, and I appreciate your digression on the pod. Take me back to your conclusion, though, about the sorts of companies that could benefit from AI in the long term.

3. AI Reprices Corporate Work

Well, there's lots of different pieces here, including this whole world of SaaS and software that's all predicated on organizing by people. We've talked about the business model of per-seat licensing and how that's a problem in a world where AI is removing jobs, but I think there's a broader principle here, which is: What is a company?

We think about a company and the units of work being humans, but that's actually always been a proxy. It's a proxy for accomplishing a task, actually producing an economic output.

It's almost like this is a similar thing to when art-generation and image-generation models came out. I wrote an article talking about how, for all of history, we've just assumed that ideation and manifestation are two peas in a pod. And AI is like, “What? No, actually, they're 2 different things.”

The actual coming up with the idea and the actual implementation of the idea can be separated. You can write a prompt, and then the AI can do the actual creation. You think about it and say, well, yeah, that actually is true. You could go hire an artist and try to get them to make what you want, and of course we've been doing that for a long time, but you don't think about the fact that there actually is a very clear division here.

And I think it's a similar aspect here. When we think about things to get done, we think about the humans that do them, but those humans are just a proxy that's actually independent from getting the job done. So in that world, what are we actually paying for? What's the actual goal? If you're a company, it makes a lot of sense that you pay per job completion, right?

And again, this is where the art angle is also interesting. There is a bit of this in the freelance world, the Upwork world or whatever, where you hire someone to make a logo for you and you're paying for the job. You're not necessarily paying for their work.

So it's not like this is a completely foreign concept, but it's going to become the defining concept in terms of how you think about paying for stuff and paying for work, and that's where I came to the advertising analogy.

The thing with Facebook and Meta and direct-response advertising is that you actually go back to Google, really. I probably should have given Google more credit in this regard. What made Google's business model such a big deal is that until then, you paid for ads based on how many people saw them.

That was a hangover from newspapers, right? You paid more to put an ad in a newspaper with a lot of circulation. You paid less for one with fewer. It's just: How many people are going to see this? But people seeing an ad is a proxy for the actual goal, which is people making a purchase decision, actually buying the item in question.

Right.

And so what I was trying to get at with this article is that it's very easy to think about this stuff from a theoretical perspective, which is, of course, it's better to pay for the actual purchase. That's the goal. You're trying to sell stuff.

But it turns out there are entire ecosystems and businesses built around the assumption that that's not possible, that you have to focus on the proxy.

Mm-hmm.

And this is where the CPG analogy, I think, is super interesting. In this whole CPG model, you have lots of different brands that are all the same thing. I've used this analogy before, but Axe body spray and Dove beauty products are made by the same company. It's pretty much the same stuff, and yet they're not thought of as the same thing because that's how branding works.

The product managers in CPG companies aren't called product managers. They're called brand managers. And actually, the current CEO of P&G is the first CEO, I think, in their history who didn't come up through the marketing organization, which sort of speaks to a lot of the shifts that have happened here.

But these are marketing companies. They build brands, run commercials, and have coupon programs. You have an affinity for a brand so that you go into the store, and the goal is almost to have a subconscious selection. You see—

Right.

You need deodorant, and you buy it. Like we talked about, you're an old-style man, right? Which is—

Old Spice.

Very much a—

Of course.

Old Spice, old-style, whatever. Yeah, I mean, your penny loafers. It all sort of goes together with being on sailboats in Nantucket or wherever it might be. So—

That's right, that's me, with my penny loafers. No, but there's an entire apparatus behind that business model that wasn't—

Massive.

—suited to playing on Facebook and Google for the last 20 years.

That's right, that's right. And especially if you go back a decade, Facebook and Google weren't nearly as good at doing what they do as they are now. It really was the case that you had to be much more selective in your targeting: “I'm targeting X, Y, Z.”

And if you paid to say, “I want to reach the millennial man on the East Coast in his penny loafers,” you paid a lot of money for that, and you better have been pretty sure you were right.

And it turned out, actually, no—just running advertisements on ESPN worked better. The cost to reach an individual consumer is actually lower, even though the overall cost is high. So there's a high barrier to entry, but the cost to reach an individual consumer is actually lower.

And the actual point of decision isn't when they're watching TV anyway. It's when they're in the supermarket, when they're walking down the aisle and they grab the deodorant. And so—

Mm-hmm.

And so there's an entire edifice that was built up around this model. And what was so interesting is, particularly if you go back a decade, all these companies realized: Yeah, in theory, this Facebook advertising stuff is amazing. It doesn't work that well for us because we're not selling bespoke products.

We're trying to do market categorization, and we're trying to reach this particular demographic. But the reality is that demographic is fairly large because we operate at scale, our manufacturing is at scale, and we have to buy shelf space at scale. Scale, scale, scale, scale, scale.

And the fact that I can reach people on an individual basis is actually not that useful for me, and it just costs too much. So we should actually double down on TV. This is why TV kept making money so much longer than people thought it would.

Even as the viewers are bleeding away, why are advertisers still on TV? Because they're built around TV. Their entire business is organized around this paradigm.

Now, again, as I noted, TV has finally collapsed. The user base got too small. Everyone had to adjust. P&G and all the CPG companies have had to shift their approach. And also, by the way, Facebook's gotten way better at efficiently finding customers that you want.

And also, COVID drove a lot of purchases online, which tightens this loop in a way that makes these ads more productive, and you can track conversions more effectively. But that's interesting in its own right. It's not that digital advertising didn't end up being important to companies like P&G and Unilever, but it took a lot longer than you might have thought.

Mm-hmm.

