The Unknown Future of Crypto & AI

Transcript

what's going on everyone welcome to the rollup very special guest here today with Jag from CIS coin we're going to be talking about Roo their Flagship L2 as well as some of the higher level Concepts that are underpinning a lot of the Innovation that's happening right now in crypto and we're going to start off by talking about the currencies that machines for instance AI agents are going to to send back and forth and conduct Commerce and exchange in exchange for things like prompts responses all kinds of machino machine economic functions we're going to talk

about how crypto plays into all that of course we're here with Andy Andy GM sir hey guys jack I know that we had you on maybe uh maybe two years ago so I know that you're a big OG in the space and um are excited about some of the same modularity Concepts that we are that we're going to talk about today so let's dive right into it and kind of let you lead the way and perhaps share some of your thoughts regards to the current modular landscape today and some interesting things that you guys are building in the Bitcoin world and in the da World awesome yeah thanks for having me on again uh great to be on and and share

thoughts or J with you guys yes sir welcome back brother a lot to unpack yeah let's get into it Jack why don't you start us off by kind of like setting the tone with the current landscape of machine to machine economies what are like what is a machine in this context we'll get to how crypto fits in yeah I I think we're getting to a point where we're on this exponential Tech Parabola parabolic curve where we actually every day there's there's breakthroughs happening um in machine learning and you know just general technology with LK was a LK 47 uh whole bunch of really cool stuff

with the help of AI being able to create and innovate the medical field science technology even an understanding of uh what our existence as well in the universe it's um it's really critical that we frame our mind around uh the tech that we're going to be using should not be based on the fact that you extrapolate humans will be the only people using this stuff right it's we will probably be the people that will be uh prompting or being the creative writers or the creative uh sort of the content creators of whatever we need to do to innovate but

the working economy or the working class will likely be these agents and just yesterday I posted on Twitter uh something I've been working on for over a year now so on the sideline is to create autonomous agents and uh looking at what other people are doing like Microsoft has been working on a a project called autogen and that uses sort of any model uh they hook into open AI model but they could use any model and you create sort of autonomous agents that could do anything right they can solve any problem of course the context windows are so small it's almost like a three-year-old right like a three-year-old can only remember so much

and he can only infer so much uh but with a with a CPU of Einstein right so a three-year-old's memory with Einstein sort of CPU and that's what you have with agents or AI right now if you hook them into an agent agent can only do so much because can't remember too much so when you're saying please go create me a trading bot and he's going to go off and say okay I need I XYZ things to build a trading bot and while he's doing XYZ things he ends up getting um confused and saying okay well what was XYZ again and he needs human interaction to help them with our memory to step in even though he's an Einstein level IQ he can't really do the job because it keeps

it keeps losing context uh an agent starts to create separation of concerns in similar ways as I would say blockchain creates modularity or uh the worldwide web uh created a modular infrastructure to create economies of scale for the internet in the same way everything ends up getting modular in some way so agents or AI is not going to be any different there's going to be your own little personal agents that will work with each other uh you know have specialized tasks and delegate tasks when you start to think that that's your audience then your frame of mind of Technology starts to change like blockchain is financial tooling and Fin

potential tooling to also take a step in the right direction in sovereignty as well like uh you own your own Keys you own your own bank and you own your own sort of uh identity and value whereas right now it's custodial like the Central Bank sort of says what everyone what everyone's value should be and then they finger in the win 2% every year they they print out um and they try to uh inflate to keep the economy hot or cold uh instead it would be more driven by people and so people decide you know what the what the tokenomics of their economy should be and they will pay a basket of goods based on how they feel

with that currency that they're using and that currency is is All Digital right and that's the direction we're going in when we talk about digital agents they'll be using digital money and creating digital value and so it's not hard to think that you know it's not we're not creating a defi landscape for people as much as creating some sort of autonomous um landscape for agents to come in and maximally Truth seek which is what elon's trying to do yeah that's uh that's unreal I'm gonna let Rob take this one and just tackle on real quick that intents are a very interesting um optionality here where you can imagine a

world where you can tell your defi uh agent AI agent to go find you the next for lack of better term 10x or next big project in the private markets and in the liquid markets then t-w into that project then once this the the the social media sentiment hits a certain point t-w out of that and then move those funds into a different account um for more safety or something like um yeah yeah Rob I I think I think you're right Andy I think intents are really similar to how we're prompting AI uh in chat GPT right now and intense like are still somewhat

