Vitalik on Techno Optimism

And with that, I would like to welcome Vitalik, who is obviously the person who coined Diak, and we'd all love to hear from him a little bit more about what the heck that means. So, with that said... Great, thank you. Yeah, I think in place of slides, I was going to just literally use my post because it has all the pictures anyway. Okay, hold on.

The other nice thing about it is... One more. The touch... Oh, okay. Sorry, I have to scroll.

Okay, it's fine. I can figure this out. I've, uh... And then, let's see if, uh, let's see if this works. Oh, wow, this is...

Okay. It is, it is a bit too slow, but that's fine. Um, okay, great. Um, so I guess, uh, start from here. Great.

So, wow, good, good to see so many people actually interested in DEAC. So I hope you guys end up enjoying this week. So technical optimism. So I think last year, I'm sure you guys probably remember, this was the time when the AI safety and EAC Twitter wars were starting to happen. And of course, I mean, I can't tell whether that whistle was for AI safety or EAC.

I hope it's for DEAC. But basically, yeah, and then at the time, of course, you know, Mark Andreessen published his manifesto. A lot of other people were like saying various things. And this made me really start to think basically about this question of what would a philosophy of radical technology is that actually makes a long-term sense to me, really concretely look like. And also, how does some of the work that we're doing, I think both in the crypto space and in the biospace and beyond fit into that?

And what does that imply about what kinds of things each of us could be working more on, right? So I started the post with this three-way diagram, right? Basically, you know, you have the anti-technology view on one side, which is basically that, you know, you have dystopia ahead and safety can come from basically us realizing that industrial civilization is bad and we need to basically go back to living in wholesome cottages and huts without the benefits of modern economies of scale. Can I have a mic, please? Okay, sure.

And then you have the accelerationist view, of course, which basically says, hey, let's turn humanity into a Kardashian-type 3, hopefully not Kardashian-type 3, that would be different, civilization as quickly as possible, and like basically accelerate on everything and especially accelerate on the AI. And so the argument here is like actually the bear is behind us, right? The bear is the 60 million people every year that are dying of aging. The bear has all kinds of challenges and risks of the present world. And acceleration is actually the basically only safe thing.

And then my view is basically combining this two and accepting that actually behind us is not a wholesome cottage, but it is a bear. And at the same time, also that like actually we have a choice in what kinds of paths forward there are, right? And some of those paths are great and some of those paths are actually not so great at all, right? And basically, of course, I'm going to start off by pointing out that like technology is actually like really, really amazing. And we have to keep appreciating that.

So this is a chart of life expectancy over the last 100 years, and basically you can see all of the worst disasters here. So World War I over here, World War II over here, Great Leap Forward over here, and you basically see how the long march of improving medical and sanitation technology over the long time scales just completely overpowers any of that. And so from that perspective, if you really care about improving human flourishing and human lives, and you really care about the long term and what's going to happen 100 or 200 years from now, then the best thing that we could care about the long term and what's going to happen 100 or 200 years from now, then the best thing that we could care about is this trend not just continuing, but even accelerating, right? So this is one side of the equation. The other side of the equation, of course, is that, you know, there are actual like bad things that industrial technology causes.

So global warming is one of those examples. Though on the other hand, there are environmental problems where one wave of technology causes them, but then the next wave of technology actually fixes them. So this is air pollution. So air pollution peaked around basically early 20th century. The diagram above here, Great Smog of London, 1952.

So if you guys don't like wearing masks, you would definitely not want to go back to 1952. But what happened since then is basically that there was a huge concerted effort to try to move beyond polluting technologies, and we have non-polluting technologies and air quality like everywhere is vastly better right I mean I remember I got to see the tail end of this in China so 2014 Beijing had to wear a mask always gray in the distance and then you know even by 2019 it had like basically receded to being almost normal right and I'm sure India is to go through that exact same transformation over the next decade as well. But one important lesson here is that this part did not happen by magic, right? This part happened because lots of people decided it's really important and they work hard to make that specific thing happen. So getting into DIAC a bit, right?

