The Exchanges

Q it seems a bit difficult to imagine, but seeing it as this emergent, accessible space, I think, is really, really compelling for not only banks, but, like, tons of different finance companies and different enterprises in general to suddenly become these transnational actors that they weren't before, or that they didn't have access to because they didn't have the resources to, like, build out that capability. Does that make sense?

A It makes perfect sense because it actually goes to the very heart of open source, in fact, because if you think about the history of open source, and we've talked about this a lot in the podcast, it's been starved for resources. And in the classic model of open source, you had these big corporate players. They needed open source because no one company could be open source because they would never be trusted. And so you have this interesting ecosystem of all these big open source funders from big companies like Cisco and Google and et cetera. And then you had like these open source consortiums and projects. And the really interesting evolution we're talking about In crypto is that we don't actually now have to rely only on those. Your point about banks is quite interesting because you're putting that same framework at the international global people transacting with each other level and creating this sort of substrate for everyone to build on to create that interconnection without having to worry about, well, that bank has a dog in the fight, so we can't trust them if they try doing the same thing kind of thing. Exactly, yeah. So let's actually now talk about how it works and what concretely we're talking about here because so far we've defined the category of stable coins. I understand the value in the larger system of You know, with cryptocurrencies, a lot of volatility, so th…

AI assessment note: “It makes perfect sense because it actually goes to the very heart of open source”

Q it seems a bit difficult to imagine, but seeing it as this emergent, accessible space, I think, is really, really compelling for not only banks, but, like, tons of different finance companies and different enterprises in general to suddenly become these transnational actors that they weren't before, or that they didn't have access to because they didn't have the resources to, like, build out that capability. Does that make sense?

A It makes perfect sense because it actually goes to the very heart of open source, in fact, because if you think about the history of open source, and we've talked about this a lot in the podcast, it's been starved for resources. And in the classic model of open source, you had these big corporate players. They needed open source because no one company could be open source because they would never be trusted. And so you have this interesting ecosystem of all these big open source funders from big companies like Cisco and Google and et cetera. And then you had like these open source consortiums and projects. And the really interesting evolution we're talking about In crypto is that we don't actually now have to rely only on those. Your point about banks is quite interesting because you're putting that same framework at the international global people transacting with each other level and creating this sort of substrate for everyone to build on to create that interconnection without having to worry about, well, that bank has a dog in the fight, so we can't trust them if they try doing the same thing kind of thing. Exactly, yeah. So let's actually now talk about how it works and what concretely we're talking about here because so far we've defined the category of stable coins. I understand the value in the larger system of You know, with cryptocurrencies, a lot of volatility, so th…

AI assessment note: “It makes perfect sense because it actually goes to the very heart of open source”

Q I think it's a very challenging question because simultaneously with the autonomous revolution, We're also seeing the online shopping revolution. So the question is, what do we still physically buy?

A So I'll share one of mine, which is, I was at a nail salon in my street in San Francisco, getting my nails done, and I realized that anytime I go, there's always people there, and when there are not people there, they'll be like, wait here, and within two minutes, five minutes at most, people show up, and I'm just like, how do they do this? I've just been fascinated by it, and then I found out that the women who Who owns a salon has a couple of shops on the street. And what she does is she transports people between the shops based on their availability. And what's fascinating to me about this is that it inverts a model that I've always thought about when it comes to cars is not just decentralizing the driving and these different, this network of, of nodes of cars, but it's actually now the reverse where it's actually bringing people to all these different places. And now you're kind of reallocating labor in a way that sort of moves with where things are.

AI assessment note: “I'll share one of mine, which is, I was at a nail salon”

Q Do you really want to take that out? I would say maybe not. So Zach Termala at Stanford has these studies showing that experts are believed more when they express uncertainty.

A I like that because I actually think that is what a true expert is. It's hubris to claim to be an expert in something you're not. At the same time, I will say coming full circle to your point, I have observed that the people who actually go out and start companies, and this takes us to entrepreneurs, are people who have such belief in an alternative view of the world, even if they're not experts at X, Y, or Z, that they're the ones who go out and do it. And I admire that. I do think it takes a certain amount of knowing that you can do that. So like, what's the difference? Is that a confidence thing? Is that a experience thing? Is that a originals mindset? I mean, where do we figure out like what's making an entrepreneur tick there?

