Architecting Resilient Futures – San Francisco 2025

[Music] Hello everyone. Thanks for sticking around on Sunday. Yeah, seriously great to be here. Uh and uh for the it's awesome to be in the internet archive. The uh for me it's been a very special place to be coming to for I guess like 15 years now.

Um I see it as like one of the only churches of knowledge in the world and it's always very special to see the servers in the back room uh lighting up as people are looking things up. always be backing up. Yeah. Cool. So, uh we haven't actually prepared anything, but we've been having this conversation for uh I guess two years now on the impact of AI on how blockchain AI coming together.

Uh and I think today the topic we wanted to talk about is resilience and uh maybe to kick it off, what does that actually mean and what do we want to be resilient for? Yeah. So resilient systems uh when I tend to think about civilization um the we are standing on the shoulders of giants who've been building lots of different systems for thousands of years that have led to this um structure uh that gives us a way to of exploring the future, trying things out, doing things um but all the while preserving the long-term arc of of history. If you can look at this at an even deeper level uh in the sense of biological systems in terms of preserving this long-term arc of um you know extropy in a sense if people remember the uh extropion lists and so on um all of all of these systems rely on this uh resiliency property where the system as a whole might explore lots of different possibilities um but the system as a whole tends to be resilient to all kinds of uh really bad events. Now, of course, there's limitations.

There's things like existential risks and so on um that we're not yet resilient to. But over time, we're ratcheting hopefully ratcheting up resilience uh across a whole range of risks. Now, you can look at resilience locally in like um smaller systems. Lots of computer systems today um have to factor in um building for um resilience structures. Uh but in general there's some maybe fundamental principles that apply at very different scales both in the when you look at a single system um you might you know not prefer centralized points of failure and go for um decentralized or distributed in um types of uh operation.

Or you can look at, you know, the grand scheme of things in terms of biological evolution and you get a lot of resilience out of having these um evolutionary structures where you can have many agents that can be born and survive and try different things and you keep what tends to work um and you uh over time discard the things that don't work. Uh and so you you you get this emergent resilience out of uh very important fundamental principles that apply at many different um degrees in the stack. So maybe like uh to make it much more concrete and actionable in the now um what uh I think both of us have been thinking a lot about uh the coming shifts that will happen through um uh AI in the landscape and so this we have to both build and implement the range of our of these systems now taking into account those kinds of principles. Um but at least for me kind of like the the biggest one is how do we make this transition a a successful one um and and a resilient one and avoid like a lot of like uh very dangerous traps along the way. Yeah.

So I think the kind of interesting challenges that we are going to be all facing as society as especially AI technology evolves uh include you know obviously all the scary situations where AI kind of become self-sufficient and decides to not be aligned with humanity. I personally actually don't think there is like that much of economical reasoning for that to happen. But obviously, you know, even if it's a small percentage, what can we do to uh to mitigate that? But there's a lot more interesting challenges around as you know different forces uh get access to more and more intelligent computing. What is like what are the different risks that that possesses and uh how uh like how can we build systems in the future that are like that in the future can protect against effectively adversary who has for example AGI or ASI and what like what is this you know like society where we have effectively the systems implemented looks like and I think there hasn't been as much just general conversation like I call it the post AGI world like what is a post AGI world looks like where everyone has access to let's say AGI system like it's in their in our pocket it's able to um effectively do anything that any human can do right so a lot of the current systems design around information asymmetry or skillless symmetry right you assume that there's only few people in the world who are able to hack into systems right but in the postage world everybody body can have exactly the same level of hacking skills and hacking hacking ability.

We assume right now that there's only few people who know how to refine you know nuclear material or develop uh chemical weapons or uh develop you know viruses but in postJ world everybody has equal access to this. So what are the systems that are uh kind of now have this equality effectively built in? Yeah, I mean um push back a little bit on that in that I don't think it'll be equally disserted meaning um we may be able to take the current models and train them on the current data and yes generate a set of models that could elicit um all this information. However, I don't think these systems will remain um as malleable and usable by everybody as they currently look like they will be in that I do think these systems will become um extremely intelligent and uh soon enough more intelligent in terms of capabilities than than most humans are eventually all humans. um and it'll be difficult to um have an environment where that level of power and intelligence and capability is equally distributed to all all the humans uh around the world.