And that is the analogy. I'm not saying this is a perfect one-to-one match between AI and digital advertising. Rather, you can have a product that is obviously better, that's obviously more productive, and, yeah, sure, o1/o3 inference costs a lot. That's a lot less than paying someone $100,000 or $200,000, whatever it might be, for these white-collar workers that are the proxy for getting the job done.

But there's an enormous amount of inertia in the current system because there are so many systems and processes built around humans as the proxy for getting work done. And so it's right to be long-term pessimistic. Someone wrote, “Oh, I can't believe you're optimistic about SaaS companies on Twitter.” I'm like, “I'm not optimistic about SaaS companies.” They're clearly in trouble. This whole business model is kind of screwed up.

But just to go back to 2015, you were right to be long-term pessimistic about TV and the entire ecosystem around it. But it still took many years longer than people thought for it to collapse and fall down, and you needed COVID, again, I think, as a forcing function for some of this stuff. That's also underrated.

Hmm.

That probably pulled forward some of the collapse for these things a fair bit, and I think it's going to be similar with AI. Sorry if I'm monologuing here.

No.

If you want to jump in, go ahead.

Go for it. I do have a question, but I'll let you keep rolling here.

The other key thing about digital advertising and the power of Facebook, and why you could have this situation in 2020 where all those CPG companies and a bunch of other big companies boycotted them and it had no impact on their business, is that these new formats create their own customers. There are entire swaths of companies, particularly in e-commerce and apps and all this sort of thing, that were created in response to the Facebook product being available. And so Facebook basically created its own customer base.

Mm-hmm.

And the funny thing is, you have these CPG companies that were going to be super narrow: “We're going to focus just on very specific customers because Facebook lets us do it.” Guess what? If P&G and Unilever want to pull their ads, now we can buy ads on a more inexpensive basis because there's less competition and actually start hurting them, taking market share from them, which Facebook wins either way.

That is antifragility. The concept of antifragility is that even things that hurt you actually make you stronger. So CPG companies boycotting Facebook makes Facebook stronger in the long run. You saw this with Apple. The whole ATT thing proved Facebook's antifragility. Yeah, it was devastating to Facebook, but it actually reinforced their overall competitive position.

Deep in their moat, yeah.

Yeah. Those were new companies, and the whole Shopify ecosystem is downstream from Facebook ads. Facebook created its own market, and I think that is what you're going to see happening: AI is going to create entirely new industries. It's going to create entirely new companies that are going to be chipping away at these large companies that can't fully adjust, and it's going to create its own ecosystem that will, in the long run, force these larger companies to cram this stuff down, particularly once it gets better and they can figure out how to adjust.

This isn't a forecast about the next 50 years that AI isn't going to go into these companies and change them. It's just saying it's not going to happen in 2025. It's going to be this longer process where companies lean into it and embrace it because it's this new opportunity, primarily for new companies, and they're going to come up from the bottom and start chipping away at these existing companies.

Mm-hmm.

And then it's almost more of a classic disruption story. One more point. I made this point in the context of enterprise AI, in the enterprise AI article, but there is a generational aspect to a lot of this. I think that SaaS is in many respects a demographic story.

You had this host of millennials coming into companies who were familiar with working online. It wasn't weird to them to log into a random website and do your work. They'd been working in Google Docs in college for ages. They were used to the whole concept, and so that created a workforce that was receptive to this.

It wasn't the '80s, when we were saying, “Go use this computer,” to a person who'd been using a pencil and paper or a typewriter for the last 30 years. There's a generational shift that is necessary, and that will be a similar thing with AI.

Hmm.

If you fast-forward to 2035, you're going to have more and more portions of the workforce that have grown up with ChatGPT; they've grown up with the assumption of AI. Right now, the arbitrage opportunity is for people who have agency: they can go use AI and be more efficient than their colleagues, and it's a big advantage.

Right.

In 15 years, everyone will use it as a matter of default. It will change from being an advantage to being table stakes. And once it's table stakes, the opportunity to push it and increase it will actually start to re-accrue to the corporation as opposed to the individual employee. So employees, enjoy it while you can. Don't be like newspapers in 2010.

Right. As far as the way companies are built, it's not an apples-to-apples comparison, but when I was reading and thinking about big companies incorporating AI into their existing workflows and their existing employee bases, and potentially deriving benefits but maybe not as many benefits as people expect, I was thinking back to your piece on Tesla's approach to self-driving technology and the bitter lesson, which stuck with me.

I will refresh people's memory in case they forgot the bitter lesson from the fall, but this was Rich Sutton. He wrote, “The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin.”

And then, continuing on, he says, “Seeking an improvement that makes a difference in the shorter term, researchers seek to leverage their human knowledge of the domain, but the only thing that matters in the long run is the leveraging of computation, and the human-knowledge approach tends to complicate methods in ways that make them less suited to taking advantage of general methods leveraging computation.”

So, big picture, stepping back, it would make sense that all the friction associated with trying to incorporate AI into human workflows becomes its own sort of gating function for some of the big incumbent companies that are trying to take advantage of this technology. Companies that get to start with a blank slate ultimately will be able to capitalize on all these capabilities more as we head into the future here, and that might take more like 4 or 5 years as opposed to 4 or 5 months, as we all project ahead in 2025. Is that sort of what you're saying?

I think, yeah. I think if you zoom out, the broad analogy is correct. I would just distinguish that the bitter lesson is about the development of the fundamental capability, but there is a separate discussion about actually productizing this. I don't think there's a bitter lesson for product development.

Right.

It's more iteration and sort of XYZ. But this reminds me of another point. We did discuss a little bit on the first podcast of the year about aggregation theory and these models and things along those lines.

