The State of AI with Marc & Ben
01:13:53 Share

Most of the content create on internet is created by average people. And so kind of the content on average, you know, as a whole on average, is average.

The test for whether your ideas, that is, how much can you charge, can you charge the value or are you just charging the amount of work it's going to take the customer to put their own rapper on top of OpenAI.

The paradox here would be the cost of developing in a given piece of software false. But the reaction of that is a massive surge of demand force of our capabilities.

And I think this is one of the thinks always been underestimated about humans, is our ability to come up with new things we need.

There is no large marketplace for data. In fact, what there are, there are very small markets for data in this wave of A I big tech has a big compute and data advantage. But is that advantage big enough to draw out all the other startups? Trying to rise up well in this episode is exciting. I co founders, mark in rison and then horwitz, who both, by the way, had a front receipt to several prior tech waves, tackle the state of vi.

So what are the characteristics that are defined successful AI companies? And is preparatory data the new oil? Or how much is IT really worth? How good are these models realistically gonna get? And what would IT take to get one hundred times Better? Mark ban, discuss all this and more, including whether the venture capital model needs a refresh to match the rate of change happening all around IT. And of course, if you want to hear more from benchmark, make you to subscribe to the ban and mark podcast. Alright, let's get started.

This is kind of the darkest side of capitalism. And a company is so greedy, though, they're illy, to destroy the country. And maybe the world like to get a little extra profit.

And they do IT like the the really kind of thing thing as they cry out. It's for safety. And we've created alien that we can control.

But we're not going to stop work on. We're going to keep building in the fastest weekend, and we're going to buy every fricking GPU on the planet. But we need the government to come in and stop IT from being up. This is literally the current position of google and microsoft are now is crazy.

The content here is for informational purposes only, should not be taken as legal, business tax or investment advice, or be used to evaluate any investment or security, and is not directed at any investor or potential investors. In any a sixteen c fun. Please note that a sixteen z and it's a filly.

It's may maintain investments in the companies. Discuss in this podcast for more details, including a link to our investments, please see a sixteen z dot com slashed disclosures. Hey folks, welcome back. We have an exciting show today. We are going to be discussing the very hot topic of A I.

We are going to focus on the state of A I is as the exist right now in April twenty twenty four, and we are focusing specifically on the intersection of ai company building. Um so hopefully this will be irrelevant. Anybody working on a anybody at a larger company.

We have, as usual, solicited questions on x formerly known as twitter, and the questions have been fantastic. So we have a full line up of listen questions and we will dive right on. So first questions for three questions on the same topic. So Michael asks, in anticipation of upcoming AI capabilities, what should founders be focusing on building right now?

When asks, how can small A I startups compete with, establish players with massive compute and data scale advantages, when alisa mckay asks, for starters, building on top of OpenAI eeta, what are the key arteries tics of those companies will benefit from future expensive improvements in the base models for as those that will get killed by them. So let me start one point. Men, joy to you.

So sambol man recently gave an interview, I think, maybe like three minor one of the podcast, and actually said something I thought was actually quite helpful. Let's see bad if you agree with that. He said something along the line in self, you want to assume that the big foundation models coming out of the big I companies are going to get a lot Better.

So you want to assume they are onna, get like other times Better. And as a start of founder, you wanted them think, okay, if this current findings models get a one hundred times Better is my reaction, oh, that's great for me and for my start up because I much Better off as a result or is your action the opposite is an, oh, shit. I'm in real trouble. So we just stop right there are and see what you think of that .

as gentle advice. I think generally that's right. But there is some nuances to IT, right? So I think that from sams perspective, he is probably discouraging people from building foundation models, which I don't know that I would tirely agree with that, and that a lot of the set of building foundation models are doing very well.

There are many reasons for that. One is there are architectural differences which lead to how smart is a model there is. How fast is a model.

There is, how good is the model in the domain when that goes for not just text models, but, you know, image models as well. There are different domains, different kinds of images, that response to prompts differently. If you ask my journey and I agram the same question, they react very different, like you, depending on the youth cases of their tune, poor.

And then there is this whole field of distillation where, you know, sam can go build the biggest smart st model the world. And then you can walk up as a start up and kind of do a dispelled version of IT and get a model very, very smart at a lotless cost. So there are things that, yes, the big company models are going to get way Better, kind of way Better at what they are.

So you need to deal with that. So if you're trying to go head head full front of assault, you probably have a real problem because they have so much money. But if you are doing something that's differently enough, or I like, you know, different of men. So for example, you know at data bricks, they've got a foundation model, but they're using a very specific way in conjunction with there kind of leading data platform.

So okay, now if you're an enterprise and you need a model that knows all, they like new answers of how you're enterprise data model works and what things mean and needs access control and what needs to use car specific data and the main knowledge and so forth, IT doesn't really hurt them if sams model gets way Better. Similarly, eleven labs with their voice model has kind of embedded into everybody. Everybody uses that as part of kind of the A I stack.

And so it's got kind of a developer hot into IT. And then you know they're going very, very fair to what they do and really being very focused in their area. So there things that I would do like extremely promising that are kind of us sensibly, but not really competing with OpenAI or google or microsoft. So I think IT sounds a little more course screen than I would interpret. IT, if I was building and started right.

was sticking with a little bit. Start with the question of do we think the big models, the god models.

are going to get one hundred times Better? I kind of think so. I am not sure. So if you think about the language models, let's do those because are probably to people are most familiar with. I think if you look at the very top models, you know claude and OpenAI and mistral llama, the only people who I feel like really can tell the different saves users among those models are the people who study them. You know like they're getting pretty close. So you know you expecting we're talking one hundred Better that one of them might be separating from each other a lot more, but the improvement so hundred percent Better in what way like for the Normal tourist and using in a Normal way like asking IT questions and finding out stuff.

let's say some combination of just like breadth of knowledge and capability.

Yeah like I think in something may are you right?

But then also just combining with like sophistication of the answers, you know, sophistication of the output quality, the output sophistication, the output, the lack of pollution, ation sexual grounding.

Well, that I think is for sure going to a hundred times Better like that. Ah I been there on a path for that. The things that are so against that, right the alignment ant problem were okay yeah they're getting smarter but they're not allowed to say what they know.

And then battlement also kind of mix some dummer in others SE. And so you do have that thing. The other kind of question mates come up lately, which is can do we need to break through to go from what we have now, if I would, category es.

Artificial human intelligence is opposed to artificial general intelligence, meaning it's kind of the artificial version of us. We've structured the world in a certain way using our language and our ideas and our staff, and it's learn that very well. amazing.

And I can do kind of a lot of the step that we can do, but are within the asm tote or you need to break through to get to some kind of higher intelligence or general intelligence. And I think if we are the asm tote, then in some ways IT won't get one hundred times Better because it's already like pretty good relative task. But yeah, like I will no more things. It'll pollute vate less on all those dimensions itll be a hundred times. But everything there's .

this graph floating around. I forget exactly what the access are, but it's basically shows the improvement across the different models. And to your point, that shows an asm tote against the current tests, the people are using this sort of like at or slightly above human levels yeah which is what you would think if you're being trained an entirely data.

Now the counter argument on that is how the test just to sample right is a little, little bit like the question, should I run the S. A. T, which is, if you have a lot of people getting a hundred ds know in both maths and burbo on the S T, is the scale too constant you need to test that can actually test for einstein.