AI assessment note: “I like that because I actually think that is what a true expert is.”

Q it turns out when you start doing features of image recognition, you're using a whole bunch of old school edge detection and contrast and finding objects and all of this stuff. And so You, you can't just, like, show up and say, now we're gonna understand, we're gonna be the unsupervised learning company, because the question everyone's gonna ask is, well, how are you gonna make whatever you're doing practical?

A What I do remember about hearing the stories of the early days of NLP, and, and observing this, parts of this firsthand as well, is how the entire field and community, they, a lot of them had very strong opinions about, you know, there's this whole phase of, like, expert domain knowledge building, and really that's the only way to actually make NLP work at scale. There was all these things they had to do because it was before the days of big data. They couldn't even conceive of the Google scale big data. And then they went to this world where, oh my God, we don't even have to have these kinds of constraints and way of doing things because we have all this data. And now it's sexy to me that you can flip that model again and almost say, you don't even need big data with some results like this kind of paper because you don't have to have anything.

AI assessment note: “What I do remember about hearing the stories of the early days of NLP”

Q Kennedy's was a moon shot because it worked too, right? I mean, if we, if NASA was not able to get to the moon, it would have been a moon shot, presumably, right?

A And this idea of loon shots, I do want to talk briefly about Google X, which has in some ways given both a good name and a bad rap. To the concept of loonshots in multiple levels. One, because they talk about what they do as moonshots. Two, because they actually have a project called Loon, which has been mixedly received, and so that's another context. But three, I think what is really interesting here is when you think about a moonshot as this destination, as you've described it, a desire to go somewhere, I think a lot of people have a hard time in organizations in terms of teasing apart what is a shot that's worth further nurturing, Regardless of the destination, whether it fails or not. Because I agree there's no shortage of ideas, but it's precisely because there's no shortage of ideas that I want to know when can we, like, pick which ones to invest in, because there's still limited resources.

AI assessment note: “I do want to talk briefly about Google X”

Q of the really cool things with these projects is the creators make them like Satoshi made Bitcoin, and then they just launch the ship out, and it's, it's own thing, it's autonomous, it's like this sort of futuristic concept, right? And people are sort of thinking more and more about what are the other ways you can use these concepts to build these, what else can these autonomous organizations do?

A Yeah, that's great to see that. It's already out there, and there already are DAOs and DAX, decentralized autonomous corporations, in existence. And it's even more fascinating, because if you think about the history of innovation, the classic theory of the firm paper, like why does a firm exist, was 1937, right? Coase's paper, Ronald Coase's paper, and it may be time to rethink the future of the corporation, but I want to take a step back and think about a company that exists today as a classic corporation, like Coinbase. And I think it's really fascinating because one of the things that we've heard about that we haven't overtly talked about today when it comes to crypto is that in a lot of ways, some people might argue it's a bit like a religion, you know, even a cult to some extent. And if you think about it, like, there is a prophet like Satoshi Nakamoto, a Bible, the Bitcoin white paper, there are factions and sects, there is a holy object like the Genesis block, there's a belief, there's rituals, mining. I think Fred Ersom, co-founder of Coinbase, actually pointed this out originally. So my question is, how do you build a company in that type of environment?

AI assessment note: “but I want to take a step back and think about a company that exists today”

Q getting chipped away from jobs today. So the reality is that every knowledge work job is going to see this encroachment of smart machines into the workplace. And it's just a readjustment that people have to make to figure out, okay, what do I do in this equation? What does the machine do? And how do we make best use of what both parts of that equation can do best?

