The Exchanges

Q Will we see more vertical integration or more horizontal specialization?

A You know, historically, we've seen both, and what's interesting is we're already seeing both now, right? Like Apple, of course, has just been historically vertically integrated. Uh, Microsoft and Intel historically horizontally. Uh, and often, companies will start horizontal and then go vertical. So like, um, you know, Google is horizontal. I mean, it was built on top of normal servers, but then they built their servers, and then, you know, they, you know, They built their own chips. They built their own networking gear, and so I, I think you always get a mix of the two. What's interesting about now is we're actually really seeing both. I mean, I would say that OpenAI is very much a vertically integrated company now with ChatGPT driving a lot of it. I would say Anthropic, a lot of the usage really is more horizontal, and they're doing a great job of that. I think we're seeing this on the model layer too. I mean, a very interesting discussion we haven't had, but it's a very interesting one is like, open source quote unquote really seems to work with these models just because you can't, as a user You know, recreate it. So like, if you look at like BFL, they've done a great job building kind of like a horizontal layer for these models. Um, but then you've got companies like Ideogram, which have built a great kind of vertical experience as well. And so I would say for AI, we've got…

AI assessment note: “for AI, we've got already this early on great examples of both.”

Q to find product market fit. Maybe you've already found product market fit. You need optionality, don't you? How do you then find it? Cause the tension it sounds like is that some startups are taking this approach where they have essentially not just two products, but two different classes of products that have associated different classes of sales organizations and structures internally for selling them. So what are the trade-offs?

A Yeah, well, I think it's highly inadvisable to have multiple products in a startup because in my experience, having now run organizations with multiple products, you basically have two companies. If you end up selling to two different people, if you're selling to the same person, same type of budget in the organization, yes, you can have multiple products. That's pretty common. But if you end up having two products that have two different constituencies, you have to have two product marketing teams. Um, You tend to have different sales cycles. You have different support models. You have, um, different PS professional services models, and then often the feature requirements are different. So you basically have two independent companies. Um, and so that's highly inadvisable. I, I, I think that the right model, at least in the enterprise, you iterate on one product, product director fit, and you can pivot and you can add features and so forth until you get that one right. And once you have access to the customer, then you introduce another one, you know, when you want to grow the biggest Jump in operational complexity. A startup will ever do is when it goes from one product to two products. So if you do introduce a second product, which of course you will, if you can align it with the same constituency, the same buyer, that's the best. I'd be very thoughtful before doing that. A n…

AI assessment note: “I think it's highly inadvisable to have multiple products in a startup”

Q There is a At the point, every company has to sell higher into the company, into the organization. Why is that though? I mean, I still don't quite buy it because if it's something is taking off popular, like by sheer popularity, why do you need to, why do you need to shore that competency up?

A Right now, often why you have to sell higher in the organization is because, um, the technology is being sold across, cut across silos. For example, when you look at cloud as a general thing, it touches many traditional aspects, compute, networking, storage, and security. You either have to go and get every one of those decision makers on board, or you just go to someone that has all of them under them and sell to that point, which is why, you know, existing companies with C level CIO level relationships are successful at those higher level sales. And like when you see companies like, like startups, for example, move into infrastructure, you want them to sell higher into the organization so they don't have to fight for different battles.

AI assessment note: “why you have to sell higher in the organization is because, um, the technology”

Q Well, I heard that you dabbled in microbiophysics, astronomy, and I was thinking, who dabbles in those things?

A Is that so funny? Yeah, so actually, I did, um, actually, I started astronomy. I was at the local university, Northern Arizona University. I was taking classes, and I was doing research there. The interesting thing is I'd spend tons of time in this lab, like, you know, and, like, basically, we'd have to, like, check how bacteria were growing every two hours, and I'd spend all this time in the lab, and, um, In the corner was this computer, and I actually didn't have a big background in computer. I was more of a math guy, and I just spent all of this time on the, uh, on the computer, and so from then I said, okay, I was probably 20 at the time. I decided to take computer science courses, and because I had a physics background, I did mostly physics, so I did kind of physics computation, so my first job, uh, out of undergraduate was at Lawrence Livermore, which I was doing, like, massive physics simulations, so at the time I was in applied computer science in physics and not in infrastructure. But I was there actually in the nuclear weapons program.

