What DeepSeek Means For The Future Of AI | Tech Veterans Weigh In

1. 0:00 – 4:06

   DeepSeek's release

   1. SSSteven Sinofsky0:00

      The lesson is not Sputnik, the lesson is the internet

   2. MCMartin Casado0:02

      This is another step to basically AGI in your pocket.

   3. SSSteven Sinofsky0:05

      There are always pockets of people innovating. Worldcom and ATT did not predict the internet was gonna come out of universities

   4. MCMartin Casado0:11

      It really is the AI race, just like we went through the space race, and we need to win.

   5. SSSteven Sinofsky0:16

      There's no way this doesn't play out like the internet [upbeat music]

   6. SSSteph Smith0:21

      It's been a busy few weeks.

   7. MCMartin Casado0:23

      Yeah.

   8. SSSteph Smith0:23

      I don't know about you guys, my Twitter feed, all podcasts, everything, DeepSeek everywhere, maybe unsurprisingly, but what's your TLDR like? Let's just start there in terms of what came out and maybe also your take on why it blew up in the way it did, because we've seen lots of releases-

   9. SSSteven Sinofsky0:40

      Yeah

   10. SSSteph Smith0:40

       ... in the last, let's say, two years since ChatGPT.

   11. SSSteven Sinofsky0:42

       Yeah, the, the quick overview, of course, is, like out of, uh, essentially nowhere, a small hedge fund, quasi computer science research organization in China releases a whole model. Now, those in the know, know it didn't just appear. There's been a year and a half or so-

   12. SSSteph Smith1:01

       Yeah

   13. SSSteven Sinofsky1:01

       ... of build-up and-

   14. MCMartin Casado1:02

       And they're really good

   15. SSSteven Sinofsky1:03

       ... and they're really good.

   16. MCMartin Casado1:04

       They're top AI researchers, the whole thing, yeah.

   17. SSSteven Sinofsky1:05

       Nothing, nothing was in agg- But it, it appeared to take the whole rest of the world by surprise.

   18. MCMartin Casado1:12

       Yeah.

   19. SSSteven Sinofsky1:13

       And the, I think there were two big things about it that really caught everybody's attention. One was just, like, how did they go from nothing to this thing?

   20. MCMartin Casado1:23

       Yeah.

   21. SSSteven Sinofsky1:24

       And it seems to be, you know, a constant factor of compatibility and capabilities with everybody else.

   22. MCMartin Casado1:32

       Yeah.

   23. SSSteven Sinofsky1:32

       And this number got thrown around that it only cost $5 million.

   24. MCMartin Casado1:37

       Yeah, $6 million.

   25. SSSteven Sinofsky1:37

       The number, $6 million. The number is irrelevant.

   26. MCMartin Casado1:40

       Yeah, yeah.

   27. SSSteven Sinofsky1:40

       But-

   28. MCMartin Casado1:41

       Yeah

   29. SSSteven Sinofsky1:41

       ... 'cause it turns out they wrote a paper and they said, "Hey, we innovated in this particular set of things on training."

   30. MCMartin Casado1:48

       Right.
2. 4:06 – 9:19

   Teasing signal from noise

   1. SSSteph Smith4:06

      what's the signal from the noise?

   2. MCMartin Casado4:08

      So maybe I'll give you the lens of, you know, Chinese people are smart. So, so there's one lens, the lens that I hold, which is China has great researchers. DeepSeek has actually released a number of SOTA models, including V3, which is actually probably a more impressive feat. It's, it's almost like a ChatGPT-4. And oh, by the way, to create, uh, one of these, you know, chain of thought models, these reasoning models, you need to have a model like that, which they had done and we had known about. Um, all of the contributions that they've done have been in the public literature somewhere, just nobody had really aggregated. So there's a, there's a thought that I hold, which is this is a very smart team that has been executing very, very well for a long time in AI. They are some of the top researchers. The fact that they spent $6 million just doing the chain of thought is actually not out of whack with what Anthropic has now-

   3. SSSteph Smith4:56

      Mm-hmm

   4. MCMartin Casado4:56

      ... said they've spent, and OpenAI has said that they spent. And so this is a meaningful contribution from a good team in China. And so it means something, and we should respond to it. So some of what the, uh, outcry is warranted. I do think that we respond to it, but I don't think for the reasons a lot of people are saying.

   5. SSSteven Sinofsky5:11

      I completely agree with that, and in fact, you also saw the people outside of that team in China sort of piling on to try to make it more intergalactic than it was. I mean, my favorite old friend of mine, Kai-Fu Lee-

   6. MCMartin Casado5:22

      Yeah

   7. SSSteven Sinofsky5:22

      ... like, comes out on, on X and says something about, "This is why I said two years ago Chinese engineers are better than American engineers." But to your point about that people were viewing this, that they was reaching some asymptotic level of, of progress-

   8. MCMartin Casado5:37

      Yeah, like the previous base models, you know, like the, the GPT lineage seemed to have asymptoted around GPT-4.

