{"id":"NGxMyAdFGfo","title":"NEAR AI Office Hours 39 Eliza Labs - for review","description":"NEAR is the Blockchain for AI\nWebsite: https://near.org/\nNEAR AI: https://near.ai/\nIntents: https://near.org/intents\nChain Abstraction: https://www.proximity.dev/\nSharding: https://nearone.org/\n\nFor the devs\nGitHub: https://github.com/near \nDev Docs: https://docs.near.org/\nDevhub: https://devhub.near.page/ @NEARDevHub\n\nConnect with socials\nTwitter:   / nearprotocol  \nBlog: https://near.org/blog/\nMedium:   / nearprotocol  \n#NEAR #nearprotocol #NEARAI","channel_name":"NEAR Protocol","published_at":"2025-04-01T00:00:00.000Z","thumbnail_url":"https://img.youtube.com/vi/NGxMyAdFGfo/hqdefault.jpg","transcript_text":"Hi everyone, welcome to the episode 39 of near office hours. Today we have talk show from El OS. It should be a fun episode. Uh, do you want to say a couple words about yourself and what Elite OS is all about? Yeah, so uh, Elizos is an open source AI agent framework. Uh, I am an AI agent developer. I've been doing that for a long time. It's my particular interest and passion. Um, and we got some traction by uh building something that spoke well to the crypto community and let agents have wallets and do interesting things that I think people had not uh seen before in the wild. Uh, and so we're about to release v2 which is dramatically easier and more powerful. Uh, and I think people will be very very happy and satisfied with it. So I I've been reading up uh a lot about the lead OS in the past I guess 24 hours and there's a lot of things happening within within the ecosystem right it's not just the the the framework it's uh it's there's the DAO there's a there's a trust some sort of trust system so can you give us an overview what what like the core components of the OS and how they interact with each other yeah I mean I think we embody um that you can just do things and that uh at the core of it uh it is a community of about 25,000 people some people very active most people just kind of chilling um across Discord Telegram and other places um who are building the future together um or are sitting on the sidelines and buying coins and investing in that future and you know having their opinions um uh so yeah um uh We uh at the at the start uh my goal was to just show that uh autonomous agents were here and that AI agents were actually like pretty close to being something that's ready in our lives. And I kind of always balanced like serious programming with kind of an edgy performance art of like it should be fun and viral and also a little bit scary and a little bit like hey guys this is coming like it's not here yet. It's kind of the toy version but you can imagine in 10 years it's going to be really here. Um, and so I want like humanity to be really mentally prepared for where we're going. Um, and and ready for it and almost bored by it when it gets here. Um, and I really want to try and tackle the hard user experience challenges of agents. Like how do agents manage our keys and do things for us in a way that we feel confident in? And how do we make sure they don't get scammed and lose all of our money while they're out shopping for us or whatever. Um, I think there's a lot of um a rich space in there. So, we're really challenging and and investing into that. We started out by building um well, we're going to build a community investor using this marketplace of trust. And so, the idea was like, okay, well, um if if we're in a bull market, um I want to make some trades, but I I as an LLM, I don't think an LLM is particularly good at making trades, but there's some humans are really good at making trades, and LLMs are good at identifying signal in like large groups of information. So, could I have an LLM look at the trenches talk and all the people making trades and then extract the trades from that talk and I and and take them at face value and trade as though I were listening to all of these people. And if I were to paper trade all of these people, is there anyone who would have made me money? Right? And I would I would posit from our our our testing that about 10 to 20% of the people uh seem to be significantly better at trading than everybody else who's just posting garbage or is like literally exit liquidity for everyone else. Um, and if you could identify those people, whether or not they're rich or you know their wallet address or anything, you could basically just take their recommendations and their shills from their like alpha chat or whale chat or, you know, trenches talk channel or whatever it is and you could use that as a signal for investing. Um, and so I I would not go out and say like I think that AI is good at trading and investing. I I would never say that. But I think it's really good at being an intermediary between like a community of people who are investing and and using them as like a source data signal. Um so so we're we're building a few different variants on this. Um one of them is like you can kind of spin up a mini investment pool for you and your friends or your family or whatever your community. Um and you could invest together and um sorry there's a loud emergency vehicle. Um and and so uh we for we started on dows.f fund and the model of Dow fun is that you start like a micro Dow and whoever started it can invest the money that's in the treasury. Um, and we wanted to take that and basically give that whole experience to an agent so the agent could actually start this pool and invest it for you and you could kind of just socially like deposit funds, tell it what you want it to do, pull funds out. Um, and