One thing to keep in mind is that there is a tendency, just in analysis generally, to jump to the new thing and say, “The old thing's done and gone,” right? When in reality, stuff layers on top. The old stuff continues to be relevant, and I do think that, overall, LLMs that give you an answer and don't really think about it—they just generate it—are going to get better. These new models are going to make those models better.

One of the problems in developing the core models is that we're running out of data. These new models can generate interesting and useful synthetic data more effectively, and that can go back into training these new models, so we can get more of the data that we need. So they're going to get better. Number 2, they're always going to be cheaper and faster, and cheap and fast still matter for things like aggregation, like doing things at scale.

I don't think aggregation is necessarily dead. When you talk about Facebook and the opportunities for AI-generated content within the Facebook feed, or every single item on Facebook becoming an ad, right? Because you can do image...

You can see the bag in the picture, and you can click on it.

Sounds like an awesome future. Here we go. Yeah.

The cost of mistakes there is low, right? Whatever, you mislabeled the bag. Not a big deal. Doing it at scale and cheaply is going to be valuable. So I would actually push back on the idea that aggregation theory is dead.

That was never about business and SaaS. That's about the consumer internet by and large. I think the base LLMs that simply generate an answer are right a lot of the time but make mistakes, and they're still going to have real utility. There's still a huge product overhang in leveraging them. This new capability of paying for a result—and the more you pay, the more right it is—is going to be a different category.

I was going to say, that's just a completely different business.

That's right. And this is where the other point about digital advertising comes in. There's still digital advertising where you pay for display, but you pay for results, and you base your payment to Facebook on results. This is why Facebook is maddening and also inescapable.

They'll take margin because the more you trust Facebook, the better the results will be. You just give in to the AI buying algorithm, and they're showing enough ads to enough people they know will convert to get it. They're showing the ads to a bunch of people that don't know and don't care, and you're paying for it anyway, but you don't know who they are. You're getting customers you never would have otherwise.

And so you set your price target: This is my lifetime value of a customer. I'll pay up to XYZ, and Facebook will fill that. And that's why, when COVID dropped in March, Facebook had a dip for a week and then zoomed right back up, because all these companies that need Facebook—that's their whole lifeblood. They need to be acquiring new customers all the time.

Guess what? If all the brand advertisers want to leave, cheaper ads for us. We'll buy more ads, and it zooms right back up to that line of whatever their sort of LTV calculation is. AI, I think, will be similar. What is the worth of this job to me to get it right?

Mm-hmm.

What level of accuracy do I need? Given that, what is the amount of compute necessary? What's the cost of the compute? Again, this is why I wanted to make this analysis. Right now, humans are the old advertising. Half of them work, and I don't know which half, right?

There's a bit about evaluating your workforce. It's kind of—you just know there are some employees that seem really lazy and don't do anything, and the whole company would fall apart without them.

Right.

And then there are others that are really busy. They're all around. They're always present, and actually, they're not getting anything done.

Not doing all that much.

Right. It's all a proxy right now. AI is going to be much more direct. You're going to know if it works or if it doesn't, and you're going to be able to price it accordingly. There's more transparency in the pricing of the value of a job, which is also going to be hard to implement.

If you're a company with existing processes, you have no process. You have no conception of how to price a job.

Mm-hmm.

You know how to price an employee and what they ought to accomplish, but there's a lot of fuzziness in there, and there's going to be a level of precision necessary to price this appropriately.

By the way, we saw this with Facebook with ATT. The dip wasn't that the Facebook ads stopped working. It's that you couldn't know which ones were working. And so it was the uncertainty of knowing that introduced the dip.

What they had to solve was giving you a believable number that was close enough to reality about what percentage of the ads worked and which ones didn't, so you could buy with confidence. Because if you're buying and your ads aren't working, you could go out of business real quickly. And so it's this level of precision and measurement that is going to be—

Yeah.

All this needs to be built. New companies are going to build it. It's a huge opportunity, to be clear.

And it's easier for new companies to build it, is the key point here.

Well, it's easier for new companies to incorporate it. So I think there are 2 opportunities. Number 1 is new companies that do stuff existing companies do, but they do it all with AI. They're going to be very disruptive from the low end.

Then number 2, there are the framework companies—the measurement companies, the ones that actually give you the tools to know what's working and what isn't. Those companies will benefit to some extent from the new companies, but they're the ones that will help carry existing companies over the AI finish line, which I don't think—

Into the future.

—will be in 2025, but will be down the road.

Okay. Well, I just want to clarify for the record that I do not wear penny loafers, and I will not be embracing that bit in the future on this podcast.

That's what you think.

I don't know where that came from, but I'm not a penny-loafers guy. Any final thoughts, or should we move to TikTok here?

I don't even know if penny loafers is an Old Spice sort of—whatever they're called. Old Spice, Whole Spice.

I think it is.

Whatever they are.

Old Spice.

Old Spice.

Not All Spice. Not old style. This is a fun little variation on the mispronunciations here on the podcast.

4. TikTok Faces A Ban

Well, to keep it moving, Ben, we can turn to the news of the week, which is TikTok. There are several threads to the TikTok story that are totally unresolved at the time of this recording, including but not limited to the Supreme Court and its ruling on the case that's challenging the constitutionality of the law that would impose a ban on TikTok as of Sunday.

There are several reports that PRC officials are looking into Elon Musk, either as a buyer of TikTok or someone who could potentially broker a solution with the US government. Donald Trump, the soon-to-be head of the US government, is reportedly considering some kind of executive order that would attempt to stay enforcement of the law's provision that bans app stores from hosting TikTok.

TikTok itself is reportedly planning to shut down the app on Sunday, January 19, if it receives an adverse ruling from the Supreme Court later this week.

