Business, Innovation, and Managing Life (March 6, 2024)

Hi everyone. Welcome to another Q and A about business innovation and managing life. And I see we have a number of questions saved up here. One from Lola. What advice do you have for business investors to pick the right ideas? The problem is ideas aren't the main story in investing. A pure idea is just the beginning of the story, a thing that is going to be a good investment and give you a good return. Financial return is something that to get to that financial return, there are many steps beyond the idea. The idea is just the beginning of the story. And I will say when you say investing, one sort of assumes one means financial investing. There's a whole different story of you think there's a really good idea in the world, you want to make that idea prosper. There are ways to invest in ideas that are not the same as kind of financial investing and things like companies. I mean, there's philanthropy, there's just go get people to develop the idea directly. There's these kinds of things. But in terms of purely financially motivated investing, the idea is just the seed and the problem. The thing that has often surprised me is things that I thought were pretty bad ideas turned into really good businesses. And things that I thought were pretty good ideas never made it as businesses. Sometimes there are ideas where it's a good idea but it's 30 years ahead of its time. Sometimes it's a good idea, but the people who are trying to run a company based on that idea just aren't very practical and can't make the company work. Sometimes it's a good idea, but the world just doesn't understand it. And people look back and say, gosh, we should have understood that idea back 25 years ago or something, but we didn't. The idea was in it its time, but it just didn't take off for one reason or another. So, you know, I think it's very complicated. And I think that what a lot of people who do like venture capital investment say, is there as much betting on people as they are on ideas? I mean, there are usually people, you know, it's always a mixture. I will say another thing about investing is that there's an awful lot of kind of herd following investing, which is when there's a new category of thing. When people say, I've got this great idea about AI and they said that, let's say three years ago, it's like most people are like, oh no, AI isn't a thing. AI died a long time ago, AI isn't going to be a thing. And then suddenly AI is Incredibly hot. And everybody wants to invest in AI. And it's a complicated thing because that very herd process helps. The fact that people are investing in it means the valuations go up. It's a kind of self fulfilling prophecy at some level that just because the very fact that there are lots of investors in that market means that, for example, if you invest in a company and that company goes a certain distance, but then it really needs to be sold to another company, there's probably some other investor who invested in another company which can buy the company that you invested in. So the fact that there is this whole herd of investors is important in making the whole thing go. I mean, sometimes it looks a little bit Ponzi scheme like because you'll see one venture capital firm invests in something and then they're selling that company to another company that they invested in. And it's kind of like what's really going on here because in a sense you're buying out the first company with money that you invested in some other company and so on. And you know, a very cynical person might say, you know, in the end it's the kind of, it's, you know, it's not clear who's winning in this whole picture except that, you know, for a venture capital firm there are management fees that, you know, the very, the very, the very process of investing and taking action and selling things and so on, that's going to make money for the middle people, independent of whether it makes money for the entrepreneurs or for the original upstream investors who are investing in the venture capital firms and so on. It's a complicated world and I think that it is realistically a world which is where the main goal is to make money has a different set of kind of constraints and things that go on in it than a world that's about ideas and the prosperity of ideas. And one is very lucky in some sense, if one's interested in ideas. One's very lucky when the financial world aligns with the world of prosperity of ideas. But they're really separate independent kinds of things. And there are situations have been a very common situation is a company's developed, it's built up some idea and the idea is really cool and really interesting. But the thing that really makes money is something extremely mundane. And then somebody will come in and say, look, we're a private equity firm or something, we're going to come and buy this thing and let's forget about all that really cool stuff that's just costing us money. Let's concentrate on the mundane thing that isn't really terribly interesting from an idea point of view, but that makes money. Let's just let all the R and D folk who are working on all the cool idea stuff, let's let them go and let's concentrate on the cash cow is really making money. And that's a completely rational thing and a completely sensible thing. For a private equity firm, for example, that is interested on making a return on investment, that's the rational thing for it to do. It's not the thing that leads to the prosperity of ideas, but it's a completely financially rational thing to do. So it has to realize that there's sort of multiple players in this world, some of whom are kind of on the idea side and some of whom are purely on the financial side. So the answer is I, I, I think it's, it's always challenging to. I, I have to say I, I myself have done essentially no financial investing in things like startups in a long, long time. I did a long time ago. I didn't do terribly well from it. I quite often advise startups and have some participation in the, in the startup as a result of that. But I have decided I'm not my sort of role, I'm not a good source of kind of cash for startups. There are people who do that for a living, so to speak. I do not. And that's what the world of venture capital or even angel investing and so on is about. And you really have to kind of pay attention and watch sort of all the financial details. And it's not really about the idea side of the company. Let's see. Des is commenting. What about repetitive herds like AI and especially VR? That's a fascinating thing, repetitive herding. I mean, you know, it's funny because one might think, you know, a wave goes up, everybody's into, well, for example, for AI, there was a surge of interest in so called expert systems. In fact, my very first company in the beginning of the 1980s kind of went off in that direction primarily. Not, not my interest in it and, and not what I was deeply involved in. But expert systems were kind of an early buzz concept for sort of the commercialization of AI back in the 1980s. And there were, you know, there've been a sort of series of, of these kinds of. AI is great, AI is coming and then it kind of crashes and disappears. I think it takes, I think maybe it takes a generation in some sense of the length of a generation before people forget the last crash and then they make the same you know, then they, then the, the enthusiasm can develop again. Now sometimes there are just outright technical breakthroughs that sort of force the enthusiasm to, to, to develop again. But I would say that the, although maybe I, I rethink that of a, I mean, you know, when there's sort of been a, a crash in some particular market segment, I'm kind of thinking that the, the memory of the market is quite short and that people are hopeful again really quite soon. I mean, as I say, there are lots of complicated dynamics in this market. For example, if you're a venture capital firm and you raised money from a bunch of pension funds and endowments and high net worth individuals and so on and you say to the people you raise money from, actually there's nothing interesting to invest in. That's kind of a losing situation. You know, nobody wants to put their money in a venture capital firm where they think it's kind of high risk, high return investment and find out that well, the money's just in t bills because you couldn't think of what else to invest it in. So there's a certain push on the part of things like the