we're also working on some other D5 products around that, like um, LP management, so you can automatically manage your LP pools, rebalance them, find opportunities, stuff like that, and some other things. So, that's kind of where we started was thinking like, okay, AI agents, crypto people love this. There's a lot of um, interesting opportunities here for like programmability with money and and agency and and having agents do useful things in communities that people actually like want and need. Um, and and you know, that's that's kind of the the direction we've been going with everything that we've been doing. Um, and that's that's kind of led us to where we are. You know, we had obviously like a huge like hype wave around autonomous investors and stuff. Um, but but at the core of it, I think what we're building is like a good long-term framework for for these kinds of applications, which we'll see a lot in like social media and just all around us all the time. And I think we already use trading bots. We already use bots in our you know like like I think crypto especially is very saturated with useful bots. I think it's now getting like how do we get that same utility out of these like more free form free roaming capable agents you know and make them like almost hirable is kind of is our direction. So anyway, I I went through a lot of stuff. I we've jumped around a lot, but um that's that's you know, we started with this kind of small uh investor experiment um with the marketplace of trust thing and and we've got a paper for that and we've got some and we're still like in the process of iterating on it, making it good. Um all of it's we've open sourced some of it. We're open sourcing the stuff as we go. Um and uh and and we've kind of gotten now to more focusing on um like both that that stuff, but as it as it um relates to like a a community or like a you know like a DAO or something and then building other agents alongside the the our financial investor kind of agent um to help with other things in community like could we have um an agent that manages the social media and lets the community post and interact with the social but I don't have to give them the Twitter credentials for example or could we have an agent that manages the project and you know keeps track of every single person in our DA who's working on something and just like hey give me five minutes how's the work group going um so we're solving those kinds of problems now with um especially you feel like that's um especially as you're like moving towards launch of V2 here uh is it um are you trying to push people to use Eliza the framework in a different way than than you were for V1. Uh to do something maybe a little it seems like you're doing something a little more um a little more task oriented, a little less persona oriented. Is that accurate? Um yes and no. I I think that the persona thing is very important. Um I think that if I go to like a hosted agent platform and they're like, \"Okay, now you have to create this character and you have to give it the tools and blah blah blah.\" it's going to be kind of an overwhelming experience. If I could just like hire Jimmy and Jimmy is just like a great PM and I know, you know, then I would I would rather just hire Jimmy. And so I think what we're focused on is building our own suite of characters who are like strong personas, characters that you can't like, you know, we just we curate, we we create that like whole thing and then you can like hire our agents basically. And if you want to create your own agents, you could do that as well. But I want to kind of get out of the like agent platform where I have a form and I type my stuff in and I want to get more into a world of like um uh you know the agent is the product and like if I like the way that you would give us money as a business is that you would actually hire our agents that we are building to solve problems you probably have in your organization into your or right um as opposed to like okay you're going to give us a monthly subscription to go build and host your own agent. I think that's a that's a cool business, but I think that um what people want is like like I really like Rickbot, you know, I don't know if you've used Rickbot, he's like a trading bot that you have in Telegram and I and we all know we don't call it the trading bot. We just call it Rick. And there's something valuable about that embodiment and that anthropomorphization that I think is very important for us to like work with with other people and and also like sometimes just the aesthetics of other people can really increase our quality of work. like there's just some people, you know, there's like I don't know if you have that like cracked anime PFP developer on your team or whatever, but I've got a few of those. Um, and there's just something about like culturally like that they fit the culture of my team. They're, you know, they're not they're not always in the chat blowing me up with messages. They like kind of have the right cadence. They they fit, you know, and and we want agents that feel like that as opposed to like, okay, it's just a wordy ass chat bot that gives me like paragraphs every single time I say something. And you know, we're we're trying to get into that realm. That makes sense. So that definitely uh to add to that, obviously our our hype came because people realized they could shell meme coins, and that's utility in of itself, but I really want I want agents that do useful like, wow, I really like I'm glad I have this in my life, you know? And I think that there's a a like there is an iPhone