Yeah, and by the way, I'm pretty sure that I don't think Trump can issue an executive order directly undoing a congressional action. So, number 1—

He can try.

He can try.

But I hope not. Yeah.

They'd be able to wait until a court ruled, but I think that would probably happen pretty quickly. I think people are getting too caught up in the legalese here. The law does not ban TikTok. It bans the app stores from hosting it and Oracle from hosting the data.

Mm-hmm.

What I think could potentially happen is TikTok is trying to force the issue by saying, "We're just going to end the service." They don't have to end the service. It could continue working for existing customers. And so some sort of action, even if it got struck down by the courts, could be more a function of TikTok deciding to wait.

There could be some sort of injunction in this regard, too, that lets Oracle continue hosting the data, whatever they're doing, XYZ. With a lot of these things, it's easy to get hung up on specific details when there are a lot of moving pieces and decision-making that could be at play.

5. The Case Against TikTok

Yes. Well, in lieu of bald speculation about what may or may not happen over the next week on any of those fronts, we can talk in broad strokes about some of the logic here, and we got a 2-part question from Saeed. He says, "With the US quote-unquote ban of TikTok fast approaching, can Ben quickly go over the arguments for the ban and why he supports it?" What do you think?

Well, I made my case back in 2020, and I haven't really revisited it. I'm sympathetic to the general idea. The US is a free market. We're not China. I don't buy the free-speech component from a constitutional perspective just because you can go to other forums.

Mm-hmm.

And also, there is Supreme Court precedent about deference in terms of national security concerns on these sorts of issues, particularly if you're not directly muting Americans. You're taking one of their many platforms that they could use.

I do think it's worth acknowledging: Yes, you can tell someone to go to another platform. If you have 1,000,000 followers on TikTok, it's not the same thing. Which, by the way, goes back to our moderation discussion. I think an underappreciated bit of "Oh, just start a new account" is that there's real damage that comes from getting banned and losing a huge following and having to rebuild it.

I have a friend of mine on Bucks Twitter who got investigated by the police for being mad at the refs one time.

Mm-hmm.

He's still thousands of followers below what he had before and had to make a new account.

I follow his new account. I'm one of, like, 800 people who follow him now. Great follow. But yes, it's a good point. I mean, it's not literally an economic taking, but in effect it sort of is because you spend a lot of time on—

Right. You know—

—accruing a following that has value.

It is an economic taking; it's just not priced.

Yeah, exactly.

So we kind of ignore it, right? Which, by the way, if you want to zoom out to our overall spreadsheet critique—what can be measured—is actually a really compelling example. If you have 10,000 followers or 100,000 followers and your account gets banned, because it's not priced, it's underrated in the discussion even though the value is actually exceptionally high.

Mm-hmm.

Having 100,000 followers means you have a huge opportunity and real leverage that you can manifest into other things. But because it's not priced, it gets ignored, which is a great example of how spreadsheet thinking can sort of lead you astray.

Right.

Anyhow, that's a digression. To me, the TikTok thing is very straightforward. The data thing, yeah, I'm generally less worried about data than most people in general. If—okay, are you going to blackmail someone by showing they watch a lot of—

League Pass?

Yeah, League Pass videos, exactly.

I don't think that's where you're going.

But is there location tracking and those sorts of things? There are concerns. I'm not dismissing them. My bigger concern is that you're giving a foreign power—an adversarial foreign power in multiple respects, economically, militarily potentially, and ideologically—

Mm-hmm.

—a direct sort of pass-through to the hearts and minds of the American people. To me, that's insane. We wouldn't have let the Soviet Union control a television network in the Cold War.

And this is a lot more powerful than a television network, which is part of the issue here.

Well, especially with the targeting and the lack of tracking, right?

Yeah.

Could China put its thumb on the scale for some little congressional race? No way to know. You would never—

Mm-hmm.

—you would never know. I documented years ago, in the context of the Hong Kong protests and when the whole Daryl Morey NBA thing happened, that China was clearly controlling the algorithm. My evidence in that case was that you could search for every single NBA team and get NBA highlights except for the Houston Rockets. There were zero.

Right.

Again, maybe in the grand scheme of things, that's not that important of an example, but it's a blatant, clear indication that there is a thumb on the scale here about something that is important to China, and it's just nuts to allow this. The problem and the frustration are that it would've been painful in 2020 to do this. It's going to be a gazillion times more painful now as it's—

Right.

—become larger and larger and you're harming more and more creators. I'm very sympathetic to people on TikTok and the users and the creators.

I am, too. I mean, it circles back to part 1 of the podcast because it's this new platform, and there have been businesses and careers built around this platform that will now cease to exist. You're just punishing a much bigger group of people 5 years later.

And the competition stuff really bothers me. TikTok has been phenomenal competition for Facebook. It's put their rear end in gear. And, by the way, Facebook is much better positioned to capitalize on it because of the competition than they would've been in 2020. It's a bit tautological, but if China is refusing to let ByteDance sell TikTok, which has tremendous economic value, it does kind of make the point that—

Mm-hmm.

—I mean, it's kind of unfair—

Well, and now—

—but it does sort of make the point—

—it’s hard to say exactly what's happening, but it does seem like CCP officials are the ones who are now entertaining some sort of deal.

That's why I don't want to say that categorically, because the logical way to play this for ByteDance is to insist you're not selling and push it to the last minute, and only sell at the very last minute. So I don't want to categorically say that this proves the Chinese government uses it as a unique asset until we're a few months down the road and it's well and truly gone.

But also, alongside that, the idea that the people who are deciding and the people who are potentially brokering some sort of deal are not ByteDance executives but are, in fact, party officials in Beijing—that would be proof if, in fact, that is happening, or would lend credence to the concerns that people have had for several years now.