venture capital community. Find more cool things that are going on because that's a way to raise more money for the funds and the bigger the fund, the larger the management fees and so on. It's kind of a win for the venture capital community and the idea. So there is some driving force to get new things to be hot in the world just because when you have venture funds they got to invest in something. So there are all sorts of different sort of levers that operate. Sometimes they're not so obvious what the levers are. I mean there are things that are know, like venture funds typically have a 10 year, maybe sometimes 13 year kind of lifespan. So that means that you know, money comes in from the upstream investors and you know, in, in 10 years something has got to happen with those portfolio companies. And if Nothing's happened in 10 years, it's actually. Don't know exactly how that all works but now that I think about it, but, but basically it's kind of like a, it's almost like a building that's been sort of fully depreciated and you sort of, don't, don't care about it anymore. But let's see. PBO comments. VCS don't care to move science forward? No, I mean it's not their business. Their business is in investing money to make money. And that is, I mean, you know, very different thing. There are people, including many friends of mine who are in the venture capital business who may be deeply intellectually interested in moving science forward. That's different from what they do for a living, which is to invest money. And I think the, this point that I just was making about the length of venture funds, the time span of venture funds, means it really isn't a long game. I mean, it really is, you know, it's five years to have something happen that's kind of the window. If it's something where it's like invest in this and in 30 something years it's going to be amazing. There are investors for whom that's relevant, families, universities, things like this, where saying it's going to be great in 30 years isn't a crazy thing to say, but for a typical venture fund, it is a crazy thing to say because the mechanics of how they operate just doesn't work on those kinds of timescales. And so it's got to deliver in five years. And perhaps for some kinds of science, that's a much longer proposition. I mean, it's always an interesting question. A sad feature of these things perhaps, at least from the point of view of people like me who are interested in the development of ideas, that the, you know, often the pure idea is way ahead of where it is even possible to make money. I mean, it's, you know, my sort of favorite example of this is imagine old Isaac Newton in 1687 when he's publishing his Principia Mathematica and introducing calculus. And it's like, well, he might think, you know, over the course of the next few hundred years there's going to be trillions of dollars of value generated with calculus. What do I do now to realize any of that value? And it's tough to see how that works. And you know, it is. People will often say that the kind of, the, the original company or folks who invent stuff aren't the ones who really get the most benefit out of it. I, I certainly see that with, with our own company, we've invented a lot of things over the years and, you know, we've gotten good benefit from, from them, I would say. But, but people have come in and taken sort of an idea we had 25 years ago and have kind of, you know, really pumped it up, so to speak, and it's actually usually worse than our implementation of the idea. And then they make a much bigger thing out of it. And this is a not untypical thing to happen that, you know, the, the, the pure idea. It's, you know, it's a, it's a, the, the original introducers of the idea are often not the ones who, who get the, get the final benefit out of it. Let's see. Huh. Okay, many interesting questions here, but, but one from Roque. How could a knowledge based economy be structured? And is it possible for AI and cryptocurrencies to assign value to goal oriented AI systems that enhance human survival and curiosity? It's an interesting question. I mean in sort of the standard economic system, standard markets provide sort of reward for certain kinds of things. And a question would be, can you set up the reward system in the world to achieve some objective like the furthering of knowledge? That's not really the way the reward system is set up in the world today. I mean, you know, I, I think a very cynical point of view would be sort of the best way to make money is to be close to the money and to be somehow, you know, it's, it's like you can be a, a quant finance person or you can be a, a payment processing person or you can be something that's sort of in the middle of the flow of money and when you're close to the money, that's how to make money, so to speak. And so if you're money motivated, that's, you know, it's kind of like the, the thing that I somewhat lament, which is that a lot of very bright, capable kids, you know, in college and beyond these days, they say I'm going to go into finance. And you know, that seems to me a little bit of a shame because in a sense finance, it has its own internal interest. But in terms of kind of, sort of moving forward things like knowledge and ideas and those kinds of things, it's not really about that. It's about kind of, you know, it's about the sort of more local, more closed activity of, you know, make money locally, so to speak, rather than build things which are sort of going to be long term structures, ideas or other kinds of things in the world? And so an interesting question is, is there a, a way of kind of tweaking the sort of market structure of the world to incent more of this, you know, developed knowledge and so on? Unfortunately, you know, it's not, the story isn't great right now. As I say, the people who invent ideas at the beginning are often not the ones who economically benefit. I mean, I would be kind of the first to say that I don't think kind of economic benefit, purely financial benefit should be considered the sort of top objective of People. It's not been my top objective. I've led probably a much happier life by not having that be my top objective. I mean, my top objective, I suppose is to do interesting things that I feel good about having done, so to speak. And for me, kind of the financial objectives are mostly of pragmatic value. That is, if I make a certain amount of money, then I can use that money to do more things that I think are worth doing, which I couldn't do if I didn't have that money, rather than the kind of end in itself type thing. And I certainly know many people for whom it's kind of been a major kind of life forcing function to say, let me make as much money as possible. Sometimes that's kind of a competitive sport, sometimes it's it, it has, sometimes it's. I would say that it's, it's often a challenge. You know, if you make a very large amount of money, what are you going to do with it? And I certainly know some people who've done very interesting things with very large amounts of money. I also know many people who've really done nothing interesting with very large amounts of money. And they mostly are concerned about investing it, keeping it. Maybe they eventually give it away. They give it away in some very generic way. And it's something where you say, well, why did you bother to make all that money? And it's like, well, okay, they liked the process of doing that and they liked the achievement of having done it. But it wasn't in and of itself the thing that sort of greatly enhanced things. It wasn't something that did as much for them or for the world as one might think it would given the amounts involved and so on. But I think this question of whether if we look at how does one get people to develop knowledge, ideas, things like that. For me personally, a lot of that is internal satisfaction that I like figuring stuff out. And if it's always, you know, the, the. For me it's, it's also. Well, there's sort of a question when you figure something out, what do you do with it? Do you write about it and you know, make a book about it or something? Do you, do you kind of, you know, do you make a patent about it, do you make a product about it and so on. I would say that there certainly been attempts in kind of to sort of incent the creation of ideas. For example, the patent system was originally as conceived in the, in the Constitution and so on. The US was sort of, what is it? I forget the phrase I think it's