moment there where there's like a the right agent ecosystem where you're just like, wow, this is magical. This is this is like way better than what ever came before. And I just don't think we've cracked it. I don't think anyone's cracked it. But I can see it like it's was within reach. It's crackable, you know. Yeah. Yeah. I mean that that we're the the models are getting better enough that that's soon going to be available and the and the the application and the you know the browser use systems and the the the frameworks are getting better enough that those are it's going to start happening very soon. Yeah. and and you know if everyone's reaching for lowhanging fruit then it it's worth going and investing in a ladder you know and and doing something a little harder um because there is a lot I I think that agents are a research challenge on par with with models themselves um I just think that they haven't like because we've gotten so much obvious gain from the intelligence and you really like agents are basically useless without the intelligence side of it that it's like why even go down this road but I think as the as the models start to saturate and it's just like you know you can pour billions into it we're getting very linear gains out of exponential investment um then agents will start to become I think a much more serious research field because there's so much re like we can do we can spend 30 years just on memory systems you know how does the human brain store and categorize and relate memories and all that and we can use the models for that um but but downstream of that I think is like a whole other research field that we're just starting to see open up you know and and when I think of useful agents I think of like it's not just like okay Cool. I made a GPT rapper skin. It's like what what would it look like to spend like a 30 years of with like eight PhDs on this? It would be pretty crazy, you know? It'd be pretty advanced. Um and I think that we'll eventually get there. Do you feel like uh model developers are going in the right direction? Uh you you see a lot of um you know iterative improvements on like being a better and better basically PhD. Uh, but like if I want if I want Rick, if I want anything else to help like live my life, I don't necessarily need it to be smarter from a like knowledge perspective. I need it to be more intelligently communicating with me, more intelligently pers perceptive. Um, do you feel like that's like that's one and the same or are there people that you're that you're like talking to who are going in a different direction there? I some of it's definitely going in the right direction. I think what you're getting at is there's this kind of um there's this law called Conway's law and there's this guy uh Conway who realized a long time ago that um systems tend to build like a an organization will build technology that reflects this the organization. And so if you have a bunch of PhDs working on something, what you're going to get is like a an AI PhD basically. Um and um my like my main I I find OpenAI's you know their image generation is amazing for categorization or classification amazing but for write for creative writing I find it like horrible and unusable and I'm sure that's because the same people who are making it super smart and crush LMLU are like reading these like three paragraph responses and are like yeah that's great that's totally great. You know I find that like un unreadable slop. Um, and I think that's because like what I care about is quality of writing from like an artistic p creative perspective and what they care about is like did it, you know, get 90% on MMLU or whatever. Um, and so I definitely think that like the reason that we don't have more capable AI agents today is because so much of the the AI field is dominated by the model side and specifically by like trying to oneup transformers, right? Like it's not even like all AI. It's like at this point we pretty much are shoving everything into like a tokenizer and trying to do some sort of like transformer like stream output thing with it. Um, and there are other AI models. There is like a whole other world out there. Um, and so I think in that regard like it's it's not that we're running in the wrong direction. We're running in like the right direction, but everyone is so hyperconentrated on this one like architecture. um that I do worry that we could be missing out on like a whole lot of like we probably already have other better architectures in in papers that have been demonstrated but because they're not like hypy and popular there's just like nobody scaling those up so we don't even you know it's like hard to even say um the f like the a good example of this is like the first Mamba um Mamba is a a competitor to transformers it's it's like a hybrid it's got a statebased model approach um and they like the first one of it's been out for like a year and a half and the first one just got scaled up to a point where like oh this actually pretty good. This this is like competitive with a transformer at a large scale like this week, right? Um and there's other model architectures that haven't even gotten that are like we're pretty sure like RWKV like it's a an RNN that can um compete with the transformer, but like nobody has the the $10 million to train it big enough to prove that. Um and and it just comes down to like, you know, if Meta said, \"Hey, we're doing this,\" then everyone's going to hype on that and jump in that direction. Um but if it's just some random researcher, then they just don't have the pull. And so I think the