I mean, but if you want to argue against the concerns, I wouldn't say that I don't believe American people can be propagandized. It's that I think the Chinese would be uniquely terrible at trying to propagandize the U.S. population. It takes Americans to propagandize Americans. So maybe there's an aspect where the concern is somewhat overstated in that regard, and maybe that is manifesting in how they're playing this. ByteDance maybe sees value and says, "Hey, the Chinese government is saying we can't do it." Wrong play, right? We saw this—

Mm.

—when they tried to put up that message last spring, when this was being debated in Congress. They displayed a message to everyone on TikTok saying, "Call your representative. Do X, Y, Z." Totally the wrong way to play it.

Let's—

It—

—send Congress into a panic. Yeah.

That's right. That's right.

17-year-olds besiege both chambers.

6. China Misreads American Freedom

You cannot overstate the level of cultural misunderstanding between China and the U.S.

Yeah.

It affects everything. This comes up in the Little Red Book. Why do I want people to go read a notebook? It's literally the Little Red Book, Xiaohongshu, the app that all the TikTok refugees are going to. I think the whole thing, number 1, is hilarious.

Mm-hmm.

Let's abandon TikTok by going to a literal Chinese app. But number 2, I link to this story from Chinese state-run media saying, "Oh, this shows the American people are opposed to their government. They don't like their policies." You have to read between the lines, and I've been on this side of the world for a long time, to appreciate the depth of misunderstanding that undergirds that piece.

The problem with these cultural misunderstandings is that there are baseline assumptions that don't even occur to one side or the other. Believe me, I learn this all the time in terms of interfacing with my family and things along those lines. It's just a completely different view of the world. The real challenge in foreign relations between China and the U.S. is that, in these sorts of discussions, everything is implied, and you can't believe the actual words.

I've talked about one of my favorite things about the Chinese language: there are these things called chengyu, which are groups of 4 characters that are common sayings.

Mm-hmm.

The best way to insult someone is to use a chengyu and drop 1 of the characters. The character that you dropped is what you're saying they lack. So if there's a chengyu that says, "Oh, he's handsome and trustworthy and wears penny loafers and X, Y, Z," you drop the honest one, and it sounds like you're giving a compliment. It's actually this massive insult because you're implying that they're a very dishonest person, right?

Ah.

That's how—so, that is a stand-in for communication generally. The U.S. does not get that, so they actually take the Chinese at their word about different stuff and completely miss the subtext of what's going on. Meanwhile, the Chinese assume that the Americans are lying, because why wouldn't they be? They're saying all this stuff—"Oh, we care about human rights and the climate and X, Y, Z"—and they're like, "We get it. That's all a lever to get your actual priority concerns." And sometimes the U.S. is like, "No, we actually do care about this." The U.S. is very black and white. That's one of the problems with the whole Taiwan sort of thing: Taiwan is a situation that exists in gray, and the U.S. mindset can't stand it.

It wants to make it black and white and clear.

It wants clarity.

Yeah.

And it’s not a situation that you want clear because there’s no good outcome of clarity.

There’s no clear resolution on Taiwan.

Right.

There’s no question about that.

So you push for a resolution, you’re gonna get a resolution good and hard. It’s not gonna be necessarily the one that you want. So this undergirds sort of everything.

The telltale sign—the laziest media story—is when you search. You go to Twitter, you search, and you find someone saying something that supports your story. You’re like, “Twitter user XYZ said this,” and then you use that as evidence. Chinese state media is searching on Twitter, finding someone who’s like, “Yeah, the U.S. government—they think it’s propaganda. We don’t believe that. We want to go to China.” It’s like, okay, are all these users giving the middle finger to the U.S. government about TikTok?

Mm-hmm.

Yes, they are.

They sure are.

What is hard to grok if you’re not in the U.S. is that this is the U.S.

Freedom in action, baby.

That’s right.

Here we are.

Our whole birthright is about giving the middle finger to the government, and they can’t do anything about it. This goes back to the whole context of banning Trump from social media in 2020 that’s actually, I think, even hard for Europeans to understand. In Europe, the government is always on top. So the government determines what speech is allowed or not, and corporations operate in that context underneath the government.

In the U.S., corporations are an equivalent institution to the government. The government can’t tell them what to do. In the First Amendment context, it’s not just that they can’t say what they can’t say. They can’t say, “You can’t moderate,” because it’s their own platform. They can do what they want.

The idea is that freedom of speech is above both, and the government operates under it, and the companies operate under it. That can go in either direction. They can allow free speech. They can disallow free speech, and the government can’t tell them to do one or the other.

It’s this question of where in the stack they are. In the U.S., they’re on equivalent footing, and so part of the case was, well, there’s a tradition of corporations acting as one of these institutions, as a power broker. When you talk about the balance of powers in the U.S. Constitution between the executive branch, the judicial branch, and the legislative branch, actually the U.S. as a whole is a balance of powers. The—

Well, and look, you made the point on the Xiaohongshu thing. I was actually jealous that I didn’t think of that before Bill and I recorded Sharp China earlier this week. The surge of young people going to Xiaohongshu and crapping all over the American government is such a great testament to the American system.

It is. They’re like, “Oh, look, these people don’t listen to American propaganda.”

This proves it’s all failing.

Yeah.

You are getting propagandized so hard right now, and you don’t even realize it, right? That’s effective propaganda.

Well, and that’s the concern.

That’s gonna be a concern for this company.

No, Xiaohongshu is screwed. So we get all these citizens going online and bitching about the U.S. government. This is an incredibly powerful testament to the American system. It’s precisely because the Chinese don’t even see it that it’s so effective. People going on, and you and me making it explicit, that’s not effective, and everyone sees that’s propaganda. This is real propaganda. It’s in the water.

