E115: The AI Search Wars: Google vs. Microsoft, Nordstream report, State of the Union

1. 0:00 – 1:49

   Bestie intro!

   1. CPChamath Palihapitiya0:00

      I have this thing with my oldest son where I don't know if it's all kids, but he's 13. He doesn't know how to answer the phone. Not to save his life. "Hello? Hello?" He picks up the phone, "Hello?"

   2. DSDavid Sacks0:12

      (laughs)

   3. CPChamath Palihapitiya0:12

      And so, and so I said, "Listen, from now on, when you call me, I expect a certain way that you pick up the phone and that's going to be practice for you how you interact with anybody else. And vice versa as well, if I call you, you have to pick up the phone and if you don't, I'm hanging up right away." So yesterday, he calls me and, and I'm like, "Jamaal speaking," you know, it's like, "Hello, Jamaal speaking."

   4. DSDavid Sacks0:35

      Right.

   5. CPChamath Palihapitiya0:35

      "Dad." (beep) Hang on.

   6. DSDavid Sacks0:40

      (laughs)

   7. CPChamath Palihapitiya0:41

      He calls back, "Hello, Jamaal speaking." "Dad." (beep) Hang up again. We did this three more fucking times and then finally I said, " (beep) when will you get it through your head? Can you please just answer like if I say, 'Hello, Jamaal speaking,' 'Hey Dad, it's your son,' or 'Hey Dad, it's (beep) .' And it's unbelievable and he's like, "Well, none of my friends pick up the phone like this." And I'm like, "Oh my God, like, don't they think they need to have verbal communication skills?"

   8. DSDavid Sacks1:08

      (laughs)

   9. CPChamath Palihapitiya1:09

      It's like, it's unbelievable. And then when I call him, he picks up the phone, "Hello?" What? Not even hello. "Hello? Hello?"

   10. DSDavid Sacks1:17

       (laughs)

   11. CPChamath Palihapitiya1:19

       What is that? It's like a grunt.

   12. DSDavid Sacks1:21

       It's minimal efforts, like the minimum-

   13. CPChamath Palihapitiya1:23

       Oh my God.

   14. DSDavid Sacks1:23

       ... minimum number of syllables.

   15. CPChamath Palihapitiya1:25

       It's, it's embarrassing. Are your kids like this? Or is it just my kid?

   16. DSDavid Sacks1:28

       Actually, I don't think I've heard him pick up the phone. I need to test that.

   17. JCJason Calacanis1:32

       I'm going all in. Let your winners ride. Rainman, David Satterfield. I'm going all in. And I said. We opened source it to our fans and they've just gone crazy with it. Love USH. Queen of Kinloch. I'm going all

2. 1:49 – 28:26

   Report about US involvement in the destruction of Nordstream pipelines, breaking away from the military-industrial complex

   1. JCJason Calacanis1:49

      in.

   2. DSDavid Sacks1:50

      J Cal was on Twitter calling me out for not denouncing the Chinese balloon in strong enough language. Although I don't think he understood the language I was using.

   3. JCJason Calacanis1:57

      Um-

   4. DSDavid Sacks1:58

      Did you understand what the word errant means? What did you think it meant?

   5. JCJason Calacanis2:01

      I, you know, I do understand the word errant.

   6. DSDavid Sacks2:03

      Was that like a teaching moment for you?

   7. JCJason Calacanis2:04

      It tem- Yeah, it generally means straying off course. Yeah, and traveling in search of adventure, at least according to Merriam-Webster.

   8. DSDavid Sacks2:14

      (laughs)

   9. JCJason Calacanis2:14

      So, I just thought maybe you're being a little... But I know you're a dove, so, but I was a little dovish thinking they were just off course. I think... Do you think they were off course or they were doing it deliberately? You buy that it was off course?

   10. CPChamath Palihapitiya2:26

       Genuflect. Go, genuflect. J Cal, go, go, go.

   11. DSDavid Sacks2:29

       Yeah, please. So, I knew J Cal would have to join in this nationwide panic over this balloon.

   12. JCJason Calacanis2:35

       I didn't panic. I have no panic over it. What does the word errant means?