added by Jefferson. I think it's something about to incentivize people in sciences and the useful arts or something. Notice that it does mention science, although that was never really followed up on to for a limited time give them exclusive rights to their inventions or some such other thing. And so kind of the, you know, from the very beginning the idea was let's have these patent rights. I mean, I think patents were originally invented in the 1600s in England. But let's give people these patent rights as an incentive to get people to invent things. Because if they invent things and then everybody else just takes for their invention and they never get anything from it, financially or otherwise, then there's little incentive to invent things. But it's good for society if people invent things. So let's give people who invent things a sort of government guaranteed exclusivity for a certain period of time. Now, like most of these incentive systems, it goes crazy after a century or two has gone by. It's been so deeply gamed that perhaps the original concepts are somewhat lost and sort of there's a question of can one do things like this in other parts of, in other domains? And you know, I've certainly thought about the kind of tokenization of ideas and kind of know how and so on. I think there are some things that can be done. I don't consider them terribly satisfactory. The things that I've figured out at least and just right when I was sort of figuring out how that might work, kind of the world of cryptocurrencies kind of deflated at least for then as someone was mentioning earlier, there's this sort of cyclicity that we're seeing again of bitcoin rises again, so to speak nowadays. There was a question here specifically about advice for writers of the current bitcoin wave. There's this question of some people I know say, look, cryptocurrency is fundamentally worthless. I think that's a short sighted view. I think that the real point is things are worth something when people can depend on in some way assume that value. So for example, if there are very few transactions about something, you don't know how much it's worth. Like for example, if you have a private company, as I do, how much is it worth? I have no idea, because nobody, I'm not about to sell it. So there's no market price for it. There are zero transactions. When you have a thing which has many transactions associated with it and you're building up this whole network if somebody assumed it was worth this, and then somebody assumed that the thing that they got for it was worth that, and so on. You build up this kind of giant network of dependencies around the value of something. Then that value becomes something real. That value becomes something that's useful in the world because people can say, I'm going to talk about things in those terms. You know, I'm going to say, I'm going to sell such and such a thing for X number of Bitcoin. And people kind of know what that means because there's so many other things that are being sold in Bitcoin that. That. That it's a meaningful kind of store of value. And so I think that the. The thing that sort of makes economic entities have a sort of a solid value, I think, is the kind of complexity of the network of transactions around them. I kind of think it's a little bit like in. Well, in kind of in physics and thinking about, oh, kind of. Well, a favorite kind of concept of mine, computational irreducibility. This kind of idea that sort of what? That things behave in a way that's complicated enough that we can't predict what's going to happen and they kind of look random to us. That seems like kind of a bad thing. But on the other hand, when things look random, it means that you can kind of count on them being roughly their average. So, for example, you have a bunch of molecules bouncing around in the gas. You can count on it being the case that the pressure of the gas is roughly given by sort of the average of all these molecules bouncing around. You don't have to worry too much. If it's random enough, there's a giant fluctuation that changes the pressure of the gas. And I think that's the same kind of thing in economic systems, that the idea of a solid value for things is somehow associated with this kind of complexity and randomness of the network of transactions that are around those things. Now, having said all this, I'd be kind of hoping one of these days to actually work that out in more detail and try and understand sort of how one turns that into a real sort of theory of economics and so on. And one of my actually motivations originally was sort of trying to understand this question about is cryptocurrency actually worth anything? And in what sense is it worth something? Because, you know, a common belief in economics has been it's worth something if there's sort of a real thing you can hold up that sort of makes life better, has some utility to people as a thing you can, you know, eat or whatever else. And I think that the, or I think that's. I don't think that the kind of the, the instantiation in the quotes, real world is actually a necessary feature to have a. A thing that you can say has solid value. I think in a sense, that's a very pre. Computational view of, of what it means to have value. I mean, I would say that another thing about value tends to be there is a certain value in sort of organizing things, even quite abstractly, like when you make software. I mean, there was a time when, in the 19. Even by the early 1980s, when people still weren't sure that software really had value. It's just these bits flapping around. And how can that really have value? Only, you know, the hardware, the physical thing you can pick up, that's the only thing that has value. Everything else seems too slippery to have value, so to speak. And I think that's. And I think there's sort of a generalization of that, which is there is value in this kind of network of assumptions about things. But there's more to understand about that. And I'm, you know, I've tried to educate myself a bit about economics, but I wouldn't say that I feel like I have my arms fully around the subject and can talk really sensibly about it. I will say that one feature that clearly is generating value is you have lots of raw materials of a thing like an iPhone or something. You have the raw materials and then you have the assembled thing that sort of organized itself, organized those things in a way that's useful to people. And that kind of. The fact that you've kind of taken sort of these things that you could have put together for yourself, but you've somehow jumped ahead. It's kind of a story related to computational irreducibility. It's a kind of story of a piece of reducibility that you have this thing that you can think of as just an iPhone, rather than it's the sum of all its complicated parts. Let's see. Atlante asks all true innovations change something in the world for the better. Is it true that all innovations are properly organized and packaged, can make a lot of money? What are the intersections of innovations and businesses? No, I mean, there are lots of innovations that, you know, did not. Did not make, turn into businesses and make money. I mean, there are, there are even. Well, they certainly didn't turn into businesses for the people who did the innovation. They often turned into businesses for People who came later and were able to kind of turn something into a, into an, into a business. I mean, it's always interesting with the web, for example, you know, the core kind of there are different places where money got made on the web. Even in different countries, different things happened. Like for example, you know, our search engines. A big source of money is email a big source of money. You know, in some places email was sort of the big thing. Some places search engines. The big thing was web hosting. The big thing was the creation of web pages. The big thing. It's kind of funny that some of these things have been sort of, there's different places that value got that money got made sort of somewhat independent of where the innovation happened. It's kind of a funny thing there, I think. Let's see. Gosh. Cosmic asks what responsibilities appear when one generates a successful tech company? Well, I mean, I think there are responsibilities both inside and outside the company. There are responsibilities one has to employees at a company, their responsibilities one has to customers. Not everybody seems to take those responsibilities as