AI assessment note: “I started astronomy. I was at the local university, Northern Arizona University.”

Q And explain some of the macro conditions that have led to this, this transit, this change in terms of the surplus of founders pursuing these ideas. Like, what are they seeing that's, that's enabling?

A Well, I mean, the obvious is, like, you know, the, the demand for AI is basically infinite, and as a result of that, every part of the supply chain is under, under duress. I mean, everything, including, like, materials used to make things like, like memory. Um, it's also very interesting, there's something unique about AI, um, which, because the demand is infinite and growth is infinite, um, uh, what you tend to worry about is the margin of companies, which is how efficient it is. Like, normally you worry about growth, like, can I just, You know, can I just get people to buy this stuff? You don't have to worry about that here. The question is, is can you do this in a way that's profitable? And a lot of the, um, efficiencies are actually strictly a physical limitation of hardware, and so even the business model of the AI Wave is really putting a lot of stress on the existing systems, because they weren't built for AI, they weren't built for those workloads, and I think there's just this, you know, this global observation that we actually need to change the core components to get that efficiency to help drive the growth and to drive the value of the businesses.

AI assessment note: “demand for AI is basically infinite, and as a result of that, every part”

Q Please. What were your reactions to what you think of the policy?

A Well, it's interesting because it seems like any time the administration touches immigration, there's a huge outcry, knee-jerk outcry, and we saw a lot of that from VCs even. Um, but it's also very interesting that Reed Hastings, who is a classic lefty, like, and, you know, and has long been, was like, I've been in, doing policy for immigration for, you know, 30 years, and this is the right approach. And this is very much my thought, which is, this system has been gamed for a very long time. It's very hard for startups. To, to hire, uh, because of the lottery system. It's locked up by the large companies, the consultants, Amazon and Google, um, and like that has to change, and I think a very reasonable way to do it is to, to set price because, you know, you've got, uh, a market and you've got, you need to allocate supply. Price is a great way to do it, so I'm very, very positive on it. I comment about that, and a lot of people seem to disagree, so I think it's an active discussion.

AI assessment note: “price is a great way to do it, so I'm very, very positive on it.”

Q Let's, uh, let's go deeper on the customer segmentation part. What are the ramifications of the fact that a lot of the growth is being driven by the prosumer market?

A You know, we should remember that every time we have a super cycle, it tends to start, you know, in these prosumer ways, right? The internet did this, right? Like, remember when Sun outlawed the, the browser, right? This is like Sun Microsystems, right? But they didn't really know how to consume it. So the enterprise doesn't know how to consume these new technologies, but there's clearly a lot of value. And so, you know, the individuals pick them up and they use it, and then we're seeing a lot of the new behavior, um, and, and what's been very interesting is that has already led into enterprise pipeline like we've never seen. So if you actually look, like, I think they just hired an AE who closed a million dollar deal on their first day, right? So they have more enterprise pipeline than we've ever seen. So the fact that these are prosumer businesses, like, like, very specifically to your point, the fact that these are prosumer, um, uh, businesses is not in some way You know, because that's what they always sell to. It's just a natural maturation of the cycle, and if anything, it looks kind of far more promising than it did during the Internet time.

AI assessment note: “that has already led into enterprise pipeline like we've never seen.”

Q Do we have a definition of, of marginal risk that, or a perspective of how to think about that, that idea?

A Well, let's just be clear what we mean by marginal risk, which is, um, computer science, or computer systems are risky. Network systems are risky. Stochastic systems are risky. We've done, we've got decades of, you know, ways of thinking about measuring, regulating, changing common behavior based on this type of risk. And so the question is, can you take all of that apparatus that's been hard won and apply it to AI? If so, like, A, we know it's effective because we've used it before and we've got a lot of experience with it, and, and B, it's ready to be done. Or is there a different type of risk that's not endemic on those systems? In which case, we'll have to come up with something that new, which is, you go down that exploration. Maybe it works, maybe it doesn't work, et cetera, right? So that, like, um, uh, that's what marginal risk is, and I just think that the problem is, is if you don't, If you don't know what it is, how are you going to define a solution?

AI assessment note: “Well, let's just be clear what we mean by marginal risk”

Q Right, right, right. I mean, MIT was a pioneer there with the MIT license and open source. What's your biggest shift?