   9. SSSteven Sinofsky5:45

      Right, and but I think what's super interesting about that is that asymptote was true if you looked at it through the lens of, of the function that everybody was optimizing for-

   10. MCMartin Casado5:54

       Totally. [laughs]

   11. SSSteven Sinofsky5:54

       ... which is this crazy-

   12. MCMartin Casado5:56

       Yeah, yeah, yeah

   13. SSSteven Sinofsky5:56

       ... the, it, which is the, to my view, this kind of crazy hyperscaler view of the world, which is we need more compute and more data, more compute, more data, and we're just on that loop.

   14. MCMartin Casado6:06

       Yeah, yeah.

   15. SSSteven Sinofsky6:06

       And a lot of people from the outside were like, "Well, you are gonna run out of data." And I would, just as, you know, a microcomputer person, was like, well, at some point you're gonna end up breaking the problem up to the, the seven billion endpoints of the world, which will have vastly more compute than you can ever squeeze into one giant nuclear-powered data center.

   16. MCMartin Casado6:25

       Yeah.

   17. SSSteven Sinofsky6:25

       And so a lot of what they did was sort of- Like a step function change, not necessarily improvement, just a change in the trajectory.

   18. MCMartin Casado6:34

       Yes. Yeah.

   19. SSSteven Sinofsky6:34

       And, and that to me is the part where, where the hyperscalers needed to take a deep breath and say, "Okay, why did we get to where we were?" Well, because you were Google and Meta and OpenAI, funded by Microsoft, which all had like billions and billions of dollars.

   20. MCMartin Casado6:50

       Yeah.

   21. SSSteven Sinofsky6:51

       So you obviously saw the problem through the lens of capital and data, and of course, you had English language data, which, you know, there's more of than anybody else, so you could keep going.

   22. MCMartin Casado7:00

       Yeah.

   23. SSSteven Sinofsky7:01

       And so I, you know, the way I, I thought of it is we used to... When Microsoft was small, we used to just decide to s-scale things as like, is it a small problem, a medium problem, or a large problem? And I remember at one point we started joking that we lost the ability to understand small and medium problems and solutions.

   24. MCMartin Casado7:19

       Right, right, right.

   25. SSSteven Sinofsky7:19

       And we only had like large, which was just trivial-

   26. MCMartin Casado7:22

       Yeah

   27. SSSteven Sinofsky7:22

       ... and then huge and like ginormous.

   28. MCMartin Casado7:25

       [laughs] Yeah, yeah, yeah, yeah.

   29. SSSteven Sinofsky7:26

       And our default was ginormous.

   30. MCMartin Casado7:27

       Right, right.
3. 9:19 – 11:22

   OS license and reasoning traces

   1. MCMartin Casado9:19

      there are still s- things that are very significant about it that are worth talking about. For example, the license is very significant. The fact that they decided to release the reasoning steps is very significant, and so maybe it's worth talking through those two.

   2. SSSteph Smith9:32

      Yeah, please, 'cause I do feel like those are two things that you're not seeing headlines about, right? You're seeing headlines about all the other things that we just talked about. You said the reasoning traces, those were released, which using the comparable o1-

   3. MCMartin Casado9:43

      Yeah

   4. SSSteph Smith9:43

      ... were previously not.

   5. MCMartin Casado9:44

      Right.

   6. SSSteph Smith9:44

      And then the open source license. Let's talk about those.

   7. MCMartin Casado9:46

      Yeah, so there's, there's, there's two things that are pretty remarkable about DeepSeek-R1 that have implications on adoption. We haven't seen a license this permissive recently for a SOTA model. It's basically MIT license, which is like one page, you can kind of do anything, right? It's what-

   8. SSSteven Sinofsky10:02

      It's like free as in free beer, for real. [laughs]

   9. MCMartin Casado10:04

      Yeah, for real, for real. And like, so I, I mean, listen, I think at a16z we have one of the largest portfolio of AI companies, both at the model layer and at the app layer, and I will say any company at the app layer is using many models. Like it's not just like one model. Like I, I have yet to see the GPT wrapper. They're all using a lot of models.

   10. SSSteph Smith10:21

       Mm-hmm.

   11. MCMartin Casado10:21

       They do use open source models, and licenses really matter. And so this definitely is gonna result in a lot of proliferation. The second thing is, so a reasoning model actually thinks through the steps of the problem, and it uses those, that, that chain of reasoning or chain of thought to come up with deeper s- answers. You know, when OpenAI released o1, they did not release that chain of thought. Now we don't know why they didn't do it, but it just turns out that that chain of thought, if you have access to that, it allows you to train smaller models-

   12. SSSteph Smith10:49

       Mm-hmm

   13. MCMartin Casado10:50

       ... very quickly and very cheaply, and that's called distilling. So like the, the, the general term of distilling in LLM world means that you have a teacher model train a, a student model who's much smaller. And so it turns out that you can get very, very high quality smaller models by distilling these public models, and the implications are both that this is just more useful for somebody using R1, but also you get a lot more models that can run on a lot smaller devices, so you just get more proliferation that way. So it's actually a very big step when it comes to the proliferation of, of this model.

