Sharp Tech with Ben Thompson - GPT in 2023: A Marriage with Microsoft, The Bull and Bear Case for Google, Compute Questions and More

Episode Date: January 12, 2023

Reactions to the structure of the rumored Microsoft-OpenAI investment, challenges and opportunities for Google as AI matures, what happens to the internet (and AI) after auto-generated text floods the zone. At the end: text vs. images for the future and follow-ups on failed Microsoft investments, YouTube's NFL gamble, and Amazon's NFL ratings.

Hello and welcome back to Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, how are you doing? Good. I mean, some of us have gotten back to work this week as promised. Others of us are making claims of sickness and incapacitation. I won't say who's who. I mean, my record speaks for itself.

You know what? You're in a glass house, Mr. Two-week winter vacation. Take the first week of 2023 off. I did get hit with COVID over the weekend. Not the best way to start the new year, but I'm now fully negative and emancipated. So it's great to be back. It felt really good to say,

welcome back to Sharp Tech. It's been a while, man. It has been a while. So let's get to it. Yeah, I guess we should get into it. I still have some lingering congestion. So we'll see how this goes. So you said, I need to wear my vocal cords here.

Got to carry this show. Exactly. We're starting off the year with eyes on the future and specifically eyes on chat generative pre-trained transformer. More popularly known as ChatGPT, a chat bot launched by OpenAI in November 2020. Maybe you've heard of it. I want to take a look at where things stand with ChatGPT, though. and what's most exciting to you as we forecast the next steps. Obviously, we can start with the news that Microsoft is eyeing a $10 billion investment in Open AI,

with OpenAI reportedly valued at $29 billion in that deal. And I have some additional context for you from the information who writes, Microsoft and Open AI have spent months hammering out a complex deal that aims to balance their respective needs. Open AI needs more money to keep improving its software, and so it is now willing to pay a large share of future profits to Microsoft and other investors, a necessary step given the low likelihood it will try to go public. But the startup, whose chat GPT chatbot has taken the industry by storm, is proposing that it transitioned to non-profit status again after it pays a certain amount of profit to the investors. After AI pays back its first investors, Microsoft will get 75% of

profits until its principal investment is paid back and 49% of profits after that until it hits a theoretical cap according to a person briefed on the terms. And then Brandon asks, Microsoft's reported deal with Open AI is pretty weird. Do you think this is the result of regulator pressure on tech acquisitions, or is this just a way of tying up open AI so they're as good as owned by Microsoft without having to deal with the SEC? What do you think, Ben,

take it any direction you want? What Brandon definitely has right is that it's pretty weird, but that is a function of open AI being pretty weird from the beginning. Like open AI started out with this, remember Elon Musk was a part of it, along with Sam Altman and, you know, some other folks with the idea that we're going to create

artificial general intelligence, AGI, that's like the, that's the artificial intelligence that basically like is smarter than humans, right? Like it trains itself. The robots who will take over the earth one day. That's right.

And the reason they're going to do that is because they're going to do it safely and thus prevent someone else from doing it harmfully. Which, I mean, I think the whole, whole thing with AGI is there's sort of a tail risk scenario here where it does go horribly wrong and we end up in pods for energy like the Matrix or something along those lines. And I'm not sure that like just because we're the ones with good intentions,

we're going to somehow stop that from happening. It kind of feels like if that happens, it happens, right? I'm not sure that there's, there's much you can do about it. But that it's important to understand that's the context in which Open AI started and it started as a nonprofit. Now, they switched away from being a nonprofit to being a capped profit entity three or four years ago. And basically the issue was exactly this.

Like, we need to raise a lot of money. And if we need money for all this compute, then we, like, no one's going to give us money unless we can sort of return a profit. Now, my feeling is, I'm, I'm. I think this whole thing's kind of a mess, honestly. I mean, maybe there was some aspect of starting that way, got a lot of really influential and smart people on board

because you're sort of selling this really aspirational sort of idea. And we're not in it to make money. We're do X, Y, Z. I tend to feel better about companies that are in it to make money. It just seems like a very sort of straightforward. This is what we are. This is what we're going to try to do. I think it would make investments much more straightforward.

And I think this, like the situation where they're ending up basically, I mean, because Microsoft, the reason why Microsoft can do these sort of deals. And to date, like, open AI is like doesn't spend any money at all. It's like all Azure credits. So like all this money that like millions and millions of dollars, it costs to trains these models, Microsoft is basically paying for. And there, it's funny because you would hear chatter, you know, in Silicon Valley previously where like, man, Microsoft really got taken for a ride here. just, you know, paying for all this and XYZ. And which always seemed odd to me because you're seeing the manifestation of being willing to do that, where they kind of have Open AI by the, by the, by the, whatever you

want to call it. Yeah. Choose your euphemism carefully. Choose your euphemism, uh, adventure here. Um, and because at the end of the day, open eye needs compute. And Microsoft has a sort of because number one, they get a build up. right? So there's a peer sort of like building up the sort of capability. They'll be, you know, when Open AI sells their services and capabilities to other companies,

Microsoft is taking a share of that. They can sell the privilege access to sell this. They can incorporate into their products. Like they have so many other ways to make money off of this. Even if they lose money on sort of open AI itself sort of consistently for a very long time, it's a great sort of orthogonal bet for Microsoft because they're getting all sorts of of benefits, even if they don't actually make any direct money for a while. And it certainly feels like that, yeah, Microsoft basically has not control, but first dibs,

that's for sure, to this transformational technology at the cost of investments that they would probably make anyways. And I think that's downstream of starting out this idealistic structure where if they started out as a normal company and we're going to raise money from investors and they started on Google Cloud actually and we're just going to buy our commute and do XYZ. Like that'd be more expensive in the long run, but it feels like they would have more freedom of movement in the long run as well. So it feels a bit like a Faustian bargain that's been made here. But again, who knows? Like this is just another thing about this entire area. We don't know. That's why it's

so interesting. That's my favorite part about this whole story as the entire tech community is obsessed with chat GPT and all the different possibilities with AI, you talk to all sorts of smart people and they all settle on that answer where it's like, well, we're going to have to wait and see. I mean, everything that's possible today, six months ago, people would have expected to take like two or three years. And it's a very exciting time. Now, as far as- Just to jump in that, though, I mean, this is the, there's a lot of interesting, you said I could take any direction. I want to take it. So, and you also said that I had to, carry the show. So I am, I am here ready to carry the show and go in lots of directions.

There is, when you go back to like theory, right? And I've, you know, I've obviously long-man subscriber and written a lot about Clayton Christensen and his theories. But one of the sort of takes there is when new products come to market, they tend to be like fully integrated. And the reason is because the product's not quite good enough. And if you, if you control everything, you can make sort of compromises in ways that are optimal for the overall product, even if it's suboptimal for like, like if you have to support third parties, right? And you have to have like a very clear distinct plug-in structure so they know exactly how to connect. And you have to maintain that over time.