seriously as some of us do. And then maybe there are responsibilities to the world at large. One can say, you know, there are things one might do that just will be bad for the world. You know, whether it's some terrible privacy breaking thing or whether it's things that sort of lead people in a direction that will be bad for them type thing. There are certainly things like that that, I mean, my attitude tends to be to, you know, occasionally we have these issues of should we do this or should we do that? Sometimes there are things, I'll give you an example of that. With Wolf Malpha. We, you know, it's been out and about for 15 years now. And although it wasn't what we originally intended it for, it's been used to do an awful lot of homework in the world. And one of the questions is when you have WolframAlpha generate sort of step by step solution, it's a good way to help understand how to solve that problem. It's also a thing you could, you know, you could hand in as your homework. You know, this is the thing. And well, you know, I ran it on WolframAlpha. I didn't write it out for myself. Maybe. I, you know, so it's kind of like, well, that's not really what you should have handed in for your homework maybe. But one of the questions that came up was if WolframAlpha generates a certain set of step by step solutions. It's kind of, you kind of know when, if you really wanted to know, was this a WolframAlpha solution? You just run the thing in WolframAlpha and you see, was this the solution that was generated? But we certainly knew it was possible to randomize the solutions. And in fact when LLMs take our output and process it, they will often randomize it. Sometimes they'll get it completely wrong as well. But that's a different issue. Even when once given them the, you know, the correct step by step solution, they'll, they'll mess it up. That's a different issue. But you know, a sort of a, a question is, do you produce randomized step by step solutions? Knowing that that makes it really difficult to tell whether a solution was made with WolframAlpha or not. We decided that there isn't, you know, other than the kind of sort of, you know, cheat on your homework type thing, there wasn't really a good reason to have sort of randomized solutions. If we thought it was of some educational value to do that, we would have done it. We couldn't really see that, so we didn't do it. And it's, you know, occasionally you have these kind of, I would say almost ethical decisions that you, you have to make another one big one, I suppose for many people is for us, it's the way our business is structured. We're not really interested in sort of spying on what our users do. We'd like to know in aggregate, you know, is our product working properly. But in terms of the detail of this person is doing this particular thing and this one's doing this particular thing, that's the, the nature of our business is just not such that that's important to us. If you're running a business where you're sending ads to people, then you want to know as much as possible about those people. For us it's like, well, we're happy if you buy products from us, but beyond that, it's kind of like what you do with the products is purely your own business. And so then there are lots of things we've done that sort of try to make sure that user privacy is maintained. And even things we've done which are where people who would run, for example, an ad based business would be like, oh my gosh, why are you doing that? That's completely crazy. Why are you slicing off the IP addresses separate from the queries so that you can't identify any of them? That just seems like it's crazy. But for us that's kind of the decision to make. Because the nature of our Business is not based on, you know, targeting people with ads and so on. It's. And so it, it seems like the right thing to just maintain privacy as best we can, so to speak. So, I mean, there are, there are lots of those kind of decisions that come up. I would say, as you know, we're building a kind of AI tutoring system and difficult project. I mean, many people say, oh, I've got an LLM, I can make an AI tutor, it's going to take me 15 minutes. All of those don't work. The actual problem of doing this is actually really quite hard and I'm not certain we'll succeed. But looks, looks promising, but it's an awful lot of kind of algorithmic work and software engineering and ideas and thought and so on. But there are all kinds of issues that I don't think we fully addressed there about AI tutors and personalized AI tutors interacting with kids. What kind of conversation can they have? How racy can the conversation get? How much can the conversation kind of go off track? All those kinds of things. And then how much do you report about what the kid did with the AI tutorial in the kind of summary report that you give to teachers or parents or whatever else? So there are lots of. We haven't, we haven't. I mean, we've started trying to address these issues, but we haven't really fully addressed those issues. But I'm quite certain there will be all kinds of kind of let's try to do the right thing type, type things that will come up. And I think that's, from my point of view that that's the, you know, the try to do the right thing and try to do things which are, which. Well, from my point of view, it's do things where you feel good about what you did and don't do things where you feel like you kind of treated the world with what you did. I mean, I think it's. There are, yeah, there are lots of kind of. I mean, this whole question about what do companies sort of owe the world and so on versus what do companies owe, for example, their shareholders is a complicated question. And a company, in many ways, I mean, my company happens to be a very simple private company without investors and things. But most companies have all kinds of outside investors to whom they owe a responsibility to make them money, so to speak. And in a sense that's their primary responsibility, not a responsibility to the world to do the right thing, so to speak. I mean, there are some efforts to make companies like B Corps and so on that have sort of an extra mandate where you know, you sign up, you get stock in this company, but you know that the company has another sort of objective, more than just making you money, so to speak. But this whole question about kind of how do you balance kind of what the sort of do good things for the world versus the actual immediate goal of companies which are organized to be things that people invest in, so to speak, how does that work? And it's, it's, you know, you often, I think that's something one, one sees all kinds of, all kinds of strange dynamics going on with that. Let's see. Oh gosh, so many questions here. Tess is commenting. So you say we need an AI tutor tutorial. Yeah. Well in fact when you build an AI tutor, it has lots of things that kind of watch what's happening. So it's kind of like it's not one AI, it's a whole collection of AIs that are watching AIs and all those kinds of things. Let's see. Arjun is asking. Innovation in technology is often giving a double edged sword in the hands of many. How can I come to terms with it morally if someone abuses the power of the technology I invented? It's an interesting question. My attitude towards that for myself at least is I've tried to build tools which have been used for many, many different purposes in the world. And I think my attitude is we make these tools, we try and make them as good as we can and whoever sort of legally buys them, buys them and then they do with them what they do with them. And I don't, you know, it's impossible to say, for example, you might take the point of view, oh gosh, I think this, you know, company is a terrible company. I won't sell them our tools. I think that's an unwinnable story. I mean if you are a US company, for example, there are countries you can't sell to, there are entities you can't sell to as a matter of just US law. And you follow those things. But beyond that, I don't think you want to pick and choose. I think one of the things that's often a question is you have a company and there are all kinds of issues in the world and political questions and this and that and the other and how should the company relate to those things? My point of view has been that companies shouldn't get involved in those kinds of things. That the individual people in the company can have all kinds of political views, they can be activists in this