The Exchanges

Q this model is going to behave weird in this weird way. Uh, like the Manchurian candidate problem. We haven't found any of that, but it certainly seems like something we'd want to keep an eye on. But being from your perspective, like what, what are the, what are the risks that we need to be aware of going into a world where China is really pushing hard into open source?

A Yeah, there's two, there's two, and you identified them, but let's, let's, let's talk about both of them. Um, so the, so the phone home thing is the, is the easy one, which is you can put up, you know, you can packet sniff, uh, you know, a network and you can tell when the thing is doing that. And you, and plus you can go in, you can go in the code and you can see what it's doing that. And so you, you can validate, you can validate that that's either happening or not happening. And I think that, you know, that's important. Um, uh, but I, you know, I think people are gonna, people are gonna, are gonna figure that out. You, you, you can kind of gate that problem practically. Yeah. Um, the, the, the bigger issue is, um, we, we have this term in the field, uh, right now called open weights. Um, and, um, open weights is a loaded term. Uh, it, it uses the open term from open source, but of course with open source, the thing is you, you can actually read the code. Um, you know, with open weights, you have, you know, just a giant file full of numbers, as you said, that you, you can't really interpret. And then what you don't, what you don't have, what, what most, what most of the open source, uh, open weights models don't have, including, you know, deep seek specifically, What they don't have is they don't have open data, right? Um, or open corpus, right? So you, you, you can't actuall…

AI assessment note: “there's two, and you identified them, but let's, let's, let's talk about both”

Q this model is gonna behave weird in this weird way. Uh, like the Manchurian candidate problem. We haven't found any of that, but it certainly seems like something we'd want to keep an eye on. But, and from your perspective, like, what, what, what are the, what are the risks that we need to be aware of going into a world where China is really pushing hard into open source?

A Yeah, there's two, there's two, and you identified them, but let's, let's, let's talk about both of them. Um, so the, so the phone home thing is the, is the easy one, which is you can put up, you know, you can packet sniff, uh, you know, a network and you can tell when the thing is doing that and you, and plus you can go and you can go in the code and you can see what it's doing that. And so you can validate, you can even validate that that's either happening or not happening. And I think that, you know, that's important. Um, uh, but I, you know, I think people are gonna, people are gonna, are gonna figure that out. You, you, you can kind of gate that problem practically. Yeah. Um, the, the, the bigger issue is, um, we, we have this term in the field, uh, right now called open weights. Um, and, um, open weights is a loaded term. Uh, it, it uses the open term from open source, but of course with open source, the thing is you, you can actually read the code. Um, you know, with open weights, you have, you know, just a giant file full of numbers, as you said, that you, you can't really interpret. And then what you don't, what you don't have, what, what most, what most of the open source, uh, open weights models don't have, including, you know, deep seek specifically, What they don't have is they don't have open data, right? Um, or open corpus, right? So you, you, you can't actually…

AI assessment note: “Yeah, there's two, there's two, and you identified them, but let's, let's, let's talk about both of them.”

Q that isn't perfect, like is embarrassing, right? Like you look at the vision pro and it's like, well, the battery's big. Steve would have hated this, right? Like how he never would have shipped this and that being constrained and not being able to innovate because you're tied to this, like, Impossible standard of being on whatever generation 17 of the iPhone and perfecting every element is a real challenge.

A So I would say there's a corollary to that. One of the things I've observed over the years is I think technology products become obsolete at the precise moment they become perfect. And what I mean by perfect basically is like, yeah, it's like the perfect idealized complete product. Like it does everything you could possibly ever imagine. Everything a customer could imagine, everything you as the technology developer can imagine. It's absolutely perfect. Um, and there's, there's been tons of examples of this over, over the last 50 years, um, where it's like the absolute perfect permanent, it seems to be the permanent version of that product. And then it just turns out that's actually the point of obsolescence because it means creativity is no longer being applied. Right into that platform. You're just like, there's just nothing else to do. You're just like, you're, you're, you're done, right? The product has been realized. And then, and then the cycle is what happens to your point. The cycle is other people come in with completely different approaches, completely different kinds of products that are broken and weird in all kinds of ways, um, you know, but, but, but are fundamentally different. And so, you know, that is one of the time-honored traditions. And, you know, one of the, you know, one of the, you know, things you could say about, you know, Tim is his, you know, his wil…

AI assessment note: “willingness to kind of break the mold of Apple only shifts perfect products”

Q Um, so I'm wondering about your thoughts on, on when you have a, you know, uh, when, when you have a platform, uh, how hard is it to resist chasing the new shiny object? Is that the right move? Or are, are there any other things that you think Apple should be, uh, you know, changing their strategy on?