A So before we talk about how people can engage, I think that's a really important question. I do want to pause for a moment on This concept that you guys are both reinforcing of augmentation versus automation outright, because I think it's really important because it's the difference between reacting to something that just sort of happens to you versus proactively thinking, okay, let's expect this. Let's treat it as a given. Let's just say like machines are going to be our colleagues. Machines are going to automate parts of our jobs, whether you're a knowledge worker and they automate certain parts of it, or you're a worker where they automate huge chunks of it. I think that's a really important idea, and it reminds me of the, of the original notion that Doug Engelbart had of augmenting intelligence and really thinking about computers, the mouse, as an extension of us versus something that compete with us, which I think is the reality of our lives. I mean, I think people treat their smartphones already as an appendage, literally, so there's a little bit of that already, but can you guys talk a little bit more about the difference between augmentation and automation and why that's so important?

AI assessment note: “before we talk about how people can engage, I think that's a really important question”

Q what they want to imagine is doctors, you know, naturally interacting With applications that are based on Watson in order to provide advice, in order to make difficult decisions. Uh, so again, this is, this is a very hard problem, and that's why I said it needs the, the proper experimentation, the proper patience, the proper personnel, frankly. I mean, hiring is, is another big dimension. Who do you hire?

A So evaluating or defining the moonshot, it's something new. It's a hard problem. It's got significant investment behind it. The right kind of hiring. Astro Teller made an argument in an op-ed for me a few years ago that it has to be 10 X better. It can't be just, you know, 10, like, it has to be not just incrementally better, but like orders of magnitude better. He also made the argument, interestingly, that there has to be some sign today that you can actually get there. So it's not just so pie in the sky, like human beings need to fly without wings, but more that there's some indicator, whether it's Moore's law is on a Trajectory or a curve that there is some indication that we can actually get there, even if it's not available right now. A question I have for you, though, Evangelos, is Watson is an interesting example on multiple fronts, but I can't help but wonder, how can it compete with what a new class of startups can do? And when you have other companies that have a whole different type of data that IBM does not have access to, like Google and Facebook, for example, and what they're doing, and that's just an example of AI. This plays out in countless domains where can a big company actually compete with startup driven innovation?

AI assessment note: “A question I have for you, though, Evangelos, is”

Q what they want to imagine is doctors, you know, naturally interacting With applications that are based on Watson in order to provide advice, in order to make difficult decisions. Uh, so again, this is, this is a very hard problem, and that's why I said it needs the, the proper experimentation, the proper patience, the proper personnel, frankly. I mean, hiring is, is another big dimension. Who do you hire?

A So evaluating or defining the moonshot, it's something new. It's a hard problem. It's got significant investment behind it. The right kind of hiring. Astro Teller made an argument in an op-ed for me a few years ago that it has to be 10 X better. It can't be just, you know, 10, like, it has to be not just incrementally better, but like orders of magnitude better. He also made the argument, interestingly, that there has to be some sign today that you can actually get there. So it's not just so pie in the sky, like human beings need to fly without wings, but more that there's some indicator, whether it's Moore's law is on a Trajectory or a curve that there is some indication that we can actually get there, even if it's not available right now. A question I have for you, though, Evangelos, is Watson is an interesting example on multiple fronts, but I can't help but wonder, how can it compete with what a new class of startups can do? And when you have other companies that have a whole different type of data that IBM does not have access to, like Google and Facebook, for example, and what they're doing, and that's just an example of AI. This plays out in countless domains where can a big company actually compete with startup driven innovation?

AI assessment note: “So evaluating or defining the moonshot, it's something new. It's a hard problem.”

Q Do you really want to take that out? I would say maybe not. So Zach Termala at Stanford has these studies showing that experts are believed more when they express uncertainty.

A I like that because I actually think that is what a true expert is. It's hubris to claim to be an expert in something you're not. At the same time, I will say coming full circle to your point, I have observed that the people who actually go out and start companies, and this takes us to entrepreneurs, are people who have such belief in an alternative view of the world, even if they're not experts at X, Y, or Z, that they're the ones who go out and do it. And I admire that. I do think it takes a certain amount of knowing that you can do that. So like, what's the difference? Is that a confidence thing? Is that a experience thing? Is that a originals mindset? I mean, where do we figure out like what's making an entrepreneur tick there?