And it restricts your freedom of movement. And so in the long run, once the technology is stable, that's a good thing. You can build a big ecosystem. But before the technology is stable, you kind of have to do it all yourself. And the reason I jumped on you as far as the timing bit is GPT3 came out two years ago and or now three years ago, or in 2020, came out in 2020, GPT4 is probably coming out this year.

And, and now, chat GPT is an evolution of GPT3. It has all this, you know, reinforcement learning from human feedback, and they call it GPT 3.5, whatever might be. But the key thing is,

is the sort of underlying stuff was there, and you could see this being there, but it took Open AI to make themselves, to make it into a product. And that sort of fits the theory, where the company that is inventing this stuff kind of has to do it all at the beginning for it to break through and to resonate. That's exactly what we saw here.

So, yeah, it's a big deal now because this November release, what this November release really was, was a consumer accessible and understandable product that made that technology that existed more accessible. And I just think that's sort of it, it's striking to me that how neatly this sort of fits the idea where it wasn't enough to have the technology out there. It wasn't enough to have an API.

Open. I had to build a product. And that's when, boom, you have this sort of widespread understanding and explosion of interest. Exactly. It's remarkable. It's crossed over to normies. There are so many different topics on Sharp Tech where I silently worry. I'm like, is this going to make sense to like the regular person who's listening to this podcast? Everyone I know knows what chat GPT is and has played around with it because it's that intuitive and just sort of spread virally for the last couple of months. And now GPT4 is like this urban legend that people are talking about on tech Twitter every day,

one trillion parameters, whatever that means. It's all very exciting. 29 billion dollar valuation for open AI. Can I get a sense of how crazy that is at this point in their life cycle? That seems very reasonable to me. I mean, the key thing to understand. understand about startup valuations is they are not a measure of like their current revenue, which is very minimal, right?

Like you don't do like a discounted cash flow valuation on a startup and see X, Y, Z. The reason why it's $29 billion, it's saying like, okay, there's five big companies right now, right? There's Microsoft, Apple, Amazon, Google, and Facebook or meta, although meta is sort of, you know, fallen off a fair bit from their previous highs as far as valuation goes, but those are sort of the big five. This is a bet that opening eye will make it a big six. And so $29 billion, if they, there's no expectation their long term value is $29 billion.

What it's saying is we think there's a chance this valuation is $2 trillion. And maybe that will happen or won't happen, but it's a, it's a bet. And we think the odds, you know, if they're like if you, if you, just to do the quick math, which I might get wrong my head, but $29 billion to $2 trillion, if we think there's like, 12 to 13% chance that's going to happen. That's a reasonable sort of amount to pay because, you know, your expected value would match up to that. So that's how to think about startup valuations in general.

And I think it's a very reasonable bet to make because one way where this whole thing could play out is at the end, and this is one of the open questions, like are, are these models going to be super widespread? And you actually don't need that many parameters. You don't need a trillion parameter. You can do relatively small. There's a ton of research about. this that's very, very compelling. You might actually be able to make really good models without

that much data. If that happens where it actually becomes trivial for anyone to do this, then Open Eye is not going to be worth that much, relatively speaking. That's one possible outcome. But another possible outcome is actually, it's really highly differentiated to have super high end, super big, super expensive models. And part of ChatGPT also, it's not just the model itself. there is a big human component to chat GPT. You know, this reinforcement learning with human feedback. Like there's humans in the loop sort of refining this model and giving it feedback. That costs money.

Like that's expensive. And so if that's the case where it actually is really expensive to build this stuff, then the optimal route for companies is going to be not to build it themselves. It's going to be to use open-Aid stuff. Right? And so then they become like the AWS of a. AI basically, right, where it's just a service where no one else can afford to build all that stuff, just easier to rent it.

In this case, it would be easier to just rent AI as opposed to sort of building your own AI. In that world, I think Google, that would be a good world for Google. I think it makes a ton of sense for Google to expose their capabilities via Google Cloud and that sort of idea. In that outcome, like, that's where this valuation will pay off. And no one knows. Like, no one really knows which direction it's going to.

go. It's probably going to be somewhere in the middle, but, I mean, TBD. Yeah, well, in near term, the information also reported that Microsoft wants to integrate chat GPT into its office products like Teams and PowerPoint, Outlook, Microsoft Word, my favorite program. I'll read one excerpt from their report earlier in the week. These goals won't be easy to accomplish, they write. For more than a year, Microsoft's engineers and researchers have work to create personalized AI tools for composing emails and documents by applying OpenAI's machine learning models to customers' private data, said another person with direct knowledge of the plan, which hasn't previously been reported. Engineers are developing methods to train these models

on the customer data without it leaking to other customers or falling into the hands of bad actors, this person said. The AI-powered writing and editing tools also run the risk of turning off customers if those features introduce mistakes. So that's just a little glimpse at some of the potential and the short-term challenges in terms of the way this might be integrated into like a regular person. I know, but just think about the possibility. Like, where else, where is a place you're more likely to want the ability to just generate a bunch of texts than in a word processor, right? Exactly.

I mean, like, this is why, I mean, Microsoft just like feels like just there is so well placed. And yeah, there's questions, all the things we talked about before, a question about accuracy and is it right? And it's it's so confident in every answer, even though it might be totally wrong, right? But at the end of the day, if you need a place to use it, that's sort of the place to use it.

And, you know, this personal data thing, the way it works is you have this full model, like imagine it's like 100 layers in the model, right? And what you do is that final layer you put on like the personal data. And so they, for like things like GitHub co-pilot, where it does the coding stuff, they basically have, it's the same language model.

It's built on GPT3. but then the last few layers are coding specific sort of stuff. And so the way they would do this sort of personalized thing is that sort of last layer. And this, I think this is also applicable to stuff that Facebook's sort of exploring, like, how do, how can we make AI to generate personalized results for individuals? And, you know, and so that that's aware a lot of this stuff they're trying to figure out. But yeah, I mean, if you're, I mean, like, we talked about like AI and homework and stuff like that. If you're writing a report and word and you can click a button, get.

get a paragraph generated. Yeah, it was a real light bulb goes on moment for me because I was sitting there thinking, you know, I don't really expect to go to a separate website to use chat GPT as I'm working. And this is back in like November and December when I'm imagining use cases for me. Chat GPT is a great little novelty,

but I wasn't expecting it to really transform my daily life. if you could make an elegant plugin that people can just use really easily in Outlook or Microsoft Word, I mean, that's like a real game changer. Yeah, remember you had to make an outline in school. Imagine if you just like make an outline of your paper and then click a button to transfer to. And then you can specify this essay has to be exactly five pages, you know, change the margins if you need to. Change the font size. But you wouldn't even do that because the AI could figure it out.