N memorized the test that we have.

and it's great, right? But you can imagine S A T that like really can detect gradations of people who have like ultra high I Q who are ultra good at math or something. You can imagine test for A I you can imagine test the test for reasoning about human levels.

Yeah, well, maybe the I needs write the test.

I need to right the test. Yeah, there's a related question that comes up a lot, an argument we've been internally with provocative, but. This question that comes up, which is okay, you taken all, am you traded on the internet? What is the internet data? What is internet data corpus? It's an average of everything, right? Is a representation of sort of human activity.

Representative activity is going to know because of the sort of distribution of intelligence, the population most of IT somewhere in the middle. So the data set on average out of represents the average and you're taking to be very average. Yeah, you're taking be very average because most of the content created on the internet is created by average people.

And so kind of the content on average, no as a whole on average is average. And so therefore the answer is our average, right? You're going na get back.

And the answer is sort of represents the kind of thing that are average chanted by q you definitely to a hundred q index, hundred at the center of the bulk of and so by definition, you're kind of getting back the average. I actually argue like that. Maybe the case for the default p today.

Like you just ask the thing, does the earth revolve around the sun or something? You get like the averages of that. And maybe that's fine.

This gets to the point is, well, okay, the average data might be of an average person, but the data set also contains all of the things written and thought by all the releasing people. All that stuff is in there, right? And all the current people who are like that, their stuff is in there.

And so then that sort of like a prompting question, which is like, how do you prompt IT in order to get basically in order to basically navigate to a different part, what they call the latent space to navigate to different part of the dataset that basically is like the super genius part. And you know, the way these things work is if you craft the process in a different way, that actually leads IT on a different path inside the data uses you. A different kind of answer is another example this, if you ask IT right code to do x right code, sort lister, you know, whatever a render image IT will give you average, go to do that.

If you say, write me secure code to do that, you will actually write Better code with fear security holes, which is very interesting, right? Because there's accessing a different purpose of training data, which is secure code. If you asking your right the image thing the way john carmack could write IT, you get a much Better result because it's tapping into the part of late and space represented by john carmo, who is the vast graphic h programmer in the world and so you can imagine prompting craft in many different domain such that you're kind of unlocking the late and super genius yeah even if that's not .

the default answer yeah now so I think that's correct. I think there is still a potential limit to its smartness in the so we have this conversation in the firm the other day where you have there's the world which is very complex and intelligence kind of is, you know, how well can you understand, describe, represent the world, but our current iteration of artificial intelligence consists of human structuring the world and then feeding that structure that we come up with its into the A I.

And so the A I kind of is good at predicting how humans have structured the world. I supposed to how the world actually is, which is, you know, something more probably complicated, maybe the in reduced able or what have you. So do we just get to a limit where, like IT can be really smart, but its limit is going to be the smart students, I suppose you smarter than the smart as humans kind of related.

Is that going to be able to figure out brand new things, new laws of physics? And so far, of course, there are like one in three billion humans that can do that or whatever. That's a very rare kind of intelligence.

So IT still makes the AI extremely useful, but they play a different role. If they are kind of artificial humans, then that if they're like artificial, you know, super, super magical humans. yes.

So let me make the sort of extreme bowing ase for the hundred because OK. So the cnc would say the sam Allen would be saying they are going to a hundred times Better precisely if they're not going to.

Yeah, yeah, yeah, yeah yes. right.

Because you've saying that basically, in order to scared people in .

the competing well like that, whether or not they are going to get one hundred times Better, sam would be very likely to say that would like sam, for those you don't know him, is is a very smart guy, but for sure he's a competitive genus. A question about that. So you have to take that account.

right? So if they we're going to a lot Better, he would say that. But of course, if they, we're going to a lot Better, to your point, he would also say that.

But also, yes, why not? right? And so let me make the bull case that they are going to a hundred times Better or maybe even know on an up per cr for a long time.

And there is like enormous controversy, I think, on every one of the things i'm about to say. But you can find very smart people in the space who believe basically everything i'm about to say. So one is there is generalized learning happening inside the real networks.

And we know that because we now have introspections techniques where you can actually go inside and look inside the neal networks to look at the neural circuitry that is being involved as part of the training process. And you know, these things are evolving, in general computations functions. There was a case recently where somebody trained one of these on a chest database.

You just my training, a lots of just games and actually computed model of a chest ort, you know, inside the neural network and you know, that was able to do original moves. And so the neural net k training process does seem to work. And then specifically, not only that, but no meta and others recently haven't talking about how so called overtraining actually works, which is basically continuing to train the same model against the same data for a longer, putting more, more compute ycl against IT. I've talked to various my people, the fields, including there are who basically thinks that actually that works quite well. The diminishing returns people were worried about, about more training.

They proved in the new lamon release, right? That's a primary technique they use yeah exactly like one guy .

is very basically told me basically is like that we don't necessarily need more data this point to make these things Better. We maybe just need more computer ycl. We just trained in a hundred times more, and I may actually a lot Better.

So on day of lading IT, turns out that supervise learning ends up being a huge boost to these things.

yes. So we ve got that. We've got all of the kind of Alice rumors and reports of various kinds of self improvement loops that had a underway.

And most of the sort of super advance spectin ers in the field think that there is now some former self improvement loop that works, which basically is you basically get an A I to do is called china thoughts. You get IT to basically go step by step to solve a problem. You get IT to the point, I know how to do that.

And then you basically retrain A I on the answers. And so you're kind of basically doing a sort of a fourleaved upgrade across cycles of the reason capability. And so a lot of the experts to think that sort of things started to work now.

And then there's still a raging debate about synthetic data, but there's quite a few people who are actually quite yeah then there's even this trade off. There's this kind of dynamic where like l ms might be OK at writing code, but they might be really good at validating code. No, they might actually be Better validate and .

code than they are ready .

and there would yeah but that also means like a air may .

be code make of your code and .

and we have this antiproton hic bias is very deceptive with these things because you think of the models and IT. So it's like how could you have an each that's Better reality code, the writing code, but it's not an yet what IT is is, is this giant late? The space is this giant neil network.

And the zero would be there are totally different parts of the nearby network for writing code and validating code, and there is no consistency requirement once, however, that the network be equally good at. thanks. And so if it's Better, one of those things, right? So the things that is good at might be able to make the thing that is bad at Better and Better.

right, right, right, sort of self improvement thing.

And so then on top of that, there's all the other things coming right, which is it's everything is all these practical things, which is there is an enormous chip constraint right now. So every AI that anybody uses today is its capabilities are basically being gated by the availability of chips, but like that resolve over time. There's also some point of like data labeling.

There is a lot of data these things now, but there is a lot more data out in the world. And there, at least in theory, some of the reading A I companies actually paying generate your data. And by the way, even like the open first data set, they're getting much Better.

And so is a lot of like dating improve that are coming and then there's just a lot of money poking into the space to be underwrite all this. And then by the way, there's also just the systems engineering work is happening, right, which is a lot of the current systems we're basic. We are built by scientists and now the really world class engineers are showing up and tuning them up into the work Better. And maybe that's .

which makes train what, by the way, way more efficient is, well, not just inference, but also turning exactly. And then even.

you know, another improvement area is basically microsoft release there, five small english monday yesterday. Apparently it's competitive as the very small model competitive of a much larger models. And the big thing they say that they did was they basically optimize the training set.