A I think the biggest shift that maybe has impacted me is like, I just remember the transition where pretty much everybody was in computer science for the love of it because it wasn't really clear where the industry was going. Often they were doing it to get something else done to basically the professionalization of an industry. Meaning it is a real discipline. People are in it to make money. People are in it for a future, which is not a bad thing. This is required. And I think it's actually quite good because it requires to really think about what it is, what people do. And so, kind of on the negative spectrum, there's, you know, people are a lot more mercenary about it than they were before. And on the positive end, I do think we have a lot of framing around it. What does it mean to have a workforce in computer science that will come and go and to handle that in a way? But for me, it's been a very, very stark difference. The people that I used to work with 20 years ago, when we were literally all there You know, for the love of solving these great problems, to now, it's like, you know, this is your job.

AI assessment note: “transition where pretty much everybody was in computer science for the love of it”

Q the flip of that question then. So what are the challenges to people getting on board with this? I mean, besides some of the obvious things we talked about, like they still use a smart card world or they, you know, they're still stuck on passwords. Are there like other big things that are difficult or things that you have to overcome to get people to more broadly adopt it?

A I can answer part of that, which is some companies are much more interested in their image than their customer security. And as a result, For them to adopt a solution that they didn't create is an acknowledgement that they can't provide the security solution themselves. And I think, like, for me, like, the crowning example of this has been Apple. Any number of security vendors will tell you, like, when we try and, you know, provide a solution on top of Apple, they don't want to admit that they're insecure. Which is a shame, because, I mean, there was a big announcement of zero-day vulnerabilities in iOS, so every company has insecurity. Having a real security ecosystem around it makes a solution better. One of the reasons that Ubico's won the hearts and minds and has been this kind of organic, you know, bottoms up phenomenon is contributions of open source, and so.

AI assessment note: “some companies are much more interested in their image than their customer security.”

Q So what does it mean when you have this horizontal versus vertical AI layer?

A So in the past, like a lot of times tooling was something you could monetize. Like Purify was a billion dollar company that basically sold the debugger, right? All the tooling and all the infrastructure in order to build application was very much monetizable. For AI, because there's so much value in optimization, there's so much value in data, it's almost like this tooling infrastructure layer is something that, you know, is being offered, or given, or, you know, many players are just offering for free, and they're out there as libraries and things on top of services, and then the actual value is the vertical application of those to whatever. So, for example, if I look across AI startups, the ones that tend to be getting the most traction have taken AI and applied it to a vertical problem. They have access to a Priority data set, or they've done a specific sort of optimization, and now there's a vertical focus towards something, as opposed to I've got this very horizontal kind of generic AI layer. I think that's the game of the big players like the Amazons or the Googles. And so I think from an industry-wide and a startup perspective, I really think vertical focus is how we're going to see the gains of AI in the enterprise, as opposed to what we've seen in the past in computer science, which is the more horizontal.

AI assessment note: “now there's a vertical focus towards something, as opposed to I've got this very horizontal”

Q Well, I heard that you dabbled in microbiophysics, astronomy, and I was thinking, who dabbles in those things?

A Is that so funny? Yeah, so actually, I did, um, actually, I started astronomy. I was at the local university, Northern Arizona University. I was taking classes, and I was doing research there. The interesting thing is I'd spend tons of time in this lab, like, you know, and, like, basically, we'd have to, like, check how bacteria were growing every two hours, and I'd spend all this time in the lab, and, um, In the corner was this computer, and I actually didn't have a big background in computer. I was more of a math guy, and I just spent all of this time on the, uh, on the computer, and so from then I said, okay, I was probably 20 at the time. I decided to take computer science courses, and because I had a physics background, I did mostly physics, so I did kind of physics computation, so my first job, uh, out of undergraduate was at Lawrence Livermore, which I was doing, like, massive physics simulations, so at the time I was in applied computer science in physics and not in infrastructure. But I was there actually in the nuclear weapons program.

AI assessment note: “Yeah, so actually, I did, um, actually, I started astronomy.”

Q you know, which were consensus, which were non-consensus. Um, and, and then Martin, what were kind of your, your reactions to, to, to that sort of broader commentary?