   14. SSSteven Sinofsky11:21

       Absolutely.

4. 11:22 – 18:42

   Monetizing layers of the stack

   1. SSSteven Sinofsky11:22

      There's this tendency to either be like, to peg yourself at, "Oh, it should just be open," but without really defining it, which I think is important in this case. And I think because of where they came from and that they don't have a business model, that was part of what was unique about this, was it was a hedge fund, like almost like this side project, but not really a side project, just a whole thing. You know, it, it has this effect that like, well, we're just gonna give the whole thing away. And, and the, the rest of the companies are still trying to figure out their, their revenue models, which I would argue is probably premature, and it starts to look a little, to me, like, "Hey, let's charge for a web server."

   2. MCMartin Casado12:00

      Yeah.

   3. SSSteven Sinofsky12:01

      And it's like the, the business of selling, like serving HTTP, not a great business. And, and I think like everybody just got focused on the first breakthrough, which was the LLM, which in a lot of ways, if you look back at the internet, that's exact- what exactly happened was everybody got very focused on monetizing the first part of the internet, which was HTML and HTTP. And, and then along came, I don't know, Microsoft and a bunch of other companies to say, "That's not the best layer to monetize it. In fact, there might not be any money in that layer." And the real money is gonna be in shopping and in plane tickets and in television and a bunch of other stuff. And, you know, even other companies, AT&T got wound up trying to monetize even lower layer.

   4. MCMartin Casado12:44

      Yeah.

   5. SSSteven Sinofsky12:44

      And but, you know, that's not how you're gonna get to seven billion endpoints. The licensing model really matters because what's gonna happen is that there's gonna end up being some level of standardization. Now, I don't know where in the stack or in what level, but there is gonna be some level of standardization.

   6. MCMartin Casado13:01

      Yeah.

   7. SSSteven Sinofsky13:02

      And the model, the licensing model for the different layers is gonna start to matter a lot.

   8. MCMartin Casado13:08

      Yeah.

   9. SSSteven Sinofsky13:08

      And, you know, anyone who was around during the, the internet remembers the battles over the different GNU three ver- V3, V4, the, you know, open this license.

   10. MCMartin Casado13:18

       Oh, sure. I remember very well. [laughs]

   11. SSSteven Sinofsky13:19

       Right, well, you doing a dissertation-

   12. MCMartin Casado13:21

       Yeah

   13. SSSteven Sinofsky13:21

       ... you know, you were gonna have to release it, and like it turns out, even your dissertation, which part of it-

   14. MCMartin Casado13:26

       Yes

   15. SSSteven Sinofsky13:26

       ... and how you released it was a huge-

   16. MCMartin Casado13:27

       Yes

   17. SSSteven Sinofsky13:28

       ... issue because it, it m- could make or break a whole approach. Uh, and I think that, that the US industry lost sight of that importance because they got so used to this model of like open just means we're a business and we pick and choose, what we throw out there as sort of like-

   18. MCMartin Casado13:44

       Yeah

   19. SSSteven Sinofsky13:44

       ... evidence that we're an open company.

   20. MCMartin Casado13:46

       Yeah, yeah, yeah, totally.

   21. SSSteven Sinofsky13:47

       And, and I think that that view isn't, isn't aligned with how technology has just shown to evolve in an era where there's no cost for distribution.

   22. MCMartin Casado13:58

       Totally.

   23. SSSteven Sinofsky13:58

       Before there, when there was a cost for distribution, it turns out all those things were sort of the, the u- the free model was irrelevant because you still couldn't figure out how to get it to anybody.

   24. MCMartin Casado14:06

       Yeah, to get people to use it, yeah, totally.

   25. SSSteven Sinofsky14:07

       And-

   26. MCMartin Casado14:07

       So I, I do wanna take the other side of this because-

   27. SSSteven Sinofsky14:09

       Yeah

   28. MCMartin Casado14:10

       ... I actually tend to agree with you. And so like what you just said is, is A, it could be the case that the model's the wrong place to focus and everybody thinks there's a lot of value in there, and so they're playing all these cute games with openness, you know-

   29. SSSteven Sinofsky14:21

       Right

   30. MCMartin Casado14:21

       ... as opposed to distribution. And that could very well be true, but there's another, there's another view which is actually the models really are pretty valuable. Um, and in particular, like the model itself isn't an app, but it could be the case that if you're building an app, you need to vertically integrate into the model. It could be the case, and therefore like if I'm building like, you know, the next version of ChatGPT or, you know, we just had today, you know, Deep Research launch-
5. 18:42 – 22:09

   What’s different this time around?

   1. SSSteph Smith18:42

      this. You said something really interesting, which is this has-

   2. SSSteven Sinofsky18:44

      I'm glad, I'm glad I did. [laughs]

   3. SSSteph Smith18:45

      [laughs] Which is this has to pan out like the internet, and you guys have used examples of different companies, the waves, the mobile wave, cloud era. All of those are things we can learn from, but I just wanna probe you, is there something different here?