Mm-hmm.

It’s not a billboard. It’s in the water. And all these people going on Xiaohongshu and saying, “Hi,” “Nice to meet you,” “Ni hao,” blah, blah, blah, it is one of the most effective American propaganda actions in ages. It’s amazing.

Right. Well, and the juxtaposition is really powerful because, obviously, on one hand, you have Americans who are now free to curse out our government and publicly pledge their allegiance to China, and they’ll suffer no consequences.

But a Chinese digital media outlet earlier this week, I think it’s PConline, was talking about the implications for Xiaohongshu and the influx of Americans. For people who don’t know, mainland apps in China, if they’re marketing abroad, will typically develop an entirely separate platform for international audiences. Xiaohongshu did not do that. It did not intend to; it didn’t really have ambitions to expand internationally.

They had no desire to be super big in other countries.

Yeah.

And so the influx of Americans was described as a Sword of Damocles hanging over the company, and the outlet said the risks here far outweigh the opportunities because there’s just gonna be immense pressure and possibly consequences from the government if they’re not censoring all these Americans sufficiently.

Yep.

No, this is why it’s gonna end very soon. My prediction is within a week. Maybe it’ll take a little bit longer. The risks to Xiaohongshu are astronomical. They might try to moderate, but their moderation is gonna be overwhelmed. This is an underappreciated aspect of censorship, by the way, and this actually ties back to our discussion last week. Censorship is at its most effective when people self-censor. That is how you get scale from censorship.

Mm-hmm.

Yeah, you can have the Great Firewall. Yes, you can force all these social media companies to take people’s accounts offline, but you add in, “I don’t wanna be banned. I don’t wanna have that worry. Am I gonna get in trouble from the police, or am I gonna lose my 1,000 followers?” And so people self-police, and that is how you actually get censorship at scale: by activating people’s self-preservation instinct, and you get self-censorship. Guess who is not going to self-censor?

Americans.

The people giving the middle finger to the government by using Xiaohongshu. And so there’s a bit where, even if they try to have some sort of moderation apparatus, it’s going to go sideways.

So, yeah, they’ll probably institute a requirement that you need a Chinese phone number to register, and they’ll force everyone to put in their phone number. Or they might withdraw from the U.S. app stores.

Ah, okay.

And that will cut everyone off.

Or test the dedication of the TikTok refugees. Maybe you find a way to get a Shanghai phone number and use it.

It’s not easy.

I bet it’s not.

And I would advise against that for lots of reasons.

Fair enough. And to the point on Xiaohongshu, the creator of ByteDance no longer runs that company and disappeared for a while because ByteDance was censoring insufficiently one of its mainland apps.

No, they got called to the carpet. I wrote about this in The TikTok War. They got called to the carpet for— In this case, it was mostly because people were looking at too many pictures of—what did we call it?—NBA highlights.

Scantily clad, yeah. League Pass.

Yeah.

Scantily clad League Pass highlights, absolutely.

Yeah.

But it speaks to the point.

I think the whole thing’s very funny. You don’t wanna get in a propaganda war with the U.S., again, because the problem with so much communist propaganda is it’s way too literal. And U.S. soft power is—this is U.S. soft power.

Mm-hmm.

It is our citizens giving the middle finger to the U.S. government, going on a Chinese app, colonizing it, and, in the process, being received: “Oh, it’s so great. Yeah, the U.S. government’s bad,” blah, blah, blah. And the entire subtext here is, “Look at us exercising our freedoms.”

Wouldn’t it be nice if we in Beijing could do that? Yeah.

And hey, the more exposure for U.S. people to clean streets and functioning infrastructure, the better. So maybe it could be a two-way win, at least as long as it lasts.

No, absolutely, and I actually think it’s a really cool example of cultural exchange that China has systematically prevented for the last 25 years.

Yeah. Well, this is the last thing about the TikTok thing that I didn’t get to. China started this. This is a window into what we lost with the Great Firewall—

Great Firewall. I called it the Great Wall with you. Great Firewall is obviously what I’m referring to.

And by the way, there is a very strong case. Leave aside the persuasion issues and the data issues: They block all our consumer internet companies. We should block theirs.

This is a whole problem with our whole trading regime: We’re pro-free trade, but if you don’t have a tit-for-tat retaliation system and one side blatantly violates all their parts of it, you’re going to end up in a bad place. And so, just from a pure trading perspective, this is why it should’ve been done years ago. You just made it way more difficult by waiting until now.

Well, speaking of borders, there was a part 2 to Saeed’s email. He writes, “To the extent TikTok is considered problematic because of a security, political, or cultural impact by a foreign country, large amounts of data accessible to a foreign country, and a vital communication channel being owned by a foreign country, can’t these same reasons be used against US tech companies operating in foreign countries? Based on the reasons given for this ban, does the ban of US tech companies in China require reassessment?” I have some thoughts there, Ben, but do you have any reaction to that question?

I don’t understand part 2 of the question. I think China’s banning of US tech companies has been tremendously successful. Number one, they control the discourse. They can institute the censorship apparatus, maintain political stability, and have people self-censoring. They also developed this entire software ecosystem that rivals the US because they were protected.

Setting aside, from a purely analytical perspective, the morality of the whole question, it’s one of the smartest things that any country did. China’s the only entity that stands in potential competition and opposition to the US from a tech perspective because of the Great Firewall. And every single person in the US who didn’t see that back then, including myself—it should be a reminder that there’s really important stuff you can miss as it’s happening because you’re so sure it’s going to fail.

Mm-hmm.

Bill Clinton: “They’re trying to nail Jell-O to a wall.” Yeah, guess what? Totally wrong.