A Well, listen, I mean, again, it wasn't meant to be a, a technical tweet where, like, the wording was exact, and so, like, like, on the face of it, it's almost like, um, An ill-defined statement because we don't know what consensus means, right? And so then everybody picks apart the, the, uh, consensus, but, but here's my reaction to the list of like the Airbnbs and this and that, which is, I think we need to be very careful not to conflate a company having a hard round with market consensus, right? Like if you look at the list, like, you know, Keith Rabois put out, which is, which is great. And I love Keith. I mean, these are like MIT founders, Known spaces. Like I'll bet if you took like the, the, the median value of their raises, I'll over the life cycle of the company, I'll bet they're way above market. Um, many of the, the companies were YC companies. And so I just think it's so easy to like come up with these anecdotal, oh, this one company had a tough raise when that's definitely not within the spirit of what I was trying to say, which is. Markets are actually quite efficient. If the market's efficient and it's a good company, the price is going to be high. And if you don't recognize that, then you're probably beating yourself as opposed to the market, right? And so like, it really comes down to don't, you shouldn't be looking for good deals with respect to other investor…

AI assessment note: “here's my reaction to the list of like the Airbnbs and this and that”

Q you know, which were consensus, which were non-consensus. Um, and, and then Martin, what were kind of your, your reactions to, to, to that sort of broader commentary?

A Well, listen, I mean, again, it wasn't meant to be a, a technical tweet where, like, the wording was exact, and so, like, like, on the face of it, it's almost like, um, An ill-defined statement because we don't know what consensus means, right? And so then everybody picks apart the, the, uh, consensus, but, but here's my reaction to the list of like the Airbnbs and this and that, which is, I think we need to be very careful not to conflate a company having a hard round with market consensus, right? Like if you look at the list, like, you know, Keith Rabois put out, which is, which is great. And I love Keith. I mean, these are like MIT founders, Known spaces. Like I'll bet if you took like the, the, the median value of their raises, I'll over the life cycle of the company, I'll bet they're way above market. Um, many of the, the companies were YC companies. And so I just think it's so easy to like come up with these anecdotal, oh, this one company had a tough raise when that's definitely not within the spirit of what I was trying to say, which is. Markets are actually quite efficient. If the market's efficient and it's a good company, the price is going to be high. And if you don't recognize that, then you're probably beating yourself as opposed to the market, right? And so like, it really comes down to don't, you shouldn't be looking for good deals with respect to other investor…

AI assessment note: “here's my reaction to the list of like the Airbnbs and this and that”

Q Let's, uh, let's go deeper on the customer segmentation part. What are the ramifications of the fact that a lot of the growth is being driven by the prosumer market?

A You know, we should remember that every time we have a super cycle, it tends to start, you know, in these prosumer ways, right? The internet did this, right? Like, remember when Sun outlawed the, the browser, right? This is like Sun Microsystems, right? But they didn't really know how to consume it. So the enterprise doesn't know how to consume these new technologies, but there's clearly a lot of value. And so, you know, the individuals pick them up and they use it, and then we're seeing a lot of the new behavior, um, and, and what's been very interesting is that has already led into enterprise pipeline like we've never seen. So if you actually look, like, I think they just hired an AE who closed a million dollar deal on their first day, right? So they have more enterprise pipeline than we've ever seen. So the fact that these are prosumer businesses, like, like, very specifically to your point, the fact that these are prosumer, um, uh, businesses is not in some way You know, because that's what they always sell to. It's just a natural maturation of the cycle, and if anything, it looks kind of far more promising than it did during the Internet time.

AI assessment note: “that has already led into enterprise pipeline like we've never seen.”

Q Something else we debated which inspired this conversation, um, was defensibility. Which is, how should we think about defensibility for this company? Are they defensible? Where does the defensibility come from? Does it come from state? Does it come from context? Does it come from brands? Some hybrid? Martin, why don't you take that?