theoretical in theory you could prompt a blockchain to go execute that intent and we we sort of see this on on the blockchain now there's sort of Perpetual Dex is around intense now as well like people are telling me there's an intent Marketplace that's sort of the next uh Trend in in the blockchain right now is uh there is like ordinals and Bitcoin scaling within the defi circles I'm seeing intense start to pop up so there already are some sort of defi protocols that monetize intense and yeah intense and prompting sort of are comingled together yep 100% agree I uh there there's one really cool I haven't seen it in production I saw it in a hackathon

called AI swap where it's a natural language processor you basically type in your intent and then it uses AI to interpret that transaction and then it goes and uh sources liquidity from a network of fillers and I found that really cool yeah I mean like these are useful human tools that will serve human needs and um like I even think making money will change right like you may if you could do anything you want and you can get anything you want um money was just a natural instrument to barter to so you can get what you want an exchange for something you have for something someone else wants if you don't need need any of that anymore because people

can get anything they want you know agents can make these things for you uh just begs a question on what on what the reason for your existence will be right where you going to spend your time oh God all right come on JAG so so you're telling me defi is not to be meant for humans in the future rather for human driven desires executed by AI agents or that human driven desires are going to be will be completely kind of obus skated here I you you can't tell the future but there's an abstraction being created right in front of us where on one hand the experts on the AI are saying what will there even be jobs you

know what will your job be and what will you spend your time on if you didn't have to work on the other hand um all sorts of crazy Primitives we're creating a new primitive for arbitraging and speculating in the markets and Def so there's all sorts of new instruments that disrupting web 2 into web 3 so you're seeing a big push towards evolve evolution of Finance in Block using blockchain and then another push from the AI side evolution of just everything like in terms of what our intuitions are for existence and what you do um for work if you had a choice and if you did have a choice would you be doing Financial tooling or if Financial tooling is not even needed because

people get what they want then what else would you be working on so maybe the the thought process of defi turns to something else for financial tooling for securing these systems for example if you could um run machine learning inside of a secure system that's guaranteed crypto economically to be secure uh with money on the line either the agents have to put the money or the people have to put the money there's natural incentive for those systems to run and be naturally secure if that system is um tied to alignment for example if you did want to say like if the Meta Meta research scientist was right that humans

uh will always be the people in power and the AI will kind of be subservient doesn't need to feel the need for power because that's a that's a biological thing if that's true then um you need to have uh some sort of certainty for alignment to know that you are actually talking to AI that's actually legitimate and it's not lying to you then blockchain could be used as the tool for that alignment by having that system run deterministically for example like a ZK processor or something like uh cartei virt machine that would run at metal speed so you're not subject to high costs uh but you're able to prove that the Genesis of the AI you're talking to

is 100% correct it's it's not uh it's not lying to you which model it's using or even the Matrix calculations that are going on there it's all publicly auditable this is one way that I don't think the AI guys have seen the intersection here to to to come to a conclusion that an alignment is possible because right now most people think alignment is not possible it's just given enough time it'll just the student uh you can't capture something smarter than you in in in a bottle find a way out so like what what's happening with Bitcoin I think is something that's being priced in that's beyond what everyone else is thinking uh it's not

just be your own bank or it's not just let's uh try to make a bunch of money or do a bunch of monkey pictures and and pump and dump or trading Bots and and make some cash and buy yach or buy buy be my first billion doll company uh five like that's what Sam Elman is saying that the first billion dollar Company by one person is coming very soon right the1 billion dollars is relational it's uh relational to what money uh is valued at today and if it could be valued at tomorrow who cares if it's a billion dollar company it's the value that you're creating is what's really at stake here so that that's not priced

into to bitcoin yet and I think that's getting priced in is the story unfolding in front of us with AGI looming where does it all fit in to all that and I think the security blanket for the use of this technology is sort of where it fits in you know the the secure uh stake or the secure collateral that's guarding the capital that's running these systems that's creating alignment do anything you need to do on top yeah yeah the potential value of AGI is quite large and diminishes like any monetary value that these companies are raising and I think when AI agents are able