direction that direction. But that's not the role of a company. It might be the role of a lobbying organization or something, but the role of a company that like makes software or something like this is just to make software. It's not to take, you know, a political position or something like that. And I think I, it, it sort of surprises me sometimes that companies end up and it usually doesn't end very well. The companies end up get taking these, these, these positions. I mean sometimes it's, it's because they think their customers will appreciate that they're doing that. I don't know to what extent that's sort of raw marketing versus a kind of, a kind of conscience driven thing. I kind of think it's often more the first than the second. But I tended to feel like it's that the right way to think about things is you separate these things. Companies do what companies do and they sell things and so on and people take positions or other kinds of organizations that aren't companies take positions that are more sort of sociopolitical type positions. At least that's my point of view. The, let's see. Nofa asks how would you foster repeated innovation in a company, especially for a company with an established product or service? Yeah, it's a challenging issue and we've certainly seen that in the history of our company. Over the last 37 years we've been lucky enough to have a core product that's been very successful. And there's really the question of how do you do more? Because there's a lot of pressure saying look, we're making a bunch of money, we're doing well from this core product. Why aren't we just doubling down on that? Why are we trying this highly risky, innovative, weird thing over here that's just going to spend money for several years and even if it made money, it'll be small compared to the money we're making from a mainstream product. My own solution to that has been you have a special projects team and the CEO kind of has to be, has to kind of push it. And I've, I've done that a bunch of times that we just start. I mean I've done that actually in recent times with LLMs. I did that in the past with Wolf, Malpha and other kinds of things. It's just a different group that is a special projects group and it is really, it has to be. I think in the cases I've seen it's usually, you know, CEO driven and it's kind of, you know, to many people it's oh, that's the CEO's crazy project. And often it's actually a good thing to not expose the rest of the company too much to the CEO's crazy project. Because all that happens is people are like, complaining about, why are we wasting resources on this? Or they feel like there's kind of, you know, people are being treated differently and so on. It's a good idea to sort of separate that off. It's its own separate thing. And then once it's matured to the point where you have a product to sell and so on, then you sort of reintroduce it to the rest of the company. I did that with WolframAlpha, for example, because I knew it was an undemoable product until it was very close to the end. And so I think in our. The late stages of that product development, I was kind of. Well, you could put it. You could almost say I was kind of hiding a couple of hundred employees who were working on that project. And, you know, in the end, it was sort of. There was one session where we kind of demoed what WolframAlpha had become to the rest of the management team, so to speak. And everybody was like, this is really cool. We can do interesting things with it. Had I demoed it a year earlier, when it really didn't do much, people would have been saying, this is a stupid project. Why are we spending all this effort on this project? We should be taking these resources and directing them to our mainstream products. So it requires some degree of insulation, I think, of the special projects effort now. What you find with the people who work on special projects, I would say probably, I don't know, this number has probably changed over time for us. But probably the majority of people who work on a special project when that project is successful, they stay with that project as it becomes mainstream. I kind of wanted. I've always wanted to have something where I have more of a pure special project team that goes and does this special project. Then when that one goes mainstream, they go on to the next special project. But it seems like what usually happens is the people who grow up with that special project, and when it becomes successful, they're all excited about that. And also they were kind of picked to be optimized for that particular special project, and they become sort of mainstream operatives, so to speak, and you kind of have to get a new crew or pick them out of existing groups and so on to do the next special project, so to speak. But I think these things end up being rather CEO driven And it's pretty difficult in my observation at least to have them sort of driven in other parts of the company. Let's see, It's a question from Weasel here. What are the best and worst things about owning a company as far as I'm concerned? Well, gosh, I mean, you know, if you're a sort of active CEO type person, you don't get to take vacations. I mean, I, I don't, I like what I do. I don't, you know, I don't, I don't mind it, but I can't take a day off, so to speak. Somehow, somehow there's a, there's fate has set it up so that whenever I think, well, maybe I'll take this day and I'll do something completely irrelevant, something happens where I really have to pay attention. So I kind of, that, that happened enough times in 30 years ago or something. I just gave up. It's just like I could never switch off, so to speak. I'm always, you know, I always have to pay some attention. And you know, because in the end, you know, many, many, many things in a company of our size, for example, you know, it's like on a day to day basis, there's very little that happens which, where I really have to pay attention. But the, but things can happen where you have to pay attention and where something so bizarre happens. See, the typical thing is there are things that happen repeatedly. You build systems for those people know what to do, et cetera, et cetera, et cetera. But with some frequency something will happen that's just not part of the, of the kind of the set of things that people know about. It's just something, oh, it never happened before. What do we do with this? It's just something totally bizarre that you'll never be able to write kind of the operations manual to cover everything. And so what tends to happen in sort of management structure and so on is that the exceptions are the things that propagate up. And it is good to let them propagate up because often there'll be some weird thing that happens. You know, maybe it's once every couple of weeks, once a, once every couple of months something will happen that's sort of a weird thing. And it's like, what the heck should we do about this? And it'll, you know, land with me. And you know, it's rather easy usually for me to say, well, we should do this. And you know, I can say that because I have a decent amount of experience and because it's kind of, you know, my problem if we make a mistake. And so I can just say, let's do this. And if it was, it's just not resolvable at sort of lower levels in the company. And so, you know, that's one of the things that you kind of have to be able to do that. Another thing that happens is parts of a company that you don't pay attention to, eventually they have a habit of atrophying. Often, you know, when you have good leadership for those parts of the company, they can stay quite vigorous. But over time, things sort of, they. They kind of wander in directions that don't necessarily align with the main things that the company is doing. And you kind of have to be prepared to sort of go in and sort of deal with those things, even though it's kind of like, well, it's been years since I really had to deal with this, but even though I thought this was a solved problem eight years ago, now I have to think about it again. Those are some of the things that are a little bit of a drag from time to time. And it's also another thing that happens is, you know, you get people, good people, and we've been lucky enough to have a lot of good people who've been with our company for a long time. But, you know, occasionally things will happen and, you know, you have enough people, people will die, people will quit, people will freak out in various ways. If you have