Dan Desjardins: The Future of Computing is Distributed and Decentralized

28:37Where did I hear this conversation? I've gone through this conversation with somebody talking about exactly this. I think it makes sense, right? We've seen this wave in computing as well. Everything was in a room and then all of a sudden it's on your desk and now it's in a room, but then it's on your desk and now it's back in the cloud in a room. And now they're calling it Edge where it's coming back again and back and forth and so on and so forth. Same with AI. It's big, huge models that are - It's centralized to decentralized to centralized to decentralized, unbundling, bundling. We're seeing it endlessly, right?

29:07Endlessly. And so AI is going to go through that phase. And so a lot of people are putting a lot of work into the algorithms and making them more efficient and able to run on smaller and smaller devices at the edge for inferencing. But we also want to build a brain that we're having constantly learned from data near the sensors that are the ones we interact with or where we're generating data like our individual municipalities. And so when you're going to have these centralized AI models that are basically cooperating with society, with part of society built into it, you're going to need compute.

29:43And I'm not pitching or standing up to distribute. This is just going to be true regardless of the compute platform that's employed. Here's a problem that cloud is not yet solved. If we're going to run these cooperative AI models sort of built into society, sort of the helper models all over the place, they're going to run on compute and data at these different edges. And up until now, cloud is one owner of its industry. And it's scheduling workloads based on hardware that it owns and controls. So it's basically performance-based scheduling. what happens when the owners of the infrastructure that come together, the different computers and the different traffic lights and the different servers and the IOMT device that come together to create that data and compute fabric to power the site are not owned by the same owner.

30:39Now you have heterogeneous ownership and not just heterogeneous hardware that you have to overcome. So this becomes interesting. And I was on a panel with Erickson and they said the biggest problem that's going to have to be overcome in order to facilitate true decentralized edge computing is going to be remuneration. which takes me to sort of like a final interesting thought process. When you get your electrical bill, Raoul, at the end of the month, it doesn't say that you bought four toaster hours and six light bulb minutes and 17 fridge days. It says you consumed X kilowatts of power. This is the electrical bill, and that's done.

31:21It doesn't matter what appliance you are using. There's a value that's been attributed to the electrical service that you consume. Well, compute. We're still stuck in that yesteryear version where we're charging people based on the availability of a particular instance type. An instance is just the name of the hardware it's running on. That's what I meant beginning with the 70 pages of these instances. What if instead it's just, look, you have a problem and you launch this workload and your results are worth this. The same way that your electrical bill is just pared down. Then we're going to have to do that, Neil, in order to deal with heterogeneous ownership and heterogeneous platforms.

32:01And by here, I'll define heterogeneous. It just means different. So you'll have phones, you'll have fridges, you'll have computers, servers, but you'll also have it. It's owned by the library, club, by residents, local households who are part of that municipality, and they can get an offset on their electrical bills or whatever if they contribute their idle compute capacity. So measurement without being able to be spoofed or fooled is critical. security the inability for a malicious workload to compromise any of those devices is paramount and simplifying access to these making it easy is the final pillar those are the three pillars I think that not us but any compute platform that claims it will put a dent in edge computing and AI will have to overcome in order to make that future a reality and get us closer to the simulation for example hey everyone we're going to take another quick break and hear a word from our partners and then we'll be right back okay there's a whole bunch of stuff there firstly when you talk about separating electricity hours by toaster hours and bridge hours and whatever aircon hours actually that whole process could be run by ai once you get to that granular level because if every internet of things is giving you that information then your own household ai can control better than your Nest can, which is a very mild version of this, that you can become hyper lean and effective and you can exactly see what's causing your problems and do you need to change that AC unit or whatever it is.

33:41That in itself is interesting. When we're talking about localized AI, you can have household level AI. Already we're very close to having LLMs on our phone that they'll need to run on the internet. So that becomes personalized AI, household AI, municipal AI, online group AI. So Real Vision is developing AI and AI, and that's for the community of Real Vision. So it's a community AI built for the purposes of that community. It becomes super interesting and it links everything. And here's my issue with the singularity in all of this is we don't yet understand, and I don't think we ever will, how large language models learn.

34:24We don't know what they know, and we don't know how smart they are. Because we're trying to speak a foreign language to something. And if you go and speak to somebody in Hindi right now, you're not going to assess whether they're smart or not. You're just trying to somehow, in a broken language, you can speak this much to them. And if I listen to Sam Altman speak, or listen to the guy who runs Anthropic, or listen to MAD from stability, nobody's got a fucking clue how this stuff works. And what is interesting to me is if you do link everything, well, AI can spread everywhere. If it has some ability to think like a virus, which it may well do, we don't know, then it has an ability to use all compute everywhere, whenever it wants, however it wants, and control everything around our environment.