It's going to give you the exact length about the exact topic and the or yeah. It's pretty wild. Yeah. And I see all these smug college professors saying, you know, chat GPT3 isn't really that great. It's so easy to spot these essays. Just wait until GPT4, you know. It's going to be next level stuff easier than ever to cheat and college. No, I mean, just again, that's another good observation, which is,

the mistake people make is, and it just happens with technology all the time, is they evaluate the product that they see in front of them and assume it's going to be like that forever. And you have to look at what's the, what's the plane of improvement, right? GPT2 came out before GPT3. And it was mind blowing at the time if you were sort of paying attention. GPT3, I think, was a really big leap. But the trajectory is that this gets sort of better and better. And I actually think, you know, the question is like, where is there going to be like startup value in here? It's not clear.

If I were a startup, I would want to target industries where accuracy is super important because the existing companies in those industries will not want to use AI because they'll say, oh, it makes too many mistakes. Whereas you come in and say, okay, yes, it makes mistakes, but it's way cheaper. So we're going to offer a relatively quote unquote crappy product for the market, but at a super low price and you bet on it getting better. And so then by the time it's really good and actually is competitive with humans, you've established sort of real market share and mine share. To me, if I were a startup in this space, that's what I would be looking at is like actually I want to target places where AI seems like the worst application because it makes it the least likely that the existing companies in the space are going to respond. on. Okay. Well, there you go. I want to emphasize two things. Number one, Sharp Tech does not endorse cheating in college or anywhere else. Speak for yourself. Integrity First podcast. And number two, Ben and I will be leaving Sharp Tech to start our AI powered law firm within the year. So get excited for that. But

Sharp Law. Yeah, well, we'll see. Don't hold your breath on that front. The other aspect of this that we continue to get questions about almost every day, Google and chat GPT and a competition between those two. I want to run through two separate emails we got. Sean says Google's competition has only been a click away for a very long time now, but are we at that moment once chat GPT exits beta? I'd love to hear you make the bare case for Google going forward. So, Ben, what do you have for Sean?

Well, it's important to know that chat GPT doesn't know anything about like current events, like what's happening sort of now, right? So, but again, I just said, don't anchor on what it is now. You want to think about sort of the trajectory going forward. I mean, so the thing with Google is I have a very poor history of predicting outcomes for Google. I've generally been somewhat bearish and it always been wrong. Like, or I would say I was bearish like back in 2014, 2015. and was utterly incomplete wrong.

And it was actually a big learning lesson for me where I underappreciated, number one, I think Google responded to things like vertical search, like very aggressively. They changed the search results, not just to be 10 blue links, but to be like contextual related to what's there, XYZ. And then distribution really matters.

Like they're on all Android phones. They have this deal with Apple to be the default on iPhones. They have deals with almost every browser maker to be the default search there. as a friend at Microsoft reminded me, they're not default on Microsoft browser. There you can get thing, but everywhere else,

you know, it's the default. And people have been using Google for 25 years or whatever it is, right? Like there's just a lot of habit that goes into that. And for some things where, especially where Google makes money, it's not clear that chat would be the best interface,

right? Like if you're booking travel or you looking for life insurance or whatever it might be, like the sort of areas, where you're doing research, which is where, and you're going to make purchases, which is where Google does make a lot of money, is that actually, do I want a chat bot actually booking like a hotel? No, I probably want to look around. I want to decide XYZ, you know. And so on one hand, and so I'm very, very nervous to bet against Google at this point. And there's the fact that by all accounts, Google's AI capabilities are better than open AIs. Now, Why haven't they released them? Well, you know, they would say, oh, you know, we're worried about safety and X, Y, Z. But also there's, there is the business model question, which is if people start using these interfaces, like a key thing, the way Google's monetization model works for search.

It was so transformation when they came out. It used to be you would pay for ads by impressions. How many people saw this ad, right? Google's like, no, all your ads will be shown for free. You will only pay if a customer clicks on them. And basically what they did was they have. We have this valuable commodity, which is user attention, and we know what they're looking for. And so all you advertisers, you can bid against each other in an auction to say, we want to put our ad up.

And who decides who wins the auction? The user does. How do they decide who wins the auction by clicking on an ad? And by the act of clicking on an ad, that is when they're like, oh, hey, that's the ad that I want. The advertiser then pays. And it was such a win-win because it's like, you know for sure. you have an interesting customer. Now, over time, particularly on mobile,

as Google's cram more and more ads into it, it does feel increasingly sort of exploitive because it's like you get a search result in the entire page looks like search results. There's a little bitty thing that says ad right there, right? And if you go down to the organic search results, it's like, oh, there's the exact same company that also felt the need to bid to get an ad up there

and Google's just sort of harvesting money. But the problem is in a chat interface, that goes away. And, you know, I wrote about it a little bit. this week. We're like, oh, why don't, why doesn't Google just insert like a text ad in there? Number one, you, you could do that. It'd be a crappy user experience, much crappier, I think, than ads in search.

But number two, you're not getting that same dynamic of users selecting the winner. And which really feeds into this entire like system of auctions and things on those lines that that makes Google work. And so Google is the one company that does feel threatened here because, and not because it's a tech thing, because the business model. That's really it. The thing about disruption is not that it's this companies are blindsided by these new technologies. It's that the new technology doesn't fit their business model. And so they're heavily incentivized to not invest in it or did not see how it works, XYZ.

Now, everyone knows about disruption. Everyone knows the story of Kodak that I referenced this week, like including Google. I suspect their response will look like their response to vertical search where they will start incorporating chat like interfaces into their search results for topics where it makes sense like particularly like historical stuff and things where they have a high degree of confidence and it will just like search will evolve and that it will probably like it'll probably work. It's like I'm going to bet on Google at this point knowing that I will miss the top when it comes.

You've been bearish because I've been wrong before, right? And no, I'm completely honest about that. I can just like, don't trust me on Google predictions. Like I'll be pretty like I'm pretty good on Microsoft. off. I'm pretty good on meta. I'm San D's on Apple. I've been bad on Google. I admit that. And and I just sort of like thinking about, I think about this week, right? Full disclosure. Yeah. No, I mean, like I feel like I've learned a lot from Google by being wrong. Like in a lot of actually like aggregation theory and understanding the power of these companies

came in part from me being so wrong about Google. Like, wait, what am I missing here? Why is it that there, there's so much. stronger than I appreciated. And, you know, like, the only way to be right in the long run is to admit it when you totally, when you totally goofed before. And, and so, yeah, I think there's a good chance they'll figure it out. And just being, you know, defaults are really powerful and habits are really powerful. And they do have good tech.