So one of the topics that I think we care a lot about is AI, right? And people, of course, have all of their different views about whether AI is going to be here soon or not soon, whether it'll kill everyone or whether it'll not kill everyone. And there's like a lot of disagreement on this and people have their own opinions. I personally have a very huge amount of just uncertainty on this topic. Right.

I think, you know, the 10 year timeline world is possible. The 50 year timeline world is also possible. And, you know, there are even longer worlds as well. And I see the arguments for like why any specific version of AI could go very wrong. But then, of course, remember that like the kind of AI that we have now would have looked completely unimaginable to people even 10 years ago, not just in terms of its level of power, but also in terms of the entire set of approaches that it uses to approach the problem.

And so we should be ready that 10 years from now, the approach is going to so we should be ready that 10 years from now, the approach is going to be totally different. And then 10 years from now, it's going to be totally different again. And so like, we should not put too much credence into like any one single galaxy brain logical argument, right? So this is like basically my view on this, right? And then one other piece, one other view that I have on this is basically that, like, even if AI doesn't actually kill us, there's this question of, like, human agency in a world, what does that even look like, right?

And there's a lot of people who basically glamorize worlds where you basically have super intelligent AIs that are making every decision. And human beings basically are reduced to just being pets. And this is like a world that I think in general, like people don't actually want to have, right? And I think in general, like, what's the point of developing the most amazing technology possible if your ability to impact the world basically at some point just drops all the way back down to zero, right? And I think even on top of that, there's a good argument that that kind of situation is pretty unstable.

And the reason basically is, well, if you're a pet, then ultimately those AIs are going to have their own interests, they're going to have their own goals, they're going to change over time. And even if you come up with a way to keep people in control, that's actually hard to do. It's very hard to control something that is vastly smarter than you. Imagine if, like, who here has a six-year-old child? If you have a six-year-old child, imagine if you legally were allowed to, and you did, come to an agreement by which you agreed to be their slave.

How many minutes do you think it would take for you to convince them to basically undo those terms and get back to where you were? Probably not too many. So I think human beings staying exactly where they are, but at the same time, super intelligence coming to dominate everything, it has risks that even go beyond safety risks. And I think we want to be cognizant of that as well. Also, the risk of authoritarian recently is basically that it's in-person protests are just becoming vastly less effective than before.

So, you know, in one of the things that often happens in Russia, for example, is basically, you know, there are some protests and then the protest happens. And then during the protest, maybe there's like a little bit of police opposition and it seems fine, but then actually like you're being photoed everywhere from every angle. And then two weeks later, you get a knock on the door at 3 AM and, uh, you know, then like, we know what happens. Right. So it's, uh, like AI also has like a lot of these, these near-term risks that are pretty risky.

So, but at the same time, it's like, this is all true at the same time as it being true that, like, all of this stuff is going to massively accelerate science and cure diseases and do all of those really, really amazing things. So, what is DIAC, right? So, DIAC, so, you know, like, this was drawn with AI that was available last year, I guess. AI that's available this year will probably be able to do the same thing but make it a video. So, it's, you know, it's, and basically, you know, there's this intentional theme of, like, solar punk castles with skyscrapers, right?

And the idea is, like, yes, we can have technology and, you know, like, we can have nature and we can be, like, well-defended and safe all at the same time, right? And so one of the analogies that I started with, actually, is, you know, like, good old Switzerland, right? And Switzerland is interesting because it's, like, the analogies that I started with actually is I mean like a good old Switzerland right and Switzerland is interesting because it's this like it for a long time for like basically ever since I became like a very online internet person it's like the one country that people would refer to at like internet libertarians would refer to as like a classical liberal utopia that actually works right and it's like you know how like peter teal back in 08 he's like yeah i've decided that freedom and democracy are no longer compatible and then uh you know in switzerland they're like no yo dog you know we got both right you know we got a strong classical liberal society we've uh you know very strong individual rights, and guess what? Everything is friggin' direct democratic all the way down to the level where in some cantons if you want to become a citizen, like even that gets put up to a vote. So basically the question is like unless the Swiss have like some super gene that makes them like morally more enlightened than everyone else, which seems unlikely according to my current understanding of biology, then basically, what might uniquely enable this environment?