So they specially the duplicated the training that they took out all the copies, and they really optimize on a small amount training data. Un, a small amount of high quality training data is close to the largest amount of low quality data that most people train on you at this up. And you've got eight or ten different combination of sort of practical and theoretical improvement vectors that are all in play. And it's hard for me to imagine that some combination of those doesn't lead to like really drag improved from here.

I definitely great. I think that's for sure gonna happen, right? Like if you are so back to sams proposition, I think if you are a start up.

And you're like, okay, and two years as I can get as good as GPT for, you shouldn't do that, right? But IT, that would be a bad mistake. Well.

this also goes to, you know, a lot of our entrepreneurs are afraid of giving example. So a lot of you're trying to figure which is okay. I really think I know how to build a sap that corners is an alarm to do really good marketing clatter. Al was just make a very similar, a very, very same thing.

And so I build a whole system for that will IT just turn out to be that the big models in six months will be even Better in making marketing clatter just from a simple prompt, such that my apparently sophisticated system is just irrelevant because the big model just does IT. yes. yeah.

How we stuck like that, like that here. You know, the other way you think about IT is the criticism of a lot of current A I APP companies is the court you GPT rappers. There are sort of thin layers of wrapper around the core model, which means the core model commoditize member display them.

That the counter of argument, of course, is it's a little bit like calling all you old often are apps database rappers, rappers around a data abase IT. Turns out like actually rappers on database is like most modern software. And a lot of that actually turned to be really valuable. And there IT turns out there's a lot of things to build around the correction. So yeah, so but how do we think about that when we run in .

the companies thinking you about building apps? Yeah, you know it's a very tRicky question because there is also this corrections gap, right? So you why do we have copilot? Where are the pilots? right? Where are the I that there is no A I pilots.

There are only A I co pilots. There is a human in the lop on absolute everything. And that really kind of comes down to this, you know, you can't trust the AI to be correct.

And during a picture or writing program or dinner, even like this, writing a court brief without making observations, you know, all these things kind of require human and kind of turns out to be, like, fairly dangerous. not. And then I think that so what's happening a lot with the application later, people saying, well, to make IT really useful, I need to turn this co pilot into a pilot.

And can I do that? And so that's an interesting in hard problem. And then there's a question of is that Better done at the model level or at some layer on top that you know kind of teases the correct answer out of the model, you know by doing things like in code validation or what have you? Or is that as something that the models i'll be able to do? I think that's one open question.

And then you know as you get into kind of domains and you know potentially represent things, I think there's a different dimension than what the models are good at, which is what is the process flow, which is kind of in databases rotis. So on the databases kind of analogy, there is like the part of the task in a law firm that's writing the brief, but there is fifty other tasks and things that have to be integrated into the way a company works, like the process flow, the orchestration of IT. And maybe there are, you know, on a lot of these things, like if you're doing video production, there are many tools.

Are music even right? Like, okay, who's gonna ite the lyrics, which A I will write the lyrics and which I I will figure out the music. And then like how does that all come together and how do we integrate IT and so forth? And those things tend to you are just require a real understanding of the and customer and so forth in a way.

And and that that's typically been how like applications have been different than platform in the past is like there's real knowledge about how the customer using IT wants to function that doesn't have anything to do with the kind of intel IT or or is just different than what the platform is designed to do. And to get that out of the platform for kind of company of a person turns out to be really, really hard. And so those things, I think, are unlikely work, you know, especially if the process is very complex and something it's funny as a firm you know, wear a little more hard work, technology or ended.

And we've always struggled with those. You know in terms of all this is like a some progress application for like plumbers to figure out this. And we're like, well, where's the technology? But you know a lot of IT is how do you encode um you know some level of domain expertise and and kind of how things work in the actual world back into the software. I often think i've .

intel founders that you can think about this in terms of Price. You can kind of work back with pricing a little bit, which is to say sort of business value and what you can charge for, which is, you know, the special thing for any technology to do is to kind of say, I have this technological capability and i'm on to say to people and like, what am I going to charge for? IT is going to be a summer between, you know, my cost are providing IT and then you whatever mark up, I think I can justify, you know, if I never monopoly, maybe the market is infinite, but technology forward, the supply supply forward to Price model.

There is a completely different pricing model for kind of business value backwards. And so you know so called value pricing, uh, value based pricing and and that you to your point, that's basically a pressing model that says, okay, what's the business value to the customer of of the thing? And if the business value is, you know, a million dollars, that can I charge ten percent of that and get one hundred thousand dollars right? Or or whatever? And then you know why is that cost one hundred thousand dollars as compared to five thousand dollars? Is because, well because to the customers worth a million dollars and so theyll pay ten percent .

for IT yeah actually. So a great example that I grieve got the um competing in our portfolio. Christa ee, that does things like death collection. okay.

So if I can collect way more debt with way your people with my in a copilot type solution, then what's that worth? Well, it's worth the heck of a lot more than just buying an open a eylan ense because IT OpenAI license is not going to easily collect debts, uh, where kind of enable your death collector is to be massively more efficient or that kind of think so. So it's bridging the gap between the value.

And I think you had a really important point, the test for whether your ideas good is how much can you charge for IT? Can you charge the value or are you just charging the amount of work it's gna take the customer to put their own rapper on top of OpenAI. Like that's the that's the the real text me of like how deep and how important is what you've done?

yeah. And so to your point of life, the kinds of bit of the kinds of businesses the technology investors have had a hard time kind of thinking about, maybe accurately, it's ort of it's the company is a vender that has built something where is a specific solution to a business problem or IT. Turns out the business problems very valuable to the customer. And so therefore, they will pay a percentage of of the of the of the value provided back, back, back in the terms for the Price for the offer or and and that actually turns that actually turns out you can add businesses that are not very technologically differentiates, that are actually extremely lucrative ah um and and then and then because because that business is so lucrative, they can actually afford to to go think very deeply about how technology grates into the business, what else they can do. No, this is like the story of a sales first star comfort example.

right? And and by the way, there's a kind of A A chance of theory that the models are all getting really good. There are open source models. They are like that are awesome. Yeah, lama mistral, like these are great models. And so the actual layer where the value is gonna crew is gonna a be like tools, orchestration, that kind of thing, because you can just plug in whatever the best model is at the time, whether the models are gonna be competing, you know, in the death battle with each other and you know, be commoditized down to the, you know, the cheapest one wins and that kind thing. So you know, you could argue that the the best thing to do is, is that kind of connect the power to the people.

So that actually takes us to the next question. And this is a two and one question. So Michael asks, and these are and I say these are diametrical oppose, which is why I pear them.

So Michael asks, 玩， making huge investments in general A I startups when it's clear these strops won't be profitable anytime soon, which loaded, loaded question, but it'll take IT. And then kr asks, if A, I defeats the cost to building a startup, how will the structure attack, investment change? And of course, good band.

This course I could you just set. So basically, the questions are dimensions opposed. Because if you squint out of your left ee right, what you see is basically the amount of money being invested in the financial model companies kind of going up to the right at a serious pace.