And it's worth noting YouTube is a monster, like, and is, shows no signs of slowing down. Again, they have business model challenges. Those are like regatory, like ATT and the E. But that's not that that that's just a that's a secular challenge like that's fine and they're actually better place to solve that than lots of other people. Google Cloud is doing better than people realize it's still losing money, but their growth is still very high and their margins are improving like very rapidly. So I think that's looking good. I think it makes sense their continuing investments in hardware. They're still bad at selling, but the actual quality of hardware is good and they make sense to have that business model coming along in the long run.

So I think they're probably okay. And also I do love this idea of Google really exposing their capabilities via API and letting other companies build on them. And if they're going to get disrupted, let it be disrupted by a company where Google's building the infrastructure for them. Yeah. Well, and Google will also have the advantage. Once AI leaps to video, that video is going to have to be hosted somewhere and YouTube's inventory. can increase 10 fold, 100 fold, who even knows what's possible.

Oh, the other thing here, sorry, this is why I wrote a 5,000 word piece on my first day back, because I have a lot of thoughts. I mean, the video stuff, like people who have gotten stable to fusion, like that open source one, to make video. Like, it's very short clips and it's not super high quality, but you can see this sort of coming. And certainly very interesting. We lead forward every four months at this point. There's like this massive leap.

Every four weeks to be. Yeah. The stable of fusion is incredible. Like just having this out there as open source, it's sort of like it's sort of been a phenomenal example of how powerful open source sort of can be. But you think about the ability to generate content, things like the metaverse, right? When you like when you can just worlds can be created on the fly. I mean, again, we're a long ways away from that.

This still takes a lot of compute power. but the compute power is a very, very sort of big question. The amount of compute power necessary is going down, will continue going down, but it's still sort of a discrete cost in a way that, I mean, cloud services like going using Google, of course, that takes a ton of compute power as well. But it's, it's just, I don't know, it like getting like power is going to be so integral to this, right?

If we actually had a world full of nuclear power and power was super cheap, like, that would actually be an accelerant to AI. If fusion ends up being a thing, it would be a massive accelerant just because, like, you buy these chips and there's a big expenditure for the chip up front, but there's massive ongoing expenditures and electricity. Like these GPUs use tons and tons of power. And so there's lots of stuff that actually ties into this world. And, you know, the cheaper it is, the lower the marginal cost, like so much the internet.

is about zero marginal cost, where one extra user theoretically costs something, but it doesn't, in practical terms, in decision-making terms, it doesn't really matter. AI is not there yet. Like, every extra user, like, like Sam Alman talked about in the context of chat GPT, like every query is costing them pennies. Like pennies is a lot in tech. Like, that's a huge amount of money, right? Like, you, it needs to be costing 0.000, 0,000, 0,000, 0,01 cent amount so small that you don't

even calculate it because it's effectively meaningless, right? AI is not there yet, and that's going to be a big question. Right. Well, that was one of the reasons I did a double take at the $29 billion valuation. Like, I understand that OpenAI is the Victor Wimbunyama of the tech startup world. But at the same time, what they're doing is very expensive and they haven't yet figured out how to make money from it. So the next steps are going to be pretty interesting to monitor. Well, also, but that's what gives Microsoft so much leverage over them.

Microsoft's like, we'll pay it. don't worry, we'll pay it. We'll pay it, but we're going to take a lot of control and make sure we incorporated into everything we already do well. Yeah, and by the way, if you ever do figure out to make money, we're getting all our money back in a lot more. It's like the NFL and YouTube, man. But so Google, I'll just say for the record that I agree with you, there is so much muscle

memory working in Google's favor among normal people. and when I think about what I use search for, it's almost always to look up a restaurant or a phone number or book a vacation, all stuff that I don't necessarily see chat GPT, like disrupting in a meaningful revolutionary way. And couple that with their advantages being built into Apple browsers and obviously the Android. Like that's a pretty big deal and going to be difficult. to displace. So that's just

sort of a word of caution to the legions of people who keep asking. I don't trust anyone, including people that know me very, very well, to book anything on my behalf. I'm like, no, I want to know exactly where I'm saying, what seat I'm seeing on the airplane, like what time it leaves. Like, I'm not

ready to trust a computer with that. Again, people now throw my words back at me. We have look at the trajectory, maybe in the long run. I don't know. We'll see. Yeah, we'll see. another question from Dixon, and this is on the chat GPT side of the equation. He says,

Ben mentioned Google search is likely to face disruption from chat GPT. I understand that thing might be very willing. I didn't word this right. No, no, no, I'm mad. If you were to outline a scenario where search was threatened, it would look like chat GPT. So likely is the word that I don't know if I said, I don't think I said likely.

What I was trying to say is this is what disruption would look like if it came for Google search would look like yeah now is that likely or not again I think we're in a different era than 20 30 years ago just like I think companies are smarter about this number one number two I think Google can figure out a way to navigate this but if you were to outline a way where Google search actually did go down this is what it would look like where you're actually in and again and this is not a new thing for like I wrote about the seven years ago guys a Google assistant, right? Google assistant from day one was better than everyone else, and it was also intrinsically problematic for Google because you're not in this user deciding part, which is

not a technical issue. It's a business model issue. And the business model, like, that still hasn't been solved. Like, what's the business model for Google Assistant? There isn't one. And the more people use Google Assistant, the less they're using Google search, which is bad for Google's business. Now, it hasn't mattered. It's still grown plenty. But the outlines of that being a concern have been clear for a very long time. Yeah, well, and you look forward. Like, it's, it's pretty clear. I mean, it's clear to me. Using my normie powers, I just feel like Google has a lot of utility for regular people and chat GPT would have more utility in a Microsoft office context than replacing Google as like the default thing that every human uses. And I could be completely wrong on that front

as well. It's just sort of my instinct with this. Well, the other thing is that, If chat she be, sorry, I think I keep interrupting you. No, no, no. I love the, the genuine enthusiasm you have after three weeks off. Well, not just that, but this is so, like, it's new in it's new in a, it's unfolding in real time. Yeah. And, like, I think I do think a real problem for search is the absolute amount of crap on the internet, right? This has always been a problem with Google.

Everyone's trying to optimize your stuff for Google. And if you search for certain topics, you just get all this. junk stuff, that's going to go to 11. Like if you see it, like the people making money off of chat TV today are like SEO spammers, right? It's like, oh, you can generate all this all the way. This, by the way, I think is a bullish thing for meta. Like, you're going to be able to go to meta and say, I want to advertise this product. And they will jet or this is a picture of it.