35:16And that's kind of an unintended consequence that I feel thrown this. While you're, one cannot avoid thinking about Terminator here and Skynet and everything else, right?

35:30For the next, for the foreseeable immediate future, we think, or at least we think here that LLMs are excellent stochastic parrots. They happen to be able to spit out things that are. But you don't know that. because when you watch how it won deep mind go right deep mind that was i don't know if you watched the documentary but it was we've all read the story we all know it you watch it it was the first four games or whatever three games it played ordinary human games it then lost a game and never played a human game ever again and the commentators were like what the fuck is it doing this is stupid but this is a dumb move and it never, ever lost again in a way that we don't understand.

36:14So I don't think it's stochastic parroting. I think that's a human protective thing that we suggest because then we can understand it. Like we use narrative and mimetics to understand the world around us. We kind of say that. It's certainly an important conversation to continuously have because the minute we relax our skepticism or our caution around these things, we'll begin to make mistakes. It's the same with CRISPR gene editing. In a way, the parallels there are for the atomic bomb. With the barbenheimer that came out just recently, with great power, we have to be very, very careful. And we obviously are with some dark humor and cynicism here in the office.

36:58We talked about maybe accidentally unleashing something like that. um we're yeah we would become in serious trouble if we start too early letting ai um launch more ai jobs and start doing stuff um so i think it's it's always important to have guardrails for sure um we have made sure that there's a big red button of course everyone does um i don't yeah but you You might be able to, but the LLMs can't because if it has infected your fridge and it can, if I can store an LLM on my phone, I can store it on my fridge. And so in which case the LLM can be anywhere and not exactly like a virus. It's just something we're not there yet, obviously, but it's just something I think there is like, yeah, this is very difficult to start if it goes certainly too far.

37:49Ex Machina, if you've seen the movie, it was such an excellent movie. The biggest question that wasn't asked explicitly was, are these AI humanoids, are they simulating emotion with really high fidelity or are they experiencing it? And it almost becomes a metaphysical question. Does it matter? That's my point. I was having this discussion about, do they exhibit emotion? It doesn't matter. just because how we see the world is human emotion it doesn't mean different things can live i mean because if we know the complexity of a bee colony or a bacteria colony these are super organisms they have a collective intelligence of which we don't understand yet it's fascinating to watch everything yeah because we can only think of in our world yeah but yeah and it's and it's fascinating to watch these bee colonies or fungi colonies i mean oh yeah the ones that that actually you know take over their hosts brains uh and by injecting spores and whatnot it's absolutely fascinating and terrifying um let's hope ai doesn't start doing that that would then we're really in trouble right yeah well you know is there a flip you know are we biological computers i mean that's the other question that biologists and physicists are trying to answer because if we are biological computers then there's no reason they couldn't and you know it's always fun to ask the reverse question you know like um we keep saying that humankind has domesticated wheat but someone actually flipped that that conversation wrong wheat domesticated us like dogs my dogs domesticated dogs domesticated humans because they now have their food given to them yeah same with wheat wheat is now an endless it's survived That's the entire job is to survive.

39:44Making sure it's healthy, it's watered, the lands are irrigated, it's in these beautiful parts of fertile land, and so on and so forth. So I guess at the end of the day, it's important to understand that we're not above or below all the other participants on this planet. We have to be caretakers and be careful. And as long as we're progressing in a way that people are benefiting, all people are benefiting, all people and non-people, I guess, then things are sustainable. Why are we inventing AI? Why is this AI craze in the first place? Do we need it? We've been C2C, we've done fine over the last several thousands of years, millions of years.

40:23Well, AI is, again, like compute. It's a means to an end. AI in and of itself is useless. What are the use cases that are interesting with AI? Well, it's all about getting some efficiencies here and there. Better scheduling of resources. If you can make sure that the right patch of cucumbers... knowledge being a scarce asset. Right? So your brain... What do you mean by that? Yeah. So your brain, what you know and what you've learned, is a scarce asset. Right? It's not easily repeatable. But we're seeing that AI can scale knowledge in a way that no single human can. Now, we're not at AGI yet, but we're at levels of knowledge which are Frankly, making a doctor, an accountant, a lawyer, and a whole bunch of professions questioning what is their economic value versus an AI.

41:22A doctor charges a thousand bucks. The lawyer charges a thousand bucks an hour because he's a senior partner. Really? How much is that worth versus the AI? Right. It's very interesting. Everything that gets digitized goes to zero in cost, which you know because that's what you're doing. Like, well, I used to fly for the military as a military pilot before. And of course, it's hit us too. Like we're asking the same questions. You know, AI can land a plane far more accurately. In fact, I think when they were landing drones on the U.S. aircraft carriers, they had to actually build in some stochasticity to its actual touchdown point because it was so perfectly landing in the same place every single time that it was starting to damage.