You know, these companies are raising hundreds of million, billion, tens of million dollars and IT just like, oh, my god, look at to this sort of capital, sort of I don't know in fero, you know that hopefully will result to buy you at the end of the process. But my god, look how much money being even these things, if you want through your right eye, you know you think, wow, that now all of a sudden it's like much sheer to build software r it's much sheer to have a software company. It's much easier to like have a small number programmer writing complex soft because they have all these A, I copilot.

And all these development capabilities are coming online. yes. And so on the other side, the cost of building an A I like application start up my crash might just be the like, you know, the sales first, the A I sales first, I M might cost, you know, at ten or one hundred, three, A A thousand and amount of money that I took to build, you know, the old database driven sales first. And so yes, so what what do you think of the dichotomy, which is you can actually you can actually look out out of either I and see either cost to the moon. As as like for start up funding or or cost actually .

going to zero. Well, like so IT is interesting. I mean, we we act of companies in both camps, right? Like like I think probably the companies that i've gone to profitability the fastest maybe in the history, the firm have been eye companies, have been A I companies in the portal, whether revenue grow so fast that IT actually kind of runs out ahead of the cost.

And then there are like, you know, people who are in the foundation model race who are raising hundreds of millions, are even billion of dollars to kind of key pace and so forth. They also are kind of generating revenue at a fast rate. The headcount in all of them is small.

So you know where AI money goes and even, you know like if you like, at OpenAI, which is the big spender in start up world, which you know we are also investors and is yes, he can rise. They're pretty small against their revenue like IT is not a big company. Have count.

Like if you look at the revenue level on how fast I got ten there, it's pretty small. Now the total expenses are gorgeous, but they're going into the model creation. So it's interesting thing.

I mean, i'm not entirely sure how to think about IT, but I think like if you're not building a foundation model, IT will make you more efficient and probably like get to profitability, cracker. right? So the the the counter .

and and this is a very blish counter argument, but the counter argument to that would be basically that falling costs for like building new software companies are a maros. And the reason for that is the thing and economics called the jew's, which i'm going to range from wikipedia.

So the jevon s paradox occurs when technological progress increases the efficiency with which the resources used, right, reducing the amount of that research necessary for any one years. But the falling cost induces increases in demand right in lcs, icy enough that the resource use overall is increased rather than reduced. Yes, it's certainly possible, right? And so this is, you see versions of this, for example, you build in your freeway and then that actually makes traffic of worse, right?

Because basically what happens is, oh, it's great. Now there's more roads. Now we can have more people live here.

We can have more people, you know, we can make these companies bigger and others more traffic than ever. Now traffic is even worse. Or you saw classic examples during the industry revolution, coal consumption, as as the Price of coal drops.

You people use so much more coal that actually the overall, overall consumption actually increased, people getting a lot more power. But the result was was the use of a lot more coal in the paradox. And so that the paradox here would be, yes, the cost of developing in a given piece of software fall, but the the reaction of that is a massive surge of demand for software capabilities.

And so the result of that actually is, although IT even IT looks like starting to offer companies, the Price is gonna, all actually it's going to happen is going to rise for the high quality reason that you're going to be able to do so much more, yeah right, with software that the product is going to be so much Better and the roadmap is going to be so amazing of the things you can do and the customers are going to be so happy with that they're going to want more and more and more. yes. So the result of the and by the way, other, other example of you have, first, I splaying out in another related industry is the hollywood C.

G. I. And theory should have reduced the Price of making movies, in reality, is in Christian, because audience expections went up, yeah.

And now you go to a holly's wen movie. And g. wald. I, and so the movies are more expensive to make them ever.

And so the result of IT know so, but the result in hollywood is at least much more the, say, visually elaborate, you know, movie is where they're Better. And that is another question. But like much more visually elaborate, compelling kind of visually stunning movies through C. G, I, the version here would be much Better software. Yeah like radical, Better software to the end user, which causes and users to want a lot more software, which causes actually the Price .

of development to rise. You know, if you just think about like a simple case, like travel, like, okay, booking a trip through expedia is like complicated, you're likely to get IT wrong. You're click on menu and this and that the other and like, you know, in a eye version of that would be like, you know, send me to paris, put me in a hotel I love at the best Price, you know, send me on the the best possible kind of airline airline ticket and then, you know, like, make IT like really special for me and like, maybe you need a human go OK like we're going to, you know, or or maybe the AI gets far complicated ancient kay.

Well, we know the personal love shock let and we're going to like, you know, fedex in the best chocolate in the world, from switzerland into this hotel in paris and this and that the other and so like the quality, you you you can the quality could get to levels that we can even imagine today just because, you know, the software tools aren't aren't what they're gonna be. So yeah, that's right. Yeah, I kind of buy that actually like an argument.

Your body is how about yeah how about i'm in a land and whatever boston, at six o'clock I want to have dinner at seven with a table full of, like, super interesting people yeah.

right, right, right.

right. You know, yeah, right, right. yeah. You know, no, no travel have today.

No, no .

right.

Well, then you think about it's got to be integrated into my personal A I and like and this know there is just like unlimited kind of idea as that you can do. And I think this is one of the kind of thing that's always been underestimating humans is like our ability to come up with new things we need, like that has been unlimited.

And there there is a very kind of famous case where john matter, keen to the, the, the kind of prominent economist in the kind of first at the last century, have this thing that he predicted, which is like, nobody, because of automation, nobody would ever work a forty hour workweek, you know, like good, because once their needs were met, needs being like shelter and food. And you don't, even if transportation is in there, like that was IT, IT was over, and like you would never work past the need for a shelter and food. Like why I would you like there's a recently, but of course needs expanded.

So then everybody needed a refrigerator. Everybody needed a not just one car, but a car for everybody in the family, everybody the television set. Everybody needs you to like glorious vacations.

Everybody, you know. So what are we gna need next? I'm quite sure that I can't imagine. But like somebody, he's gone to imagine that. And it's quickly going .

to be become a need. Yeah, that's right. By the way, kings have famously was economic prospects for our grandchildren, which was specially to, mark said, another version of that full of the quote.

So the society, when we win the Marks as utopia, socialism, is, is achieved. Society regulator, general production, that makes the possible for me to do well. Well, to hunter, this is a quote to hunt in the morning, fish the afternoon, rear cattle in the evening, criticize after dinner.

What a glorious life. What a glorious life. Like if I could just list four things that I do not want to do, hand finished, rear cattle and criticize .

yeah right .

and by the way, that says a lot about Marks. That was also his four things. Well.

criticizing being his favorite thing. I think it's basically communism in the natural and I don't want to get two political but yes.

yes, hundred to percent. And so yeah sorry, this ah you think what they have came with kids and Marks this incredibly constructed, incredible, constricted view of what people want to do and then and then correspondingly, you know the everything is just like, you know people people who want people who want to have a mission. I think probably some people who want to fish and yeah but you know a lot of lot of people want to have a mission. They want to have a cause.

they want to have a purpose. They want to want actually a good thing in .

life that turns out IT turns IT turns IT turns out, yeah, IT was started international events. Okay, so yeah, so yeah, I think that i've long felt a little bit of this often. Is the world thing a decay ago? I I always thought, I always thought that basically demand for software is sort of perfectly elastic, possibly to infinity.