And I think one of the answers is basically that's what someone gets defense for free. It gets such a high level of defense for free that it was even able to survive like friggin World War II with its entire self surrounded by hostile territory. And basically, you know, it's, you know, it has the mountains, you know, and of course, no, what's happening? And of course, the flag is a big plus. Yeah.

So basically, right, the, and then of course, the flag is a big plus. So, basically, right, and then, of course, what's the opposite of Switzerland? It's, like, the friggin' Siberian steppes, right, that just, like, get overrun by incredibly big armies. And, like, that's where all of the craziest and all of the most undemocratic stuff in the world happens. So, basically, the question question is like, well, what if we take this as a metaphor and we try to like make a world where the underlying kind of base level rules of the world look as much as possible like this and as little as possible like the opposite of that.

And I would argue that this is good for safety and at the same time it can be good for preserving a free and democratic society, right? And so the question is like, well, what does this actually look like, right? So basically, yeah, the question is, well, like, what are the rules of the world that we can change? So we cannot change the laws of physics, unfortunately. I mean, maybe we'll live in a simulation and like maybe if we, you know, just like put up the right words on the beach in the right font, in the right language, like it's going to unlock a cheat code and then we'll be able to just like, you know, like change the cosmological constant.

Maybe we can like set the speed of light to only 100 meters per second, and then just, like, see what social changes that'll result. But unfortunately, we don't have that ability, right? So we can't change the laws of physics. And at the same time, obviously, you can change the laws of politics, but there's lots and lots of people that want to change the laws of politics in all kinds of directions already. And so what if we think about the question of what are the benefits that we can get by changing the laws of technology?

And so here I have this. And so the D in D-ACC is meant to stand for defense, but also standing for democratic, also standing for decentralization. So it's a type of defense where instead of defense being done the way that like actual medieval castles did it, where the king controls it, and then that becomes like a choke point that the king uses to extract even more value from the economy, then instead, and also unlike the vision of defense that sometimes gets propagated by large country governments where the vision of defense is basically things are defended because you can trust me. This code is safe because this code is made by trustworthy entities. The hardware is safe because it's made by people in the right countries.

Why actually have a model of defense that's, like, more inherently democratic and inherently distributed. So I basically have this, like, four-sided typology, right? And so top level is, of course, the world of atoms and the world of bits, right? And within the world of atoms, I split it into micro and macro, And so macro is generally when people talk about defense. OK, so there is this whole other separate topic where when people talk about defense, like about a third of the time, that's actually a euphemism for offense.

So not doing that. But then the other thing is that to me, this also includes things like resilient physical infrastructure, right? So this includes, like, solving the problem for if global supply chains break, how do we, like, reduce the number of people who get hurt even in that scenario by, like, a factor of 10, right? So resilient physical infrastructure is one. And then World of Atoms micro basically means biodefense, right?

And in biodefense, there's like actually this huge amount of unclaimed opportunities to make society more well defended. And these unclaimed opportunities actually, like most of these, they do not look like what happens during COVID, right? What happens during COVID is centralized biodefense, right? Like, first, you know, you had centralized epistemology, and you had the World Health Organization, like, basically publicly say that, like, you know, you should not wear a mask. And, of course, that decision was, like, originally a result of, like, trying to overfit to this one specific situation where they wanted healthcare workers to get masks first.

But then of course, the world is a chaotic place. And like, this is why you're supposed to just like pursue truth stubbornly instead of being galaxy brains about everything. And so instead, like basically that ended up creating this whole meme. And then even in 2021 and 2022, you had people just like wiping surfaces with alcohol all the time and like not actually putting proper resources into making sure the air is clean, right? And then what happened in the second, right?

Basically, you know, we had travel restrictions, quarantines, lots of things that were, I mean, realistically, like that's like a massive social experiment on people's mental and psychological health. But if you put it through the FDA or an IRB, they'd say hell no. But we just did it on the whole population. So a lot of the stuff that happened there was actually high cost. Things that I propose for biodefense are all not high cost in that way at all.

Most of them are invisible. So a very basic one is just air filtering, right? Basically, so if you think about this room, right? So when you breathe out, your breath contains roughly, it's like 96% regular oxygen and nitrogen, about 4% carbon dioxide. And then there, there's like tiny fractions of stuff that also includes some of whatever viruses you have, right?