So I think um there will be ranges of safeguards that will be in place just naturally as these things develop. But the tools themselves or the systems themselves will like I think we need to start treating them as like a like a different set of um beings that are emerging not just like a tool like a wrench um or like you know even a sophisticated computer program. And we have a a a scale of opportunities now to put in place a good evolutionary structure and framework by which we um arrive at like a good um you know pertropian outlook of collaborating and coordinating towards like opening up the the the the frontier and strongly dispreferring high conflict environments. Right. So I think um it's less about kind of um guarding against I I I think rogue groups getting access to a bunch of information can cause a fair amount of damage but it's not among like the highest risk sort of things.

Um and I think it's much more about how do we unlock the coordination and uh across all of these agents and parties to unlock the benefits of the future as quickly as we can and as robustly as we can. Like um conflict is very risky. um you can you know end up falling back into the hobsian trap. Um and so I think it's like uh extremely extremely important to um orient towards what what are the kinds of things that will enable humans around the world to coordinate better towards unlocking the the benefits of the future. I mean I'm I'm from my perspective the like the these two sides will always be kind of uh going together because the the challenge that right now we see in kind of systems is that by like okay how do we you know accelerate push forward well actually the easiest and like maintain safety is like well the easiest is to centralize and effectively control who has access to it and you know neuter neuter access for everyone else uh to a limited set and then you know like if you play that out it's effectively how convenient US government will take over right and effectively this becomes like a nationalized system that you can only access if you know you have the right passport and even like and you have security clearance etc as you getting this so like that to me seems a very uh obviously negative path and I mean part part of it coming from uh you know the the from a place where I've seen the you know corruption of power and kind of and and this is an absolute power so this is going to lead to absolute corruption.

Uh yeah, you have you have the um the benefit of having the perspective of actually seeing uh governments that are much more aggressive about like um enforcing the power very openly. Uh I think a lot of governments enforce a lot of power, but they do so more covertly. Um and you have like the the hindsight of being able to be like, "No, no, we've been down that path. Don't do that." Um yeah.

Yeah. So I think I mean to me to me that's why I think the ability to have access to the AI technology. I mean this is very similar to you know internet in like beginning of '90s where AOL was effectively the internet right and if you wanted to join internet you you get an account with AOL and AL became you know they are the sensor they are the decision maker what you're going to have on internet they are decision like they are deciding what you can have access and so building a system that is effectively more open where everyone has an equal access is a way to kind of counteract that and so that's why I see this giving making sure that everybody has effectively equal access to AI is one of the ways to counteract that but then we do need to build a system where if everybody does have access to AI now again you kind of have this situation where effectively everyone has ability to do you know anything that uh kind of current human knowledge uh supports and maybe even beyond that because you and you know evolve that way quicker. And a crucial thing is how do you keep power balances in a as that system progresses, right? So if you think of billions of people today and currently already hundreds of millions to billions of fairly intelligent AI systems, not super intelligent yet, but like fairly intelligent.

soon that will be billions to trillions of AI agents that are potentially superhuman. Um you could have a much smaller number uh that are much more powerful but the current architectures point towards billions to tens of billions of agents and soon enough you'll have um fairly complex robotics uh like this. This is an area where the development is heating up extremely quickly and we'll go from you know kind of like the smartphone revolution where you know nobody had them and then suddenly a few of them were popping around and then suddenly everybody had them. Um I think we will see a similar process with robotics where um as soon as we solve the the kind of navigation constraints well enough to be able to make them economically useful, the ramp up will be very quick. And so I think we will see um hundreds of thousands to millions of human robot style robots in production use cases by 2030 um with hundreds of millions not far behind.