They can generate images and text and A, B, test it and find the perfect one. And instead of humans generating all those images and humans generate all those images and humans generating all that text, you have maybe like 10 options. They're like, we used our AI to run through 10,000 options. We came up with the perfect ad that works best with this sort of thing. And like that's going to be a phenomenal tool that I bet is going to show up sooner rather than later. And that works very well for meta given the type of advertising they do, which is top of the funnel. You try to get people's attention about a product they don't yet know about.

For Google, that's a problem because Google is looking for the right result. and they're just going to be inundated with crap generated by AI. And so probably the biggest threat of AI to search isn't just people using AI instead of search. It's AI spoiling search to an even greater extent than it's spoiled today. Well, and on the Microsoft end of the search spectrum, there's all this talk about Bing incorporating ChatGPT, but Dixon wanted to know how will Bing monetize it? And how does anybody monetize a chat interface implementation of search? I think that Microsoft doesn't care.

That's the problem for Google, right? This is what happened with Android. Google released Android for free and quote unquote open source in contrast to Windows phone licensing model, which tried to replicate the PCOS licensing model. And the reason is that Google cared about protecting their search business, Right. And if Microsoft is in control the phone, they were worried they'd be cut off. And you can see that on iOS. Google has to pay Apple billions of dollars a year to make sure there's still the defaults, right? They don't have to pay anyone because they're the default on Android because they make Android.

Right. And so like Android, the Android investment is paying off in the tune of billions and billions of dollars a year in money Google does not have to pay to anyone else to sort of have it be the search. Now, they do give money to the OEM makers and cetera. but it's not nearly the extent to be if Microsoft was in control, for example, right? Or they might be completely cut off because Microsoft would make Bing sort of the default. So in that case, Android was a great investment for Google, even though they didn't monetize it directly

because it built this big moat around search and it harmed Microsoft. Microsoft will think the same way about Bing, right? We don't need to make money on it. We just need to harm Google and have an obvious inroad to our services, where you use Bing, you use it for free, you can get other sort of stuff.

And then you like, oh, you could incorporate this into Word. Oh, you want to use Word? Oh, you need to get an office subscription or Microsoft subscription or whatever it might be. Like, so it's definitely, it's like an inverse of what we saw happen a decade ago. And that's a problem for Google. It's always tough when a very well-funded competitor does not care about making money. That sucks for a small company. It sucks for a big company, too.

Does Bing need to get a new name if it's going to be taken seriously as a threat to Google going forward. Because Sachi and Adela, we talked about it on the holiday episode. He had the foresight to recognize that teams needed to be named teams and not Skype. And Bing is just so, it's, it is such a, sorry, three, two, one. No, keep this. You organically laughing because you're so disgusted with the name Bing is great content. It's just such a viscerally mediocre name that I feel like there has to be some sort of departure if they're trying to make it successful in the mainstream.

I mean, we're talking about our Steve Balmer statue. Oh, that reminds me. I got, what's funny is what are the acquisitions that got mixed up with an HP acquisition. HP is the one that sued because they said the company the acquired was rotten. Microsoft didn't need to sue. The company just stunk. So my apologies for getting that wrong. But what was the viscerally mediocre?

Or viscerally mediocre? Is that going to be the title of our Steve Balmer statue? It was a great run. But look, Satya Nadella is creating quite a legacy for himself at Microsoft with yet another seeming victory on the way here with Open AI. I mean, even if it's not a victory, it's such an obvious and reasonable bet, right? Like maybe it won't work out. Process over results. Yeah, no, exactly.

I mean, yeah. And his record really is. is really remarkable. Yeah, Bing, I agree. And Bing feels like 2000s era Microsoft, right? Like, you know, with the, what was their MP3 player called? Yeah, Zoom. Right.

Zoon, Bing, all feel of sort of like the same generation. I'm with you. I think whenever they incorporate all this sort of stuff, come out with a new name. I like it. Okay. So next question, Robin says, I was listening to Ezra Klein's AI Skagit. skeptic show today, and it occurred to me that no one seems to be talking about the possibility that these AI models that are built on text from the internet like GPT will start to get very

weird if the cost of writing plausible text goes to zero. That is, the internet is likely to be flooded with bullshit from AI models, and if you're using the internet to train AI models, the quality of the output of those models could go down over time as the influence of bullshit it goes up exponentially as a portion of the training corpus. What do you think of that possibility, then? I think people are thinking about this. And this kind of ties into the Google search being inundated with crap sort of thing. It's the same sort of concept.

This is, I think, a bullish thing for larger entities like Open AI or Google or whatever might be because they will have the resources to make better inputs, right? And I think one thing, there's a ton of open questions on AI. This is another one. I think the extent to which the internet could be used as data was shocking, but something that chat GPT is shown is that actually, no, refining the data to get it like to makes a huge difference, right? Like that's, and so that to the extent that matters, the better it will be for companies with large resources to have better inputs, right? Garbage in, garbage out sort of thing. And, and so yeah, I think that that's an interesting question.

And I do think, you know, people talk about, like, Microsoft had a, was it volley or whatever, the voice, like, you give it three seconds of someone's voice and it can, like, reproduce, like, anything in it. Yeah. I think it's probably going to take a while, but be healthy for society to realize probably everything on the internet is bullshit. And, like, you, I mean, it's interesting. I think it's good for, I want to have to be bad for someone like me, right? Someone's going to, like, just release a podcast that's my voice. That doesn't sound great. On the other hand, someone that you know and trust and rely on becomes that much more valuable. It's as some random article you sort of found sort of on the internet.

I think there will be returns. Like brand is going to actually become more important on the internet, right? You go back to like CPG sort of companies. All this stuff, shampoo is the same, the toothpaste, the deodorants, like the plastic bags. You differentiate via brand. And I think that's going to happen to more and more stuff. Stuff that is a commodity will be differentiated by brand to a greater and greater extent. And that seems like a way this might play out in the long run.

We'll know the real Ben Thompson when he mispronounces the word bag on the podcast. That's why Wisconsin accent. I'm not apologizing for that. I was back of the day when I was teaching English ages ago, I was at the school that was very, very, strict on pronunciation. And I complained, I think I was doing a tank or bank. I'm like, no, the predisposities guy's wrong. Like, no, you just say it weird. I'm like, oh, that's that's on. Plastic bag. Yeah, all the AG and A&K and A&G endings is a very distinct Wisconsin accent. You know what? I'm proud of it. So there you go. There you go. Go Bucks. All I will say is that Luke fickle.

It's been incredible. There you go. I saw the fickle tweets earlier this week. So it took about 40 minutes of the podcast for me to transform from really excited about all this to really nervous about all this. Once you start throwing out AI generated voice and, you know, deep fake podcasts and stuff, that's when I throw up my hands and say, all right, let's be really careful about what we do here. Because, you know, you can see we're not that far off from it becoming very difficult to distinguish between what's real and fake, whether you're Google bots trawling the internet for links or an AI learning model or a regular human just trying to navigate like the internet. Yeah, I mean, the great thing is AI bots don't get COVID.