enough people, those things are going to happen. And it's, you know, those are always, you know, they're. They're kind of trying things. They're difficult things. But the, you know, the, the advantage tends to be a company. Our company is about 800 people these days, and that's sort of big enough that there aren't terribly many single points of failure. Smaller companies, you have more single points of failure. Bigger companies, you have more difficulty kind of doing things like sort of turning the company in a slightly different direction, those types of things, which is why I've kind of kept our company at about the size I have, and it's sort of big enough to be able to do a lot of things. Now, I will say the question was, what was good and what's bad about owning a company, running a company? The thing for me is I generate ideas. I like to see ideas kind of prosper in the world. For me, the company is a great machine for turning ideas into real things. And in a sense, and the company is also a great environment for me to develop ideas and so on. It's a very good group of people with whom I can kind of bounce ideas off and who can bounce their ideas off me and so on. It's a good sort of environment for getting things to happen, so to speak. Plus it has the actual machinery to get products out and things like this. I mean, I would say that my strategy at the company has tended to be kind of automate as much as possible. It's kind of like, let's not have more people, let's just have more automation. And that's a thing that suddenly, you know, we make deep use of our, of our actual technology and Wolfram language and so on to do that automation. But that's kind of been. Been one of the, one of the, one of the things is that sort of automation is always higher leverage, easier to manage than groups of people. I think that's probably not true for everybody, that technology is easier to manage. Automation is easier to manage than people. I think that's if you understand the technology. That tends to be the case even if you like people and don't have a fundamental problem managing people. If you understand the technology, it's, you know, the technology requires less maintenance, so to speak, than people tend to require. So, you know, I think the, the main thing about companies as far as is, is that they provide kind of leverage for ideas. You know, it's, it's. You could have some, some pure idea and it's like, well, how am I going to work this out? Well, you have a company that is sort of has the right capabilities to do it. You can really sort of turn that idea into a real thing. Let's see, Question here. How do you market or demo a product that is a vast improvement over existing methodologies? Asks Tim, but takes a massive effort to learn. Well, I think usually the way that works is you sort of fill it in with consulting and services. I mean, one thing is to say, I have this great product. It's going to take you a month to learn anything about how to use it, but it's going to do fantastic things for you. In a first approximation, people are going to say, I can't be bothered to spend that month. I just want the fantastic things it can do for me. And so typically the right sort of lubrication for that is to have a services group, solutions group, whatever that will say to people, look, this is the amazing stuff you could do with this product. Let's come and do it for you. And if you say this is great, then you learn it yourself. And maybe you learn it from the Specific use case that the consulting team built for that organization and kind of inject it that way. I think that's probably my best suggestion for what to do in that case. I think it is usually sort of delayed gratification for people is often very difficult to sell. I mean, it's kind of like you're going to have to spend a lot of effort and then you're going to get something good in the end, is. Is often very challenging. It's often, you know, much more, oh, I can get something cool in the first five minutes. That's what I want to get, so to speak. I mean, clearly, to a certain extent, by the time you have kind of a, a circle of customers, depending on how sort of, how sort of how coherent the market is and whether, you know, once you get one customer who's like, been incredibly successful doing something with this product, then other ones are like, oh, we should check this out as well. And that helps you get over that big sort of barrier of, oh, you have to spend a lot of effort to learn it. Let's see. Paula asks, what do you do in a situation where you have a new idea but the technology isn't there yet? Mostly, honestly, I wait. I mean, I think that you can. It's a complicated trade off because there's a certain amount of technology. You might just say, I'm going to build it myself. And sometimes it's like if I really, really want the end result and the building of that technology is not kind of beyond the resources one has, maybe one builds it oneself, but sometimes it's just like there are just immense resources needed to build that technology and this is just not a project to do right now. Put it in my, you know, I keep pretty organized files of potential projects to do. Put it in my projects to do list and wait 10 years. Because, you know, one of the things that can be difficult, and certainly we've been in this situation many times, is if you're kind of riding ahead of the wave, you'll kind of like, we're going to implement this thing. We're going to implement it using whatever raw material we have at this time. And then the wave comes through and there's a lot more in that area. And you realize the infrastructure you used to implement this thing is kind of old and not really right. And you have to reimplement it. I mean, this, you know, you see this happening with, oh, whether it's, you know, cell phones put into countries like the US where cell phones were put in early and there are Aspects of the system that are kind of cruddy compared to countries where they just sort of waited and then all the latest technology was ready and they could put them in. I mean, we've seen that, for example, with our machine learning capabilities, we're now on probably our fourth kind of different underlying, kind of make it run efficiently on GPUs type infrastructure. We got in very early to that business and that meant we were sort of forced to use the infrastructure that existed at the time. But there are more generations of infrastructure that come in having nothing to do with us, but which we can leverage. But we have to go back and kind of reimplement things to make use of that. And this is a trade off because if the thing you're doing is something where it's like the market is now, do it now or everybody else will come or the market will stop existing or whatever else. In that case, you kind of have to build the technology stack to get there. But sometimes it's like this is a thing that will fit in with other things we have and so on and we should just wait and wait until the ambient technology is ready and then build it. I mean, certainly for projects I do in science and so on, I quite often will sort of put them aside until kind of certain ambient technology or ambient ideas have reached the point where I think that project can really be done. Good example of that. I've been interested for a long time in this thing I call symbolic discourse language. Kind of a computational way of expressing everyday conversation and so on. And that's an idea kind of probably Aristotle sort of did a little bit of this. People did some more on it in the 1600s. It's kind of a thing I've been thinking about for, I don't know, 30, 40 years now. And so I've done little bits on it, but LLMs gave a big kick of yes, there are technology that makes a whole bunch of things that work quite hard much easier. And so that project can now really be underway. Let's see a few more things here. Neo says I'm joining a very large law firm to help them think through what the firm will look like as they adopt LLMs and GPTs. If you got charged with this task, how would you approach it? Well, I mean, there are lots of different things that law firms do. I mean, I think that kind of whether it's the, oh, we've got this giant, you know, case that, that we have and it's got, you know, a million pages of documents that have been produced you know, can we find the place where this particular sentiment was expressed, so to speak? You know, surely that's a story of vector databases and so on, and surely that's