A Yeah, so I gotta say, like, the, the, the actual data on this stuff is really noisy, uh, because everything's doing well, so it's kind of hard to have a theory, but, um, if you, if you actually kind of dig into it and you watch this thing for three years, like, something seems to be pretty clear, and that is, Um, a really hard thing about building, you know, any startup or software is the bootstrap problem. Like, how do you get, like, the first 102 hundred customers? And, like, AI actually solves that problem. It just solves the bootstrap problem. It's like, these models are so magical. So, like, You know, you wrap one of these models, uh, you know, you make it available, and people think it's amazing they show up. But what's also clear is, like, that doesn't solve your retention problem if you're a, if you're a software company, right? And so, like, it solves a very hard problem but doesn't solve another problem. And so, you know, and arguably there's actually a lot of perverse economies of scale that are actually in play with these AI companies because, like, the models that commoditize very quickly, anybody can kind of use them, et cetera. Um, and so what we found out is the pattern that seems to work is, you know, a startup will come and they'll do a model, And, you know, they'll get a bunch of users on that model and that'll be great, but then they have to kind of revert t…

AI assessment note: “they have to revert to traditional software to build traditional modes”

Q Let's get deeper into some examples. It's obvious OpenAI and Anthropic have tremendous growth opportunities ahead of them, but why are we excited for opportunities across, for leadership across the stack?

A Yeah, so I'll jump in. So, like, there was this view that, like, OpenAI would win everything, or these large models would win everything early on. Um, but if you actually look at the, the, the history to now, the last three years, it's been the opposite. Like, so if you remember, like, what was the first use case that opening I did, right? It was, it was code. It was co-pilot, right? But they lost that. And then they were actually the first to image, really, with Dali. They lost that, right? Mid-Journey came out. Um, they were kind of the first to, like, real video with Sora, and they lost that. And yet, they've gotten tremendous amount of value, uh, out of text. And so, like, like Sarah said, I think this is right. Like, the primary takeaway is these markets are larger, um, and they're growing faster than we expected, and so you result in fragmentation. So things before that we would have said, like, oh, this is like some sub thing, and OpenAI will get it, or this is a minor market, or whatever, ends up turning to be Large enough to multiple companies with tremendous growth and tremendous value, right? And so we think kind of the only crime, this is going to be caveated later on, is zero sum thinking, right? Like, like anybody that likes decried, oh, defensibility isn't going to work has been wrong. Um, anybody that's decried, like, it's all going to aggregate has been wrong. …

AI assessment note: “these markets are larger, and they're growing faster than we expected, and so you result in fragmentation.”

Q So what is the steel man of the, the critique of, of open source that they were making, uh, a couple, couple years ago?

A That it, you know, this is like a nuclear weapon. Would you open source your nuclear weapon plans? Would you open source your F-sixteen plan? So the idea was that somehow like this was like, and, and, uh, you know, Nuclear weapons are not dual use. Nuclear energy is dual use, right? An F-sixteed is, is not dual use. Like, a jet engine is dual use. But a lot of the analogies that were used at the time were something that, you know, if you squint one way, parts of it are dual use. They could be used for good or for bad. But like, the examples were clearly the weapons. And that's what they would say. They would say, listen, these things are incredibly dangerous. Would you open source, like, whatever the plans for an F-sixteen? And then, you know, the other side, which slowly decided, like, this conversation is ridiculous, we gotta go ahead and set up. It says, you know, no, you would not do this for an F-sixteen, because that is a fighter, you know, jet. And however, like, a lot of the technologies used to build it, yes, this is, you know, fundamental. It's not like people aren't gonna figure it out anyways, and we need to be the leader, just like we were the leader in nuclear. And we were, then, by the way, in nuclear, like, If you go historically, when that came out, we invested incredibly heavily in it. The things that we thought were proximal to weapons, of course, we made sen…

AI assessment note: “That it, you know, this is like a nuclear weapon.”

Q maybe they were asking, where should the burden of proof be? Because it's hard to prove that there is risk, but it's also hard to prove that there isn't risk. And so this question of what's risky is, is it riskier to just go full steam ahead, or is it riskier to kind of slow down until we better understand, um, sort of these models, you know, interpretability, et cetera?

A I mean, I think it's really important to ground these hypothetical discussions on what we've learned as an industry. I mean, the disc are, Course around tech safety has been around for 40 years, and we went through it with compute, like, remember when we're like, okay, Saddam Hussein shouldn't have PlayStations, because you can use, uh, GPUs to simulate nuclear weapons. That was actually a pretty robust and real discussion, um, but that did not stop, you know, us from having other people create chips or video games, right? I mean, we went through the internet, we went through cloud, we went through mobile, and so we've been through all of these tech waves, and we've learned how to have this discussion in a way that, that, For the United States, interest balances these two things. And, you know, listen, we've had kind of areas that were very sensitive to national governance. Think about like Huawei and Cisco, for example, and we as a nation did start to put in kind of import and export restrictions as a result. And so I just feel these almost platonic, you know, polemic questions like the one that you just posed aren't rooted in 40 years of learning. So all I ask is, if we're gonna make a departure From a posture that was developed from 40 years, we better have a pretty damn good reason. And if we don't have a good reason, then I think we should probably learn from that, that ex…