So they actually have a consistent production schedule. There you go. Availability is the best ability. That's right. That's right. All right. One other question on chat GPT. How well do AIs like GPT work in other languages, Nick wants to ask?

My guess, he says, is that because English is the default language of the internet, more English content is incorporated into the model, and therefore it does better with English? If AIs are quote unquote better at English, does that mean the improvements in productivity that could come with AI are going to benefit English speakers more than other languages? Will this have implications on the growth of English-speaking countries versus others? What do you think, Ben? I mean, my assumption is that yes, it's better in English. Yes, that will benefit English speakers. And the dominance of sort of the English-speaking Internet, which is by and large the American Internet and big American companies,

is probably going to continue. I mean, I think that, yeah, I think that's probably the case. I mean, or maybe, I guess it's definitely the case of the sheer amount of bullshit coming out of English-beat countries. We'll skyrocket. We're probably the winners in that as well. We've mastered the art of bullshit. All right. So the last question I have for you is you wrote about the Big Five and their integration of AI earlier this week.

One company you mentioned was Apple. What do you expect to see from them? as they incorporate all this into their business going forward. I've been, it still blows me away the Apple stable diffusion release, which wasn't just an optimization for the stable diffusion library to use their sort of APIs, but also, you know, like they had new operating system releases. So that like, like, so, you know, they're, those APIs call things within the operating system, right?

And they made changes under the surface to accommodate these operations. optimizations that they did. And what's really interesting is if they make changes to their chip designs going forward to sort of optimize this sort of thing. And number one, I think it says a lot about Apple in a really positive sense that, yes, this was an obvious thing to do. But there's lots of big companies that do not do obvious things, right? Like they get old, they get slow. They get, you know, oh, what are concerns about this? And there's, you know, been pushed back on stable diffusion.

And the fact they did this and they did it so quickly is. just it says good things I think about Apple's culture generally, number one. Number two, it's really interesting because I don't think it's necessarily like Apple has all this compute power. They're making their own chips, right? And they have all this capability. It's like, what is we actually using this for? Right. He's like, hey, we got a better camera.

And then you know, next year, hey, we got a better camera. Right. And then three years later, hey, we got a better camera. Right. Right. Like the, I welcome the improvements. I use the iPhone camera all the time. but a world that is hungry for more compute is a good world for Apple.

And what's so compelling about if some of this stuff can start running locally, right? So there's the whole lens of thing, right? Where you're your own, I think. They had to upload that to servers that were running on, you know, on GPUs. And they had to pay for that. And it got, they got overrun. And so like when they're at the high of their popularity, like when they start is like, oh, you get it back in 10 minutes.

That is 20 minutes. That it was like two days. Right. Because like, you know, there's just there's a scale building challenge. Suddenly LeBron James wanted his lens of photos. That's right, right? Whereas, like, people, imagine that running locally. So it's running on your phone.

And suddenly that completely fixes this marginal cost problem because you're basically distributing those marginal costs to individual people on their own phone using their own electricity, which is an Apple's problem, right? And they have very efficient processors and all these sorts of things. And is that going to ever be as good as an Nvidia GPU in the cloud? no, like that it's so much larger, he's just so much more power, so much more capabilities. But the massive advantage in not, in basically getting that compute for free from a developer perspective is huge.

And Apple is so is far better placed than anyone to pull this off because because they're fully integrated, they can, like you can imagine a future where just the stable diffusion is like built into your phone. And so a developer can write to it. and they don't need to download this huge one gigabyte file or whatever might, which again is relatively tiny, all things considered.

And they just get this capability for free. And they don't have to pay an ongoing thing. And in this world, Apple's the big beneficiary because their phones become very differentiated because there's all these cool apps that are on Apple's phones that are not, maybe they work on Android, but maybe not as well. There's just probably a window of large differentiation. It's bad for a lot of the other companies. that are sort of like in the middle.

And if you're a small developer, it's like, wow, I can have access to this capability that I don't need to pay for because customers are running on their phones. Again, this is another huge open question is what, how much of difference are going to be between these local models and these sort of like cloud based ones, local compute versus cloud compute. But it's incur it's right by Apple to be pushing in that direction because it's a great bet that that could really pay off for them. Yeah, I mean, it speaks well of, like you said, their leadership just because a lot of

companies with as many advantages as Apple has end up being pretty complacent about the way they approach their business. And it's cool to see them take risks and try to make this work. It's funny, though, as a layperson, my assumption with tech is always that it's only going going to get more and more efficient and it'll be really easy to run these programs that take all this energy and literally require, like you said, just huge amounts of power now. My assumption is that, you know, a couple years down the line, everything just becomes a lot more easier, a lot easier.

And so I wonder, like, is there a limit in terms of what the efficiency actually looks like with this process? I don't know. That's an open question. And your assumption is right. That's the way it's always worked, right? I mean, there's a long-running sort of thing in tech about thin versus thick clients, right? Like, is everything central and then you access it via what's called a thin client or is everything distributed to the edge?

And the reality is the pendulum sort of tends to swing back and forth. And right now, mobile has actually been a fairly centralized area where you're using all these cloud services. And even if you're accessing like Facebook or Twitter on your phone, that's really all running sort of in the cloud. out. And whereas you go back to like the PC era, like you had all, all these applications were like Photoshop and Word all your stuff. We're running on your PC, right? We talk about Figma, for example, right?

Figma was transformational because Photoshop was a thick model where you're running Photoshop locally and then you have to figure out how to sync stuff, right? Whereas Figma's all is a thin model. It's all, it's running on a centralized service. You get all the sync for free. Everything's already all in one place. And getting that to work was really, really powerful, right? and you see this pendulum sort of going back and forth.

And Apple is heavily motivated by their business model and by their capabilities to push AI to a thick model where you're doing lots of stuff sort of locally. Because in a thick model, they sell the device that actually does that. And their devices will probably do it better than anybody else's. And so that, but is that, are these going to be good enough, right? It's just going to be so much better in the cloud that, that, that, in the reality is all this stuff, it's probably going to be, it depends. It's sort of somewhere in the middle. But this is... But it's not a foregone conclusion, which is what my assumption was.

Yeah, and one of the biggest things, this was a, what the biggest thing is about 2022 in stable diffusion in particular. Like, it was just sort of the assumption that, yeah, this stuff is going to be super centralized and sort of need to run, you know, these big data centers. And that's how it's going to be accessed. And that's not the case for images. Now, text is more difficult.