42:05like the tarmac that was landing on this was like pinpoint precision. So they actually had to build in some variability to its landing spot. So absolutely. And a lot of people are worried about losing jobs. It's not that we're going to lose jobs. We're going to evolve new jobs, more creative ones, more guidance. There'll be a transition and whatnot. And I don't think it's going to happen too, too quickly. I mean, I think there's a scare now because there's been a leap, obviously with chat GPT and other similar technologies. But there still will be a reasonable transition between our current way we do things because it comes down to trust.

42:44And as you said earlier, we don't know exactly what's going on at the end of the day, deep down within these models. So here's a question about that and distributed networks is there are two battles going on, which is the same battle we talked about earlier. That is centralized AI. Let's call it open AI. Let's call it Google. let's call it anthropic and there's a few others and china has their own a few other people and then there's the open source ai models how the hell do you stop it why can't it's like you can't stop atomic um energy development or nuclear weapon development really it's somewhat easier to control the uranium but you know we've seen it with crisper gene editing any of this stuff it just go to a different country.

43:30It's the prisoner's dilemma. If you don't do it, someone else will, and you're kind of stuck, and therefore you must.

43:39Entire doctrines are being rewritten, in fact, are already rewritten, how if your enemy or any naval or air force fleet that harnesses AI will have air superiority. It's imperative to develop these things, because if you don't, your enemy will. So how do you stop that? That's really hard. I mean, I guess you, you cut off. I mean, you can only go after algorithms, data and compute if you want to stop it and communications. You've distributed compute. The algorithms are distributed because they're open source. I mean, what'd you stop electricity? But that's even that's distributing now, right now we've got solar panels and we have, you know, we're moving away from centralized grids to decentralized grids.

44:27It's just, So, it's almost depressing to ask the question, you know, do we, the human species, have we matured? Have we learned anything since the detonation of several hundreds of nuclear bombs and the devastating effect they've had? Have we learned anything when we're creating potent nerve agents, bioweapons? Have we learned anything about doing anti-sublight kill kinetic weapons where we're creating frapnel orbits at 7 kilometers per second that will basically age us here? I'm sure you're aware of this. If you look at lower Earth orbit, it's littered with dead debris and satellites. And one of the ways you knock out your opponents, I don't know if we've gone in this sort of military slide here, but if you want to knock out communications take out the satellites and take out the undersea cables cutting an undersea cable isn't really going to threaten our civilization at least not in the long term but if you start taking out satellites with kinetic weapons creating again like these gigantic irreparable perapinal clouds that travel at speeds that will you won't even feel it would just go right through you and you'd be pulverized.

45:43So we're at risk of creating a very physical jail around the planet if we don't let cooler heads prevail. So the question is, can cooler heads prevail or is it the prisoner's dilemma? It's also interesting to me, at the center of all of this, of all of this conversation we're having, is Elon Musk. because not only has he put up more satellites than all the other governments combined, and he's done it as a single person, but he's created an entire network communication, distributed network communication, which is Starlink. Yeah. But then I look at the cars and they are distributed compute. And they are using AI to drive the robots because that's what they are in the end.

46:38their computers. Oh, yeah. Everything is a computer. I mean, this, I mean, this AirPod case will, you know, I can find it. I can locate this chips in it. It's broadcasting its position, obviously. So it's this phone, this laptop, this smart TV, that laptop, this thing. Everything is basically a computer with chips and a sensor, right? So Elon Musk, when you take a step back and you look at all the companies from solar power, battery storage obviously launch system satellite network and now communications platform i always wondered why you bought twitter um training the ai it was right well this training the ai but the optimist in me thought when i was looking back and say okay you know i had all the boring company right for i thought he was setting up might still be i have no idea i'm just speculating here's my the curious uh wannabe astronomy i i always thought he was building his own space program but compartmentalized he is right well no question but if you can deploy these subsalene tunnel diggers to the moon now you don't have to carry all the the materials to build these bases you can just dig them out and especially if you're also if you want to base on mars it has to be underground that's it so like he's got all the things put together and maybe at the at the last second he noticed you know what's missing the the imagination and the will of the public right So if you can influence the media and get people to dream again, as Elon Musk started, and I hope he still is, as an optimistic dreamer influenced by Jules Verne and all the explorers.

48:10And he built all the technology. And the last thing that was missing was maybe the will of the people, the boldness, the desire to dream big. Now it becomes almost social commentary. I feel like the human race has been preoccupied with less than exciting topics, I think, these days. And in fact, probably, that are damaging to our collective well-being. And nobody's talking about, or few people, relatively speaking, are talking about nuclear-powered space travel. And what's below the surface of Mars? Or how much water is actually below the poles on the moon? or helium-3, if you mine that out of regular dust, you can create anutronic fusion fuel, which power a teaspoon of it would power a city for weeks.