   13. CPChamath Palihapitiya2:39

       (laughs) It had a Minolta camera attached to it. (laughs)

   14. DSDavid Sacks2:43

       J Cal, I got some really bad news for you. There are these things called spy satellites-

   15. JCJason Calacanis2:48

       No, we know about those.

   16. DSDavid Sacks2:48

       ... and they can see everything.

   17. JCJason Calacanis2:50

       Oh, no, it's obvious people have been sending these balloons over here for a while. I just thought you were framing as it was errant, as in it like strayed off course.

   18. DSDavid Sacks2:58

       Well, we don't know.

   19. JCJason Calacanis2:58

       Do you think it was e- I'm sincerely asking, do you think it was errant, it was an accident?

   20. DSDavid Sacks3:01

       Well, here's the thing, it's such a harebrained scheme to send a balloon flying over American territory that the Occam's razor explanation here is that they somehow lost control over it.

   21. JCJason Calacanis3:14

       Okay.

   22. DSDavid Sacks3:15

       And these things are not steerable, so like my guess is that it probably wasn't deliberate just because of how stupid a plan it would be and how like obvious it would be. But it could be, I don't really know.

   23. JCJason Calacanis3:29

       Okay.

   24. DSDavid Sacks3:29

       What I do know is that the whole nation got in a lather and a tizzy and started hyperventilating about this balloon and it just shows how reflexively hawkish the media is. You know, it's like, hey, can we just wrap up this war in Ukraine before we start another war?

   25. JCJason Calacanis3:44

       That's one-

   26. DSDavid Sacks3:44

       With an even more formidable superpower.

   27. JCJason Calacanis3:46

       Yeah, I think that's fair. The media-

   28. CPChamath Palihapitiya3:47

       The balloon, the balloon got more attention than us blowing up Nord Stream.

   29. DSDavid Sacks3:50

       Exactly. (laughs)

   30. JCJason Calacanis3:51

       Okay, well, hold on. We've already jumped into it. No, what I would say about that is, uh, the media insight is the valid one. I was just responding to the errant and I was curious if you actually thought it was an accident. I don't think it's an accident. I don't think there are accidents.
3. 28:26 – 33:13

   Bestie refresh!

   1. JCJason Calacanis28:26

      Welcome to The All In Podcast. With us again, David The Dove Sachs-

   2. DFDavid Friedberg28:29

      (laughs)

   3. JCJason Calacanis28:29

      ... Chamath Palihapitiya, and The Sultan of Science, who is on his podcast. Oh my God, so many podcasts you're doing. The Sultan of Science is in hot demand. What are all these podcasts you're doing, Friedberg? All these science podcasts pulling you in.

   4. DFDavid Friedberg28:42

      I did a podcast with Brian Keating last week who was really kind enough to reach out. He's had some awesome guests. He's a-

   5. JCJason Calacanis28:48

      Hmm.

   6. DFDavid Friedberg28:49

      ... uh, cosmologist at UCSD, professor down there. And we were supposed to record that day and then we canceled I think last minute, right?

   7. JCJason Calacanis28:57

      Yeah, yeah.

   8. DFDavid Friedberg28:58

      So I was only supposed to be on with him for an hour and I'm like, "Oh, well, my next thing just got freed up." So ended up doing like three hours. It was-

   9. JCJason Calacanis29:04

      Well done.

   10. DFDavid Friedberg29:04

       ... I was so... I was, like, exhausted that day so I looked really hungover on the video and probably-

   11. JCJason Calacanis29:09

       It's par for the course.

   12. DFDavid Friedberg29:10

       ... stumbled a lot. Yeah.

   13. JCJason Calacanis29:11

       But no Lex Fridman for you, so Chamath-

   14. DFDavid Friedberg29:13

       (laughs)

   15. JCJason Calacanis29:13

       ... and I have done Fridman, Lex Fridman, but you have not. He's invited you yet, Lex? Has Lex invited you?

   16. DFDavid Friedberg29:18

       He's not? No? No, he's not.

   17. JCJason Calacanis29:19

       You're uninvited.

   18. DSDavid Sacks29:20

       Where's my invitation? Where's my invitation?