Hi virus. Okay. That's okay. Maybe I'm the virus. Maybe we're all the virus, right?

So, right. So basically, as it turns out, one of the things that we learned is that if this event was happening like literally outside right now, like imagine this roof disappeared, right? Then like actually virus risk for everyone here would reduce by like a factor of more than 20, right? It's like pretty massive, right? Basically, the majority of the spread happens through aerosols, relatively small particles that just stay in the air for a relatively long time.

So what we can do instead is one thing we could do is better ventilation. But then the problem is that in Chiang Mai, the temperature during the daytime is around 30 degrees. And so if we actually removed this and then we literally replaced it with glass, well, if we replaced it with glass, glass would literally have a greenhouse. It would be worse than 30 degrees, right? But we don't want that, right?

Instead of like trying to overturn the world, one simple thing you could do is just air filtering, right? Basically, with every, any room, you can put in a couple of boxes, filters, and that by itself can just massively reduce the risk. Another option, ultraviolet light. Ultraviolet light is, like, much more convenient. And, like, we've actually done some experiments of just, like, putting it up in, like, very small-scale events.

And it's just, like, completely unnoticeable to people. I mean, there is, I mean, it is more convenient. And arguably, it's also a higher-risk technology because, like, ultraviolet light does all kinds of different things. And so right now, it's looking like filtering will be the primary, and ultraviolet is more the backup option. But basically, if you have things like that, then you could plausibly reduce transmission in indoor environments by a factor of 5 or 10 or potentially more.

Then on top of that, if you do all kinds of other things, right? So early detection of pandemics, if you do much better testing. So, you know, 2021, to get a PCR test, you have to go to a lab. 2024, if you want a PCR quality test, you could do it with an at-home machine that's like basically about like maybe twice as big as this mouse. And then maybe in a year or two later, you'll have like a test that you can just like breathe into, right?

It's like, oh, hey, hey, what's up? What's up, officer? Oh, were you concerned that I'm drunk? Oh, no, sir. I just want to see if you have virus.

Nope, no virus. Carry on. Oh, wait, wait, you know, it's pretty, basically, like, these are things that become convenient enough that you're able to just, like, have them at home and just, like, add a couple of things that if we were to do them every day, the inconvenience level would literally be less than toothbrushing. But it alone, like, the combination of this stuff alone would basically transform us from being vulnerable to pandemics like COVID to even existing COVID, potentially just being a disease that just does not manage to spread. Right.

And so if you want to ask, like a lot of the time, like we if you want to take this back to AI safety. Right. One of the questions that gets asked is, well, OK, let's let's like not talk like get into this like galaxy galaxy brain mode of, oh, the AI is obviously going to have an IQ that's over 9,000, and so it's going to figure out some magical way of killing everyone, and we're not going to be able to stop it, right? It's like, well, no, let's get into the specifics and try to identify what are actual threats, right? And when you actually try to identify the threats, there's basically two things that tend to come up.

One of them is bio and the other is cyber attacks, right? And with bio, basically, and so, well, cyber attacks and also persuasion, right? So convincing the people to kill each other. And that just happens to be like, you know, these three planks, and then the actual killing step, also the fourth plank, right? So with bio, basically, the worst thing that you can do is you basically make a virus, and you make a super virus, and then the virus, like, stays in your body, does nothing for about, like, two years, and then it spreads to everyone, and then suddenly it kills everyone, and it's too late to do anything, right?

And so arguably, actually, short-term anti-COVID research, in addition to being useful directly, is like actually very useful even against the science fiction threat. But why? The reason basically is that, like, what is the one mechanism that we know today about how to make a virus that lasts for a long time and then is unnoticeable for a long time and then after half a decade kills you. Well, it's called the human immunodeficiency virus. You know, people have probably heard of it.

And like actually over the last couple of decades, we roughly have gotten vastly better at being able to treat it, right? And like actually people make arguments that long COVID actually has some, like has a lot of similar, and that's why it ends up persisting, right? So research into treatment of existing viruses also helps us against, like, realistic science fiction threats, right? So basically, like, it's not an either-or between, you know, like, worry about the science fiction threat versus worry about the threat that's right here in front of us. Like, actually, one is, the things that work against one are also very good things to work against the other.