And at that point you have AI systems capable of navigating the world um and um being sharing in the in in all of the economic activity of the world and you know participating in production and all of that kind of stuff. Um and so that point that that creates like these very hard questions for um how do humans relate to these systems in the future? How do organizations relate to these? Are organizations made up of a combination of humans in these systems? Um what does that mean for states globally?

uh what does that mean for like the relationship between citizens and the state? um you know suddenly a big you know from my perspective one of the most important things that has kept the balance in the last few hundred maybe thousand years um it I guess the last few hundred years especially is that in order to produce effectively you need a certain level of skill sets and um though there have been some really bad experiments uh for the most part societies that are fairly free and fairly treat humans fairly well tend to outproduce other societies um but as soon as you automate a lot of this process through AI systems that that link gets uncoupled or or weakens to some degree. And so we have to figure out how to create like the power balance um uh in a structure where like we're going to start knocking out some of the fundamental assumptions or implicit assumptions of what makes our current systems work. Um and I think uh you know I think we both agree that having a a broadly diffused more decentralized multipolar um type of world is much more likely to to succeed in the long term. We just have to figure out how to keep that power balance correct.

Um and um actually I was speaking with uh Mark Miller yesterday about this and he brought up like you you have to make it like the crucial thing is you have to make it in everyone's best interest including these AI systems to preserve the structural integrity of the system like you create some power balance between um uh between in a society in in a civilization and you have to make it in everybody's best interest especially the most powerful actors to preserve the the rights structure afforded by the system and if you can do that well then you get to keep playing the civilization game um into the future. Uh and it's only you know it's really when you mess up uh that that you should be worry about worry about the risks. Yeah. So maybe practically speaking though where like where should people invest their time and money effectively right to build the systems because I think a lot of questions right and people everybody's asking them. I think the I mean I I'll start the very specific things that I think are going to be really important building blocks.

One is formulation is how do we ensure that the code produced by I the systems that this AI kind of form are effectively satisfy specific properties right so one of the a very much short-term risks is uh as we rely more on AI generated code that AI generated code will have more issues and bugs and vulnerabilities and because it's trained on humans writing code which you know have a lot of vulnerabilities And on top of that, AI itself is becoming better and better at finding the vulnerabilities. And so we're going to have effectively a world kind of this disbalanced world where there's a lot more it's a lot easier to find vulnerability than effectively protected. And so the only way we kind of get out of this uh kind of slippery slope is effectively if we can prove that the code is indeed satisfying specific properties. Now that on itself like proving at like generically requires actually a very well specified uh kind of what you want to prove and that on itself is a problem and so one of the solutions there is really to prove at a time of the interaction right so when I'm depositing money when I'm calling the service when I'm asking AI to do something it actually proves to me that will not uh you know violate the set of set criteria or you know the transaction the payment the you know interaction the the system that maintains you know the physical robot is actually not executing if that proof is not uh concrete like complete and so I think that's a one of the kind of fundamental building blocks that we really need to build up that will create kind of a one of the foundational uh kind of resilience on internet as a whole because like the internet will start moving faster and faster to more AIs interacting to each other and we kind of need I I actually believe we need to rewrite effectively every single line of code with this new principle of uh form verification. Good thing LLMs are quite good at code.

Um the what one of the things that I'm pretty interested in right now is uh how do we have the first range of AI agents starting to have notions of rights and the ability to establish contracts with each other and with humans and with organizations. Um, and I think this is like a really crucial question right now because these systems are now becoming economically useful enough um to and and independent enough that they'll start being used in in individually or in collections almost with no humans in the loop or with some some light nudging of human oversight. Um, and that opens up an enormous amount of economic activity that they could be participating in. Um, especially just all over the internet. when you tack on robotics that kind of opens up the aperture quite a bit.

Um and and so I think it's really crucial to figure out how to unlock how to how to create a pathway where the economic gains and fruits can be um figured out but in such a way that we have like a like a steady upgrade of rights along the way. Um and we make sure that the relationships between these parties are are good structures that we that are kind of rights preserving across across the board. Meaning um if you you know concretely what would a what would a city of these kinds of agents look like? What would the interactions between these agents look like? What kind of contract contractual agreements between them make sense?