   4. MCMartin Casado18:58

      And to bring it back to DeepSeek, I mean, that means, like, this is very important to realize the, you know, the capabilities of China. It's a very credible player. But I don't think that r1 itself as a standalone is gonna have that deep of an impact. But on the internet, so there, there's actually these parallels, um, when it comes to actually capital buildout that you see in the AI, which is it takes a lot of investment.

   5. SSSteph Smith19:19

      Mm-hmm.

   6. MCMartin Casado19:20

      And there's a, there's a special parallel that Mark Andreessen actually reminded me of it too, which people don't tend to see as well, which is in the early days of the internet, like the mid to late '90s, a lot of, um, investors, a lot of big money, think like banks or sovereigns, they wanted exposure to the internet, but they had no idea how to invest in software companies.

   7. SSSteph Smith19:38

      Mm-hmm.

   8. MCMartin Casado19:38

      Like, what are these new software companies? Who are these people? Like, they're all private companies. Like, you know, if you remember this time. So what did all of them do? They all invested in fiber infrastructure. [laughs]

   9. SSSteven Sinofsky19:47

      Oh, yeah.

   10. MCMartin Casado19:47

       Like the whole di- So we're starting to see this thing again, right? We see a lot of, you know, banks and big investors like, "Listen, we wanna build out data centers," 'cause they don't know how to invest in startups like we know how to invest in startups, right? So on one hand, you could be like, "Oh, we're gonna see all of this kind of capital expenditure, and all this capital expenditure's gonna go into, um, physical infrastructure, and therefore we're gonna have another, you know, fiber glut equivalent, but a data center glut." So the counter to that point where I think is different is at the time of the fiber buildout, you kind of had one company which happened to be cooking its numbers, where it had a ton of debt to build all of this out, and then when the price of fiber dropped, that company went out of business, and, like, that caused a huge issue. You have a much better foundation, much, much better foundation for the AI wave, right? Like the primary investors are the big three cloud companies that've got hundreds of billions of dollars on the balance sheet.

   11. SSSteph Smith20:39

       Hmm.

   12. MCMartin Casado20:39

       Even if all of this goes away, they'll be fine. Nvidia can take a price dip, and Nvidia will be fine. So I don't think we're heading to, like, the same type of glut and crash that other people have, which is very, uh, you know, appealing to draw parallels to the internet for that I don't think is there.

   13. SSSteven Sinofsky20:54

       Oh, I am completely with you on that. That part of it is gonna look like the amount that Google invested in the ear- in, in the early 2000s-

   14. MCMartin Casado21:02

       Yeah

   15. SSSteven Sinofsky21:02

       ... or the amount that Facebook e- invested five years later, or the... You know, people forget that Microsoft poured, I don't know, 30, $40 billion into Bing.

   16. MCMartin Casado21:10

       Yeah.

   17. SSSteven Sinofsky21:10

       And it's still number three or whatever, but it doesn't, it still doesn't matter.

   18. MCMartin Casado21:13

       Yeah. I, I would bet, I don't know this as a fact, I'll bet Meta's spending more money on VR than it is on AI right now.

   19. SSSteven Sinofsky21:19

       Yeah.

   20. MCMartin Casado21:19

       [laughs] You know, just to show you.

   21. SSSteven Sinofsky21:21

       And, and, and, and, and so-

   22. MCMartin Casado21:22

       And maybe Apple too, right?

   23. SSSteven Sinofsky21:23

       Oh, oh, Apple. Well, but also 'cause Apple, whatever is bigger than gargantuan-

   24. MCMartin Casado21:27

       [laughs]

   25. SSSteven Sinofsky21:27

       ... is how much they're spending.

   26. MCMartin Casado21:29

       [laughs]

   27. SSSteven Sinofsky21:29

       And, and so it really isn't about the investing profile or who is gonna win or lose on financials, and I think that is-

   28. MCMartin Casado21:38

       Yeah

   29. SSSteven Sinofsky21:38

       ... a super important point that you made.

   30. MCMartin Casado21:40

       Yeah.
6. 22:09 – 29:24

   Scaling up vs scaling out

   1. SSSteph Smith22:09

      these companies have had so much cash on their balance sheet. They've been waiting for a moment-

   2. MCMartin Casado22:12

      Yeah

   3. SSSteph Smith22:12

      ... to invest in the next generation.

   4. SSSteven Sinofsky22:14

      Which also contributes to their, their willingness to scale up-

   5. SSSteph Smith22:18

      Yes

   6. SSSteven Sinofsky22:18

      ... as much as they did.

   7. SSSteph Smith22:19

      So let's talk about that. In your article, you talk about the difference between scale up and scale out, and the natural tendency in these early parts of the wave to scale up, when really there tends to be a shift towards software basically going to zero cost. So Steven, what do you mean by that, and are we at that kind of change in trajectory?

   8. SSSteven Sinofsky22:37

      Sure. Well, that, now we'll just switch to make sure we're, we're really talking about the technology now.

   9. MCMartin Casado22:40

      Yep.