And the theory they are basically is if you just continuously ing down the cost of software, which he's been happening over time, that basically demand basically is is like basically correlates uh over and in the reasons because you know of as we've been discussing, but is kind there, there's always something else to do in software. There's always something else to automate. There's always something else to optimize.

There's always something else to improve. There is always something to make Better. And you know in the in the moment with the constraints that you have today, you may not, you know, think of what that ideas, but the minute you don't have those constraints, you will imagine what IT is. I just giving example that give an example of playing out A I right now, right? So there have been know we have companies to do this there.

There have been companies that have made A I there there made of our systems for doing security cameras forever, right? It's like for a long time, I was like a big deal to have some ford that would do like, you know, we have different security, cambridge es, to store them on A D V R and be able to reply them and having never face IT lets you do that, I was like, you know, A I security cameras all of the student can have, like actual, like thematic, acknowledge what's happening, the environment and so they can say, you know, hey, that's bad and then they can say, o, hey, you know, that's been but is getting a gun right? right? And by the way, that's Better he's Carrying a gun but that's because like he hunts on in thursday and friday as compared to that's mary and SHE never cares a gun and like you like something is wrong. She's really mad, right? She's got a yeah really steam depression on her face and we should probably be worried about IT, right? And so there is like an entirely new set of capabilities you can do.

This is one example for security systems that were never possible, P. A. I. And the security system that actually has a semantic understanding of the world is obviously much more sophisticated. Than the one that doesn't didn't might actually be more .

expensive to make, right, right? Well, I just imagine health care, right? Like like you could wake up every morning and have a complete diagnostic. You know, like how my doing today, like whether all my levels of everything, and you know how should I interpret them, you are Better than and this is one thing where A I is really good is yeah medical diagnosis because because it's a super high dimensional problem.

But if you can get access to you know your continuous glue, cos reading, you know, maybe sequence bad now, again, this and that the other, yeah, you've got an incredible kind of view of thanks. And who who doesn't want to be health here? Yeah, like now we have a scale that basically what we do yeah, maybe maybe jack your heart right or something, but like pretty primitive stuff compared to where we could go.

Yeah right. OK good. All right. So let's go to the next topic. So on the topic of data, so a major tom asks, as these A I models allow for us to copy existing at functionality, minimal cost preparatory data seems to be the most important mode.

How do you think that? Will that proprietory data value? What other motes do you think companies can focus building in this new environment? And then jeff wise hopped asks, um how should companies protect sensitive data, trade secrets from Price data, individual privacy in the brave new world of A I so let me let me start with a provocative let me start with a provocative statement and see if you agree with that, which is you start to hear a lot the sort of statement of class is like data, the new oil.

And so it's like, okay, did you know data? The key input to training AI making all the stuff work. And so there for data, basically the new the new resource is the limiting resources, the supervision thing.

And so you know whoever has the best data is going to win, and you see that directly in how you train the eyes. And then you know you also have like a lot of companies, of course, that are now trying to figure out what to do. They are and a very common thing you hear from companies as well.

We have prepared tory data, right? So I you, i'm a hospital chain or I know whatever, any kind of business, insurance company, whatever. And I ve got all this preparatory data that I can apply, that i'll be able to build things with my preparatory tory data.

A I, that won't just you be something that anybody will be able to have. Let me argue that basically, let's see. H, I let me argue like almost every case like that, it's not true. It's basically with internet kids would call cope. It's simply not true and and the reason is just not true is because the amount of data available um on the internet and just generally in the environment is just a million times greater um and so well may not you IT may not be true that I have your specific medical information.

I have so much medical information of internal for so many people in so many different scenarios that IT just wants the value of quote your data no IT is just like overwhelming and so your your propriety data, as you know, company act will be a little bit useful in the margin, but it's not actually gna move the little and is not really going to be to be to entering most cases. And then only site is proof for the for my my belief that this is mostly cope is there has never banned nor is there now any sort of basically any level of ort of racism histiaeus kettle for data market for data. There's no, there's no, there's no large marketplace for data in the, in, the in.

In fact, what there are, there are very small markets for data. So there are these call data brokers that will sell you you know large numbers of like IT, you know information about uses some internet or something and they're just small businesses like there are just not large IT just turns out like information and lots of people is just not very valuable. And so if the data actually had value, you know, I would have a market Price and you would see a transacting and you actually very specifically don't see that, which is sort of yeah sort of quantity proof of the data actually is not nearly as well. Those people .

can where I agree. So I agree that the data, like just as here's a bunch of and I can sell IT without doing anything to the data, is like massive Operated, like I definitely agree with that.

And like maybe I can imagine some exceptions like some you know special population genomic database or something that are that we're very hard to require that are useful in some way that and that's not just like living on the internet or something like that. I could imagine where that super highly structured, very general purpose and not widely available, but for most data in companies is not like that. And that IT tends to not they're widely available or not general purpose.

It's kind of specific. Having said that, right? Like companies have made great use of data. For example, a company that you're familiar with, meta uses its data to kind of great ends itself, feeding IT into its own A I systems, optimizing its products in incredible ways. And I think that you know us entries and horror actually, you know.

So we just raised seven point two billion dollars and not a huge deal, but we took our data and we put IT into an A, I system. And R, L, P, S were able there is a million questions. Investors have about everything we've done, our track record, every company we've invested in.

So for and for any those questions, they could just S, C, A, I, they could be wake up a three morning. Do I really want to trust these guys and go in A, C, A, I, A question? And boom, they get an answer back in some way.

They have to wait for us. And so for so we really kind of improved our investor relations product tremendously through use of our data. And I think that almost every company can improve its competitiveness through use of its own data. But the idea that is collected some data that I can go like sell or that is oil or or what have you that's yeah that's probably not true, I would say.

And you know it's kind of be interesting as a lot of the data that you would think would be the most valuable would be like your own code base and right, you're software that you written so much of that lives and get up. Nobody is actually I don't know of any company. We do work with the you know whatever thousand and software companies do we know any that's like building their own programing all on their own code like IT or and would that be a good idea? Probably not just because there's so much code out there that the systems have been trained on so like that, not so much of advances. So I think it's a very specific kind of data that would have value this.

Let's make an action then if I if i'm running a big company, like if i'm running an insurance company or a banker of hospital changes, I like that. How are the consumer package? F, C, or something? Like what? How should I validate? Like how should I validate that I actually have a valuable property or data set that I should really be focusing on using VS maybe versus in the alternate, by the way? Maybe there's other things. Maybe I should be taking all the effort I was spent on trying to optimize use of that data and maybe I should use IT entirely trying to build things using internet data instead.

Yeah so so I think I mean, look, if you're ready hear in the international business, then like all your actuarial data is both interesting and I I don't know that anybody publishes there are actual actual al data. And so like I am sure how you would train the model on stuff off of the internet. Can I change out?

So that would be a good thing. That would be a good test cause I an insurance company, i've got records on ten million people and actually tables, they get second, they die. Okay, that's great.

But like there's lots and lots of actual general actual da on the internet for a large scale populations? No, because governments collect the data, they process and they publish reports and there's lots of great. There's lots of academic studies. And so like is your is is your large data set giving you any additional actual information that the much larger dataset on the internet isn't already providing you? Like are you are your insurance science actually actually really any difference than just everybody?

I think so cos on intake on the do you know when you get insurance, they gave you like a blood test, they ve got all these things and off if you're smoker and so for and in the I think in the general data set, like yeah you know who dies but you don't know what the fact they did coming in.