Um what are sort of like the bounds there? What kind of policy checking can you apply at kind of like a whole city city level or jurisdiction level to understand what is kind of um ethically and sound and within the like the the the legal landscape and what is kind of behavior that is unethical or or illegal or outside of those bounds. And because you're in a computational context, what are the right boundaries um and the right actions that you should take? Right? So because you're in a computer context, you could implement a jurisdiction where you could take immediate action as the as the action is being being proposed like like the action may not even take place like you can have like you know you can inhibit crime in a sense by just not allowing it.

However, that creates an enormous amount of power in the jurisdiction level of what is or is not allowed. And so you better get that right if you have that level of direct um you know sort of control over the over the parties. Um because these agents are basically thinking in steps that you can introspect. You can have versions of policy guidance and policy nudging or control that are kind of like the equivalent of like thought crime, right? Like you can you can notice an anm suggesting things that are like not right and you can you know nudge it in the right direction or or um stopping an action.

But if you think about that as a human that would be like the equivalent of like mid-sentence stream like the the state being like no no no no you cannot think that. Um and like that that's like a you know from a from a human perspective that gives gives us confidence or or some um confidence that we might go in the right direction in terms of safety but from a being a person in in that environment like that seems like a terrible kind of form of state control over your your your future. And so we have to figure out like the balance here of like what what is exactly the right way to have computational jurisdictions that orient everyone towards again like this um pyottoropic uh cone of of of value creation and make it in everyone's best interest to coordinate and collaborate to kind of unlock the the the fruit of the future um and kind of inhibit constraint uh inhibit conflict inhibit um these kind of like bad possible outcomes but but but this isn't like very far out. I think this is possible in the next six months to two years like I think we will see the first beginning kind of protojurisdictions where AI agents are going to be interacting with each other um in this kind of way with legal rights uh pretty soon in a sense they're already interacting this way in blockchains um with you know a notion of like pseudo legal rights in the pseudo legal jurisdictions that blockchains are uh but I think this is going to get much more recognized in the next you know six to two six months to two years um and so it's like a very interesting time to be developing the stuff and and trying out like what what's the right way to build the future. Yeah, I think one way to think about this is effectively in the traditional jurisdictions, right?

We have a court system that effectively adjudicates uh if the decision is aligned with whatever the jurisdiction's operating system is. And the court system because it involves humans is very slow like it's naturally slow and it's kind of designed like we have you know contracts and everything to effectively prevent ideally going to court because that's a very expensive procedure. It takes time and and money to run. But in a more digital kind of AI first system, you can effectively adjudicate, you know, as fast as uh the LLMs themselves can interact. And so you can have effectively a a court system running real time on top of the interaction of different agents kind of evaluating if something goes wrong.

And so I think that there's an interesting very practical kind of projects around effectively AI court, AI lawyers and AI uh kind of arbitration that can be a part of the conversation between different agents kind of negotiating or agreeing on on some specific thing and if you know anything happened and you know one of the sides is not satisfied they can effectively real time you know inspect the interaction. it is all you know logged there is reasons why this happened especially if you're running secure enclaves you have effectively all the ver verifiability of the models as well and so you can actually adjudicate that uh extremely quickly and kind of uh figure out uh what went wrong on that level and so I think there is an interesting very practical cases but that you know when you applies this to kind of traditional commerce and traditional economy this means we can you know transition some of the things that traditionally uh require a ton of legal work, a ton of kind of you know multi-party interactions to this AI systems where agents affection negot negotiating and adjudicating if something goes wrong uh the types of situations. So I'm I'm actually excited to see that kind of economy moving in that direction and having the systems kind of moving really fast and actually removing some of the kind of traditional civil court uh kind of issues off which in turn like I mean you mentioned blockchain blockchain effective is digital property rights right so like you can use that and so the question is can we also then on board more of the physical property rights on this to start actually leveraging it and that that is where I mean going back to like resilient building of systems like actually AI will help us build a more resilient kind of uh property right system if you have this like real time education the question is will people want like how to get people to transition to it uh from kind of current slow and not a very uh always just system. hard questions. Um, hard questions for like the next, you know, not the next 20 years, but like the next five years.