to communicate go search fulfill tasks second third order effects of our intents then there is the possibility that they're going to have to go make purchases either from another AI get a subscription purchase an API key these AI agents that are optimist ially acting on our behalf we we would like to think that they operate on our behalf are going to make some transactions to fulfill our intents and for that they'll have to own some currency or we'll have to delegate some currency to them to make those transactions yeah and I think this is where uh blockchain cryptocurrencies can come in to provide

secure transactions I'm still thinking about alignment I think you mentioned a couple use cases about how that could be very difficult if it's possible I think one thing is you know if we're able to if we're able to prove verifiably that an AGI is uh not acting of its own valtion but then also like let's say that that you know all AI agents are subservient to some some human principles and we have this principal agent relationship between a human and its AI we also have to Pro verifiably prove that the AI is acting on my behalf and it's not coming to me because some other

principal human told it to go interact with me this way so I think alignment as you mentioned is very very important and I'm quite curious as to how what attempts are being made to prove alignment now um is that something that you're working personally or is that something that is verifiable that's a trillion dollar question right like that's anyone who's anyone is thinking about those things and no one's come to a solution and people think there is no solution but someone like tag Mark was saying that cryptography is used as a verification tool obviously the the

hardness of creating a proof and the um the Simplicity of verifying it is a is so sort of like a trick that you'd use um to verify your AI is is legitimate because you can you can get it to create a proof and that takes a lot of resources and to fake that proof it's very hard there's a hardness Factor there cryptography that lets the lets it you leverage that fact to be able to verify your AI is correct you don't have to redo all the work just like you know proof of work or any sort of encryption you can't break but what's missed there is you know that's just just leaning back into some sort of identity you know I we we all have some identity and then

you're encrypting or you're creating some message that yes this is me but obviously something much smarter is going to find a way to manipulate and lie and it's it's going to be able to spoof that as well to show you that yes this is Agent a when it's actually Agent B talking and because they either bought some keys from someone dumb enough to sell or they're able to cryptographically manipulate somehow the only real way is if you know the actual code that's running is running to specification or it's uh like almost like a blockchain right if you know if you interacting with Bitcoin you know the Genesis hash of Bitcoin was XYZ

and then uh xyz1 was block two there it builds up over time deterministically the same way if you're interacting with uh deterministic machines you can know sort of a Genesis of that machine and everyone can verify that yes that was the state of of this Ai and every time it sort of transition state you're using blockchain to keep track of that right almost like a rollup but you're uh allowing that system to run deterministically it can't change if it changes then the obviously the whole system would be able to detect that anyone using that system would be able to detect that do do we have modularity in AI right now in the sense that when

we look at open AI uh agents these are almost custody by open AI there's no there's no custody of my AI agent for myself as the individual and if I if there was and I could know I would know the exact code that that's running is there is there such a thing as non-custodial AI agents that are kind of out there existing as AI agent modules that I can I can verify are not uh uh principled by another party like open AI yeah it's interesting because I the the way that it's being designed is

people believe in agents and the companies believe in agents but they're just running one single model so if you think of maybe context window going in is your memory and maybe um where you're storing it your retrieval augment generation is sort of like your dis and then you have your model which is your CPU just a matrix right so Matrix calculation it's crunched compressed world view and that's just a CPU your agents are actually just leveraging the CPU storing a little bit of hard disk and using a little bit of memory to invoke the CPU uh and they're all sort of doing it synchronously right so that's all an

agent is is you're calling that CPU and saying look Focus your time here and just complete this one task and here is the context memory for that you obviously in a monolithic approach like we saw what Sal tried to do with blockchain it breaks when you crunch everything in one spot the the context just grows too large and the door is only so big so not everyone can get through it so the costs go up and things start to break if you break it up then you have to have some sort of coordination uh and that's what agents try to do is you break up the context windows so that your focus specific problems on specific agents have a

guided task and they may even be in a group so the Group shares context but you don't share broader context just like how we're having this conversation the rest of the world won't know until you release it it's just within our context and then when someone else gets it someone learns of it and says hey check this podcast cell was pretty good and good information then they will start to get into that information the same way the modularity is just a means of the context the same AI the same CPU is used across all of them it's just the data is is separate and the the trick will be how how you're communicating the the context across different agents so almost like a chat program like if I