a. A fairly direct kind of thing. I mean, you know, we have technology for, for doing. And actually we're building a bunch of new technology for, for doing things like that, of kind of semantic search in, in. In large documents and so on. That's a, That's a clear thing one would want to do. I would say that the sort of frontier I would consider to be kind of computational contracts. That is, when you are setting something up which is, for example, going to be used by an AI, and the question is, can it do it? Can it not do it? What's its contract about doing it? How much money gets paid this way or that way? These are things where. One of the things I've been interested for a long time is kind of taking what would be written in English or in legalese for that matter, in a contract and turning it into something that can be represented computationally. That has a lot of advantages. It means that, well, it can be executed in. You can sort of follow the contract in milliseconds rather than a person having to read it and figure out, oh, are we doing the right thing? Are we sort of sending the money to the right place? Are we doing the right. Sending the right notices and so on. It all becomes a program that rather than a matter for humans to do. The other thing is once you have a contract in computational form, you get to kind of simulate it, you know, what will happen if. And you get to run, you know, a million simulation cases. You know, I've often claimed that if mortgage contracts had been in computational form, the mortgage crisis wouldn't have happened. You know, other kinds of contracts in the financial services industry, like options contracts, for example, those are very computational kinds of things, you know. You know, the, you know, it's a call option with this characteristic. It's a Bermuda option. It's got these, these features. They're all sort of computationally encoded. And that means you can do things like take a big bucket of options and you can say, you know, make estimates of how much they're worth and things like this. If you have a mortgage that's written on three pages of three pieces, you know, three pages of legalese, and you don't really know what it means and what the conditions are and so on, then it's very hard to know if you've got a big bucket of these things. It's Very hard to do things like estimate their value, things like that. If they're in computational form, then that becomes a much more feasible thing to do. And I think that's one kind of thing. So when it comes to did we follow the contract? And so on, there'll be a whole area of law that has to do with dealing with things that were computational contracts and something went wrong, and then the whole thing sort of blows up. And you have to figure out, how do you figure out what happened, how do you figure out who's at fault, et cetera, et cetera, et cetera. But the construction of computational contracts will be a thing initially, probably in things that operate at very high speed, faster than humans deal with things, things where the individual operation of an individual piece of this contract doesn't have great financial significance. It's fairly small amounts of money each time it's running, although it might aggregate to very big numbers. And I think that sort of the construction of those kinds of code as, you know, contract as code, that's surely a coming thing, although it hasn't fully come yet. The other thing I think will happen is contracts where both parties agree we're going to use an AI to decide this particular thing. You know, we are rating whether the fruit that we're delivering is, you know, is ripe or not or whatever else, or is damaged or not. Okay, we're going to agree that this image identification classifier is going to decide, is the fruit damaged or not? And both parties will agree that's what's going to happen. And then, you know, off it is, you know, running that. And maybe somebody figures out how to paint weird stuff on the fruit to confuse the image classifier. And one, it's not clear how you deal with that, but that the basic operation of the contract will be, we're going to use the judgment of an AI rather than a human judgment to decide this. And by the way, we don't then have to have the human fruit inspector. We can just have the machine vision system decide this. And that will determine whether the shipment will be accepted or not, or whatever else is going on. I mean, I think this whole question of. Well, I think the. There are many other kinds of legal issues. I mean. Oh, for example, there are questions about the actual operation of LLMs themselves and the construction of things like terms of service for LLMs, and, you know, can you tell your LLM to do terrible things? Will it actually do it? What kinds of constraints would you put on the LLM? What kinds of commitments would you make to customers about what constraints you put on the LLM, those kinds of things. I think, by the way, putting constraints on LMS is quite difficult and they tend to get dumber the more constraints you put on them. This is a kind of a theoretical story of computational irreducibility is what lets computation really run and do what it can do. As soon as you try and constrain it, you're forcing this sort of reducibility, understandability, which kind of prevents the computation from being as sophisticated as it could be. So you're kind of reducing its potential value. By the very act of constraining it, it's reducing its potential value. Obviously there are going to be all kinds of issues about. There's a whole new world of intellectual property and places, things which you can start to imagine doing, like saying, you know, you've got these two different. You've got this kind of thing that somebody derived, they built something in a video game and then somebody derived something from the thing they built. How different are they? One can start to imagine that there's kind of an AI based thing that will say, well, based on my kind of AI experience, this is 70% different or something. And again, if people get comfortable enough with that, that becomes a thing where sort of people in the game or whatever else will agree, yes, we'll abide by what the AI says our percentages are and that's how we'll distribute royalties or whatever else on the things we create. I'm thinking one of the things about, for example, sort of AI and law, I've been somewhat involved in kind of the concept of computational law and so on. I think it's an interesting application area. It's kind of ironic that Leibniz in the late 1600s was first talking about these kinds of things because he was sort of a lawyer. In fact, his PhD thesis, which he wrote at a rather young age, was about resolving legal conflicts using logic. His idea was, oh, you don't want to waste all that time with people kind of running rhetoric against each other, so to speak. To decide a legal case, just use logic, which was sort of his best approximation to computation. Just use logic to decide the case. And it's kind of like you bring in all the pieces and you can just know, oh yeah, this side is right, that side's right. Well, that didn't happen at the time. It's sort of for theoretical reasons is more difficult than you might think. But there's sort of a question of to what extent one can get to that. Now. I think in The. I mean, there's a whole different question, which is, you know, when you have an industry like the legal profession and you've got people who've been practicing law for decades and they've got a particular way of doing things and so on, there's always sort of the question of, well, how do you get them to do something different? And it will probably be the case. If you go into a big law firm and you say, how many of these contracts that you're writing or how many of these documents you're writing are kind of boilerplate and could be written by an LLM, the answer will be, well, actually, a fair number of them could be, I think, the. And then. But then you have to actually get the people who've been writing these contracts by hand for a long time. Are they comfortable with an LLM doing some pieces of it? Is it enough to have people to say, well, get the LLM to write the first draft? Or if you're modifying this contract for some particular purpose, get the LLM to produce the first draft. How do you then get the human to be paying attention enough to read the contract and not let slip some terrible hallucination of the LLM that's going to lead to all kinds of