AI assessment note: “and this takes us to entrepreneurs, are people who have such belief”

Q of the really cool things with these projects is the creators make them like Satoshi made Bitcoin, and then they just launch the ship out, and it's, it's own thing, it's autonomous, it's like this sort of futuristic concept, right? And people are sort of thinking more and more about what are the other ways you can use these concepts to build these, what else can these autonomous organizations do?

A Yeah, that's great to see that. It's already out there, and there already are DAOs and DAX, decentralized autonomous corporations, in existence. And it's even more fascinating, because if you think about the history of innovation, the classic theory of the firm paper, like why does a firm exist, was 1937, right? Coase's paper, Ronald Coase's paper, and it may be time to rethink the future of the corporation, but I want to take a step back and think about a company that exists today as a classic corporation, like Coinbase. And I think it's really fascinating because one of the things that we've heard about that we haven't overtly talked about today when it comes to crypto is that in a lot of ways, some people might argue it's a bit like a religion, you know, even a cult to some extent. And if you think about it, like, there is a prophet like Satoshi Nakamoto, a Bible, the Bitcoin white paper, there are factions and sects, there is a holy object like the Genesis block, there's a belief, there's rituals, mining. I think Fred Ersom, co-founder of Coinbase, actually pointed this out originally. So my question is, how do you build a company in that type of environment?

AI assessment note: “I want to take a step back and think about a company that exists today”

Q there, and there are going to be a lot of potentially dystopian kinds of futures that might come up that, and so, uh, I'd say that for Even if you think that you're not collecting data about yourself, it's being collected anyway. And so it's important to think about some of the longer term implications of that. You can avoid it. So how do we think about moving forward positively?

A I'm so glad you brought that up because, you know, I, I, one of the things that I don't like is when we preemptively decide things before there's a sort of this permissionless innovation and creativity that can happen. But to your point, we absolutely have to be more thoughtful about Those consequences. I mean, one great example on a very evolutionary scale of this conversation, we had Yuval Harari on this podcast, and he's the author of Sapiens and Homo Deus, and one of the big things that he pointed out is, like, we're, as we enter a world where we may be able to augment ourselves with technology, it could actually entrench inequalities, because a certain class of people might be able to afford a certain class of devices or augmentation that other people might not, and in fact, what was previously, you know, an inequality of You know, societal, cultural level becomes a technological one then, and then one that actually really affects outcomes.

AI assessment note: “we absolutely have to be more thoughtful about Those consequences”

Q I think it's a very challenging question because simultaneously with the autonomous revolution, We're also seeing the online shopping revolution. So the question is, what do we still physically buy?

A So I'll share one of mine, which is, I was at a nail salon in my street in San Francisco, getting my nails done, and I realized that anytime I go, there's always people there, and when there are not people there, they'll be like, wait here, and within two minutes, five minutes at most, people show up, and I'm just like, how do they do this? I've just been fascinated by it, and then I found out that the women who Who owns a salon has a couple of shops on the street. And what she does is she transports people between the shops based on their availability. And what's fascinating to me about this is that it inverts a model that I've always thought about when it comes to cars is not just decentralizing the driving and these different, this network of, of nodes of cars, but it's actually now the reverse where it's actually bringing people to all these different places. And now you're kind of reallocating labor in a way that sort of moves with where things are.

AI assessment note: “I was at a nail salon in my street in San Francisco”

Q to start outweighing the benefits. And I think you can only usually see that by looking at the results. So, you know, in the case of expertise, right, the question is, okay, when you start to generate ideas, are you finding yourself trapped by what you already know in the field? As you're, you know, evaluating different kinds of ideas, do you consistently gravitate toward what's already accepted and proven?

A It's really funny that you guys tried overfitting the U-shape curve to all these different things as well. I think that there's a gender or racial background or other background effect that can play out here differently. I'm thinking of Cases where a lot of women, myself included, will sometimes underplay their expertise, um, because, you know, I've seen a lot of my former male colleagues, like, they would be experts in things that they necessarily weren't, but they would have the confidence to say that they were. And to me, that wasn't a sign of confidence. I'd actually get really irritated when people said, like, be confident and say you're an expert in that. And I'd be like, I'm not going to freaking claim to be an expert in something I'm not. Like, I don't think that's confidence. I think that's just being full of crap. I think it's really interesting because I think some of these things also play out where there's an interaction effect between people's background, whether it's gender, race, or privilege, or other things that have influenced how they grow up. Like, how did you see that play out in thinking about originals?