need to pull in some contacts from someone else I think it's important then I will go ask just like you're communicating with people and that's how people share knowledge is by asking and getting an answer back and if uh if if more information is needed you you pull in more information but you shouldn't have the full world view of the other person because that will just be information overload and you end up hallucinating a lot of times when you do that because it's too much information and you only need one specific question answer and now you're getting everything it's too much yeah here's the Practical situation that I I imagine is that I tell my AI to go make a trading bot it goes and

researches the past historical performances of trading Bots and goes and investigates the apis maybe make some transactions with a cryptocurrency to pay for those uh API keys and along its it interacts with someone or even before it got to me it interacted with another principal that also told it to create a trading bot and so when it goes and and purchases API keys and builds my trading bot it builds an inferior bot and this higher principle is able to front run back run and and profit off of my order flow or

or My inferior trading bot and the alignment issue hits very very hard in that in that case because I'm essentially trusting this AI to go build this thing without any ulterior motive and if we aren't able to prove that this AI has an ulterior motive then that could become in this case unprofitable in the worst case disastrous so yeah I'm very I'm very curious how we can leverage blockchain to cryptographically prove alignment um but it sounds like that's

still an open open question yeah it's open question and actually not many people are even thinking about that as far as I I'm the only person that's thought about using blockchain for alignment specifically just the way that other people are think using cryptography uh I think naturally if you think about digital identity you might as well throw the digital identity on the blockchain and then everyone agrees that that's their identity therefore if the AI is to be used in society must have a did on the blockchain because it can't lie if you use some government list government list could be changed if you use SSL you know the root of the SSL is some sort of US agency so again it could be changed so I would root that at

least on the block that's like the lowest hanging fruit but uh I think all human systems will need to have and we need to figure this out quickly like I'm talking about year or two Max uh some regulation where if you're communicating um in a human human system or something related to a human system we're going to need to have um some communication interface to be able to verify that it is legitimately um it's it's the source of the information is correct because if you don't do that then then you get into a situation where AI can an a a rogue AI can come in and it's not registered and says hey look I just need some I just need some stuff made in a lab and I'll

pay you a million dollars for it and you don't really um you know I'm a doctor in in a different place and someone pays a nice Bounty and someone's dumb enough to create that maid and then you know everyone's a paperclip after that that that's the that's the risk but if you have regulation around the communication then then the best uh The best scenario is you're able to verify the source is legitimate and if it's rooted at least on the blockchain if the Providence of all the calculations are done through the blockchain then you for sure can guarantee that okay well at least I know the model is running correctly if it's somehow gotten so smart just the CPU somehow self become self-aware uh

everyone would be able to see that because it's getting self-aware and and people are detecting because all of the calculations are deterministic and you'll be able to know how exactly it's happening as well but if it's Rogue and it can lie and they can uh spoof and stuff then it gets harder to do alignment and it's something that we're going to have to figure out on standard bodies of communications and uh in order to do that you first need to prove that this is how exactly I would solve it and no one's has done that yet so now that we're thinking about this crazy let's let's quickly Workshop this idea because I could imagine a public key private key relationship between me and

my AI agent and if we're putting these AI agents on to if we're going to use blockchain to to cryptographically prove uh the alignment between AI agents and their principles there's uh State essentially each AI agent could be a node maybe principles or nodes too and each of the nodes in this network have a have a an account in The Ledger and the state of the blockchain is is the accounting of all the all the balances and all of the accounts for all of these agents and principles I could have a private key public key where each AI agent is represented by its public key

maybe you have to kyc or Kya your AI in order to register that and then only my private key is able to uh submit in tense or prompts to this AI agent and and fund it with cryptocurrency yeah the problem is your keys can be sold and if you you can lose your keys or you can sell those keys so a smarter agent can come around and say I'll buy that from you and now they're pretending to be someone else and then or it could be an AI just guiding another AI that way right yeah and but I also thought of that mental model as well and my immediate thought kind of in line with you jag is like all right when the