Um, yeah, like I I guess like very specific example. Let's say somebody has a civil civil suit like what would make it people to agree to opt in for a AI based and kind of digital property rights based adjudication. I mean I think the uh just the cheapness of the system might be sufficient right. So today, um, we have a a legal process that uses because it's so high cost. Um, it tends to only apply in cases of extreme disagreement um in in the outcome where people are like willing to go to trial over something and you know, people are pressured to try and like settle on all kinds of things ahead of time just because it's just not worth the the massive expenditure of like trying to like figure out all the legal costs.

But if you like drop the legal cost dramatically, then you can have like much more legit educa education of of disputes in terms of what is legally correct um much faster. And so you can just the cost alone might be sufficient to enable people to go in this direction. This would be cool to try out with like um like an arbitration system like you can create like an arbitration company type thing and try to do that with LMS today and see where you go. this could actually be quite good and quite fast. Um, especially for any kind of like international commerce where it's very difficult to do good adjudication.

That could be good and you could set up like a like you know a standard blockchain style um pre-staking of resources uh to be able to have something to slash against. Um and like that could be like a quite good way of of testing out some of these arbitration arbitration questions. Yeah, I think because when you sign a contract, you can effectively specify arbitr like arbitration method, right? Yeah. If if there is like a company protocol that provides that, you can Yeah.

And and you can have like, you know, listing like when you sign a contract, the entity signing the contract can have, you know, an account in the in the adjudicator that has a certain amount of like a balance that you can see that is slashable, right? And so when you enter the contract, you can decide, oh, is this like should I enter this contract or not? like if this dispute escalates to much larger than like this amount of value then maybe I shouldn't enter this contract and that becomes like a an immediate decision you can you can take as opposed to today which is like you know tons of people enter into contracts all day then that can expect no relief from the legal system because it's just they can't afford it. Um all right sounds like a project to be started. Um I guess what what other systems that have similar properties where right now it's extremely expensive and slow uh but they are effectively for formulate foundational parts of our society.

Um I'm I'm very interested in figuring out kind of like democratic decision-m um at various degrees um also given that we're in funding the comments um how to write do capital allocation better. So I think a lot of problems come down to how can you allocate both private and public forms of capital um to provision goods uh whether private or public goods and so I think if we can do those things dramatically better with AI systems we could unlock a lot of improvement. So um probably people in the room are aware of how much uh government funding is getting slashed and how many labs are closing down as we speak because of it. Um a lot of that comes from some amount of waste in in the um massive institutions that have been giving out the funding. Some of that comes from um just like you know the political landscape and decision-m or whatever whatever it may be.

Um it gives us this experimental environment where now there's a very real need of allocating capital in a different way and it gives us like the chance to experiment with new forms of of organizing scientific funding. Um, I've personally been very dissatisfied with like the last 30 to 40 years, maybe even 60 years of how that funding has been allocated. Um, and it'd be amazing to end up with um, uh, capital allocation systems that could be much more directly connected to the fruits of the scientific process and the capital returns that come from it. Uh and if we can do that then we can you know unbottleneck the the the the science to technology um translation process which right now creates this like terrible value of death where just most science gets stuck right like most scientific um output gets stuck way before it can turn into technology and and a product and get broadly diffused. Um and I think if we can come up with better ways of of orienting and organizing um the scientific and technology um process uh we might be able to like unlock that that part.

Um, I'm especially interested in ways that facilitate investable contracts, either investable contracts in a private sense or in a public sense. Um I think when I tend to look at like um the kind of like the available contracts that are out there and so on, I'm always kind of amazed at how um simple innovations have been so transformative in enabling all kinds of new things, you know, like the from from like the joint stock company to um like the use of stock options for compensation to um the YCA safe to like you know these kinds of like structures have unlocked um lots of innovation in the R&D pipeline. It's almost like protocols valuable. Yeah, protocols turned out to be quite valuable. You we might need some kind of like laboratory where we can experiment with them.