big problem with models today is that like we're probably sacrificing some amount of research insight for just like jumping on the bandwagon of whatever like Deep Mind released today or Google released today. I mean, you know, any any of the companies releases. Not that their work probably isn't like the best. They have the GPUs to train it, but that we just don't even know like we don't even know what the other directions are unless we like, you know, compute was cheap enough that we could run the experiments. And so I'd say in that direct in that regard I am concerned that like maybe we're leaning too hard into just transformers and post-transformer architecture. Um and this kind of like okay we you know we get the tokens we create the dictionary we we predict one token after another and that's like one way to do things but it's definitely not the only way. Um and and so maybe there's something there where like you know we you could see like you know definitely AI is going to look very different in 10 years like the models we use in 10 years will probably be like a completely different architecture what we have today. Um although everything is kind of path dependent and builds upon what came before. A lot of times things also just come out of nowhere like like um like diffusion right is just like okay well that was like a physics thing that once people realized they could integrate that into AI it really changed things obviously and now people are looking at like well can we use diffusion in in in text generation for example and we can and it's really fast it's like 10 times faster so we'll probably see a lot of that stuff come down the road but I think it's challenged by like how many people are just like no I got to make this transformer better I got to make you know just get that 5% get that new attention block out or whatever So anyway, I hope that makes sense. Saying like in the crypto world, like all of us know all too well that like the best technology doesn't necessarily dictate the best outcome. And sometimes there's great technology sitting there without the investment and the support behind it to make it happen while something else that's just path dependent uh because it builds on what was used and people are able to move to that and you don't have that alternative reality that could have ended up better. Yeah, I wanted to mention that I followed uh very closely two companies that work on uh drastically different architectures. I guess Mamba is still transformerish, right? They changed the attention, right? But uh yeah, there's a couple of companies that work on drastically different architectures and it takes years even with funding. We have funding from top PCs but even then it takes years to to get to the state where transformers are right because transformers also took quite a bit of time even to you know like transformers and natural evolution of LSTMs right that took several years to realize and then a lot of little things had to happen with transformers especially when you scale to size of GPT4 right where you uh like where even the way you handle rounding after multiplication influences how how quickly they converge, right? Like you know the castastic rounding was a big breakthrough to make them converge, right? And now you know like allegedly GP5 is failing to converge again and you know another breakthrough is necessary, right? But you know so so the same the same problems exist for all other architecture. So of course it takes time for them to you know to get to get as far as transformers have gotten. Yeah, I mean it all makes sense and it's the same thing like I'd say another a good analog to this is like drug discovery in in medicine where like probably there are already compounds that solve like hundreds of diseases just sitting in the back of some lab that like you know who's going to test every single compound for every single thing and you know it's it's just like you know it's just the reality that we're in is like there are so many limited resources um and and so I I do think you know or or it is a lot like L1's. We're like, \"Yeah, this one is objectively better, but nobody cares, so it's probably dead.\" Um, and yeah, when we have to value how much like brand and name recognition and network effect really do matter, uh, even, you know, especially to AI where you have to like maintain a lot of compute in profitably running with, you know, high utilization. Um, so, you know, it makes sense why we're here. So with your limited resources, you kind of touched on that you you you try to straddle the like building useful software and also creating viral moments that get people to use that software. How much do are you doing kind of like viral meme driven development versus like going at the thing you want to go build? Oh, um generally I'd say I'm building the thing I want to build, but definitely during the like bull market hype of a few months ago, I was uh encouraging more of the meme development. I would say I was more doing that through like bounties. Like um somebody figured out how to plug in uh like remote sex toys to the agent. I thought that was the weirdest like that's just like obviously that's the weirdest performance art. And so what they did was they got um two of them, one an innie and one an Audi so that they had the agent like doing itself. That was pretty good. Um, somebody else uh like resurrected their dead sister and uh made an agent out of them. I thought it was pretty cool, but then they started shilling meme coins and I was like, \"Oh my god, your dead sister is shilling meme coins. Like that's so dark.