   19. DFDavid Friedberg29:21

       It's 'cause there's-

   20. JCJason Calacanis29:21

       I don't know what's going on here. Collectible 4, Lex. What are you doing? All right, let's-

   21. DSDavid Sacks29:25

       No- no Davos and no Lex Fridman.

   22. JCJason Calacanis29:28

       Wait, what's going on? You would never-

   23. DSDavid Sacks29:29

       Am I too anti-establishment? What's going on?

   24. DFDavid Friedberg29:31

       No, I'm too anti-establishment, Sachs. That's- it's characterized me-

   25. JCJason Calacanis29:35

       Nobody- nobody's inviting this quartet to anything. (laughs) Full- full stop. We're not doing All In live from Davos. (laughs) It's not happening, folks. Sorry.

   26. DSDavid Sacks29:45

       (bleep) that.

   27. JCJason Calacanis29:45

       They don't want that heat.

   28. DFDavid Friedberg29:46

       (laughs)

   29. JCJason Calacanis29:46

       I agree. (bleep) that.

   30. DSDavid Sacks29:47

       I'm with Groucho Marx, I don't wanna be part of any club that would have me as a member.
4. 33:13 – 45:06

   The AI Search Wars: Microsoft pressing hard, Google has a rough week

   1. JCJason Calacanis33:13

      Search wars. Microsoft versus Google. Okay, it's been a rough couple days for Google. We've all seen it. Google and Microsoft both did live demos of their new generative AI, yada, yada, yada. You guys all know about ChatGPT. But now, Bing is integrating it into their search engine, getting there before Google. And Microsoft CEO Satya Nadella, he is going ham. He looks great. He's fit. He's wearing a tight T-shirt, and he is saying he's gonna make Google dance. Dance, Google, dance. He is getting up in their business and uh, listen, he is in a, a distant second place, so it makes sense. On the other hand, uh, Google's AI demo was, frankly, a bit of a disaster. Poorly received. Stock, they dropped 12% since this event. And their presentation did not include the chat bot Bard, because in search, because it wasn't working. It seems like there was an error in it when they said, "What new discoveries from the James Webb Space Telescope can I tell my nine-year-old?" And Bard answered that it took the first pictures of a planet outside our solar system, which is false, which, of course, we all know about ChatGPT. It's only right half the time, uh, and it's a little woke, uh, on the margins. So, um, anyway, there was a screenshot circulating today, which is probably false, but (laughs) it says the following: "Me and a bunch of coworkers were just laid off from Google for our AI demo going wrong. It was a team of 168 people who prepared the slides for the demo. All of us are out of jobs." I can't imagine that's real. Uh, but if it was, that would be a hardcore moment for Google, to fire a bunch of people for screwing it up. Listen, you worked in the belly of the beast, Freeberg. What are your thoughts on Bing poking the tiger and telling Google, "Dance." You know, "Sundar, dance."