49:03No one talks about this. We're more busy contemplating politics and whatnot. So I was sort of hoping that Elon Musk was going after the last piece of his entire space program, his disassembled space program, to actually influence the people, nudged him towards being curious. He's just going to do it anyway. now again this is not just elon must is a good person or a bad person or he he does everything to everybody pisses everybody off and inspires everybody at the same time and everyone thinks he's a crook as well as a genius so that aside i think what he did was say okay i want to go to mars the mission statement and i want to build a colony there so go back to first principles what's the hardest thing well we know we could probably get there the technology was there he was paying for it no government could afford to do it so then if you break it down to the component parts and get each one of those businesses to be cash flow positive so the car company let's say let's say the ai training which is twitter let's say um obviously spacex all of this they all throw cash and before you know it they're all building this end goal and then it just kind of happens because everything is in place and that's what that's what i you know even when i look at the optimus robot is like well you can't send humans to mars to start with and if you're going to have to do stuff you'll do it with robots how do you train robots that are acceptable to humans that don't freak you out is train it on humanity so you can your political views will match with your robots or whatever it is, which is Twitter, which is why he doesn't give a shit about freedom of speech.

50:44He just wants an unbiased model so he can train the robots to be like you and I. Once upon a time, I'm sure he was dreaming of space travel. I'm sure he still does, but it's important not to become too cynical, I think, right? Otherwise, why are we trying to go to Mars in the first place. This is another question. Like, why? Why go? Are you just going to set up a copy of ourselves over there just to do more of the same? Because in that case, it's not AI. That's the virus. We're the virus, right? Well, maybe that's true, right? Yeah. I mean, I don't know. But the other question I want to throw at you in all of this is, okay, we've kind of looked at steady state of compute, saying, okay, we're distributing.

51:29We can scale it. AI has come. It's a game changer. requires more compute everything's just going to be more compute more compute that's endless talk me through quantum quantum is going to make a lot of conventional compute obsolete um so it's a guessing game is the one that's going to be ready and i was having this conversation just recently conventional compute computers have come about decades ago but it's only in the last 10 years 20 years where you're starting you're starting to see like coding really built deliberately built into even high school curriculum, right? So from the moment of invention, it took decades.

52:09Like of course it was used immediately, used to build movement and capital trajectories, but by the time it's actually fully mainstream, like baked into our educational curriculum, everything else, it took several decades. As a physicist who's taken a quantum mechanics class or two, there are still a lot of hurdles that we're still working on overcoming to make the first quantum computers more than just cool, but viable for big problems. And I know there's a lot of big efforts that you, again, can count on two hands that are doing some pretty amazing stuff right now. You've seen the wormhole on a chip video that came out by other magazines is formidable.

52:49But there's still a long ways to go before that technology is mature enough to be deployed on many desks, let alone everyone's desk. and it's going to take even longer before they're known as part of the curriculum in school. Again, as a physicist, I was trying to look around. How do I program a quantum computer? How do I write a quantum algorithm? The very first textbooks on this are just starting. Everybody should know right now what a for loop is or anyone who's done a comp sci or a science degree or anything touching the sciences. Everyone knows what a for loop is and how to write one. But if you ask someone, hey, how do you write the quantum version of that?

53:29Or is there a quantum version of that? Like, as someone who's written a lot of code myself and who's a physicist, I have no clue how to think about writing. What does it do to everything when you've got kind of, so we talked about knowledge being scalable. We're talking about what you're doing is making compute scalable. But now we've got, with quantum, and we're not there yet. So let's talk 30 years in the future, infinite compute and AI. That feels like that is the singularity of all things because there's no constraints on anything that we understand today. So in two words, I'm going to say more triangles.

54:14And that's an inside joke for people doing finite element analysis. when you design a model whether it's a car or an airplane and you want to see the impact forces and how it distorts the metals and breaks things you break that model down into tiny little triangles 3D versions of triangles and then you apply the boundary conditions and see how forces distribute everything else but the more you turn up the fidelity so the smaller you make those triangles the more triangles you have the more equations and the bigger the matrices that he computed it upon in order to get your answers so basically if you give someone if you double the amount of compute that that they have access to they're they're going to more triangles they're going to turn up the resolution they're going to push the batteries they're going to come up with a new ai algorithm instead of doing like 10 billion tokens they'll go to a trillion tokens i like if you if you make more clean pasture available to um to rabbits well they're not going to just be sustainable and responsible they're just going to multiply and then they'll eat all that grass so So to me, the cynicism to me thinks that if you give someone infinite resources, you can find a way to use them.