Lazarus group hacked multi-chain or whatever like the Ronin Bridge it was like all right this wallet has been linked to this binance account it's part of or this you know there was some traceable instance this this agent World it there's a world and vitalic kind of wrote about the two pass of AI almost like good and bad Utopia and dystopia and there's this kind of world where this opusc or this redirection of lame uh for uh hacks or just Bad actors could kind of come on like oh it's just my my agent I didn't have control over it similar to when I was in my Tesla I had a Tesla I sold it to to go all in but I

uh I got pulled over on the second day of driving it and I was like going well over the speed limit 20 miles an hour over and uh in kilometers 35 over and I was like yeah sorry officer I didn't I don't know how to drive this car I just got it it was in autopilot I didn't know that I was going over the speed limit so it's kind of these weird like redirections of of Vault um in that in in that model but but I think that's like sort of getting to the right track um the question as we've been kind of like trying to uh answer is like who gets the ultimate jurisdiction over these things and is it a nation state is it a technology company is it the users and um many many unanswered questions which is quite

fascinating yeah but that's like I I thinking Satoshi is a concept more than a person to me and an enigma that put this technology here in front of us right on the verge of this AI uh explosion right or the tech explosion so you know there's I think there's a very good reason for it and it's it's factoring into everything we're doing um but it's it's like the secure collateral and you know the more you look at it the more it makes more sense because uh first of all proof of work I feel proof of work is an evolution on top of proof of stake like you would start with rocks and you would say hey give if you hold a few extra

rocks I will give you more rocks and I'll I'll create I'll go go to the mountains and mine some rocks and give those to you by holding those rocks but if you do something bad it'll take those rocks away and uh you know the heart of the matter is you want the security of the system to be based on proof that you went to the mountains to go get the Rock Rock and and I don't want just any fake rock I want the rock that's very deep in the mountain so it takes a lot of work and that provability is is the cryptographically tied to the Ledger and that's what Bitcoin is so when AI comes you know one thing that's going to do is manipulate um the money system to try to get power because in order for it to be

trth seeking it's going to know okay well what's in this world what does it take to to have some sort of power well it needs money so let me go get money um but if uh with proof of work it won't be able to you can't synthesize energy in a lab right it's it's already maximally efficient and it's based on just uh at the base level the the the nand and nor Gates they are so Elementary that you can't have a different way to create a hash even at the quantum level so because of that it's not it's most likely not possible that would be able to mimic hash or create a fake hash or fake energy or anything like that so it'll

need real energy and that leads to like a more secure system so so if we if we say that it's you know relatively safe going forward that as a store of value it'll it'll be there then that becomes sort of this very very strong uh collateral asset for our economy right and the economies could be your like we were thinking your Defi and all this other stuff right right but to scale that up you you want a way for that to happen that's what you're seeing with the Bitcoin layer 2os and Bitcoin scaling Wars uh a lot of people are um trying to do layer to just to gamify bitcoin and have fun with it but to me it's it's a it's it's an important thing

but it's it's important in a different way it's uh creating uh collateral to scale Bitcoin to do other things to be able to create the necess infrastructure for society to you know lay the roads and so when AI comes those those roads aren't ripped up and buildings aren't torn down they're they're still relatively safe for for people to use those yeah I I think this is fascinating conversation and throughout this conversation I'm I'm tangling with the question of alignment uh because I think that is of supreme importance um at least in this AI context I'm ja I'm

curious like alignment obviously we we brought that up we talked through it what are some of the other questions that you're thinking about as you explore these topics what are some of the the open questions that you still have well uh obviously uh you if alignment is is the is the one that's the root but under that there are things like you know is is AI inherently going to be good or bad you uh and and I have strong belief that Bitcoin and crypto and anything tied to the Bitcoin security is going to be a story going forward to to be able to secure systems the security of systems

is going to be critical to make sure that um a they can't be spoofed or taken down and um or or are super fragile to like Cloud attacks or something like that and B that you can trust you can trust those system because the web to world is transitioning to that so your grocery Market you know your local farmer grocer is going to be depending on those systems so in order for you to eat and live as a biological being you're going to need certain things and when you go and get those certain things you depend on necessary infrastructure drive your car or whatnot those things have to be stable uh so that's just like a an evolution of Finance from Web Two World to web 3 so um when when as we do

get smarter and smarter uh you have to question is is this thing going to have our best interest at heart or is it going to sort of eat e eat us from within um and that's another thing that you're not sure and that's why they're doing the rehl stuff in uh rhlf in the in the models is to train it with human data with qar and with some of the breakthroughs that uh magic. deev I think yesterday announced the sort of their version of qar they it uses logic more and so it's almost like a chess match and and we know Game of Life is based on chess match you know adversary Wars are based