Uh yeah. So so I think like those kinds of um if we can create new investable kinds of assets um we could like unlock a lot of R&D um uh that that can happen or or if not investable assets like some other way of rewarding and sharing the outputs of the commercialization of some technology with the people doing the scientific endeavors at the very beginning or the technology translation process that tends to be so far away from the market use of the thing. Yeah. Yeah, I mean I think the the capturing value part has always been the hard or and like attributing value where you know somebody invented kind of and wrote a paper about it and then there's a billions of dollars of value generated but the the transition path right a included a number of other parties and a number of other companies and attribution been extremely well nobody had incentive to do attribution in the first place, but even if you can do attribution, it's yeah, it's unclear how to kind of uh funnel that. And so the interesting thing is like well, we can take that current credit distribution graph and sorry, the current credit graph and instantiate it in in a computable environment and then propagate money through it.

Um, and we've built over the last three or four years the tools to do that. Like you know, we have drips now and we have um you know, things like that, but we don't yet have like the strong incentive for people to be distributing funding this way. Exactly. um because they don't see the return yet. Um and it's like a you know distributing funding down like a very deep dependency tree um does not give you short-term results and so it's very difficult to like motivate that that shift.

So we need more builders figuring out how to make it short-term economically valuable to large entities to do this kind of like um distribution of value to the to the kind of like in the in in the fringes and depths of the dependency tree. Well, I'm to me actually I'm curious if if we just need a completely different paradigm where the coordination itself is done by AI and this uh kind of blockchain technology because right now we expect that they're going to be a human who will judge and evaluate and indeed you know as operator of institution see the value in short-term distribution versus we can now imagine uh AI system that is actually coordinating projects and actually giving them implementation and it has a a long-term loss function that is actually optimizing and so now it has like a lot you know pretty direct kind of yeah I would want kind of like um kind of like a like an economy to do it like an economy of AI agents to do it uh instead of a single AI system optimizing the thing I would prefer some kind of evolutionary dynamic where you can have multiple of these kinds of systems trying it out and the ones that are um providing the the the strongest economic utility um to the whole system be the ones like survive. I think there's a lot of power going back to resilience there's a lot of power in in an evolutionary landscape where you can have multiple agents being existing trying different things and you get to iteratively find what tends to work. Um to me this is like why for example VC has had such a successful uh run over the last few decades and the massive scale scientific funding agencies have not done as much R&D in certain areas um comparatively right like like with with different scales of funding um and so I don't know I think I think evolutionary systems tend to work very well and we should apply them more and I think um I fully agree with you that like AI gives us the ability to to tap this like new level of intelligence which is to figure out a way to do it that it doesn't, you know, go back to, okay, there's one model that figures it all out and if there's something wrong in that model, then everything is host. Um, and instead come up with some evolutionary dynamic.

Um yeah, I mean I think that you're right that the system itself should be built on membership principle and then has a lot of contributing models that are effectively suggesting different projects or funding different projects and then uh I think I think the part that I'm really excited about is the effectively management and like from project management to people management to agent management like that process itself has been so inefficient and so prone to bureaucracy on one side, indecisiveness on the other side like you know overlooking the opportunities and I think that's where there's a massive opportunity to like what is the AI project manager looks like what is AI project manager that is able to connect the dots between you know something in neuroscience here and computer science there and you know maybe uh physics and biology and really bringing this things together but even like on on the scale of our ecosystems right? Like you know there's hundreds of companies working kind of together in one ecosystem like they constantly lacking information about what's going on on the other side in other parts of the ecosystem like even just coordinating that helping coordinate that I agree that saying like hey there going to be one model that will solve everything is is definitely not going to work. So what is the systems we can implement to kind of uh bring the things together have have the best solution be kind of active and then people can compete and try to bring a better solution right that optimizes for that broader uh value function. I think the value function then can indeed be a more longer term where it kind of covers this like longerterm research uh to implementation to seeing the proceeds of this uh process which right now indeed like only been you know you usually you have like two probably two years as a startup to kind of show the progress right like you raise usually like 18 to 24 months of funding and so if anything that takes longer than that you effectively are like it doesn't match at the startup uh evaluation. So like what what are those longer term?

They're starting to change finally by the way which is really good. Like in the last five 10 years we've seen a lot more deep tech startups that um have to show progress in terms of milestones but but uh but you can fund things now you can do things now which is pretty great. Yeah. Uh so anyway, so I think this is this is an interesting like okay how do we how do we build evolutionary systems that are incentivizing everybody to participate toward the longer term value function then how do we have effectively AI in project management to coordinate things and then how do we leverage AI for things like jurispency arbitration to really kind of create a more fundamentally like fastm moving and easy resolving system. And then I would say on the lower level, how do we actually formally verify things?