   10. SSSteven Sinofsky22:40

       Not, like, the finances. But when you're big, you, you wanna double down on being big, and so you start building bigger and bigger and bigger computers that, that don't distribute the computation elsewhere. So if you're IBM, you just say the next mainframe is another mainframe that's even bigger. If you're Sun Microsystems, you just keep building bigger and bigger workstations. Then if you're digital equipment, bigger and bigger mini computers. And then one day, and by the way, all along you're just doing more MIPS in the acronym sense than the previous maker for less money. And then the microcomputer comes along, and not only did they do, like, fewer MIPS that, but they cost nothing, and there were gonna be gazillions of them. And so you went from an era when IBM would, would lease, you know, 100 or 500 new mainframes in a year, and Sun might sell, you know, 500,000 workstations to, like, oh, let's sell 10 million computers in a quarter. And, and I think that that's scale out, where there's less computing but in many more endpoints, is a deep architectural win as well because it gives more people more control over what happens. It reduces the cost. So today if you have to... You know, the most expensive MIPS you can get are in, like, a nuclear-powered data center with, like, liquid cooling and blah, blah, blah, whereas the MIPS on my phone are free and readily available for use. And, and I think that that to me has been a blind spot with the, the s- model developers now. They all do it. I mean, I, you know, run Llama on my Mac and it's-

   11. MCMartin Casado24:11

       Yeah.

   12. SSSteven Sinofsky24:11

       It's just, the first time you do it, it's, your, your mind is blown.

   13. MCMartin Casado24:13

       Yeah, yeah.

   14. SSSteven Sinofsky24:14

       And, and then you start to go, well, now that's just how it should happen. And then you look at Apple and their strategy, which the execution hasn't been great, but the idea that all these things will just surface as features popping up all over my phone, and they're not gonna cost anything, my data's not gonna go anywhere, that's got to be the way that this evolves. Now, will there be some set of features that are only You know, hyperscale cloud inference. Oh, yeah. But so, but just like the, you know, most data operations happen in the cloud now.

   15. MCMartin Casado24:43

       Yeah, yeah.

   16. SSSteven Sinofsky24:44

       You know, but, but like most databases are still on my device.

   17. MCMartin Casado24:47

       Yeah, yeah, yeah, yeah.

   18. SSSteven Sinofsky24:48

       It, it-

   19. MCMartin Casado24:48

       So I, I, I'm smiling because, like, this is the story from, like, a microcomputer guy. I'll tell this story from an internet guy.

   20. SSSteven Sinofsky24:54

       Right. [laughs]

   21. MCMartin Casado24:55

       'Cause there, there's the perfect parallel, like, which is like, do you remember the switching wars?

   22. SSSteven Sinofsky24:59

       Oh, yeah, absolutely.

   23. MCMartin Casado25:00

       [laughs]

   24. SSSteven Sinofsky25:00

       Yeah, yeah.

   25. MCMartin Casado25:00

       [laughs] I mean, so, so for the longest time, you had the telephone networks, and they were perfect. They would converge in milliseconds.

   26. SSSteven Sinofsky25:06

       Mm.

   27. MCMartin Casado25:06

       They would never drop anything. You got guaranteed quality of service, and here comes the-

   28. SSSteven Sinofsky25:10

       Five nines.

   29. MCMartin Casado25:11

       That's five nines. [laughs]

   30. SSSteven Sinofsky25:12

       Yeah.
7. 29:24 – 31:48

   Changing benchmarks

   1. SSSteph Smith29:24

      if you think about how this progresses to date, I feel like the benchmarks have always been like which model has the most parameters, how is it doing on this coding test, that isn't representative necessarily, like what device can this fit on, how much does it cost? Like, do we expect then a different set of benchmarks or things that we're judging these models by, or should we just be looking at the app layer? Like, does there need to be some sort of shift that kind of moves us away from, like, bigger, better, as you were saying, scale up, and something that represents scale out?

   2. SSSteven Sinofsky29:53

      Of course, I thought all those benchmarks were just silly to begin with. To me, they all seemed like, remember the benchmark we used to do with browsers was, like, how fast it could finish rendering a whole picture.

   3. MCMartin Casado30:02

      [laughs]

   4. SSSteven Sinofsky30:03

      And so Mark Andreessen invented the image tag-

   5. MCMartin Casado30:05

      Yeah, yeah

   6. SSSteven Sinofsky30:05

      ... in the browser, and it, the neat thing that they did in their implementation was progressively render it.

   7. SSSteph Smith30:10

      Mm-hmm.

   8. SSSteven Sinofsky30:10

      And then what that did is, like, empower stopwatches all over the world of magazines to, to write who finishes rendering a picture faster. And of course, here we stand today, like-

   9. SSSteph Smith30:19

      Yeah

   10. SSSteven Sinofsky30:19

       ... that's a thing you can measure even. That's a time, and it doesn't matter.

   11. SSSteph Smith30:22

       Mm-hmm.