And so what you really are looking for is like, okay, for this profile of person with this kind, with these kinds of lab results, how long do they live and that that's where the value is. And I think that you know interesting like go you know think about like a company like a queen base where right they have incredibly valuable assets in the terms of money. They have to stop people from breaking in.

They've done a masse amount of work on that. They've seen all kinds of break and types. I'm sure they have tons of data on that.

It's probably like weirdly specific people trying to break into crypto x changes. And so you know like I think that could be very useful for them. I don't think they can sell IT anybody. But you know I think every companies got data, uh, if you know, fed into an intelligence system will help their business. And I think almost nobody has data, they could just go sell. And then there is this kind of in between question, which is what data would you want to let microsoft or google or OpenAI or anybody get their gabbi little fingers on? And that are not sure that that that I think is a question enterprises are restless with more than it's not so much.

Should we go like solar data but IT? Should we turning on model just so we can maximize the value? Or should we feed IT into the big model? And if we feed into the big model, do all of our competitors now have the thing that we just did? And you know, where could we trust the big company to not do that to us, which I kind of think the answer on trusting the big company not to f with your data is probably, I want to that if your .

competitive .

depends on anybody do that, there at least reports .

that certain big companies are using all kinds of data that they should be using to train their models already. so.

Yep, I think like I think those reports are very likely true, right? Or or they'd have open data, right? Like this is now we've talked about this before, but you have the same companies that are saying they're not ceiling.

All the data from people you are are taking in an an authorized way, refuse to say open their data, like when that tells where your data came from. And in fact, they're trying to shut down all, open the snow, open to snow, open weight snow and data and open nothing. And the go the government trying get to do that. You're not a deef. Then why are you doing that right?

What are you heading? By the way, there's another twist and turn here yourself. For example, the insurance example, I kind of deliberately loaded IT because it's actually illegal to use genetic genetic data for insurance proposes, right? So there's a thing called the geno law genetic information that on discrimination act of two thousand and eight and basically IT IT, basically banks, health insurer in the U.

S. ROM. Actually using internet data for the purpose of doing your health assessment, actually a al assessment of of in which, by the way, and because now the genomics are getting really good, like that data probably actually is, you know among the most accurate data you could have if you are actually trying to predict like when people .

are going na a second die and they're literally there are literally not allowed to use IT yeah IT is I think that this is an interesting like we this application of good intentions are in a policy way that's probably going to kill more people than ever. Get saved by every kind of health. F D A is our policy that we have, which is.

You know, in a world of A I having access to data on all humans, why they get sick, what their genetics are, the data, the oil like that is about to the health care oil, as you know. If you could match those up, then we'd never not know why we're sick. You know, you could make everybody much healthier, all these kinds of things.

But you know, to kind of stop the insurance company from kind of overcharging people who are more likely to die, we've kind of locked up all this data. A kind of Better idea would be to just go OK for the people who are likely ted, like we subsidize health care like massively for individuals anyway, just like differential differentially sub, you know subsidized and you know and then like you solve the problem and you don't look up all the data. But yes, typical of politics and policy. And the most of them more like that. I think, yeah, there is interesting questions like insurance.

Like basically there one of the questions people asked about insurances. Like if you have perfectly predictive information, individual es does the whole kinds of of insurance actually still work, right? Because the whole theory of insurance risk pooling, right, you, you, you, you is, is precisely the fact you don't know what's going to happen in the civic case. That means you build these statistical models and then you risk pool, and then you have variable payments depending on exactly what happens. But if you literally knew what was gna happen in every case, because, for example, you have put all this predicted genomic data, then all of us said that would make sense to risk because you just say, well, no, this person gna cost to x, that person is going to cost why there's .

no help insurance. You already doesn't make sense in that way, right? Like insurance, the idea of insurance is kind of like the insert with crop insurance where, like, okay, you know, microbe fails. And so we all put money in a pool in case like microbe fails so that you know we can cover IT. It's kind of design for the risk for a catastrophic, unlikely incident like everybody's guy go to the doctor all the fucking time and some people get sicker than others in that kind of thing. But like where health insurance works is like all medical gets yeah paid for through this insurance systems, which is this layer of loss and bureaucracy and giant companies and all that stuff went like if we're going to pay for people's health care, just pay for people's health care, that what do we do, right? Like and if you wanted distance that people from, like going for nonsense reasons and just up the copy, like it's like what are we doing from from a justice and point .

from a fair stand point like what IT make sense for me, what IT make sense for me to pay more for your health care if I knew that you were going to be more expensive than like, you know him, i'm directly know if if, if everybody knows what future health care causes for person. Yeah, there is a very good predict model for IT societal willingness to all pool in the .

way that we do today. I might really diminish yeah yeah. Like you you can also, if you know, like there are things that you do genetically and maybe we give everybody pass on that, like you can try your genetic, but then like there's things you do, behavior like that like dramatically increases to the chance of getting sick.

And so maybe you know we intensified people to stay healthy instead of just like paying for them not to die. The there's a lot of systemic fixes. We could use the health care system and IT IT couldn't be designed in a more ridiculous sway, I think.

But I could we design a more ridiculous way? It's actually more ridiculous. And some other countries, but it's pretty crazy here.

Nathan Nathan od asks a one of the strongest common teens between the current native A I and web one point o and so let me start there. Let me give you a theory. You you think so. I guess this question, you know, because of my role with me at escape. No, get this a lot because of our role early on thy with the internet.

So you know the internet bone was like a major, major event technology and is still within a lot of you know people's memories and so know this sort of know people like to reason from analogy. So it's like, okay, the air boom must be like internet OOM. Starting a company must be like starting an internet company.

And you know what? What is this like? And we actually got a bushing questions like that, you know, that are kind of analysts tions like that, I actually think and on and then, you know, you, I were there for the internet boom.

So we live through that, the bus in the boom in the bus. So I actually thinks that the l gy. Doesn't really work, for the most part, worse in certain ways. But IT does not really work for the most part and the reasons because the the internet there, that was a network, whereas A I is a computer.

Yp.

okay, yeah so so so people to start.

we're saying so the like the PC boom or .

the PC people, even I would say the microprocessor like my best love is to the mice processor yeah, or even to the like the reasonable computers like back to the main frame. A in the reasons because he had looked at the internet did was the internet know obviously was the network, but the network connected together many existing computers, and then, of course, people built many other new kinds computers to connect to their net.

But fundamental their net was the network. And then and and that's important because most of most of the third of industry dynamics, competitive DNA ics, start up dynamics around the internet had to do with basically building either building net ks are building applications that run top of the networks and the internet generation start of cessile consumed by network fax, and you know about this part. So positive feedback loop you get when you connect a lot of people together and you know things like so called meta law, which is sort of the value network experience, the way expense is you more people to.

And then you know there are all these fights. These fights, you know all the social networks are whatever, fighting to try to get network of facts and try to do each other. These users because the network effects.

And so it's kind of is dominated by network effects, which is what you respect from from the network business. A I like there. There is a netware effects. N A, I, that we can talk about.