A Yeah. So look, Apple's always had this, you know, very clearly defined strategy that, you know, Steve, Steve and Tim, you know, working together figured out a long time ago, which is, you know, they, they, I forget the exact term, but it's, it's something like basically they, they, they invest deeply into the core of what they do. You know, they'll basically work internally on things for many years. They all, they only actually release things when they feel like they're kind of fully baked. Um, right. And, and, and so as a consequence, they have this thing where, and Tim says this, right. Uh, you know, they're rarely first to market with new technologies, you know, They're, they're more often in the category of what, you know, Peter, Peter Thiel calls last to market. You know, they're, you know, they'll, they'll come out whatever, three years later, whatever, five years later. You know, they're, you know, there were tablets for years before the iPad. There were, you know, smartphones for years before the iPhone. Folding phones.

AI assessment note: “they're rarely first to market with new technologies”

Q How do you get a job as a venture capitalist in twenty-twenty-five?

A Um, so I, I mean, look, the, the best way, the best way to do it is to have a track record early as somebody who is like in the loop specifically on new product development. Um, and so somebody who, you know, be, be like deeply in the trenches, um, at one of these new companies in one of these spaces, um, you know, participate in the creation of, of, of a great new product, uh, and, and, and a great new company and, you know, really demonstrate that you know how to do that. Um, you know, there's, there, you know, there, there are great VCs who have not done that, but, you know, I think that is sort of a foundational skillset. Uh, you know, for working with the kinds of founders that, that you want to work with, who are going to, who, you know, are going to want you to have, you know, kind of very interesting things to say on that. Um, as I think that, you know, still the, the best way to do it.

AI assessment note: “the best way to do it is to have a track record early”

Q How do you get a job as a venture capitalist in twenty-twenty-five?

A Um, so I, I mean, look, the, the best way, the best way to do it is to have a track record early as somebody who is like in the loop specifically on new product development. Um, and so somebody who, you know, be, be like deeply in the trenches, um, at one of these new companies in one of these spaces, um, you know, participate in the creation of, of, of a great new product, uh, and, and, and a great new company and, you know, really demonstrate that you know how to do that. Um, you know, there's, there, you know, there, there are great VCs who have not done that, but, you know, I think that is sort of a foundational skillset Uh, you know, for working with the kinds of founders that, that you want to work with, who are going to, who, you know, are going to want you to have, you know, kind of very interesting things to say on that. Um, as I think that, you know, still the, the best way to do it.

AI assessment note: “the best way to do it is to have a track record early”

Q today. It feels like it's still on the horizon around these sort of like, you know, eyewear based computing, you know, potentially net new devices that we're, that, that we'll see from, uh, you know, companies like open AI over time, but where do you, Like, like, how, how real is the threat, you know, this year, uh, versus 10 years from today, and, and kind of, what's your framework?

A Yeah. Well, look, I mean, I think the biggest ultimate danger, I mean, the biggest ultimate danger is very clear, which is just like, at what point do you not carry around a pane of glass in your hand, you know, call the phone, um, you know, because other things have superseded it. And then, you know, look, everything, you know, everything becomes obsolete at this point. Um, so there will, there will come some time to make sure when we're not, you know, carrying phones around and we'll, we'll watch movies where people have phones and we'll be like, yeah, look at how, look at how primitive they were. Right. Because, because we'll have moved on to other things and whether those things are i-based or You know, uh, you know, the other kinds of wearables or whether it's just kind of, you know, computing happening in the environment, um, or just, you know, entirely voice based or, you know, who knows what it is, but, um, you know, there will come a time when that happens, you know, is that time three years from now? Cause there's like some, you know, huge breakthrough, you know, from, from some company that figures out the, the, the product that absolutes the phone right away, or is that 20 years from now because the phone is just, you know, such a standard platform for everything that we do in our lives and everything else, you know, kind of remains a peripheral to the phone. I mean…

AI assessment note: “I think it's highly likely that we'll, we'll have a phone for a very long time.”