AI assessment note: “if we're gonna make a departure From a posture... we better have a pretty damn good reason”

Q Sam Allman once had the advice to startups, uh, last year. He was like, if you're worried about us improving our models, you're in a tough spot. But if you get, if you get more excited about your business, um, by us sort of improving our models, then, then you're in a good spot. Do you think that's a helpful framework or?

A I think it's helpful for open AI. I want to say that too. Like, you know, like if, you know, A-sixteen Z, if you think us investing in this company isn't good for you, then like, you know, but like, if you want to buy from our companies, like that's great. I mean, there's a very open question. That's actually a technical question. It isn't a business question, which is how much does general Training generalize, right? So we know in, like, the pre-training world, it generalized really well. So you'd create one model, and that model was just as good at code as it was at, like, writing a poem or, right? So we know that it was very general, and in that world, sure, as the models get more powerful, then they can do all of the things, so they compete with all of the things, right? But it seems clear to me, and again, this is an observation, and it may not be correct, that as we get more into the RL world, That you make some trade-offs, and then let's say I IRL something for code, it's not gonna be as good as something else, and like you're making these trade-offs, and in that world, then it's not the case that the model's gonna generally be good, so I think it's great to compete at the model layer, and so again, I think this is maybe a reasonable rubric, certainly for OpenAI to have people believe, maybe a reasonable rubric if you believe that these models are gonna be generally grea…

AI assessment note: “I just don't think it holds up to how things are gonna play out.”

Q And then you, you started up, or it was sort of one of the leaders of the, of the infra fund. At what point did we develop a clear infra practice at ASNZ?

A So we always, um, had strong infra people, right? So like Ben Horowitz is an infra guy, even, you know, and honestly, I would say Mark is. Like he kind of, you know, masquerades as a consumer guy, but he's actually like revolutionized the way we use computers in this Deep infrastructure way. And by the way, many things came from that JavaScript, et cetera. We had Peter Levine, you know, Zensource. We had Scott Weiss. And so like, there's always been deep infra, but when I joined the firm, we didn't think of it as infra. We thought of it as enterprise, right? And so, you know, you're either in the consumer team, you're in the fintech team, you know, or you're in the enterprise team. And the thing about the, about just classifying this stuff as enterprise, The go-to-market motions for something that touches technology and being able to reason about that is so different than reasoning to the go-to-market motion that's purely through, like, sales, right? So, and, and we've just learned over time that, like, we do deep market diligence as a firm, uh, and we do deep diligence on companies before we invest in them, and that the type of diligence we do if it was deep infrastructure just required a different, you know, type of Junior partner and type of analysis. Over time, we realized that, like, companies where you can evaluate more by, like, the business model, the market buyer, the …

AI assessment note: “those kind of class of companies are just sufficiently distinct, so we decided to just pull them apart”

Q What do you mean by pre-chasm market?

A There's no market category. There's no budget. The customer isn't educated by what you're doing. And the reason that it doesn't work out is because like VAR is a value added reseller. So like a lot of the enterprise actually purchases from a reseller, not from like the vendor directly or MSPs managed service provider. Which will actually provide services. And the thing is they don't have the sales force to carry pre chasm products. They're good at distributing things where there's a known budget, but if you're doing something. Fundamentally new, there's no way that a VAR can actually pitch, educate the customer and so forth. And so normally you have to create a pull based market before you can actually engage partners.

AI assessment note: “There's no market category. There's no budget. The customer isn't educated”

Q If we do need to rethink some fundamental assumptions, what may that look like?

A Well, I just think that like, um, people like, you know, Steve and myself have built these deep intuitions on how systems function and how they hit the industry based on. 4050 years of like watching this stuff. And I just don't know, like things like will value go to the model or to the app? How much capital can you apply to this stuff? What classes of problems can you solve versus not solve? Um, what guarantees that can you provide? Uh, how does this impact productivity? There's a lot of things that we've got intuitions on, and for me, the big question is, do we have to, like, reshape those assumptions or not, and to what extent do we have to? Because lots of physics feel a little bit different. I'll just give you one example. I mean, I've said this many times. I think it's so important. 20 years ago, if you're a startup of 10 people, and I gave you a billion dollars, what would you do with it?