Language models have to be much larger. There's a huge, much larger sort of memory requirement in particular. And so images are in some respect simpler, right? Because we just fill in a lot of the details sort of naturally. And so will that happen with text models? Maybe, maybe it'll take a little longer. Again, there are bottles out there in like research papers that are shockingly good while being shockingly small. So that's going to be a really interesting to sort of watch and observe.

It may be the case that most language models are more centralized and image is more distributed. But the other thing, this is my most, this has been my most emoji. Galaxy brain. No, emoji mind exploding thing. And this came up in the interview I did before break with Nat Freeman and Daniel Gross. Nat was telling me about this app that basically will create music from text. And it did that by analyzing the sonic waveforms from other music, which it basically changed music into an image and then learned. music and then can now sort of generate music, right?

Going backwards, go from waveform, then back into music. This is where I start shifting uncomfortably in my seat, unsure of whether I want to be part of this future. But yes. But I mean, this idea, what's so fascinating about that beyond these fascinating of itself, but this idea that images may be much more powerful than we appreciate. Beyond the image is worth a thousand words or whatever, so a picture's worth a thousand words sort of idea, right? like the reason why everyone assumes the large language model is going to be the most important because everything's text right you look at it around the world everything we do is text but why is everything text like the reality is is text is an accessible communication form for almost everyone right almost everyone can learn how to read and write not almost everyone can learn how to draw a picture right there's just a the barriered entry for image creation has been far far far far far higher than for text creation. And so if you need a technology that is accessible to everyone, reading and writing is much more accessible to everyone.

And downstream from that is that text matters for everything. What happens when images are not just equally accessible, but actually more accessible because they can be run locally, because you actually don't need as much resources to do images, then you do text. What happens when all of our communications doesn't dribble down to letters, but dribbles down to images, right? Like that's where it's really crazy stuff starts to happen.

That's where this becomes a transformational technology that first, where it's not just a big six. There's an explosion in huge new companies that this is where the current Big Five are in trouble because they're predicated on the world as it is, and this is a completely new world. Now, if you look back on transformational, well, the internet or things like that, This takes decades, like 15, 20 years.

So this is what I'm talking about is if there's something here. And it's very, it's not tangible. It's hard to sort of put your finger on. But we're talking, we look back in 2035 or 2040. And it's like, man, remember when we thought text was like a big deal? Like, none of us can even read anymore, right? Like, or just like, so that's, that's, I think the biggest thing that's like, wow. I mean, where this could really be something.

what happens in a world where images actually are more important than the large language models. Okay, so if I'm going to step into the galaxy brain pain of the meme with you and imagine what the world will be, don't we still need text to generate the AI photos? Yes, right now we do. So it's a good point. Great catch by you. I'm a word guy. I'm team words for the future and certainly today.

Yeah, the way these image bottles are built is because it's not just getting images off the web. It's images and then the accompanying text that describes what that image is, right? There's this tag in images like the alt's tag that's text that describes what that image is. And that is what actually makes all these models possible. So yes, text is not actually going to go away. But this idea that we've, humans do understand. and our visual creatures, but we've never had the ability

to generate images on a zero marginal cost basis, and there's been a high barrier to intrigues, you had to be able to draw. And even if you can draw, it takes a long time. What if you can, everyone can draw and they can draw instantly?

Right. Yeah, there will be an abundance going forward, and the question is how quickly we get there. All right, a couple questions at the end, unrelated to AI, just because we missed the podcast, earlier in the week. We could go a little bit long here. Dan says in the last episode, Ben mentioned that a big problem with Windows phone was its lack of a good app ecosystem and that

Microsoft spent $7.2 billion propping up the phones through the Nokia headset acquisition. My question is, how do you have any app problems when you're willing to spend that much money? Just hypothetically, Microsoft could have found the top 100 apps and the top. 100 games in the app store, approached their developers in about 2011 and said to each of them, here's $36 million in guaranteed income if you

port your app to Windows phone and keep it there for five years. Did Microsoft try this? If so, why didn't it work? Do you have any thoughts, Ben? Yeah, my thought is, I don't want to know how old Dan is because I feel like it's going to make me feel very old. Of course Microsoft did this.

Microsoft offered money to all sorts of people. I was on the team that did this. The reality is that you don't, building an app is a big commitment. You have to build it. You have to maintain it. It has to be an ongoing thing. And it doesn't matter yet. Sure, Microsoft got some apps on their store. They have apps in the store today in large part because they paid the developers to get them. But the like it's a from a developer perspective, if you're not getting users and it's not contributing on an ongoing basis to your business, it's a bad choice to take like an upfront payment and then you're stuck maintaining this sort of thing for forever like software as much as we

talk about zero marginal cost like at the end of the day like software breaks all the time is brittle it's an ongoing thing you have to be changing things and it's just not worth it if there aren't users there it like you have to have the demand to pull in the supply when that supply requires sort of that ongoing investment on an ongoing basis and you can't sort of brute force that the the like it may seem like a chicken and an egg problem. Like, do you know, do you need the apps or you need the users? I'm firmly convinced you need the users first. And that will pull the developers in.

It doesn't work the other way. And we know that precisely because Microsoft tried that. And it didn't work. If there was a problem, Microsoft was throwing piles and piles of money at it. Like, it doesn't work. Okay. Well, to keep it moving, Gabe says, I think you and Ben, overlooked a crucial point as to why YouTube would spend so much money acquiring the rights to

Sunday ticket. It's because YouTube can charge advertisers far more money because of their ability to target each individual viewer. The traditional ad model makes an educated guess that I like trucks, beer, gambling, and have erectile dysfunction. Game too much information. Google knows that I'm a balding Mormon with a soon-to-be-six-kid family who needs a minivan and likes to drink sparkling water. It won't happen overnight, but Google's increase in advertising revenue per viewer will be profound. What do you think?

Do you agree, Ben? I think that's reasonable. I think, you know, this has been an ongoing sort of assumption for a long time that these platforms will figure out targeted ads in a meaningful way that will make a difference. There is. So, I don't know, we'll see.

I mean, I think one of the, one of the tensions that's going on with ATT and with sort of the EU sort of stuff about targeting advertising is the less targeted advertising is, the better it is for sort of large entities. Because they're sort of serving a broad-based market, their business models built around that, you know, at the end, you know, whether it be beer or whether it be trucks or whatever. Getting neat, the other thing with getting niche is if you're selling a whole bunch of stuff, like how niche do you want to get? because the cost of the ads goes up and like making something super custom goes up. Like there's there's lots of interesting tradeoffs between being, selling stuff at scale. Like, like, right.

And I wrote an article a while ago, a TV advertising surprising strength that was really about this. That was why is TV advertising saying stronger, longer than people think? And the reality is a lot of products and companies are predicated on TV advertising, not in that they make mass market stuff that depends on scale, right? If you're a CPG company, you know, if you're your P&G selling your Swiffer mop or whatever, like you like the way the whole model works is by selling a lot of them.