So that's bio, right? Basically, there's like an entire spectrum of things that we can do where if they were deployed, the level of convenience, like loss of convenience, the level of like any kind of centralized power that would be required is like actually quite low, but it would make a huge difference. And this is like basically an unclaimed gain that like we really could gain with a surprisingly low level of resources. So that's bio. Then world of bits, right?

So I think basically everything that we do in the crypto space, very relevant to the world of bits. So this is like cyber versus info is kind of an artificial distinction that's unique to me. And the way that I define the distinction is basically that cyber defense is defense against things where all reasonable people agree who the attacker is. Right. So, computer hacking is like one basic example.

Right. Like the code obviously doesn't agree who is the attacker because if the code agreed who is the attacker, the code would just not let them in. But all reasonable people do, right? Info defense is like, it's much more challenging, because of course, one man's misinformation is another man's unjustly suppressed valid point. And it's like inherently much more subjective, right?

So basically, cyber defense, a lot of this is already happening in crypto. There's a lot of this happening on traditional Web2. There's a lot of work that can be done with encryption signatures, zero-knowledge proofs, the whole hardware side, certain types of blockchain applications, secure operating systems, there's a lot of points of intersection between keeping your coins safe today and keeping humanity safe from potentially, like, worst-case AIs if they do crazy things 10 or 20 years from now, right? And, you know, arguably, of course, the two are actually very connected because if an AI starts doing crazy things, what's the first thing it's going to do? It's going to get a bunch of money by stealing it from people's unsecured wallets, right?

And in TradFi, it'll be able to do the same thing as well. So that's cyber defense. And then info defense is like anti-scam tools, tools for collective truth finding, tools for identifying misinformation. But particularly, you know, this is a D-Act, it's not, you know, centralized, I'm the king, trust me act. And so the goal is to try to find mechanisms that do this in a way that do not involve, like enshrining one particular centralized elite that decides for everyone else what's true and what's false, right?

Because we've seen how those kinds of things are, like one, they're just flawed, and two is that they've, like, in the current climate, they're just completely incapable of, like, actually getting consensus and being listened to by more than half the population. So, and as we've seen, you know, community notes, like, people from both sides of the spectrum really, you know, like, follow it and generally tends to appreciate it. And the reason is that, like, it freaking goes after everyone, right? You know, it goes after the reds, it goes after the blues, it goes after, like, authoritarian governments, it goes after corporations. Oh, the most fun one is, like, I'm sure some of you guys have seen this, right, where sometimes, like, you get this advertisement and the advertisement says like, hey, this is an amazing product.

Buy it for $24 at this link. And then you have a community note that just says, hey, this is just a reseller. Here's a link where you can buy the same thing for $5. Like who here has seen this, right? So like basically, like this is I think one of the best things that Elon does.

He actually allows the community notes democracy to just totally destroy his advertisers. So the mechanisms that actually are democratic and that are open, community notes is fully open source, you can verify the outputs and inspect them yourself, actually do stand a chance of getting like past this kind of single tribe barrier and getting buy-in from everyone. So these are like roughly the four planks, right? And I mean, obviously, I think they all feed into each other in various ways. And there's definitely like some points that are kind of somewhere in between them, right?

But the core idea is basically that if you just ask the question of sort of how do we develop technology so that the underlying, at least kind of like middle level laws of that the world operates under, are laws that favor defense over offense and that favor individuals and communities over like very powerful centralized actors, then like these are some of the natural answers that comes out of that and there is a lot of value in working on each of those individual things. So thank you. Thank you. It seems that you guys have absorbed the mic. Can you raise your hand if you are holding it?

Over there. Okay. Are you up for answering a few questions? Amazing. Can you guys pass the mic forward for me?

Awesome. Thank you so much. So I believe that in order to defend ourselves, we need to know what we should be defending for. And in the current world, we're kind of in an accelerating phase of developing technology. So the complexity of the society kind of grows exponentially.