AI assessment note: “things like will value go to the model or to the app?”

Q So let's trace how the conversation has changed because we don't see Vinod tweeting about open source anymore. Obviously, open ad has changed your tune, especially right now. What, um, is it really just deep seek? Is that, or how do you trace kind of how, how the sentiment shifted on open source?

A Let's, let's go through a few theories. I'm not really sure what happened. I almost felt like it was almost culturally in vogue to be a thought leader on the negative externalities of tech, and it kind of started with Bostrom, but it was picked up by Elon. It was picked up by, um, uh, Moskowitz. I mean, a bunch of, like, these intellectuals that, like, we all respect and still do. I mean, they're just really the titans of, Our industry in our era, they were asking these very interesting intellectual questions around, like, do we live in a simulation? What happens if AI can recursively, uh, self-improve? And then actually, you know, that created whole kind of cultures and online social discourse around this stuff. And so, I think to no small part, that became a bit of a runaway train, and it's just catnip to policymakers.

AI assessment note: “it was almost culturally in vogue to be a thought leader on the negative externalities”

Q maybe they were asking, where should the burden of proof be? Because it's hard to prove that there is risk, but it's also hard to prove that there isn't risk. And so this question of what's risky is, is it riskier to just go full steam ahead, or is it riskier to kind of slow down until we better understand, um, sort of these models, you know, interpretability, et cetera?

A I mean, I think it's really important to ground these hypothetical discussions on what we've learned as an industry. I mean, the disc are, Course around tech safety has been around for 40 years, and we went through it with compute, like, remember when we're like, okay, Saddam Hussein shouldn't have PlayStations, because you can use, uh, GPUs to simulate nuclear weapons. That was actually a pretty robust and real discussion, um, but that did not stop, you know, us from having other people create chips or video games, right? I mean, we went through the internet, we went through cloud, we went through mobile, and so we've been through all of these tech waves, and we've learned how to have this discussion in a way that, that, For the United States, interest balances these two things. And, you know, listen, we've had kind of areas that were very sensitive to national governance. Think about like Huawei and Cisco, for example, and we as a nation did start to put in kind of import and export restrictions as a result. And so I just feel these almost platonic, you know, polemic questions like the one that you just posed aren't rooted in 40 years of learning. So all I ask is, if we're gonna make a departure From a posture that was developed from 40 years, we better have a pretty damn good reason. And if we don't have a good reason, then I think we should probably learn from that, that ex…

AI assessment note: “if we're gonna make a departure From a posture... we better have a pretty damn good reason”

Q So let's trace how the conversation has changed because we don't see Vinod tweeting about open source anymore. Obviously, open ad has changed your tune, especially right now. What, um, is it really just deep seek? Is that, or how do you trace kind of how, how the sentiment shifted on open source?

A Let's, let's go through a few theories. I'm not really sure what happened. I almost felt like it was almost culturally in vogue to be a thought leader on the negative externalities of tech, and it kind of started with Bostrom, but it was picked up by Elon. It was picked up by, um, uh, Moskowitz. I mean, a bunch of, like, these intellectuals that, like, we all respect and still do. I mean, they're just really the titans of, Our industry in our era, they were asking these very interesting intellectual questions around, like, do we live in a simulation? What happens if AI can recursively, uh, self-improve? And then actually, you know, that created whole kind of cultures and online social discourse around this stuff. And so, I think to no small part, that became a bit of a runaway train, and it's just catnip to policymakers.

AI assessment note: “Let's, let's go through a few theories. I'm not really sure what happened.”

Q And then you, you started up, or it was sort of one of the leaders of the, of the infra fund. At what point did we develop a clear infra practice at ASNZ?