And you have deals with retailers. So it's on an end cap. People see it that's right there. And there's this whole sort of ecosystem. And all those large companies have a very hard time transitioning into super niche targeted sort of stuff. Because they're kind of selling mass marketed products. So why do they even want to spend so much to target people? Like TV is actually much more efficient for them in advertising in that way.

And so then it's like, well, there's going to be new CPG companies, right? The Harry's Razors or whatever. And they're going to come along and peel off. And that did happen to an extent. But also what happened is those companies became so dependent on Facebook in particular to find them their customers. That Facebook just took more and more of their margin, right? And it's like, well, you would not find this customer without Facebook.

It's all additive to your bottom line, so you can't really complain. But I'm just like, man, paying a lot for this. And I think that was the, I wrote another article about like email address and razorblade, something like that. That was about this idea that all of these fanciful projections about super niche targets missed the fact that they had a complete and utter dependency on Facebook. And Facebook was going to take their margin because of that. And so same thing could sort of happen with Google here. Again, you need companies that are structured and prepared to take advantage of that niche advertising for it to sort of manifest itself. And we'll see to what extent that happens over time.

I think just in general, I was very bullish on a lot of this niche sort of CPG sort of stuff in general. But at the end of day, like there are real returns to scale. I think the status quo is more powerful than it feels. Team status quo here. Absolutely. Yeah. Team Google in your browser. Yeah, we'll see. I mean, I'm skeptical of Google making money on this deal just because their ad rates are so high. I think it's much more likely they lose money. I think they'll get more subscribers than people think because you don't even get a satellite dish. Like that's a big sort of barrier to entry. But I do think this really only makes sense in the context of them wanting to develop the YouTube ecosystem sort of broadly and this being a lost leader. But I mean, hey, if we come back in a few years and Google's like, yeah, we

lost the money off this. We have to remember that Gabe the baldy Mormon who does not have a erectile dysfunction disorder was right about this. Well, I was really frustrated after we finished our YouTube TV segment because we had a mutual friend point out that Sunday ticket is actually less valuable to football fans today than it's ever been because of how diluted it is by Red Zone and all the different games that have been spun off throughout the week. and that sort of distorted the way I was seeing the deal

because I don't really watch football on a regular basis anymore. Growing up, I wanted Sunday ticket because I was a fan of the whole league and wanted the ability to watch every game. But now I don't know that Sunday ticket would be essential to football fans unless you live out of market and want to follow your favorite team. Like Red Zone is a pretty good replacement that seems to satisfy like millions and millions of football fans.

Yeah, but that is a reason, though, for, I think for the NFL to go with someone like, I think making Sunday direct ticket more accessible is good because you want to keep the people that move out of market. You want to keep them invested, right? And requiring them to get a satellite dish. Like, yeah, the most hardcore folks will do that. But again, there were scenarios we didn't have to get satellite this. But by and large, I think this is good for the NFL. And satellite was also an easier sell to the rest of the family. like 20 years ago, whereas now you can try to convince people to get excited about Direc TV,

like, good luck. The only other note I wanted to hit is that the Thursday night football audience saw a 41% audience drop from the 2021 average since moving broadcast primarily to Amazon Prime. That was reported by the Sports Business Journal last week. That's pretty interesting. I would not have guessed that it would be that big of a dip. and I'm unclear on how much Amazon cares about a dip like that and or what Amazon is looking for from its NFL rights. But I was wondering if you had any reactions off the top of your head.

Yeah, that's definitely much larger than I thought. Because I think the first game was like surprisingly good. It was only down like 10% or something, which is, which is not great. I think the NFL or the positive argument would be that the share. of younger viewers was was either stayed up or was maybe even higher and which makes sense people who sort of cut the cord and then but they have Amazon Prime right they're used to watching things online you definitely heard a lot of anecdotal stuff uh I think John Gruber talked about like his dad like just didn't watch because it was too much of a hassle to sort of figure out like I ought to

do it which makes sense there's like it's easy to sort of watch things I think from the NFL perspective, the benefit of YouTube relative direct TV is it's so much more accessible. It's easier to just sign up on YouTube than to get a satellite dish. This is the opposite case. It's so much easier to flip to a channel than to learn how to use sort of the streaming thing, sort of X, Y, Z, which, you know, is sort of an interesting lesson, an interesting takeaway. The big question for sure is, what?

What will streaming services do with sports? Like to what extent will they invest in them? You know, that's going to be the sports leagues are really invested in tech companies wanting sports rights. So that they can't. Absolutely. Can't flame them. Yeah.

Yeah. And, you know, we talk with the NBA like, you know, and, you know, Time Warner talking about cutting costs, X by Z. Well, if the NBA can say, you know, bring Amazon to the table, that's, that's a compelling thing. The other thing is this, this is another argument for team status. quote, right? Like, people are used to watching stuff on TV and it's, they're not used to streaming it. It's also an argument for streaming services to get really compelling content because if you have to train users how to do it, you need something that's going to inspire them to learn. And that means the

most expensive stuff. Was this mostly just a trial balloon on Amazon's part? Yeah, I think it's a good question. I think from Amazon's perspective, this is sort of a long-term bet. on being a major sort of network of the future where the cable bundle goes away, right? What happens in the world where the cable bundle goes away? There's still a lot of money to be made there. And you make money not just on the NFL and selling, you know, targeted ads like Gabe's sort of idea, but also by then, oh, I want to now subscribe to HBO. We're subscribed to HBO.

I already watch TV on Amazon Prime. I can subscribe to HBO on Amazon Prime. And then I'm paying 30% of my HBO subscription fee to Amazon to Amazon. Amazon for sort of an ongoing basis, right? That's pretty compelling. And so I think this is, from Amazon's perspective, it's about solidifying prime as a place to watch TV so that it could be a marketplace for selling other stuff, not necessarily just their, just their own stuff.

And I think that that's a reasonable bet to make. There's different reasonable bets in it working. Like, that's the nature of a bet. But we'll see. I mean, it's, it's, this is another. I think what's happening with streaming is really, really interesting is going to be another super compelling bit to watch in in 2023. We have the NBA stuff coming up soon. Bob Iger's back at Disney.

There is. Can't wait. There does feel like a bit of a, you know, the empire strikes back aspect to some of this stuff for sure. There you go. Well, we will cover all of it in 2023 as the empire continues to strike back. We went a little bit longer tonight. I want to thank Sudafed. And I look forward to coming back next week with two episodes.

Yeah, no free ads. But it's great to be back. And I can't wait for more. Yes, we were supposed to take a holiday on Monday, but you know, we missed this week. We'll give it to you next week.