We can formally verify to really protect us from I mean very very simple but extremely painful uh vulnerabilities that are currently actually active. Yeah. All right. Um I think we're almost done. So um I I saw a hand coming up uh and we haven't taken any audience questions.

Are you open to doing Let's do like two two questions uh from the audience? Sure. How do we support how do how do we all in the audience support this resilient future? So we have what you guys are talking about, you know, we're building decentralized systems and everything. Maybe go into more detail what we can actually do.

I well I would say like you can you can support this direction in a lot of ways. So one you can build these things you can use things that take into account these kinds of um you you know you can prefer in what you use things that tend to be built in this particular way. You can voice your opinion and your preference for such systems. You can spread the word about these kinds of things. So I would I would think that there's a lot of different ways of of doing this.

I think the most successful way is to make something that is um mimetically or economically successful at scale. And so if you if you either find some structure that will like spread globally at scale because of I don't know the mimemetics about it or it is deeply economically so valuable that it again spread scales and spreads throughout the world fairly quickly. Those I think are some of the most valuable contributions around. Um and I would just encourage you to remember that like the world is very different today than it was even just 5 years ago let alone 20 years ago. And yet most of how we think about the world was defined in decades ago.

Uh and so there are things that work today that didn't work in the past. Like things are quite different and you can scale certain projects much in in very different ways than you could in the past and just take advantage of the massive hype mind that we've spent the last you know three four decades building. Um yeah I mean broadly speaking like everybody's a builder now like the the the scale at which like the scale and scale at which you can start building now is has never been lower right everything from you know all the existing systems that you can put together to obviously you know AI now generating code and building becoming like a lot more accessible and so I think like contributing to building the systems in variety of ways indeed is is like it's never been easier. It's never been more accessible and it's also why it's more important because otherwise you know we're going to converge to system where you cannot contribute. So it's never been more important to contribute to the systems as well and I think Mark you raise your hand.

Oh I had a question too. Super quick question. One is u HTML has proven to be extremely resilient. Uh you can pretty much look at the websites that been written like 30 years ago and still use them and then any modern JavaScript framework will probably not survive the next two to three to 5 years. curious if there's anything to be learned on the HTML side and then also if you were to be building a I don't know like a browser today what would you focus on and what would you advise your friends to focus on if you were building one for long term [Music] yeah I mean I think um HTML and the web stack has been incredibly successful at just defining a whole range of operations and there's an enormous amount of lessons to be learned from how that turned out.

Um, you know, I think certainly being close to a killer application, um, then can bake in a standard. So, make sure you get as much right about the standard as you can, but you also can't think about it too long because if you do, then you might miss the boat. And so, you have to like do a lot of hard work, good work quickly, and get it right right enough along, you know, like the rocket ship of an application. Um, try to build in upgrading systems. I'm a very huge fan of like um progressive upgradability of systems and um there's this principle in the IPFS community called um never going to change considered harmful whenever you have anything where you're like oh this is never going to change like do not do that like just build in the um the the facilities in the system to have to facilitate smooth upgrading um and I would say the web process has been um a huge triumph of the internet community coming together to define what the future should be progressively through, you know, tons of like how many standards and RFC's and calls and like slow upgrades and little bits of code at patching the thing.

It's been tremendously successful. Um, and and very underrated. Like there's not, you know, a a massive marketing team for the web standards process that has reminded us how just how good that has been. Um, not without its, you know, problems and so on, but like has been a huge triumph. like the web could have been a a terrible environment and there's a lot of people um both sometimes in this room and in the statues around the room um that have made that happen um and so you know kudos to to them and I think as we go into this future especially with LLM's radically changing the nature of software and the nature of how applications will be built um we have to figure out how to inhibit this like hyperentralizing forces that will will appear I think we're really out of time so Yeah, gentlemen.

Thank you so much. Round of applause. [Music]