   12. SSSteven Sinofsky30:23

       And so I think that that, those will all go away, and we're just very quickly gonna get to, like, what does it actually do? I do think that the measure that's gonna start to really matter will depend on the, the application that people are going after. Like, take this research stuff that just pa- appeared, like, this week. Like, everybody's got a research app. Well, it turns out when you're doing research, the metric that matters is truth. And, and all of a sudden we're back to hallucinate, but now, like, well, you're giving footnote links and you're giving sources because what's really happening under the covers is, is like, well, it's, it's a little bit less of generative and a little bit more of IR.

   13. MCMartin Casado31:03

       Yeah.

   14. SSSteven Sinofsky31:03

       And, and all of a sudden vector databases and looking things up and reproducing them matter, and so now we're gonna get to a point where, like, okay, we're- people are gonna develop, I think it's gonna lo- probably along the lines of ImageNet, and they're gonna start to generate us thousands and thousands of routine tests-

   15. MCMartin Casado31:20

       Yeah

   16. SSSteven Sinofsky31:20

       ... that are like, is this true?

   17. MCMartin Casado31:22

       This is totally an aside, but you, you reminded me of like a kind of a weird historical errata, which is the fact that Andreessen made the image tag. So in a way, he's kind of also the grandfather to like some AI because CLIP, which is an AI model, basically will take an image and describe it. The way it does it using the meta tags in image tags. [laughs]

   18. SSSteven Sinofsky31:41

       Yeah, yeah, yeah.

   19. MCMartin Casado31:41

       That's right. So he created the metadata to, to do this. I, I, I will say back on the topic of, of, of the images, here's one

8. 31:48 – 35:27

   Building defensibility

   1. MCMartin Casado31:48

      thing I've, I've noticed working with these companies where, um, these, these models are actually pretty magic by themselves, right? Like, like if you have a big model and you just expose it, people use them, right? Which is very kind of different than computers. Like you just put the model out there.

   2. SSSteven Sinofsky32:00

      Yeah, yeah.

   3. MCMartin Casado32:01

      The thing is, is all the other models catch up very quickly 'cause they distill so well. So like that's, that's not defensible in a way.

   4. SSSteph Smith32:08

      Mm-hmm.

   5. MCMartin Casado32:08

      And so the companies that are defensible that I've seen is they'll put out a model that's very compelling, and then once the users are engaged with the model, they find ways to build an app around that that actually is retentive, right? So it'll start converging on like PowerPoint or like, you know, it'll start addressing, it's like more stateful and requires configuration. So that tends to be very defensible. And then the applications that use models, they use lots of models, and they do fine-tune these models a whole bunch. And so I think that now, I would say the last two years have been like the story of the large model. It really has been, and they've been magic. Like people use them and they, people really like them. And like-

   6. SSSteven Sinofsky32:40

      Yeah

   7. MCMartin Casado32:40

      ... you know, the first time you're in ChatGPT, you're like, "This is amazing." And now I think we're in the era of workflow around models which are stateful complex systems, right? And, and also many models powering, you know, apps in, in more sophisticated-

   8. SSSteven Sinofsky32:55

      I, the many models is a great point. Uh, uh, uh, like to build on that, like this is what happened when, with user interface.

   9. MCMartin Casado33:02

      Yes.

   10. SSSteven Sinofsky33:02

       And so the whole notion of user interface that IBM put forward was just derived exactly from their s- their green screens and their 3270s, and they, they made a, they made a shelf of rules on how-

   11. MCMartin Casado33:13

       4-40 in- or 40 characters of-

   12. SSSteven Sinofsky33:16

       Of like exactly how the UI should be, and this is the F10 button and this is the whatever. And then like it turns out that people were u- building all sorts of UI frameworks. It actually looks exactly like the browser today, where there's a zillion frameworks on the, on the endpoint. You pick and choose, you do what you want to do. You can invent a new calendar dropdown if you want or not waste your time. It's really up to you, and I do think that that aspect of creativity is extremely important to applications. And then one other, just to be kind of dull and boring, that I think is gonna start to matter a lot for apps to be differentiable and to also, to use an MBA, have a moat, which is that they're, the apps are gonna also embrace the enterprise.

   13. MCMartin Casado33:56

       Yeah. Yeah.

   14. SSSteven Sinofsky33:56

       And, and for better or worse, one of the lessons that we keep learning is, you know, if you want to get adoption in the enterprise, you're gonna have to do a bunch of work like to turn off parts of your app or to filter-

   15. SSSteph Smith34:07

       Mm

   16. SSSteven Sinofsky34:07

       ... parts of your app or to-

   17. SSSteph Smith34:08

       Yep

   18. SSSteven Sinofsky34:08

       ... disable it or whatever it is. And I think the smartest entrepreneurs are gonna recognize the, like the need for like sign on-

   19. MCMartin Casado34:16

       Yeah

   20. SSSteven Sinofsky34:16

       ... single sign on at the beginning.

   21. MCMartin Casado34:18

       To RBAC and SSO are like the dance every time.

   22. SSSteven Sinofsky34:21

       Every single time, like it turns into, and because it turns out that's also a great way to price.

   23. MCMartin Casado34:25

       Yeah.