   2. DFDavid Friedberg35:00

      You know, what's interesting is Google's had like an incredible AI competency, uh, particularly since they bought DeepMind, and it's been predominantly oriented towards kind of, you know, internal problems. You know, their, they demonstrated last year that their AI improved data center energy efficiency by 40%. They've used it for ad optimization, ad copy optimization, uh, the YouTube follow, uh, video algorithm, so what video is suggested to you as your next video to watch, which massively increased YouTube hours watched, uh, per user, which massively increased YouTube revenue. You know, what's the right time and place to insert videos in YouTube or insert ads in vid- YouTube videos? So, you know, autofill in Gmail and Docs. So, so much of this competency has been oriented specifically to avoid this primary disruption in search. Obviously, now, things have come to a bit of a point, because, you know, this alternative for search has been revealed in ChatGPT. And you guys can kind of think about search and, you know, we've used this term in the past. Larry and Sergey, the textbook that they read, you know, one of the original textbooks that's used in, um, internet search engine technology is called information retrieval. Information retrieval. So, information retrieval is this idea that you, you know, how do you pull data from a static, uh, data set? Um, and it involves scanning that data set or crawling it, and then creating an index against it, and then a ranking model for how do you pull stuff out of the index to present the results from the data that's available, based on what it is you're querying for. You know, and doing that all in a 10th of a second. So, you know, if you think about the information retrieval problem, you type in the data or some rough estimation of the data you want to pull up, and then a list is presented to you.And over time, Google realized, hey, we could show that data in smarter, quicker ways. Like if we can identify that you're looking for a very specific answer, we can reveal that answer in the one box, which is the thing that sits above the search results. Like if you said, "What time is it?" You know, "What- when does this movie show at this theater?" So they can pull out the structured data and give you a very specific answer rather than a list from the- the- the database. And then over time, there were other kind of modalities for displaying data that it turns out were even better than the list, like maps or shopping, where you can kind of see a matrix of results, or YouTube where you can see a, you know, longer form version of content. And so these different kind of, um, you know, information retrieval, you know, media were presented to you and it really kind of changed the game and created much better user satisfaction in terms of getting what they were looking for. Th- th- the challenge with- with this new modality is it's not really fully encompassing. So if you can kind of think about the human computer interaction problem, you want to see flight times and airlines and the price of flights in a matrix. You don't necessarily want a- a- a text stream written to you, uh, to give you the, um, you know, the- the answer that you're looking for, or you want to see a visual display of, uh, shopping results, or- or you do want to see a bunch of different people's commentary 'cause you're looking for different points of view on a topic rather than just get an answer. But there are certainly a bunch of answer solutions for which ChatGPT, uh, type, you know, uh, natural language responsiveness becomes a fantastic and better mode to present answers to you than the matrix or the list or the ranking and so on. Now, the one thing that I think is worth noting, I did a back of the envelope analysis on the cost of doing this compared to ChatGPT. So- so Google makes about three bucks per click. You can back into what their revenue per search is a bunch of different ways. One way is three bucks per click, about a 3% click-through rate on ads. Some people estimate this is about right, about five cents to 10 cents revenue per search done on Google, or anywhere from one cent to 10 cent, let's say.

   3. JCJason Calacanis38:54

      Even if they don't click the ads because one out of 100 people click an ad and that's where-

   4. DFDavid Friedberg38:59

      Yeah.

   5. JCJason Calacanis38:59

      ... the money comes from.

   6. DFDavid Friedberg38:59

      So let's- let's just call it five cents, right? And you can assume a roughly 50% margin, uh, on that search, which means a 50% COGS or cost of goods or a cost to run that search and present those ads. So, you know, right now Google Search costs them about, you know, call it two and a half cents per search, uh, to present the results. A recent estimate on running the GPT-3 model for ChatGPT is that each, um, result takes about 30 cents of compute. So it's about an order of magnitude higher cost to run that search result than it is to- to do it through a traditional search query.

   7. JCJason Calacanis39:39

      Today.

   8. DFDavid Friedberg39:40

      Which makes-

   9. JCJason Calacanis39:40

      Today.

   10. DFDavid Friedberg39:40

       Today. That's right. And so- so that's the point. Like it has to come down by about an order of magnitude. Now, this is a ve- this then becomes a very deep technical discussion that I'm certainly not the expert, but there are a lot of great experts that... And there's great blogs and Substacks on this, on what's it going to take to get there to get a 10X reduction in cost on- on running these models. And there's, uh, a lot related to kind of optimization on how you run them on a compute platform, the- the type of compute hardware that's being used all the way down to the chips that are being used. So there's still quite a lot of work to go before this becomes truly economically competitive with Google. And that really matters because if you get to the scale of Google, you're talking about spending eight to $20 billion a quarter just to run search results and display them. And so for ChatGPT type solutions on Bing or elsewhere to scale and to use that as the modality, you're talking about something that today would cost $80 billion a quarter, uh, to run from a compute perspective, if you were to do this across all search queries. So it's certainly going to be a total game changer for a subset of search queries, but to make it economically, uh, work for- for these businesses, whether it's Bing or Google or others, there's a lot of work, uh, still to be done.