But um it's it's more like a microprocessor is more like a chip, it's more like a computer in the it's a system that basically right, if data comes in, data gets process, dat comes out, things happen. That's a computer. And information processing system is a computer, is a, is a new kind of computer. It's a, you know, we like to say the the the sort of computers up until now, we we're called anomalous machines, which is to say determinist I C computers, which is they are like you, hyper literal, and they do exactly the same thing every time. And if they make a mistake, it's, yes, the programmers fall, but they're very limited in their ability to interact with people and understand the world. You know, we think of a eye and large language models as a new kind of computer, a probably tic computer and their own network based computer that, you know, by the way, is not very accurate and is, you know doesn't give you the same result every time, and in fact, might actually argue with you and tell you that that doesn't and answer your question.

yeah, IT would trace very different in nature than the old computers. And that makes you kind of compatibility, the ability to build things, big things, out of little things, more complex.

right? But but the capabilities are new and different and and valuable and important because they can understand language and images, and you know all these those things that .

you see when you never solve with deterministic computers, we can now go after, right? Yeah exactly.

So I think I think that I think the analogy and I think the lessons learn are much more likely to be drawn from the early days the computer industry, or from the early days the microprocessor, than the early dates the internet does. Does that not right?

I think so. Yeah, I don't think so. And that doesn't mean there is no like boom and best and all that because just the mature of technology, you know, people get too excited and they make IT to depressed. So there will be some of that. I'm sure there will be over built out to you know potentially of eventide of chips and power and that kind of thing. Don't we start with the sort of but but I agree, like I think networks are fundamentally different in the nature of how they involved in computers and and the kind of just the adoption curve in all those kinds of things will be different.

yes. So then and this kind of goes to wear how I think the industry is going to unfold. And so this is kind of my best theory for kind of what happens from here of this kind of this giant question of like is the induction going to be a few god models are a very large number of of models of different sizes.

And so for so the computer, like famously, the the original computers, like there is onal, I be a mainframes. The big computers, there were very, very large, expensive, and there were only a few of them. And the prevAiling view actually for a long time, was that all there, whatever be.

And there was this famous statement by Thomas washin senior, who was the creator by bm, you know, which was the ominous company for the first, like, you know, fifty years, the of the computer industry. And he said he, he said, I I believe that I actually to hood. I don't I I don't know that the role ll ever need more than five computers. And I think the reason for that, I was literally IT was like the governments going to have two. And then there's like three big insurance companies and then that's IT.

Who else would need to do all that? Bh, exactly.

yeah. Who else would need? Who else needs to attract a huge amount numbers? Who who else needs that level of calculation capability? It's just not a relevant you know, it's just not not not a releve concept.

And by the way, they were like big and expensive. And so who else going to ford them right? And who else can afford the headcount required to manage them and maintain them? And in the days, I think these things were big.

These things were so big that you would have an entire building that got built on a computer, right? And you'd have, like they do famously have all these guys and White lab. Literally like taking care of the computer because everything had to be cut, super plane, or the computer was stop working.

And so, you know, he was this thing where, you know, today we have the idea and A I got model, which is like a big foundation model. We, the a of, like, I got mainframe. There be these things.

And by the way, if you watch all science fiction, IT almost always has this sort of concede. It's like, okay, there is a big supercomputer. And IT either is like doing the right thing, doing the wrong thing.

And if it's doing the wrong thing, you know that that's often the plot of the of the science fiction movie. Is this you have to go on and try to feel you figure how to fix IT to feat IT. I sort of this idea of like A A single top down thing, of course, that held for a long time.

I got held for, you know, the first few decades. And then, you know, even when computers, computers start to get smaller. So then you had so called mini computers with the next face.

And so that was a computer that, you know, didn't cost fifty million dollars. Instead, IT costs, you know, five, five hundred thousand dollars. But even still, five hundred or thousand dollars is a lot of money.

People are not putting many computers in their homes. And so as like midsized companies can can buy many computers, but certainly people can't. And then of course, with the PC, they shrunk down five hundred dollars.

And then with a, they shrink down to five hundred dollars. And then, you know, sitting here today, obviously you have computers of every shape, size description all the way down to, you know, computers, the cost of penny. You've got a computer in your thermostat that you know basically control the temperature in the room and you know probably cost to penny.

And it's probably some embedded ARM ship with for more on IT. And there is, you know, many billions of those all around the world. You buy a new car today and has something new car today, have something on the order of two hundred computers in them, maybe maybe more at this point.

And so you you just basically assume with the chip today, sitting here today, you just kind of assume that everything has a trip in IT. You assume that everything, by the way, draws electricity or has a battery because that needs to power the chip. And then increasingly, you assume that everything on the internet, because basically all computers are assumed to be on the internet, or they will be, and so, so, and so is a kind.

What you have is the computer in ister today is this massive pyramid. And you still have a small number of, like the supercomputer clusters with these strange main frames that are like the god model, god mainframes. And then you've got, you know, a large number of many computers, you've got a larger number of PC, you've got a much larger number smart phones, and then you've got a giant number, the systems and IT. Turns out like the computer industry is all of those things.

And you know what? What is the what? You know what? What is size of computer do you want? Is based on what, what exactly you trying to do and who are you and what do you need? And so if if that analogy holds IT basically means actually we are going to have A I models of every conceivable shape, size, description capability, right, based on trained on lots of different kinds of data, running at very different kinds of scale, very different private, different policies, different security policies.

Know you? You're just gonna like enormous variability and variety, and it's going to be an entire ecosystem and not just a couple of companies. Yeah, let me see you think about well.

I think that's right. And I also think that the other thing that's interesting about this area of computing, if you look at prayers of computing, from the main friend to the smart phone, a huge source of lock and was basically the difficulty of using them. So you know, nobody ever got part for buying IBM because, like, you know, you had people trained on, you know, people knew how to use the Operating.

So some like IT was you know he was just kind of like a safe choice due to the massive complexity of like dealing with a computer and then either with a smart phone like the way, you know, why is the apple computer smart phones so dominant? You know what makes IT so powerful as well? Because like switching off that is so expensive and complicated and so forth.

It's an interesting question with A I, because A I is the easiest computer to use by four. Speak english, like time to person. And so like, what is the luck in there? And so are you completely free to use the size, Price, choice, speed that you need for your particular task? Are you locked into the god model? And you know, I think it's so a bit of an open question, but it's it's pretty interesting that could be very different. And prayer generations.

yeah yeah, that make sense. And then just to complete the question, what would we say? So band, what? What would you say? Our lessons learn from the internet era that we live through that that would apply that people to think about.

I think a big one is, is probably just the the bomb st nature of IT that like, you know, the demand, the interest in the internet, the recognition of what IT could be, was so high that money just kind of poured in and buckets. And you and then the underlying thing, which in an internet age was the telecom infrastructure and fiber and so forth, got just unlimited funding. And unlimited fiber was built out.

And that eventually we had a fireball collect. And all the till I come, companies went bankrupt. And and that was great fun. But you know like we ended in a good place.

And I think that that something like that probably pretty like like that happened in A I where like you know, every company's gna get funded. We don't need that many A I companies. So a lot of them we're gonna bus is gonna a huge you know huge investor losses there will be and over built out of chips for sure at some point.

And then now we're of too many chips. And yeah, some chip companies will go bankrupt for sure. And then you know and I think probably the same thing with data centers.

And so for like, well, be behind, behind, behind and they will overbuilt t at some point. So that all be a very interesting, I think that and that's kind of the that's every new technology. So CarOlina p.