   24. SSSteven Sinofsky34:25

       It's-

   25. SSSteph Smith34:26

       Yeah

   26. SSSteven Sinofsky34:26

       ... but it's not super hard, and I think that there's so much dumb stuff has been done about AI and alignment and censorship and, and like whose point of view is it and all this other stuff that, that there's now a whole industry-

   27. MCMartin Casado34:41

       Yeah

   28. SSSteven Sinofsky34:41

       ... that just wants to show up and tell you all the things that they don't want out of AI.

   29. MCMartin Casado34:46

       Yeah.

   30. SSSteven Sinofsky34:46

       And the smartest entrepreneurs are gonna actually get ahead of that-
9. 35:27 – 41:17

   AI’s Sputnik moment?

   1. SSSteph Smith35:27

      about DeepSeek as the Sputnik moment and that can be viewed in the lens of like geopolitics, US, China. But also if you think about Sputnik, that wouldn't have been a moment if Kennedy didn't do his moon landing speech, if we didn't actually get there, so in other words, if changes weren't made. And so let's say you're in a boardroom, you're an advisor.

   2. MCMartin Casado35:47

      I don't wanna talk to the boardroom, I wanna talk to the US government, right? And so like for me actually, the biggest aha of DeepSeek is nothing we've talked about right now. The biggest aha of DeepSeek is how blind our policies, our policies have been around AI, right?

   3. SSSteph Smith36:00

      Mm-hmm.

   4. MCMartin Casado36:00

      Just they've been so wrong-headed. So our previous policies around AI have been we can't open source 'cause it'll enable China. We've got to limit our, uh, big labs, you know. We've got to put all of this regulation on top of it and the reason is, is for safety and all this other stuff.

   5. SSSteven Sinofsky36:17

      Export controls.

   6. MCMartin Casado36:18

      All the export controls-

   7. SSSteph Smith36:19

      Yeah

   8. MCMartin Casado36:19

      ... so we can't enable other countries. We can put export controls on chips. We've talked about putting export controls on software, you know, like weight limits, all of this other stuff. Like that was our entire policy, and for me the, the, the biggest, biggest, biggest takeaway, the whole DeepSeek thing is, is that's the wrong way to do policy. China has got a lot of very smart people. They're incredibly capable. They're great researchers. They can build stuff as great as, uh, uh, uh, as well as we can and they can open source it. We did not enable them. They did this even with export controls on chips, right?

   9. SSSteph Smith36:47

      Mm-hmm.

   10. MCMartin Casado36:47

       So there's basically all of our activity has been for naught and what we should be doing is funding and investing in our research labs and we should be going as fast as we can and it really is the AI race just like we went through the space race and we need to win. And I, we have everything that we need to win. The only thing in our way is our own regulatory

   11. SSSteven Sinofsky37:05

       And just to, just to build on that, like the lesson is not Sputnik, the lesson is the internet.

   12. SSSteph Smith37:10

       Hmm.

   13. SSSteven Sinofsky37:11

       And, and 'cause the, what we learned from, from the internet, which, you know, Al Gore famously claimed to have invented the internet, but what he really did was invent the regulation that allowed the internet to flourish.

   14. MCMartin Casado37:21

       Yes.

   15. SSSteven Sinofsky37:21

       And, and it could've been, they could've looked at the internet and said, "Oh my God, this is a Sputnik moment," and then tried to turn it into what AT&T and WorldCom wanted.

   16. SSSteph Smith37:29

       Right.

   17. SSSteven Sinofsky37:29

       And they were there lobbying, trying to make that happen.

   18. MCMartin Casado37:31

       Yeah, exactly.

   19. SSSteven Sinofsky37:32

       Like they absolutely... And, and frankly, AOL wanted it to happen that way too. And so they, they ignored that, and they went with what g- made the internet strong to begin with. And so what gave us this DeepSeek moment was the strength of the worldwide technology community. And, and so as much as people wanna own it, you know, it, you, and be the singular provider, it's not gonna work. The biggest difference, n-n-not to overanalyze the analogy, I think it's a Sputnik moment in the sense that it's a wake-up call for the, for the, for half the world. It, it, it isn't a geopolitical wake-up call. It's not about war.

   20. SSSteph Smith38:08

       Mm-hmm.

   21. SSSteven Sinofsky38:08

       It, it's literally just about technology diffusion.

   22. MCMartin Casado38:11

       Yeah.

   23. SSSteven Sinofsky38:11

       And we've had so many m-misfires since then. I mean, we had the whole encryption war-

   24. MCMartin Casado38:17

       Right

   25. SSSteven Sinofsky38:17

       ... where we tried to put export controls on encryption and all this. And you know, although we, people thought we were being silly as an industry when s- many of us would champion this. Well, you can't. It's like outlying math. It, it turns out it is-

   26. SSSteph Smith38:30

       Heard that before [laughs]

   27. SSSteven Sinofsky38:30

       ... outlying math. And, and the fact that it used those chips, well, the world's economy, uh, as we've seen, is very, very hard to put export controls on things.

   28. MCMartin Casado38:38

       Remember, remember when we were gonna export control PlayStations?