55:26Why not? I mean, that's how innovation happens. We're constantly pushing the barriers. If we had a space drive that could go at 90 % the sphere of light, we're going to look for a way to go at 97 % the sphere of light. So we can go see more parts of the galaxy. It's just the way we are, I guess. So what does it mean? I think we're going to have someone in the future called Raul2 and Dan2 will be having exactly the same conversation about the next thing after quantum computing, talking about what happens when we truly have infinite computing, find more ways to use it. Okay, final question, something you and I have talked about over the years as well is as we're getting towards distributed compute and payment systems, I'm thinking at simplicity level for people to understand, it is the Uber of compute.

56:18computing, right? So I can get paid by having excess capacity. Like if I wanted to use my car for other stuff, right? I can do that and I can get paid. We're also seeing that there's machines that doesn't have to have humans. So it could be my fridge, right? In the future state, and you've tested that technology out, worked pretty well. Anything with a screensaver, if I take that screensaver, I've now given permission to, you know, you to use my compute. So surely that sounds like crypto payment, right? Because that's the easiest way of making machines, machine payments and globalized payment system where you're not having to use currency.

56:55We need to push. We need to make sure. You're a Canadian. I don't want your Canadian dollars and I can't have got no bank account to put them into. So I have to convert them. It's ineffective and it's not fun. The moving around small amounts of money or value, different currency changes, going through traditional systems is definitely a bottleneck. Absolutely. But I just want to back up as you said some interesting stuff. The idea of having the Uber for compute, I think that's a very good top-level way to look at remuneration for providing a service, access to your access to compute power. Distributive has a much more holistic vision of this.

57:37It's not about, hey, who wants free money? Put your hand up and sign up your computers. Many companies have tried that. I think it's boring and I think it's unsatisfying to humankind. What we want is for active participants in the innovation journey and sort of like staking a claim or taking ownership. Compute is one form of value, but the other one is also like private data. Your fridge in the future won't just contain your food. It will have a few hard drives in it. It will store and analyze footage from your local security cameras. that will look at if a person is starting to have changes in their gait as they're aging or if they're dizzy or if they're calculated or if they fell or something like that.

58:18Like, if you have an ownership, say, on your data and on the compute and all the results of it and aren't just uploading it to some monolithic cloud services provider, then all of a sudden you're a participant and not just a subject in this digital age. So some people, yes, sure, they can make some money. You have agency. Because your data is part of you, right? It's a byproduct of you. That's it. And so participating in these mesh networks is the best way to stay involved and maintain a voice and some control. And so the future that we are, that since we've met four years ago that we've made good on producing, is one in which participants are truly participating.

59:02They're not just getting paid for the computer. They're finding these Los Angeles pairs. is that they're elucidating all the formation mechanisms for galaxies, dark matter, and helping uncover some of these things. Or Cartridge, Kellogg's, serial sales data. So let's say it becomes also cause-based, right? Humans love a mission, they love a cause. So I'm going to say to whoever it is, is running a huge project on something that matters to me. Pancreatic cancer and genome, editing to remove that. So I can then say, well, you can use my compute for that. So as opposed to giving you money to the Pancreatic Cancer Society, I can say, no, just use my compute as well.

59:50And so I just permission then. So I'm now doing a mutual good, societal good out of my own free choice. And it's not about money exchange. It's about I'm giving you something of mine that's valuable. Well, a perfect analogy for that. And then I'll come right back to it. Like I agree on a percent, Bank of Montreal gave$5 million cash to the AI hub at the University of Toronto. They turned around and they gave that$5 million cash to Microsoft Azure. So Microsoft got$5 million. Bank of Montreal, BMO, got the$5 million tax receipt. And U of T got to do some AI compute what if you provided five million dollars worth of compute to uft and the value of that five million dollars was the tax receipt but the the green cash as they call it never moved now bmo has hundreds of thousands of enterprise servers full of excess compute most of the day they could have donated five million dollars of the computing power or that's the enterprise version you could asked 5 million people donating a dollar worth of compute.

1:00:57Real Vision has hundreds of thousands of unique visitors per month. Something distributive is done. And I haven't really gone heavy on the tech and innovation we've done, but this is important. We've created five lines of code that we can put in your website that would make all multiple hundred thousand visitors turn into compute nodes if they accepted. That would provide power simultaneously just by putting those five lines of code in the website. So if the New York Times put our five lines of code. So we would say to our members, let's say, listen, if you want to help us continue to build our AI, which is for the community, by the community, essentially, we can either give that money to Azure or you can be truly part of this.