Rez has a great kind of, and you know, amazing work on this, where like that is just the nature of a new technology, is that you overbuild you understand, then you overbuild you know. And there is hype ycl that funds to build out and a lot of money is lost, but we get the infrastructure. And that's awesome because that's when IT really gets adopted and change this world.

I want to say, you know, with the internet, the other the other kind of big kind thing is the internet went through a couple of phases, right? Like he went through a very open phase, which was unbelievably great. IT was probably one of the greatest boom to the economy, you know, that is certainly created communist growth and power in america, but you know, kind of economic power and soft cultural power and and these kinds of things.

And then, you know, IT became closed with the next generation architecture, with you know, kind discovery on the internet being on entirely by google and, you know, kind of other things, you know, on by other companies. And you know A I I think could go either way. So that could be very open or like, you know, with kind of misguided regulation, you know we could actually force our way from something that you is open source, open weights.

Anybody can build IT. We'll have a plethora. This technology will be like use all of the american innovation to compete or will, you know well cut IT all off, will force IT into the hands of the companies that kind of on the internet today and you know and well put ourselves at a huge disadvantage, I think, competitively against china in particularly, but but everybody in the world. So so I think that that that's something that that finally, you know that we are involved with trying to make sure doesn't happen, but is a real possibility right now.

Yes, the certain ironies, that network used to be a property. Yeah.

they opened up yeah yeah right lanman apple talk, that boy that bias exactly. And so these are .

all early proprie that works for all individual specific ters. And the internet appeared and kind of T C P, I P. And everything opened up. Ah yeah is trying to go the other in the big company is trying to at the other way. I started out as like open .

just like basically just like right right。

And now they're they're ying to lock IT down so it's it's it's say it's a fairly deferred .

ous turn of events yeah very ferial say, you know, I can. It's remarkable to me. I mean, this is kind of the darkest side of capitalism, and the company is so greedy that they're willing to destroy the country.

And maybe the world like to get a little extra profit, but, you know, and they do IT like the really kind asking thing as they claim it's for safety. We've created naing and that we can control, but we're not going to stop work on. We're going to keep building in as fast as weekend.

We're gone to buy every trick in GPU on the planet, but we need the government to come in and stop IT from being open. This is literally the current position of google and the microsoft f right now. It's crazy. And where I going to secure IT.

So we're going to make sure that like chinese bias can just like steal our chip plants, take them out of the country. And we will even realized for six month .

yeah has nothing to do with security IT only has to do with monopoly. yes.

The other you just been and going back on your point, speculations, there's this critical that we hear a lot right, which is like, okay, you idiots basically it's like your idiot, idiot entrepreneurs. You is just like there's suspected a bubble with ever new technology. Like basically like when are you people onna learn to not do that? yes.

And that is there is no joke. There is no joke to relaxed this, which is the foremost dangerous words and investing are this time is different, the proof most dangerous words and investing are the foremost dangerous words are in investing are the time is different, right? Like, so like the history repeat, does that not repeat the my sense of your reference lot of Prices book, which I agree is good, although I don't think that works well anymore.

We can talk about some time, but but is a good at least background piece on this IT is like it's just controvertible. true. Basically, every significant technology advance in history was greeted by some, some kind of financial bubbles.

Based financial markets existed this year, by the way. This includes, like everything from, you know, radio, television and the railroads, lots and lots surprised. By the way, there was a, there was actually a sok.

There is a like trones s boom bust in the sixties, got is the chronics, every every company of the name chronics. And so there was that you, there was like a laser boom base cycle. There are all these boom cycles.

And so basis, like any new any new technology, that's what economists college general purpose technology, which say something that can be used in lots of different ways like you inspire a sort of a speculative mania and you know, he looks like the critiques like, okay, why do you need to have speculative mania? Why do you need to have a cycle? Because like, you know, people, some people investing the things, so who's lot of money? And then there's this bus cycle that you know causes everybody depressed, maybe to waste the roll out. And it's like two things. Number one is like you just don't know like if it's a general purpose technology like A I S and is potentially useful many ways, like nobody actually knows up front like what the success all use cases are going to be, be what's successful company is are going to be like you actually have to you .

have to learn by doing, you're going have to that venture. Captain, yes.

exactly. Yeah, exactly. So yeah, the true venture capital model kind of is the core venture ital, the kind that we do. We sort of assume that half the companies fail, half the projects fail. And and you know .

if if if any of us, if any, are completely .

like the money is. And so like and of course, if we are any of our competitors, you know could figure out how to do the fifty percent that work without doing the fifty percent that don't work, we would do that. But you know, here we set sixty years into the field and like, nobody figured out.

So there is there is that that unpredictably to IT. And then the other the other kind of interesting way to think about this is like, okay, what would that mean? I have a society in which new technology did not inspire speculation.

And IT would mean having a society that basically is just like inherently like super pessimistic about both the prospects of the new technology, but also the prospects of entrepreneurship. And you know, people invented ing new things and doing new things. And of course, there are many societies like that on planet earth know, just like fundamentally, I don't have the spirit of adventure and adventure that that a place like lei does and know, are they Better off for worse off.

And you know, generally speaking, the worse off, the less future or answer IT less, less, less, less focused on on building things, less focused on on figuring how to get growth. And so I I think there's at least my site, there's a comes with the territory thing like we would all prefer to avoid that outside of elective base cycle. But like IT seems to come with the territory every single time. And I at least I have not no nobody I no society I am aware of has ever figured out the capture good .

that also the yeah and like why would you I mean, it's kind of like you. The whole western united states was built off the goldrush and like every kind of treatment and like popular culture of the goldrush kind of focuses on the people who was, who didn't make any.

But there were people who made a lot of money, you know, and found gold, and, you know, in the internet bubble, which you know was completely ridiculous, you know, kind of every every movie, if you go back and watch any movie between like two thousand and one and two thousand and four, they're all like how only more did a dot come and miss and met the other. And they are all these funny documentary and support. But like, that's when amazon got to started.

The know, that's when ebay get started. That's when google got started. These companies you know with that were started in the bubble in the kind of time of this great speculation. There was golden in those companies.

And if you kid anywhere, those like you funded you know probably the next set of companies, you know, which included things like you know facebook and accent, you know step in all these things. And so yeah, I mean, like that's just the nature of bit. I mean, like that what makes IT exciting.

And you know it's just it's an amazing kind of think that you know look, the transfer of money from people who have access money to people who are trying to do new things to make the world a Better place is the greatest thing in the world like. And if we some of the people with access money lose some of that excess money and trying to make the world a Better place, like, why are you mad about that like that? That's the thing that I could never like. Why would you be mad at eo, Young, ambitious people trying to improve the world, getting funded, and some of that being the guided? Like, why is that bad?

Right right? As compared to yeah compared as just like as compared to everything .

else in the world and all the people are not see you. I like we just buy like you know lots of manches and boats and just what right?

exactly? I don't need money to help run causes just once at the other news, right? okay.

So aren't we are a minute twenty, we made IT all the way through. Four questions are going great. So let's call IT here. Thank you, everybody, for joining us. And I believe we should do a part two of this, if not a part three, three, six because we have a lot of questions to go. But thanks, everybody for joining us today.

Alright.

thank you.