AI assessment note: “I think that there's a gender or racial background or other background effect”

Q even in that big company, you better make clear as to who is involved and make that decision. You might even change the circle of the people. In other words, okay, it may be the top 20 people. Are they radically truthful and radically transparent and operating an idea meritocratic way? They can do that. They can get together and say, how are we going to be with each other?

A Great point. Okay, so we've talked a lot about how a lot of these frameworks for decisions apply to principles for life, business, relationships, family, spouse, any, all different kinds. Let's talk about, at a very macro level, how do you think this plays out when you think about Competition and innovation around the world. Like, one of my favorite moments is in the book, you talk about your early days and how there is a sort of aspirational and inspirational period in the U.S. in the seventies, or I think it was the late sixties, in the post Kennedy era. And I'm curious for where you think we are now and how it's going to play out, especially given your love of places like China.

AI assessment note: “Let's talk about, at a very macro level, how do you think this plays out”

Q it turns out when you start doing features of image recognition, you're using a whole bunch of old school edge detection and contrast and finding objects and all of this stuff. And so You, you can't just, like, show up and say, now we're gonna understand, we're gonna be the unsupervised learning company, because the question everyone's gonna ask is, well, how are you gonna make whatever you're doing practical?

A What I do remember about hearing the stories of the early days of NLP, and, and observing this, parts of this firsthand as well, is how the entire field and community, they, a lot of them had very strong opinions about, you know, there's this whole phase of, like, expert domain knowledge building, and really that's the only way to actually make NLP work at scale. There was all these things they had to do because it was before the days of big data. They couldn't even conceive of the Google scale big data. And then they went to this world where, oh my God, we don't even have to have these kinds of constraints and way of doing things because we have all this data. And now it's sexy to me that you can flip that model again and almost say, you don't even need big data with some results like this kind of paper because you don't have to have anything.

AI assessment note: “What I do remember about hearing the stories of the early days of NLP”

Q Who created certain data, certain data item? Who did modify that data item? Uh, who did look at those data items? Uh, and, and this, uh, the goal is to provide this service across multiple data sources, across multiple, uh, data storage systems.

A I love this. It's like data provenance as a service. You know, I asked you, um, about how you, you guys came up with the names for all those projects, because I'm always fascinated by the naming of things. One of the funniest stories that Michael Franklin told me about Amplab is that when you guys were coming up with the name that Dave Patterson wrote an email going, why are you guys being all backward about this? That you guys should instead take a first principles approach to what you want the qualities to be, and then come up with the acronym instead of, you know, the other way around. You must have had some debates around how to end up on RISE.

AI assessment note: “I love this. It's like data provenance as a service. You know, I asked you”

Q getting chipped away from jobs today. So the reality is that every knowledge work job is going to see this encroachment of smart machines into the workplace. And it's just a readjustment that people have to make to figure out, okay, what do I do in this equation? What does the machine do? And how do we make best use of what both parts of that equation can do best?

A So before we talk about how people can engage, I think that's a really important question. I do want to pause for a moment on This concept that you guys are both reinforcing of augmentation versus automation outright, because I think it's really important because it's the difference between reacting to something that just sort of happens to you versus proactively thinking, okay, let's expect this. Let's treat it as a given. Let's just say like machines are going to be our colleagues. Machines are going to automate parts of our jobs, whether you're a knowledge worker and they automate certain parts of it, or you're a worker where they automate huge chunks of it. I think that's a really important idea, and it reminds me of the, of the original notion that Doug Engelbart had of augmenting intelligence and really thinking about computers, the mouse, as an extension of us versus something that compete with us, which I think is the reality of our lives. I mean, I think people treat their smartphones already as an appendage, literally, so there's a little bit of that already, but can you guys talk a little bit more about the difference between augmentation and automation and why that's so important?