1:01:40Yeah, while they're watching your content, their computer is basically sleeping. Other than what it's displaying on the screen, their processor, some proportion of them, could be helping work on Real Vision's models. And I'm using real, I'm not picking on Real Vision here, but the same thing for the New York Times, Global Mail, and basically any community with hundreds of thousands or millions of online viewers. I mean, if you put these five lines of code into the New York Times HTML page, you have six million compute nodes, just like that available. That's more computing power, like I said earlier, than all of Canada's national research infrastructure put together just by putting six lines of code in the index.html file of that website.

1:02:20Think about YouTube. YouTube, bought by Google. Every time you upload a video, it's being encoded on Google servers. But instead, they could be distributing that video encoding into pieces onto computers of people that are already watching YouTube while they're there. And you could use this truly to eliminate ads or provide an alternative. Six lines of code, get rid of ads. It will help us run our models on you, the patrons, instead of in the cloud, that'll be far more cost-effective than these ads. And also, we could pay people for that as well. We'll say, thank you, you get X dollars off your subscription next year because you've lent us compute to help train the model, so you get an economic incentive out of it.

1:03:10And so everything we've talked about, just to summarize all this, we've talked about a lot of things here. We've talked about the ability to donate, compute, and get a tax receipt from Children's Hospital in Eastern Ontario, for example. They're looking to recognize philanthropy, not just in cash, but donation of time and services, and now compute. We're in the digital age. It's time to recognize the donation, the value, the donated value of compute. And thankfully, we have the cloud service providers who've laid out 70 pages plus worth of the actual value of compute. So I think it would be very easy to recognize the value of compute because it's a market commodity that's been well established.

1:03:46So that's the donation model. But then, again, there's this web collaborative online web community model where just being on a page allows you to participate in the content creation or AI model creation that serves you anyway. You to them. We'll just talk about mail, like anybody with these online presences. and same again to bring it back to these participatory ai models at the edge if you want these heterogeneous compute fabrics and sensor fabrics made up of different owners they all have to be incentivized fairly and they have to trust each other and it will power the model that again gives value back to them from detecting you know when when roots are ingressing into the water lines, they have to replace a certain structure of pipe or optimizing the clusters or what have you.

1:04:34So that's the future that we see. And again, I'm not saying distributive is going to be the one to do it, but I certainly think the future is... Compute can be a common good. And people have been saying that for years and years and years, but I think it's more than that. I think it's a deliberate ownership cultural mindset of the people with their hardware and infrastructure and data where they're participating in this digital age and not just passively collecting income, but they're picking a cause and they're dynamically interacting with it. I think it'd be wonderful to have a YouTube powered where all the video content uploaded was being encoded and done by the community of watchers itself.

1:05:16So it becomes a self-sustaining thing and we can cut down all these data center costs all together. And does it become more decentralized in terms of editing and everything else in terms of sorry, in terms of any centralized entity stopping something. So if the compute for all of the videos is being done on a distributed network, it's almost impossible to shut down the network, right? You could shut down individual worker nodes, but the work would just go somewhere else. So it becomes very resilient in that sense. Yeah. Okay, two final questions. One is, I know a lot of people are going to be listening to this and thinking, or going to comment about Filecoin.

1:05:58So what are they doing? How is that different? Just so people understand how fundamentally different it is. Filecoin is tackling an equally gigantic farm, which is more from the storage side of things. So there's a lot of unused storage space out there. And if there's a way to intelligently stitch together those pockets of excess storage, then you can create a decentralized storage cloud. And so it's probably a really good way to use them as an analogy if you want to understand what we're doing. We're on the other side of that fence. We're taking up all the unused pockets of compute, excess compute capacity.

1:06:42Both of these are AWS, right? So they do the compute and they do the storage. And what's happening here is there's two distributed models that are attacking the centralized model. And more, yeah. Yeah, I mean, AWS is more than just computing storage. They're networking and there's software layers and platform layers, and there are a whole bunch of things. Like they've had a lot of time to build a lot of things, but they won't tackle, by definition, they can't tackle the web platform community sharing aspect in real time because they're not web-based. And B, they would be attacking their own business model by allowing people to use computers and servers that are not theirs.

1:07:24and so they have too much concrete fiber optic cable and and silicone that they as some cost that they won't be able they won't be they'll be hesitant to implement technologies that allow people to become self-sufficient of course put it only there's a couple of other interesting ones in the crypto space just because it tends to be more centralized hive mapper is one which is a decentralized version of google maps super fascinating and And then there's the other one, Helium, which is mobile phone networks, Wi-Fi networks, decentralized. So it's this centralization, decentralization mega trend.