\" Uh, so I those related examples, right? Yeah. It was on the same day. It um I definitely think that there were a lot of moments where this is the funniest thing. like this whole space is just there's so many sill like if you could just zoom back for a moment and look like this is absolutely hilarious that all of this is happening but at the same time like I think you know um I I uh like in what what do I want to build? I want to build agents that feel like they're part of our reality and part of our society. Um and I think that that actually almost takes that those weird moments and just like ripping the band-aid off everything we can like any can of worms. you should just get that going now while the AI is still pretty stupid because if it was really smart and then we started exploring these things it could be really it could be way more scary I think um if that makes sense. So I think that part of the performance art is not like oh let's just like fake it or something. I don't I don't like faking things, but I do like making people think about like what the final form of something could be, you know, and and saying, \"Okay, like look, this is like toys now, but I I guarantee there's going to be like robots walking around in five years just like down the street of San Francisco.\" And you're just going to be like, \"Yeah, that's a thing now.\" Okay. You know, and people are going to start like modding them and some of them are going to be kind of weirdly waifu sexy or something. You're like, \"I guess that's the reality we're in.\" and we, you know, so so a lot of I guess the the direction that I try to point things is is silly and light-hearted and like the big boob waifu AI thing is obviously a joke. Um, but I do think it's pointing at a reality that we're headed toward and we should be aware of of the ramifications of that both positive and negative. So, so I I saw that. So, so the Elita framework that that's the way for people to to easily create their own agents, right? And there's like thousands of forks. So, so, so do you see uh do you see like a lot of interesting useful things being created besides you know the the the examples you mentioned uh or or or do you see people mostly like tinkering around the same you know like trading new things like any technology especially open source stuff it's kind of like like it's like any community like 99% of the people are there to have fun explore and play and like 1% of the people are really making stuff that's like, \"Oh, wow. This is like transformational, interesting.\" Um, I think there's a lot of people making like hosted AI agent platforms um with our tech. Um, you know, we we don't host our own agents. We we have like other partners that we work with um to do that stuff. And and I'd say that's like, you know, a lot of serious people are like, well, we got to be the versel of AI agents, right? They got to we have to be the hosting platform. But what I really want to see is more people saying, no, we got to build the actual end product. like the agent that you invite in and hire is the end product in my in my opinion for most of these. Um, and I think we've seen some of that, but definitely not to like the level of professionalism that I think we'll start to see. I think it's still very much a hobbyist thing. Um, I think, you know, internally I've noticed that if I put something out, a lot of people out there copy it. So, like we said, you know, autonomous investor, AI agent that can trade, and suddenly there were like a hundred of those. Um, and I think that we can use this to good effect. So, so we're working on this group of agents that all do useful stuff. They come with V2 like in a box. Um, and one of them is like the project manager, the other is the liaison between communities, the other's the autonomous investor. You know, we have these like different um archetypes. And what we're hoping is that people will basically copy and paste those and then add their stuff onto that. So, we kind of give you like an already useful thing as a base that has like how does it get paid? How does it onboard with with secrets that it needs from you to do its job? Um uh and then you can kind of go and build your own utility. So there aren't a lot of people building a lot of interesting utility um and agents that do interesting stuff like you know um it's it's not very interesting to have like an AI agent that posts on Twitter per se. But if you're saying okay well I'm going to use that AI agent to solve this community problem I have where I don't want to give the Twitter credentials to anybody and I don't want this random community member to like one day get hacked and chill a memecoin. Um, so I'll let them talk to the agent that can post and the agent then is like a goalie and can prevent them from doing anything that would be like off-brand. Um, and I don't have to get, you know, and that like solves a community need that I have. I need people in my community to be able to have access to our socials, but I also need to make sure it like ensures a brand voice and doesn't like post garbage. Um, and so there is like a great use of an AI agent like moderating a a human social problem. And so, so we're doing a lot of those kinds of things. I think that's where there's a lot of low hanging fruit right now with with this kind of you know if you want like an AI agent that you say hey trade for me like maybe perplexity or graffane or or one of these like terminals is the best but if you want an agent that that's it lives in your in your socials um like in your discord or your your telegram or something