AI assessment note: “So before we talk about how people can engage, I think that's a really important question.”

Q to think, wow, it would be really interesting To have a tool that is able to look at code and sort of just very different than previously, which was just literally looking at the syntax, but doing millions of examples of generating code and finding bad examples, would it be able to do a better job at finding bad examples in the next piece of code that flows into it?

A I mean, one of the points you guys made last time in our last podcast is that at the end of the day, these things aren't working in isolation. It's not like there's one magic approach. You know, there's always a combination of techniques that come together to actually build real products. How would this sort of fit into that? Because one of the thoughts that I had is that clearly this kind of approach, even if you don't have clearly defined rules, will always be more beneficial in places where we don't have any data, like any big data, just like humans, like kids learning from N equals two, like their parents, or N equals one if there's a single parent.

AI assessment note: “How would this sort of fit into that? Because one of the thoughts”

Q Kennedy's was a moon shot because it worked too, right? I mean, if we, if NASA was not able to get to the moon, it would have been a moon shot, presumably, right?

A And this idea of loon shots, I do want to talk briefly about Google X, which has in some ways given both a good name and a bad rap. To the concept of loonshots in multiple levels. One, because they talk about what they do as moonshots. Two, because they actually have a project called Loon, which has been mixedly received, and so that's another context. But three, I think what is really interesting here is when you think about a moonshot as this destination, as you've described it, a desire to go somewhere, I think a lot of people have a hard time in organizations in terms of teasing apart what is a shot that's worth further nurturing, Regardless of the destination, whether it fails or not. Because I agree there's no shortage of ideas, but it's precisely because there's no shortage of ideas that I want to know when can we, like, pick which ones to invest in, because there's still limited resources.

AI assessment note: “I do want to talk briefly about Google X, which has in some ways”

Q to think, wow, it would be really interesting To have a tool that is able to look at code and sort of just very different than previously, which was just literally looking at the syntax, but doing millions of examples of generating code and finding bad examples, would it be able to do a better job at finding bad examples in the next piece of code that flows into it?

A I mean, one of the points you guys made last time in our last podcast is that at the end of the day, these things aren't working in isolation. It's not like there's one magic approach. You know, there's always a combination of techniques that come together to actually build real products. How would this sort of fit into that? Because one of the thoughts that I had is that clearly this kind of approach, even if you don't have clearly defined rules, will always be more beneficial in places where we don't have any data, like any big data, just like humans, like kids learning from N equals two, like their parents, or N equals one if there's a single parent.

AI assessment note: “How would this sort of fit into that? Because one of the thoughts that I had”

Q So there's, there's no rules. There's no perfect understanding of the search space. There's like, what's the last function? Like, why would you, how would you even write a loss function?

A I do want to push back a little bit though, because Far be it from any of us in here to hype this up. But there's something unique happening here, which at least I perceive this in the paper, which is that I was struck by the analogy to evolution. Like, this is how human beings have evolved. This is evolution that we learn by trial and error on a massive million scale. So I don't want to completely dismiss the idea that we can get to some kind of generalized intelligence. I mean, of course, I understand and agree, but what are the limits and what are the possibilities that can actually take us there? And where are we constrained? Just to break that down a bit more.

AI assessment note: “I do want to push back a little bit though, because Far be it”

Q So there's, there's no rules. There's no perfect understanding of the search space. There's like, what's the last function? Like, why would you, how would you even write a loss function?

A I do want to push back a little bit though, because Far be it from any of us in here to hype this up. But there's something unique happening here, which at least I perceive this in the paper, which is that I was struck by the analogy to evolution. Like, this is how human beings have evolved. This is evolution that we learn by trial and error on a massive million scale. So I don't want to completely dismiss the idea that we can get to some kind of generalized intelligence. I mean, of course, I understand and agree, but what are the limits and what are the possibilities that can actually take us there? And where are we constrained? Just to break that down a bit more.

AI assessment note: “I do want to push back a little bit though, because”