1:08:06And it's healthy. It is healthy. It's good. Final question. How many people are using your technology right now? And where are you allowed to discuss it? What kind of users and how are they doing it? We have 2 ,000 people right now. I was just looking at the numbers. So it's still small, but we're in a dozen universities. I was saying earlier, it's free for academic institutions. In Canada, States, Kenya, Brazil. Brazil, for example, it's doing live forest fire detection using all the computers that are on the campus there. There are 8 ,000 cameras across the Amazon. And if you can detect smoke and fire at the earliest onset, then you can fight it more effectively.

1:08:51But if you were to run 8 ,000 cameras persistently, AWS would cost you millions of dollars a month. So it's not even possible or it's not worth it. Force fires create hundreds of millions of dollars worth of damage, so you could argue it is. But if you could do it for thousands of dollars instead of millions of dollars, then it certainly makes it more palatable. We're in six hospitals creating private compute clusters in hospital in order to optimize certain schedules. and now we're starting to get into some genomics. Actually, one more use case here. So I meant to summarize with users. This one is new and it's unique.

1:09:30If we can create a compute cluster by an hospital, as we happen, and you create a compute cluster by a different hospital, there's ways where you can run models. You've heard of federated machine learning, federated models, where you can run analytics on data, on computers, behind each hospital's firewall, and you can fuse the results that leave the raw data behind each firewall. Now think of a hundred hospitals, think of a thousand hospitals, each with, for example, small patient cohorts with psoriatic arthritis. There are new genomics approaches that allow the right medicine to get to the right person on the first try, unlike the trial and error attempt that we do nowadays.

1:10:11That would save hundreds of millions to billions of dollars worth of the wrong medication going to the wrong person on the first try because it means they're not going to be working within months because they're off work while they're going through these treatments. But you can't take all that small patient forward data and put it together, whether it's the cloud or somewhere else, because it's not allowed to leave the building. Data sharing is the biggest barrier to innovation in the precision medicine landscape. So I say again, what if you can send compute the data instead of bringing data to the compute, which is the cloud model?

1:10:47So bringing them to the data, and now you have each of these hospitals acting as their own mini clouds. But if they cooperate in the sense where they run their respective pieces of compute and then fuse the results together, now you have something different. You have hospitals or individuals or whatever maintain ownership of their data, making some revenue from the fact that they're selling the results of those computations, but not giving up the underlying data. And to bring it back to Brazil, we're chopping down the rainforest right now. I think four-fifths of medicines around the world come from the materials, the biological materials that we find in the rich forests there.

1:11:28And we're losing that, right? So imagine instead setting up in organizations with some schools for all we know, for what it matters. And these could be many repositories of collected genetic material, sample material, and computers. The computers can be used for students and the pursuit of knowledge during the day. And they could be running all kinds of data pipelines, data science pipelines at night on these data things, generating revenue locally to replace deforestation. And so this idea of data ownership, doing the compute, selling the results and creating a revenue stream, it's basically, it's a distributed, decentralized version of a cloud services provider.

1:12:15Imagine a virtual version of a hyperscaler where everybody gets a piece of the pie. Providers of data, providers of compute, the marshal of communication, the organization, the orchestration. This is what I mean by participatory. participatory notes. If you think about it for a pharmaceutical company, it's a lot cheaper for them to distribute out the scientific research to the universities. Everyone's running their own data. They can sell it back to the pharmaceutical company. They just take what they need from that data. They can develop new drugs, et cetera. They pay for it. Everybody's a winner.

1:12:45Everyone saves costs. Well, in this model, in the prior model, pharmaceutical companies would just buy the data from the get, like the data's collected and refined at the source. They buy the data once and boom, the people who provide that data, they're gone. It's like selling your IP. It's gone, right? Instead, they could hold on to their genetic sample data and the pharma companies can just launch job. They can write the code that they were going to write anyway. They write their analytics pipelines, their AI models. They write their queries. They write whatever it is they were going to do if they have that data and they launch it.

1:13:22and then it runs remotely at those sites and then they get the results back. Money or revenue is generated for the custodians of that data. It allows local economies. The pharma companies might not be happy because the first thought is, hey, if we have the data, we can run all 10 ,000 analytics. Now we have to pay 10 ,000 times for the analytics that we'd only have to pay once to get that data. But if you think about it, they were just going to upload that data to the cloud and then pay the cloud 10 ,000 times to run it there. So it's still an order of magnitude cheaper for them to not buy the data from where it's coming from, but just pay to deploy these pipelines at a reduced rate, which creates local economies over there.

1:14:06And of course, we can expand this. But this is what I mean. Push away or let's create room for everyone else to participate in this revenue creation opportunity at the local economic level. Well, you don't have to answer this, but surely you've spoken to Google about this. I mean, surely they would want to put this in every browser and give it a permissioning system. It's a game changer because they're the third runner in the cloud providing thing. They're big, but not as big as the other two. This is kind of what Microsoft did, the nuke button, right? Microsoft pressed the new button and said, fuck it, let's do the AI one to overtake Google and destroy search.