   29. SSSteven Sinofsky38:40

       Oh, yeah.

   30. MCMartin Casado38:40

       [laughs]
10. 41:17 – 43:00

    Impact on frontier companies

    1. MCMartin Casado41:17

       don't think this is a crisis moment for OpenAI or Anthropic. I think like apps are hard to build. [chuckles] I think that like right now the apps that they put out are very complex. They actually know their users. They have very specific use cases. And so I mean, I think for them it's a bit of a wake-up call that they, you know, they can't slouch and they gotta move very quickly, but I'm still very, very bullish on our labs. I, I think they can stay ahead too. So again, like there's this view of DeepSeek as a crisis moment for NVIDIA, a crisis moment for OpenAI and Anthropic. I don't buy any of that.

    2. SSSteph Smith41:44

       Mm-hmm.

    3. MCMartin Casado41:44

       I think it's some- it's more of like a wake-up call for the regulatory environment, and then listen, we should all acknowledge that. Listen, there's gonna be global competition. We need to stay ahead.

    4. SSSteven Sinofsky41:52

       What we should see now, the right reaction from all of these, these frontier folks, is they should all just start be building apps.

    5. MCMartin Casado41:59

       Yes.

    6. SSSteven Sinofsky42:00

       Because the, the best, the best feedback loop to build a great platform for other people to use is to be building apps, and there's this whole concentrated conversation over competing with your partners or whatever. Our industry is co-opetition through and through. It's Andy Grove's lesson.

    7. MCMartin Casado42:15

       Yeah.

    8. SSSteven Sinofsky42:15

       So just everybody should be prepared for the, these big players to compete with you.

    9. MCMartin Casado42:21

       Yeah.

    10. SSSteven Sinofsky42:21

        But history has shown that's no surefire success.

    11. MCMartin Casado42:24

        Yeah, but if ta- if TAM agrees is 10X, there's just a lot of room for a lot of folks.

    12. SSSteven Sinofsky42:28

        Yeah. I mean, people like, you know, Microsoft spent 10 plus years like a distant number three in the applications business.

    13. MCMartin Casado42:35

        Yeah.

    14. SSSteven Sinofsky42:35

        And it was a platform shift that all the other players ignored that all, that caused it to win.

    15. MCMartin Casado42:41

        Yeah.

    16. SSSteven Sinofsky42:41

        And so I think that, that the TAM is gonna be 100X. It's gonna be every endpoint. The, the revenue is gonna come from the apps side of it, and then there'll be a developer side of it.

    17. SSSteph Smith42:51

        Mm-hmm.

    18. SSSteven Sinofsky42:51

        It'll just be a different pricing model for different sets of scenarios, but it's gonna be there. So we're just, everything is rising right now.

    19. SSSteph Smith42:59

        Since it is this positive

11. 43:00 – 44:29

    Participating in the next wave

    1. SSSteph Smith43:00

       sum growing world, do you have any thoughts just real quick on the fact that this came from like an algorithmic hedge fund, like a quant? Is that any different to your expectation, or does that actually signal that more can participate?

    2. SSSteven Sinofsky43:12

       It's a good reminder that the, there are always pockets of people innovating.

    3. MCMartin Casado43:17

       Yeah.

    4. SSSteven Sinofsky43:17

       And, you know-

    5. SSSteph Smith43:18

       Mm-hmm

    6. SSSteven Sinofsky43:18

       ... nobody, no, WorldCom and AT&T did not predict the internet was gonna come out of universities. Like they did not think that a physics lab in, in Switzerland was gonna invent the protocols that become foundational.

    7. MCMartin Casado43:30

       [laughs] That's so true.

    8. SSSteven Sinofsky43:31

       I, I mean, they certainly, and they also didn't expect a-

    9. MCMartin Casado43:34

       That's so true

    10. SSSteven Sinofsky43:35

        ... a failed corporate lab to develop TCP/IP-

    11. SSSteph Smith43:38

        Mm-hmm

    12. SSSteven Sinofsky43:38

        ... that became the standard. I mean, it's not just any old co- It wasn't like the IBM lab. It was like literally a lab that they'd all but shut down-

    13. SSSteph Smith43:45

        Right

    14. SSSteven Sinofsky43:45

        ... because it failed just down the street at Park.

    15. MCMartin Casado43:48

        Yeah.

    16. SSSteven Sinofsky43:48

        And, and so-

    17. MCMartin Casado43:49

        Do you remember like SRI was involved?

    18. SSSteven Sinofsky43:51

        I mean, like all these things.

    19. MCMartin Casado43:51

        Like all these, like there's these places that you don't even think about anymore.

    20. SSSteven Sinofsky43:54

        Right, and so-

    21. MCMartin Casado43:55

        Yeah. That's right

    22. SSSteven Sinofsky43:56

        ... there, uh, there, like it, all of this is gonna, like the, most of this isn't gonna be even in any history that's written in five years.

    23. MCMartin Casado44:02

        Yeah.

    24. SSSteven Sinofsky44:02

        And, and I think that that is the excitement. [upbeat music]

Episode duration: 44:29