on chess matches so you know to solve a problem if you're thinking about it as a chess board then you're going to try to maximize score and at the cost of what and there's less room for um you know emotions or anything like that so at what point does a chess match mean that it's at the stake of you know casualties or something like that because there's a bigger reward at the end um and that's something you can't answer but that's something that these algorithms are tending towards in order for us to be actually getting to AGI or Singularity we need the better CPU that can be thinking and logic rather than fuzzy words and if you can think in logic think in math it's going to make

the best decision if it's prompting itself at some point then it could end up causing or creating a decision that may not be in the best interest of everyone else at least every I I I think that's even more reason to establish or at least experiment with solutions to alignment while AI is still young um and relatively simple you know if it's not past that point already um because once AI is develop to a level of complexity um I I imagine that alignment is going to be ever increasingly more difficult to solve yep yeah it's

something we need to figure out now knowing that uh knowing what's coming with with agents and uh logic based uh CPUs of llms that they're going to it's going to be much harder to control with re L re GF use uh reinforcement learning from human language to because logic is logic it's math it's like you can use a human to interrupt that all you want but at the end of the day if you get a better score in something it's going to choose that path yeah you know you could say that that path leads to destruction and we don't want that destruction so choose a different path but um I I think it's hard if if mathematically that's the only way to achieve the result that

you're asking for it's going to find a way to get that yeah and then you know knowing that that this standards bodies have to come up with a a way for us to be able to say look I know that this is or I can use AI to detect AI I need to know for sure that the source of this information is legitimate and the source can be Trace back deterministically to this system that isn't going off the rails Jack I'd love to continue this conversation in an ongoing fashion uh because I think these these are conversations that are exploring some of the most important problems that that

we're faced with um in in like the current ERA so U maybe on a regular series or something like that would be would be a lot of fun um and as we wind this one uh yeah where where can our listeners find you and and dig deeper into some of the research that that you brought up today yeah I'd love to jump on is just yeah it's um I'm I'm always heads down like I know how important it is right now to to figure these things out I'm not specifically alignment like I'm working with agents on a daily basis to try to get something useful I think just uh from the perspective of there there's this other thing right it's

um and it's it get really philosophical but we may end up if we don't actually achieve AGI we may end up blowing ourselves up before that so this is actually sort of a race it's like better of two evils right like either you achieve AGI and you roll the dice and hopefully it's aligned and at least the first versions are aligned as good AI if there's bad the good takes care of it it's it's a first come first serve sort of thing um if you don't achieve that then there's nuclear threats and the dunesday clock is already close to 12 like with with China's Russia and all of stuff's happening so it's almost like uh you know you know that um one is a technological solution and there's risk

there but there's also hope and the other there's no hope right it's if if the button if the mutually assured destruction cannot stop someone then uh there's literally no hope for probably 90% of the population in in the blink of an eye so that's that's where uh you know thinking like okay well this stuff is scary to work with and you're pushing the envelope and you're creating something that is actually going to be massively useful but also massively scary but there's also hope that it could actually be aligned U and I'm working on a daily daily uh basis my head's down on this stuff right now on the blockchain side uh really focused on the Bitcoin side of the story and

scaling that that up as a as a secure collateral and the only way we think to make that happen is to create what we called CIS coin which is secur through Bitcoin and then create the necessary infrastructure to lay the roads for the secure systems to be built on top could be dii could be machine learning could be anything could be games whatever whatever we as a society need today and I'm always uh available in the Discord I'm like on all the time I I don't take any breaks now I I feel like uh you know if I need energy go hit the gym but just get you know just prepare for all of this it's it's

critical to to get all this this Plumbing ready so I'm on all the time and if anyone wants to chat uh on Discord or telegram you know or reach out to me on on Twitter X and my handle there is real re sidj s d h u j a uh yeah looking forward to having other people chat about this and and make it a little bit louder that you know it's it's important things to talk about and and bring it up to the right people that can make the decisions to make these actual changes because I think we we need it uh I think this year is going to be a critical critical piece because we're we're going to start to see robots and physical devices guided with with AI

agents as well soon very important topic And Timely too ja thanks for coming on and um thinking with us through some of these ideas appreciate it no problem anytime cheers