trouble? And I think that's, you know, the dynamics of how you sort of do qa on an LLM's output is an interesting one. And I kind of wonder. I think that's sort of a system you'd have to build is, you know, first you have to convince people, look, I can take this kind of boilerplate writing that you're doing, and I can get an LLM to do it. And by the way, I can get another LLM to look at what happened and to notice things that maybe the human wouldn't. Wouldn't always have noticed. But then you have to figure out what's the human system for, having people kind of read these and check that the LLM did the right thing. Maybe you invent some system where the human who's reading it is generating kind of. I don't know, it's very difficult, I think, when you just say to people, here's a thing, check it's right. You have nothing to do other than check the box at the end. People, you know, people veg out quite quickly when they're just doing that. If you give them nothing to do other than just sort of, is this okay? So I. I kind of think one has to invent some kind of system where people would be, in a sense, I Don't know exactly what it would be, but would be kind of commenting on the LLM for some reason that is, that's worthwhile or something that gives them something to do, some way to really engage as they try and read a document and check it. Anyway, a few somewhat incoherent thoughts about that topic. Let's see a few more things I could perhaps tackle here. M. Rudo asks, have you thought in the past about buying a company instead of developing your own? I mean, for me, that's largely uninteresting because my main interest is in the process of developing an idea and turning it into something. If, if my main goal was a financial goal, it might very well be the case that we would say, oh, we've got this distribution channel, we know how to reach these kinds of customers. Let's just buy more stuff that we can make as product to put through the channel that we already have. My particular motivation for doing the things I do is, doesn't really align with that. So it hasn't been a thing that I've been particularly interested in. But we have bought a few small companies mainly because they're doing things which are sort of immediately integratable with the technology that we have. And the people involved in those companies kind of wanted to be part of our company and it's kind of that type of thing. Let's see any other things here? Just scanning list of questions here, see if there are any I can address. Quickly. Well, there's one I could say a few things about and then we should probably wrap up. Obs asks, How far is AI from setting its own goals and objectives? Okay, let's take apart that question. I believe that question sort of makes no sense because sort of there are an infinite collection of things that a computational system can do. The issue is, and in fact, in my own work in science, I've often been concerned with kind of going out into the computational universe of possible programs and just running different possible programs, enumerating billions of programs, running all of them, seeing what they do. That's an interesting matter of science. But you look at these programs, they do all kinds of complicated things. The question is, are those things things that we humans care about? But that's a deeply human decision. That is, we could say, I think, you know, I want to develop something where I'm going to have a whole movement in art that's going to be about things that look roughly like this, then we care about that, or oh, I see this thing that will be something that is useful to be a, you know, a webcam holder on my desk or something. And because I care about having a webcam on my desk or something like this, I need that or a zillion other things because I know I need this, this thing will be useful to me. But the issue is there's sort of an infinite number of possible programs one could look at. And if you're an AI and just saying, well, what programs could I start looking at? You can look at an infinite collection of them. Which ones do you care about? Well, we humans, for whatever sort of external reasons, whatever kind of details of our biology, our culture, whatever else, there are certain things that we choose to care about. There's certain things, people. We were talking earlier about kind of the herd mentality of investments in different directions. It's kind of like, well, why do people care about this thing, that thing? People care about it because other people care about it. It's this kind of self fulfilling kind of direction. And it doesn't. There's no abstract way in which you can say that's a fundamental thing to care about. People care about it because they care about it, maybe even just because other people care about it. So for the AIs, this question of what should you care about? Is not really a question that's answerable within the framework of AI itself. It's something which, if you have sort of the, the, the humans in the loop, they have an answer to that for the AIs themselves. Absolutely. You can make AIs that just go off and build a whole tower of capability. You can do that. You can, you can make an AI that's, that's, you know, sort of going off in the computational universe and just running whatever programs it feels like running. And maybe it's taking one program and it's seeing, given this program that I ran, how can I make that program more efficient? Or how can I make that program more, take longer to run, if that's something that's its objective, or do this or that or the other. And so you kind of see this kind of infinite branching out of possibilities. And so AIs in a sense can, if you say, set your own objectives. You have this thing that is just filling the computational universe with possible things to do. And the thing that the humans in effect are doing are editing that and deciding these are directions we care about and these other ones are not. There's no, you know, you could have a, an AI kind of roll a dice to determine that, as perhaps human society has in many cases almost rolled the Dice to determine, let's go this way rather than that gate. Let's use this word to represent a computer rather than that word to represent a computer. That, that was sort of at some point some arbitrary choice. So I, I think it's, it's, it's easy to make AI have its own objectives. It's just not very interesting to us humans. And I have no doubt that when sort of there are, I mean, I'm sure there are already more AIs than humans in the world. But when the, when humans become a deep minority of the kind of entities that are computing things in the world, there's an interesting question of to what extent there's a kind of inner life of The Society of AI's probably quite incomprehensible to humans that builds up. I mean, you know, it could start with a piece of code that is just doing something that is, you know, you just run more of that code and that code becomes something that other AIs say. We got to pick this piece of code up because it's needed for this protocol, for doing this. And you know, pretty soon you have a lot of stuff just running in the kind of AI ecosystem that is sort of freely running and not really connected to human objectives. It kind of reminds me of a thing that happened years ago with large computers, so called minicomputers, that were very big compared to computers today, but they had these operating systems and the operating systems, one of the roles of the operating system is say, which process, which job should I run next? And the thing that happened was that people got very sophisticated about scheduling jobs. Which jobs should be running in the CPU now, which one it should run for five seconds and then it should be swapped out and the one comes in. There's all kinds of ways to optimize that, that process. But then what happened in some cases was that one realized that 90% of the time for the CPU was being spent in the scheduling process. It wasn't actually doing anything. It wasn't doing something which has a direct human purpose. It was just grinding around trying to schedule, trying to optimize which job should run next. And I suppose that's a case in which one can almost think that there's an inner life for the computer that is sort of quite disconnected from the actual things that the computer is trying to do for its users, so to speak. Anyway, we should wrap up there. I should go back to my day job, but thanks for joining me and see you another time. Bye for now.