Q today. It feels like it's still on the horizon around these sort of like, you know, eyewear based computing, you know, potentially net new devices that we're, that, that we'll see from, uh, you know, companies like open AI over time, but where do you, Like, like, how, how real is the threat, you know, this year, ah, versus 10 years from today, and, and kind of, what's your framework?

A Yeah, well, look, I mean, I think the biggest ultimate danger, I mean, the biggest ultimate danger is very clear, which is just, like, at what point do you not carry around a pane of glass in your hand, you know, called a phone, um, you know, because other things have superseded it, and, you know, look, everything, you know, everything becomes obsolete at this point, um, so there will, there will come some time to make sure when we're not, you know, carrying phones around, and we'll, we'll watch movies where people have phones, and we'll be like, yeah, look at, look at how primitive they were, right, because, because we'll have moved on to other things, and whether those things are i-based, or, You know, uh, you know, the other kinds of wearables, or whether it's just kind of, you know, computing happening in the environment, um, or just, you know, entirely voice based, or, you know, who knows what it is, but, um, you know, there will come a time when that happens. You know, is that fine three years from now, because there's like some, you know, huge breakthrough, you know, from, from some company that figures out the, the, the product that obsoleses the phone right away, or is that 20 years from now, because the phone is just, you know, such a standard platform for everything that we do in our lives, and everything else, you know, kind of remains a peripheral to the phone. I m…

AI assessment note: “I think it's highly likely that we'll, we'll have a phone for a very long time.”

Q question like looking back over the, you know, maybe 10 or 10 or 15 years was, was what moments did you feel like there just was not a lot of action happening? Because this summer is just the pace, uh, from so many different teams has been absolutely insane. Everybody's like trying to keep up. And it didn't used to feel that way, at least from my point of view.

A So my, my view on it always is there's like these, there's this, this, these disconnected, you know, kind of patterns or trends. There's, there's sort of the, the sort of day to day phenomenon where like engineers show up every day and they make things a little bit better. And then every once in a while, you know, you, you get a technical breakthrough or a new platform and, and, and that process kind of this, you know, kind of sawtooth kind of up to the right kind of process kind of plays out over time, kind of regardless of what else is happening in the world. And so it, it keeps happening through recessions and depressions and wars and like all kinds of crazy. Crazy, crazy stuff that's happening, but basically, you know, the, the technology keeps getting better. So there's, there's kind of that curve. And then, and then there's the, the sort of enthusiasm curve and, and the, and then the adoption curve, you know, which is basically like, when do these things actually show up in the world? And then by the way, when are people actually ready, uh, you know, for, for, for the new thing? Um, like, if you talk to the people who worked on like, I'm sure you guys have talked to people who work on language models, they will tell you that they were surprised the chat GPT was the breakthrough moment. Cause they thought everybody already knew what these models could do for, you know, thr…

AI assessment note: “arbitrary disconnection between what's actually happening in the substance and then what people are seeing”

Q know, a friend, things like that. And he mentioned that currently. Your chats are not privileged. They can be used in, in, in a, in a lawsuit or, or other, uh, situations. Uh, how, how optimistic are you that our sort of legal system in the U.S. can get some of these issues right where maybe it can't just be, you know, total free markets, kind of lawless, whatever goes?

A You know, so in the case of training data, I think that there, I mean, there's a bunch of these copyright, you know, kind of lawsuits happening right now. There's, you know, the big New York Times opening, I won, and there's, you know, been a bunch of others. Um, I, I think in that, for that particular problem, my guess is that problem ultimately has to be solved through legislation. Um, it's, it's, it's ultimately a legislative question. The reason is because it goes to the nature of copyright law itself, you know, which, which is legislation. And, and, and of course, you know, the, the, the content industry is already claiming that, of course, you know, using, using copyrighted data to train You know, without permission or without paying is, is, is sort of, you know, they, they believe illegal on his face, you know, due to violation copyright law. The counter argument to that, which, you know, which we believe is, well, it's not copying, right? There's, there's a distinction between training and copying. Just like in the real world, there's a distinction between reading a book and copying the book, you know, as a person. And so there, there, there's going to need, I, I think, you know, the courts are trying to grapple with that. There's a whole bunch of cases. There's jurisdictional questions, you know, probably ultimately Congress is going to have to figure out a, um, You kn…

AI assessment note: “You know, so in the case of training data, I think that there”