But for ourselves and for our institutions and for our states, the pace of learning goes kind of sublinear or at most like a linear pace. So there's gonna be a bigger and bigger discrepancy. So if the world is getting more and more complex and it creates more and more attractors, runaway scenarios, and feedback loops that we are not aware of, how could we actually defend ourselves in this situation? Yeah, so I think the thing that's hard and the thing that's probably unrealistic is, like, getting the entire world to continue, to, like, continually update their knowledge and habits, but to be able to, like, each individual personally defend against everything, right? The thing that is more realistic is solutions that can be worked on by a, like, relatively large and, like, open, but still compared to the world, relatively small group of people, and then those technologies can be be developed and they're developed in a format where they can be released they're open and it becomes easier for like institutions and like basically tools that be and places where people already are that already exist to adopt them right and so like you try to increase the like basically try to develop things where the path to adoption is a path that requires as little effort from almost everyone as possible.

Awesome. We'll go over to Star Wars here. Howdy. So I was thinking about this kind of binary you set up between the world of bits and the world of atoms, and I was curious if there's maybe room for another adjacency here. What I'm thinking about in particular is things which happen that are in kind of a psychological realm or a sociological realm, and even to a certain extent the anthropological realm, when we think about the interactions between humans and the problem of, like, wetware in the individual and then collective wetware.

And so I was curious, we have to reduce, right? So we can pass on information, but I'm curious about just how much broader does this go or do you, are you maintaining the binary on purpose? Yeah. I mean, I think like, to me, I would say anything sociological, I would generally consider it parts, part of the bits because these days most of society happens through bits. And I think like everything in the info defense bucket, I would definitely call already in that category.

One other thing that I talk about a little bit later, right, is that like I think this post is meant to be not just about DEC, but also about like techno-optimism in general, right? And so one of the things I talk about is basically the value in like accelerating development of social coordination tools and that has value not just in the context of like of defending against attackers but also even sort of in peacetime, right? Or at least within the context of communities that are already reasonably aligned with each other, how do you actually come to consensus quicker? And I think that also is part of the answer to the question of how do we actually accelerate the adoption of these technologies and avoid both people not adopting them because they're lazy and people not adopting them because they're concerned about, is this some weird thing that like, is this some, like, weird thing that's being pushed on me by, like, world economic people that want me to eat bugs, right? And, like, I think, you know, there is a technological part to that, which is basically that all of these technologies, they're open source things where, like, lots of people can look at and verify them, right?

But at the same time, there is actually this, like, sociological part of, like, being able to much more quickly identify, like, points of common agreement and actually execute on them. One of the things that I think we can do both, you know, like as the Ethereum community and as the Zuzula community is, like, really actually help to dog food some of these things. And also, I think, another important social technology to dog food is the technology of actually funding the development of all this stuff. So I think there's definitely a lot that we can do for that. I have a question about the sense of urgency, especially on the defense-offense part, because if we're not even talking about AGI, but just AI that is used for, you know, like drones with facial recognition or systems, surveillance systems, or better control mechanisms, it might be just too late if these technologies, if the, like, DEAC technologies are not adopted and scaled fast enough.

Do you agree with that? And if yes, like, what's... Yeah, I mean, I think there's, like, two different versions of it's too late. Right, one of those versions is basically, like, it's too late right one of those versions is basically like it's too late We're screwed and nothing we can do about it in which case like we should probably just time you know Retire and just like go and enjoy the time that we have left But maybe for a lot of us actually the best way to enjoy time is to nerd nerd out on really cool problems in which Case there's really cool problems over here, so you might as well join the other camp. And then the second possibility is basically that it's too late in the sense that, like, only using sort of relatively soft methods of, like, hey, let's, like, you know, like, get a bunch of crypto people to donate a couple hundred million here, like, make a meme coin and, like, donate a couple hundred million here.

Appeal to a government, donate a couple hundred million here. And, like, maybe get a few billion here. Like, that's not going to work. And, you know, we need something that's sort of more muscular or controversial, right? And this, you know, like, we saw some of this play out in the AI safety debates, right?

Where you have people who literally think that we're in a 10-year timeline world. And so there just is a tail risk that literally at any time, an AI could just go totally crazy, right? And that's why things like SB 1047 was pushed, though ended up failing in California. And then there was a couple of relatively similar bills that are being pushed in other contexts, right? I think also aside from SB 1047, the other thing that's going to come on the horizon is hardware regulation, right?