   11. JCJason Calacanis40:54

       The great part about this, Chamath, is that Bing gave 10 billion to, uh, our friends Sam and ChatGPT to invest in Azure, uh, which now has the infrastructure and will be providing the ChatGPT infrastructure to startups or corporations, big companies and small alike. So that $10 billion should do enough to grind it down between software optimization, data optimization, chip optimization, and cloud optimization. Yes, you would think so or no?

   12. CPChamath Palihapitiya41:23

       The ability to run this at scale is gonna happen because we're getting better and better at creating silicon that specializes in doing things in a massively parallelized way. And the cost of energy at the same time is getting cheaper and cheaper along with it. When you multiply these two things together, the effect of it is that you'll be able to run these models, the same output today will cost one, one-tenth as long as you ride the energy and compute curve for the next few years. So that's just gonna naturally happen. I have two interesting takeaways and one is maybe a little bit of a sidebar. So the- the sidebar is, if you guys were sitting on top of something that you thought was as foundational as Google Search back in 1999, would you have sold 49% of it for $10 billion?

   13. JCJason Calacanis42:11

       Hard no.

   14. CPChamath Palihapitiya42:12

       I think the answer is no. I think the answer is no.

   15. JCJason Calacanis42:15

       Not in an environment where you have unlimited ability to raise capital.

   16. CPChamath Palihapitiya42:19

       This is something that we've said before, which is that ChatGPT is an incredibly important innovation. But it's an element of a platform who will get quickly commoditized because everybody will compete over time. And so I think what Microsoft is doing is the natural thing for somebody on the outside looking in at an entity that has 93% share of a very valuable category, which is, how can I scorch the earth? And so Microsoft effectively for 10 billion bought almost 50% of a tool...... and now we'll make that tool as pervasive as possible so that consumer expectations are such that Google is forced to decay the quality of their business model in order to compete. So that, as Friedberg said, you have to invest in all kinds of compute resources that today are still somewhat expensive. And that will flow into the P&L, and what you will see is that the business quality degrades. And this is why when Google did the demo of Bard, the first thing that happened was the stock went off 500 basis points. We, they lopped off $100 billion of the market cap, mostly in reaction to, "Oh my God, this is not good for the long-term business." So the real question-

   17. JCJason Calacanis43:28

       Explain why it's not good for the long term business on a mechanical basis. When you get an answer, you don't have to click the links?

   18. CPChamath Palihapitiya43:32

       No. Right now, if you look at Google's business, they have the best business model ever invented on Earth, ever, for a for-profit company. It just rains money. This is a business that this year will do almost $100 billion of free cash flow. It's a business that has to find ways, and we kind of joke, but they have to find ways to spend money. Otherwise, they'd be showing probably 50 or 60% EBITDA margins, and people would wonder, "Hey, wait a minute. You can't let something like this go unattended." So they try to do a lot more things to make that core treasure look not as incredible as it is. They have 120 billion of cash. This is a business that's just an absolute juggernaut and-

   19. JCJason Calacanis44:18

       They have 10 times as many employees as they need to run the core business. That, that says enough, right?

   20. CPChamath Palihapitiya44:22

       Well, I don't, I don't know what that is, but my point is that it's an incredible business. So that business will get worse if Microsoft takes a few 100 basis points of share, if Meta takes a few 100 basis points of share, if Tencent does, if a few startups do. Quora, by the way, launched something called Po, which I was experimenting-

   21. JCJason Calacanis44:39

       Yep.

   22. CPChamath Palihapitiya44:39

       ... and playing around with last weekend. If you add it all up, what Satya said is true, which is even if all we do collectively as an industry is take 500 or 600 basis points of share away from Google, it doesn't create that much incremental cost for us, but it does create enormous headwinds and pressure for Google with respect to how they are valued and how they will have to get revalued. And that's what happened. So the last thing

5. 45:06 – 1:12:41

   Sundar's next move: How should Google counterpunch? Google's troubled business model, will we see successful lawsuits over training data?

   1. CPChamath Palihapitiya45:06

      I'll say is, the question that I've been thinking about is, what does Sundar do? Right?

   2. JCJason Calacanis45:10

      Hmm.

   3. CPChamath Palihapitiya45:10

      So what's the countermeasure?