1:14:52It feels like somebody like Google or Microsoft or somebody is going to say, you know what, we might as well just use this technology and just completely take this over. I mean, in Google specifically, if they find that idea interesting of leveraging YouTube patrons while they're watching their content in order to do some of the processing for them in lieu of the ads that we keep getting every 20 seconds on YouTube videos, which would free up Google's data center capacity or GCP capacity to sell to more customers. I mean, they're... I'm kind of thinking bigger than that. If your little code is embedded in every single Chrome browser around the world, so now my computer, every time it goes on screen saver, is available for compute.

1:15:43I give permission. Google takes a slice. They become the world's largest compute provider by doing distributed compute. It would be massive. If they were to do that, and it would have to be an opt-in basis, right? because people don't want to inadvertently have compute cycles taken from it. Even though it's a favorable, it doesn't touch your data, it has nothing to do with pictures or contacts. It's purely just compute power. It's not storage or files or whatnot. If Google were to do that, it would be the most powerful supercomputer in history, period. Why the hell would they know? Why the hell would they know?

1:16:25because nobody has invented yet a pure web-based distributed computing platform until us. Well, that's what I mean. I mean, all I'm thinking is like, as you scale this business, you've proven it out, you're starting to have great success. I'm just thinking through, okay, well, where's the massive disruptor here? Yeah, you can do every university, you can do all of these research things, But really, if every computer in the world that runs a Chrome browser can do this on an opt-in basis and everybody gets paid, including Google, it's like there's free money for everybody. And everybody saves the costs and it destroys Microsoft and it destroys Amazon, which is what they want to do.

1:17:10Well, I don't want to be conned between their wars here. No, that's right. But no, especially with the current chip shortage and everything else, I think recycle, reuse. I think there's a lot of compute out there. I think in order to tap into it, you need a system that's secure and trusted. And web technology has been around for decades now. And if you're just surfing the web without downloading stuff, it's one of the most secure platforms in terms of executing untrusted code ever developed. It's the most ubiquitous technology on the planet. It's fast with WebAssembly and everything else. um it i think we think it's a no-brainer so so yes if google put these uh experimented putting some of this code into any of their their pages that they host they would have instantaneously millions and millions of extra compute nodes and they could remunerate those compute nodes um for the participation so there's a lot of good stuff here i mean i look forward to um redistributing value of compute so computer if you buy in north america's two thousand dollars If you buy it in Kenya, it's still$2 ,000.

1:18:13The difference there is the average salary per annum. So there are places in the world that are being excluded by cost from participating in digital economies and AI. Right now, they say the rich get richer. Well, the computee are getting computeier, right? And that's the price of admission for doing AI. So AI is also accelerating the divide, right? Those who have the means to do it and those who don't. so there is another play it's not just making people passive income but it's also making and therefore accessible from an ease of use, what have you we're excited we're very excited about where this is going thank you so much for the exciting questions the conversation, hopefully we'll avoid Skynet who the hell knows, one thing I will do is I will check in with you in a year's time or so and see where you are because I've been following this journey and it's just a fascinating journey.

1:19:10I just love what you're doing and it's just very, very big and it's very disruptive. So I love it. Great to see you, Dan. And as I said, I'll get you back soon. Thanks, Raul. Take care. Okay. So there was a lot to get our heads around there. Dan is very modest in what he's done, but to understand that any computer, via a screensaver, can permission decentralized compute, whether it's private or public, that is going to change the structure of how society uses compute, how AI works, how everything works. The scale of this is simply gigantic. And he's just starting in this journey. We're privileged enough to see it early on.

1:19:59As I said, I've been following this journey for about four years now and seen how far he's got and where this is going. And right now we understand that all compute is owned by very few people, but maybe that's not going to be the case. Can he truly disrupt by 92 % the cost of computing power? What does that mean for other countries? What does that mean for all of us? But there's some dystopian sides of distributing compute and distributing AI and having it in your fridge, in your neighborhood, in your house, everywhere. What does that mean for humanity? And when you add the quantum computing as well, it means another game changer overall.

1:20:41The exponential age is relentless. It's not going to stop. And the game theory that he talked about is going to continue to play out. So it'll only accelerate. And there will be regulation on many of these elements here. but it's almost impossible to put a distributed genie back in the bottle. So whether we use it to our advantage or fear it, that's up to us. But anyway, super interesting, mind-blowing conversation. What's up, revolutionaries? Thanks for tuning in. For more content like this, head over to realvision.com and get unfiltered access to the very best, brightest, and biggest names in finance.