Because if you have AI, like basically it does look like AI for the AI paradigms that'll be around for the next while are hardware dominated in the sense that the version that you can do with 100x the hardware is like a very significant step up from then probably you could do with 1x the hardware and so one other plausible regulatory thing you could do is like basically get you know us China Taiwan like major actors to try to say like actually come together and like agree on a bill that says, at least large scale industrial training hardware gets tracked and then you maintain the capability that basically both at the same time can order a global shutdown of at least large scale industrial hardware. And lots of things are gonna slip through the cracks and individual GPUs that people have for their own purposes are going to slip through the cracks but if you can if we have the capability of like reducing hardware by a fact uh compute hardware by a factor of 100 for like one or two years then like this is something that a like a lot of AI safety people who are like of the very worried camp have responded positively to right so these are the kinds of things that you would talk about if you were in the more worried camp on ai in particular right and then if we do that then your arguments would be that like we need these more muscular things in order to make these shit hits the fan timeline a little bit more than 10 years instead of 10 years and uh but then at the same time, there's the question of like, well, ultimately, if you're buying time, you have to buy time for something, right? And like a world where you're like, basically, AI keeps getting easier and easier to make and like crazy stuff keep getting easier and easier to make. But then, like, basically, things keep becoming more and more unstable, then, like, at some point, you need to do, like, something else, right? And so you need a DAC anyway.

And then, of course, I mean, like, on the whole AI side, there's, like, you know, there's definitely a lot of people who oppose, like, both 1047 and, I think, what would realistically oppose hardware regulation and other things as well, right? And the, like, the two places where that comes from is like one is just like strongly believing that this like very risky 10-year timeline world is just not the world that we live in and two is like basically yeah seeing downsides of that in terms of like centralized control and potentially in terms of just like slowing developments down when like actually yeah ai development developments could do a lot of good right but then i think if you believe if you're the type of person who believes those things and i think for you it's uh like you would want to support dx stuff and then uh like basically yeah you know like and the dx stuff is valuable like in part as a hedge right and in part part to reduce the chance that something crazy is going to, like something medium-scale crazy will happen that will just push the world to start actually doing the more muscular stuff, whether it makes sense or not, right? So I think, yeah, in the intermediate world, basically you would need more, whether regulation, whether, like, much on, like, specific technologies or potentially other kinds of things. And, like, that's one of the things that you would actually, yeah, start looking into. But then, you know, like, I think the specific combination of these things just ends up depending a lot on like what in which combination of risks you're worried about at different time scales and this is something that's like like lots of people have very made-up opinions about in all kinds of directions we have two more questions and then we'll leave it at that so I'm gonna pass the mic to the next person.

So going back to Switzerland as a model for democracy, how do we prevent democracy from eroding individual rights? So they say that democracy is two wolves and a lamb deciding what to have for dinner, right? So how do we prevent that outcome? Yeah, I mean, I think, you know, there are lots of countries where democracy has turned into that. And Switzerland seems to be the ones where like empirically relatively less of that has actually happened, right?

So maybe come to Switzerland next year, and this will be maybe one of the topics of one of the salons and you'll have a great time. Great. So next question for you. Two quick things. One, on the pets note, it's been very prevalent in a lot of our discussion in the future here.

My cat leads a pretty banging life and also is influencing my agency pretty heavily, right? Influencing my travel schedule, this kind of thing. And so when I think about this playing forward, you know, the inverse case of letting the chimpanzees align the humans is like also suboptimal. So in a lot of ways, I think about this as like, how can we gracefully pass the power off towards positive, super intelligent entities down the road. So I think from a chimpanzee's perspective the better alternative might be like instead of passing things on and becoming what a chimpanzee is today, which is a zoo animal, you could why not become the human.

And so this is where the whole BCI roadmap ends up coming in. Alrighty. I think that we're going to have to cap it because we do have to be mindful that tensions game is happening after this. And I'd like to give everybody that's been standing in this hot room a little bit of a break. So let's give Vitalik a huge round of applause.

Thank you so much for coming. And yeah, just mingle, take like 20 seconds.