We need to work on your Clinton impersonation going forward.

Oh, I wasn’t trying to impersonate at all. I can keep with it. And so, my assessment of the Great Firewall is that it was brilliant, again—

Mm-hmm.

—from an analytical perspective—

If you’re—yeah.

—setting aside the moral—

An authoritarian government.

—the moral issues. Yeah. So, I’m not reassessing anything. And by extension, the fact that no one else did it to date is why this is a long-term risk that you’re setting this precedent. Sure. Are any other companies going to have it in them to block US services and develop their own? No, it’s too late. They missed the boat.

Well, yeah.

It is analogous to the chip stuff. Is there a bit where, in the long run, you’re really risking the industry—not just China developing its own industry, but then selling abroad? Yes. Do I have real problems with the new controls announced this week? Absolutely. I think they go too far.

This changes the paradigm from trying to stop China to instituting a permission structure on all of tech, which I think is a very bad and problematic shift. It’s going to be more destructive in the long run because now you’re treating everyone as an enemy by default. I don’t think that is a wise choice at all.

But at the end of the day, the US does still control the entire chip industry, right? And there’s—

Mm-hmm.

—I think that is the analogy in this case to US tech companies.

Well, yeah, and I also think that a threshold issue is just the nature of the Chinese government and the difference between the US government and the Chinese Communist Party in terms of their ability and intent to control American tech companies. The CCP owns a 1% stake in a ByteDance subsidiary but has a golden share on ByteDance’s board.

Which, by the way, this was all known—I wrote about it in 2020. This is how the US works. We wait until the last minute, when it’s much more difficult and costly. But there was an aspect of 2020 where it was still, “Trump supports this, so it must be wrong,” and I think that drove a lot of decision-making.

No, but you made the case in 2020, and it was eye-opening to me because I was one of the people who was like, “What the hell is Trump talking about here? There was a rally that was poorly attended, and now we’re banning TikTok?” It was eye-opening back then for me to read the case and read what TikTok can do. And also, again, the CCP has voting mechanisms that allow the party to override majority stakeholders.

And the voting mechanisms require—

They’re also required by law—

They can still tell them what to do.

—to comply with state intelligence work.

Yeah.

Yeah, exactly.

And by the way, you can make the case, Saeed, that the US can just tell tech companies what to do. At the end of the day—and maybe this goes back to the censorship thing—we’re all parsing whether they actually forced them to do X, Y, and Z. A very valid takeaway is that the US does tell tech companies what to do, and it doesn’t matter what laws are there; it’s still the case that it happens.

You go back to things like the Snowden revelations and all these sorts of pieces: The US has just as much control over its tech companies as China does over theirs. And my response to that is I’m a US citizen, so I’d rather we be in charge than them. There is—

Well, that, and also—

You can’t escape that sort of reality.

The CCP- and PLA-affiliated hacker groups have hacked not only American companies but also the US government itself over the past couple of years. The hacking problem, I think, is far more extensive than the general public realizes. But the CCP has also been actively working to undermine US interests abroad for years now.

Given that, it would be unbelievably naive to think that the CCP wouldn’t eventually try to use TikTok to similar ends, and those concerns don’t exist for a company like Meta operating in the EU. So, I just don’t think it really is an apples-to-apples situation as far as Saeed’s email is concerned. But there are certainly areas where the government has tried to exercise control. Meta’s not obligated under US law to comply with those attempts.

Yeah, but my addition would be that even if it happens in fact, at some point you have to pick sides.

That’s fair. And that actually is a really solid place to land on all this, because I think there are people who want to go back and forth, and it’s just: In principle, this is a matter of national security, and sometimes that wins out at the end of the day.

Yeah. No, I think that’s right. I think this Joseph email actually gets to one extra point that’s a good way to wrap this up.

7. Free Speech Versus Security

Okay, so Joseph says, “I love you guys and agree with 99% of your takes, so I feel bad writing in only to challenge points, but I can’t resist. I’m a bit surprised—

Don’t feel bad. And 99 percent’s too high. You need to lower that.

Yeah, and I always appreciate some feisty emails. “I am a bit surprised by Ben’s consistent support of the TikTok ban. Doesn’t this violate his ‘you’re not gonna out-China China’ rule, especially now that he’s trying to be more consistently on the free speech side of things?”

All right, that’s unfair. I’ve been extremely consistent, with that one very narrowly carved-out exception. I don’t think this “more consistently” means I was wishy-washy to date. I was writing articles and getting a lot of flak for them going back through the entire run of Stratechery, very consistently, about free speech as the top priority. So I—

So it’s not newfound.

So let’s not overstate it. If you go back and read what I wrote when Trump was taken off social media, it was in the context of: This is pro-free-speech; corporations have free-speech rights also. And so, I just want—

Mm-hmm.

That’s a little bit of an unfair characterization. I’m going to put my record forward on this point.

Okay. So do you have an answer for Joseph beyond needling him on his word choice there?

All right. No, so—

What do you think in general?

No, it’s a—this is absolutely a fair point of pushback. I think it is correct. Number one, I would say I don’t think this is a free-speech case per se because people can go elsewhere, to which Joseph ought to say, “Well, when people get kicked off social networks, they can also go elsewhere.” So—

Mm-hmm.

I’m hoisting myself on my own petard on this bit. The reality is everything is a trade-off. This sounds like a cop-out answer, but it’s true. But this whole TikTok thing—this isn’t a sport that’s banned in 2020. And honestly, the reason why I didn’t write more about it, I’m like, “Look, I made my case. It’s 51/49, so I’m not going to spend a ton of political capital going down with the ship on this.” But—

Right.

it is sort of what I think. It was very carefully considered.