A So we always, um, had strong infra people, right? So like Ben Horowitz is an infra guy, even, you know, and honestly, I would say Mark is. Like he kind of, you know, masquerades as a consumer guy, but he's actually like revolutionized the way we use computers in this Deep infrastructure way. And by the way, many things came from that JavaScript, et cetera. We had Peter Levine, you know, Zensource. We had Scott Weiss. And so like, there's always been deep infra, but when I joined the firm, we didn't think of it as infra. We thought of it as enterprise, right? And so, you know, you're either in the consumer team, you're in the fintech team, you know, or you're in the enterprise team. And the thing about the, about just classifying this stuff as enterprise, The go-to-market motions for something that touches technology and being able to reason about that is so different than reasoning to the go-to-market motion that's purely through, like, sales, right? So, and, and we've just learned over time that, like, we do deep market diligence as a firm, uh, and we do deep diligence on companies before we invest in them, and that the type of diligence we do if it was deep infrastructure just required a different, you know, type of Junior partner and type of analysis. Over time, we realized that, like, companies where you can evaluate more by, like, the business model, the market buyer, the …

AI assessment note: “we decided to just pull them apart”

Q So what does it mean when you have this horizontal versus vertical AI layer?

A So in the past, like a lot of times tooling was something you could monetize. Like Purify was a billion dollar company that basically sold the debugger, right? All the tooling and all the infrastructure in order to build application was very much monetizable. For AI, because there's so much value in optimization, there's so much value in data, it's almost like this tooling infrastructure layer is something that, you know, is being offered, or given, or, you know, many players are just offering for free, and they're out there as libraries and things on top of services, and then the actual value is the vertical application of those to whatever. So, for example, if I look across AI startups, the ones that tend to be getting the most traction have taken AI and applied it to a vertical problem. They have access to a Priority data set, or they've done a specific sort of optimization, and now there's a vertical focus towards something, as opposed to I've got this very horizontal kind of generic AI layer. I think that's the game of the big players like the Amazons or the Googles. And so I think from an industry-wide and a startup perspective, I really think vertical focus is how we're going to see the gains of AI in the enterprise, as opposed to what we've seen in the past in computer science, which is the more horizontal.

AI assessment note: “actual value is the vertical application of those to whatever”

Q I think it is important to take a step back and I should talk about how computing has changed so much. Things that we take for granted. I mean, we have way more data than ever before. It's real time and faster than ever before. Um, talk about what's coming next. Like, what do you think is how things are changing? Like one thing that fascinates me is microservices architecture.

A That is, of course, the question. And I actually think, um, The major vectors aren't necessarily technical in the way we like to think. So I think, I think it's cool in listening. I mean, like, you know, like being a CTO for a long time and, and, and having a, you know, technical background, I love to, to, to think about like all the cool new stuff that's happening. But if I look at what are the major shifts in the industry, it's not super technical in as much as the following. We're seeing this massive shift. And then I asked myself, what is this shift? Well, it seems to me that we're seeing a couple of trends. Number one, Uh, app developers are starting to consume infrastructure as pieces of software. That's not really a technical thing as much as, as a shift in responsibility away from something that used to be operated, um, by one guy to becoming basically an object in a program created by another guy, right?

AI assessment note: “app developers are starting to consume infrastructure as pieces of software”

Q There's been a lot of focus in the last few years by, by several companies, but also by the brand industry around this idea of alignment. Um, have we made any progress on alignment or what is your assertive perspective of what are they trying to do? Is that a feasible goal? Um, help us understand what they're trying to solve for.

A So at an almost tautological level, alignment's an obvious thing you'd want to do. I have a purpose. I want to align the AI to this purpose, and it turns out these models are Problematic, generally unruly, chaotic, whatever, ah, adjective you wanna use, and so, like, you know, understanding how to better align them to any sort of stated goal is, is very obviously a good thing. And so, I think we'd all agree that alignment to whatever the goal is to make them more effective at that goal and do that thing is good, especially given these models who have, tend to have a mind of their own. The subtext, Um, that certainly I bristle to is that, is that the people doing the alignment are somehow protecting the rest of us from whatever they think their ideal is as far as, you know, dangers to me or thoughts I shouldn't have or information I shouldn't be exposed to. Um, which is why I think we need to be, even when we come up with policy, we need to be very careful not to impose like a different set of, um, you know, ideological Rules on top of these. I, I just, I just think like alignment is something we should all understand. Actually aligning them to me is, is, is kind of where I take issue from any sort of kind of top-down mandate.

AI assessment note: “I want to align the AI to this purpose, and it turns out these models are Problematic”