   4. JCJason Calacanis45:12

      Yes. This is what I was gonna get to, yeah.

   5. CPChamath Palihapitiya45:14

      I think the countermeasure here, if I was him, is to go to the board and say, "Guys, we're going to double TAC," right? So TAC is the traffic acquisition cost that Google pays their publishers. It is effectively their way of guaranteeing an exclusivity on search traffic. So for example, if you guys have an iPhone, it's Google Search, that's the default search in the iPhone. Google pays Apple. This year, this renegotiation for that deal could mean that Apple gets paid $25 billion for giving away that right to Google. So if these, Google does all these kinds of deals. Last year they spent, I think, 45 billion or so. So about 21%.

   6. JCJason Calacanis45:53

      In a way, when you think about that, Chamath, Google basically paid Apple, which was working on sec- search technology, they were working on a search solution.

   7. CPChamath Palihapitiya46:00

      Yeah.

   8. JCJason Calacanis46:00

      They paid them to stay out of the business.

   9. CPChamath Palihapitiya46:01

      And they're, and they're paying everybody. So I think the question for Google is the following. If you think you're going to lose share, and let's say you go to 75% share, would you rather go there and actually still maintain your core stranglehold on search? Or do you actually want 75% share where now all of these other competitors have been ceded? Well, you can decay business model quality and still remain exclusive if you just double the TAC. And what you do is you put all these other guys on their heels, because as we talked about, if you're paying publishers two times more than what anybody else is paying them, you'll be able to get publishers to say, "Hey, you know what? Don't let those AI agents crawl your website because I'm paying you all this money. Remember that," right? So do not crawl in robots.txt equivalent for these AI agents. And I think that that'll put Microsoft and all these other folks on their heels. And then as, as you have to figure out all this derivative work stuff, all these lawsuits, Google will look pristine because they can say, "I'm paying these guys double because I acknowledge-"

   10. JCJason Calacanis47:04

       Absolutely.

   11. CPChamath Palihapitiya47:05

       "... that this is a core part of the service." So that's the game theory I think that has to get figured out, but if I was Sundar-

   12. JCJason Calacanis47:09

       I love the second part.

   13. CPChamath Palihapitiya47:10

       ... I'd double the TAC.

   14. JCJason Calacanis47:11

       I love the second part.

   15. CPChamath Palihapitiya47:12

       I think, I th-

   16. JCJason Calacanis47:12

       Because, hold on, let me get Sax about this.

   17. CPChamath Palihapitiya47:14

       Yeah.

   18. JCJason Calacanis47:14

       I love the second part, Chamath, because in this clip I'm about to show, Nilay Patel from The Verge did an awesome interview with Satya, and he basically would not answer this question, at least to my satisfaction, which is, "Hey, what do the publishers get out of this? You've ingested our information. How do we get paid?" Watch this clip. It's very telling.

   19. GUGuest47:36

       ... in the answer or even in the chat session.

   20. But if I ask the new Bing, "What are the 10 best gaming TVs?" And it just makes me a list, why should I, the user, then click on the, the link to The Verge, which has another list of the 10 best gaming TVs?

   21. Well, I mean, that's a great question. But even there, you will sort of say, "Hey, where did these things come from?"

   22. Mm-hmm.

   23. And "Would you want to go dig in?" Like, that-

   24. Yeah.

   25. ... even search today has that. Like, we have answers. They may not be as high quality answers, they just are getting better. So I don't think of this as a complete departure, uh, from what is expected of a search engine today, which is supposed to e- really respond to your query while giving them the links that they can then click on, like ads. Uh, and search works that way.

   26. JCJason Calacanis48:18

       In my mind, there's a terrible answer. He needs to address how they get paid. He punted the answer-

   27. DSDavid Sacks48:22

       Yeah.

   28. GUGuest48:22

       Mm-hmm.

   29. JCJason Calacanis48:22

       ... and just said, "Hey, listen, search works this way." Sax, will the rights to the data, will s- will Google just say to Quora, "Hey, we'll give you a billion dollars a year for this dataset if you don't-"

   30. CPChamath Palihapitiya48:32

       They should.

Episode duration: 1:49:44