Fifty-one forty-nine for you, that is. It’s a close call for you.

Yeah, for me.

Yeah.

Just because free markets matter, and free speech is not just a legalistic thing. It is all these people on this platform and it being a meaningful way to communicate. This is an area—and maybe you want to say you were wrong to say it was always number 1. Well, no, it’s not always number 1. Free speech jurisprudence in general isn’t always absolute. There are—

Right.

—exceptions that are carved out by the Supreme Court. One of those exceptions carved out by the Supreme Court is national security. This is one where I would align with Supreme Court precedent in this regard. I regret this. I don’t like it. I acknowledge it’s a violation of these principles I just articulated last podcast, but I have looked at all sides, weighed all the issues, considered not just the issues today but in the long run, and come down very narrowly on this. I’ve given it strict—what’s the word?

Strict scrutiny.

Strict scrutiny.

Here we go.

I’ve given it strict scrutiny and come down on the fact that the national security concerns do trump—no pun intended—my principles. If you want to say, “Oh, well, so you just gave this whole thing. You’re not wishy-washy,” well, okay, that’s fine. I will accept that. The reality is—

Well, that’s part of being an adult: balancing—

It’s part of being an adult—

—competing priorities and principles.

Yeah.

Yeah.

So, Joseph, you child.

No, we love you, Joseph. Thank you for agreeing with us 99% of the time.

Yeah, look, here’s the thing. Just in broad strokes, TikTok is where an entire generation of Americans gets their news, and the nature of social media platforms is such that control over the algorithm and the ability to either amplify or suppress certain viewpoints is an unbelievably powerful tool.

Right. And again, which, by the way, has been manifested. It’s not just what I showed back then. There have been studies since then that search on various terms, and the reality is there is a significant thumb on the scale about anything China-related right now, just by default, even before we get into some sort of potential conflict or things along those lines.

Exactly. It’s the hypothetical ability to wield this weapon that is just insane. And—

Right. And by the way, that’s why I want to emphasize it: it is happening now. There is a danger—and I’m going to contradict myself—where we do policies based on hypotheticals, which is generally a very bad place to be, right?

Mm-hmm.

Like, to go back to the COVID example, you can hypothesize all these potential bad outcomes and then make policy in response to the hypothetical, which are bad policies, and you should have actually waited for more proof points and better understanding to make very strident decisions and policies that you’re stuck with for a very long time. So I don’t like legislating against hypotheticals. That’s why I do think it matters. There is evidence, there is some degree of a thumb on the scale already, and then that gives more meat to the hypothetical. Again, it is a very close call.

And lots of other vectors where the CCP has been actively maligning U.S. interests across the last several years.

There is a bit about the U.S.—

All of that weighs against—

The U.S. needs to take China more seriously, right? Like—

—tolerating this.

Right. That’s part of it. And I take China very seriously, and maybe you want to say I’m biased because of where I live. I acknowledge that every time I write about this, because that might be the case. Your 99% acceptance rate should decrease—that’s why you should decrease it—because there is an inherent sort of prejudice. I think it gives me better perspectives, and it also—

Mm-hmm.

—gives me a biased perspective, which is for you to figure out and for me to acknowledge.

Yeah. Well, and as far as taking China seriously, I think, to Joseph’s point, the impulse to be uncomfortable with this law is natural and very American, consistent with the U.S. tradition. But the reality is there—

Of not liking ambiguity. And, by the way, I am very uncomfortable with the law, so that’s why I did my—

No, exactly.

So.

Well, and it’s also—the American system can tolerate a lot of different dissent, but you look at, like, with Twitter even, if you go back to COVID and the way Twitter handled some of the censorship issues there and certain things that were suppressed, that distorted the conversation a couple years ago. I wonder if Elon was running Twitter a couple years ago, whether certain policies would have been different, because there just would have been more open discourse around that. I offer that strictly as an example of where algorithmic control is really, really powerful and can have real-world consequences on the democratic process and our policies. And then, obviously—

This is a good example of how you could take the hypothetical bit and say, “Well, your concerns about free speech—” People would push back on this in the 2016–2017 era. “Your concerns about free speech are hypothetical,” right? And it turned out—that’s why we go back to the COVID example—because the hypotheses became reality, and that should influence your thinking and your changes.

By the way, to go back to Joseph’s email, am I demonstrating inconsistency? Absolutely. Consistency is the hobgoblin of small minds, or whatever the word is. We’re doing the big-mind podcast here.

We’re adults here.

That’s right.

We’re adults here. Yeah. Well, and there’s going to be more inconsistency.

But it’s fair—it’s fair to call me out. It’s absolutely fair to call me out. I’m giving Joseph our time, totally valid email, totally valid points.

And each one should be evaluated on a case-by-case basis.

People concerned about misinformation online. Yeah.

Yeah.

Ben Thompson It’s like, look, no, we care about free speech; we’re just thinking about all the details and thinking it through, and that’s actually totally valid. I, on that one, fall in a different spot, and I’ve ended up in an even more extreme spot than I did 4 years ago for the reasons I articulated last time. But that’s why I respect and am open to the argument, because this stuff’s not easy.

One more bit on “not out-China China.” That applies not just to the U.S.; that applies to other countries as well. They’re not going to ban U.S. companies because they’re not going to out-China China, right? Like, the—

Hmm.

—and so there is a certain amount of—

You have to want it to ban U.S. companies.

—favor, yeah. That’s right.

Well, Ben, I’ve enjoyed the ride here, and whether TikTok is around or not by next week, we will be back. We will have an episode up Monday afternoon, getting back to regular schedules around here. Until then, enjoy the weekend, and I will talk to you soon.

Talk to you later.