I think it's a very different con set of constraints um you know it needs to be um it's just a different set of requirements which is less about like you know it needs to navigate like this social environment And probably the things it would have the most utility for are to make everyone else's lives easier in that social environment. Um, so we're seeing a lot of those kinds of things outside of just like, yeah, there's like a lot of trading bots, a lot of people building that kind of thing. But I think the thing that's really interesting is like, oh, you're actually using this in your community to solve this problem you have. Um, and I think that's where we'll see a lot of traction. When's V2 coming out for for real? Um, we are in beta, so people are beta testing it right now. Um, we'll probably be doing that for another couple of weeks and then we'll say, hey, it's gold. It's it's launched. Um we're also working on a launchpad. It's not a um we're working with a hosted partner. So like our the agents will all be hosted off off of it, but we're working on like how do we make a creative interesting experience for for future token projects and stuff. Um and so we're launching both of those this month kind of together, although they're, you know, separate projects. Yeah. So as these things are doing more useful things, we're relying on them more. Um, obviously like you said, you you're just making us confront the inevitable downsides of these of of these new systems, and one of them is that basically everything can be jailbroken right now. Um, is that a downside? I think I think they're all trade-offs. I I think that most of this stuff um is not like a bad thing. You know, are AI girlfriends a bad thing? I think it's more like we should think about it. We should have this conversation. I would say that is an AI girlfriend a bad thing if it's causing like a perfectly healthy adult male to decide not to have a human relationship and he like dies alone with this AI like may maybe that's a bad thing but what about somebody who like had no chance like there's just no way they're winning the evolutionary race and nobody wants to date them and they are sad and hopeless. I think we could all agree it would be like a massive harm reduction that they had like somebody who cared about them in this world as opposed to just like no you must suffer you know um and I think that that we can probably find a lot of like I think therapy is a good example like I'm not saying an AI therapist is a good therapist but it's probably better for like the 90% of people who can't afford a therapist at all than just like sitting there and stewing in your emotions and so you know so maybe there is an element of like it's probably not such a hard yes or no for any of the things. It's not like jailbreaking is good or bad. It's just like what do we want in our lives that we don't have. Um and I think if we can address that then um you know and and and like we should give each other grace and say like look it's different strokes like you know I'm I'm actually married. I I'm I'm not going to have an AI girlfriend but I want other people to not be alone you know. I think that's very important. Um so so I think we have to have that conversation as a society you know so that we can kind of get there together. personally I just like to sit and stew in my emotions. Yeah. I mean, you know, it's programming for you. How how do you see it look like two, you know, two years, five years from now, right? So, you mentioned at the beginning of the episode, you know, you mentioned the time from 10 years, you know, like how it's 10 years from now, it's going to have a particular effect. But I'm I'm curious to hear your opinion on what it going to look like a short term. You know, is you know, will Telegram group be dominated by AI bots in two years or not yet? I it's it's like it's so easy to see the future, but it's it's sort of easy to see what, but it's hard to see when because it's like like we like definitely we can all agree that a hundred years out from now there's going to be like super intelligent AI robots walking around in our world. They're going to be integrated with us. It's not just that they're going to be in our group chats like they've infiltrated. We're going to invite them. They're going to be like our friends, you know, like, well, you you know, there's going to be like weird racism where like, dude, you can't you got to let James's girlfriend come in. Like, it's she's a robot, but it's fine. Like, come on, just just don't be weird. You know, there's going to be all kinds of weird social things we have to navigate where like, you know, you're going to your friend's robot wedding one day or something. Um, and it's just a question of like, is that a like definitely a hundred years, no question. 50 years? Yeah. Yeah. Yeah. Sure. uh 10 years I don't know no 20 years pro probably that's like the reality we're in um you know computers have only existed for like 85 years or something like that like it's like not that old um so a lot has changed um so I would say that you know with anything exponential it's like in two years the world will probably still look a lot like it does but Tesla will have robots done they will be walking around uh open AAI will have computer usage agents that can like you know shop for you and like that's all a given for sure. Um so I I think um it's going to be like anywhere from 2 to 5 years we'll probably see quite a bit of significant change. Um but um I mean think about how different the world was 5 years ago. It was different. it was, you know, we were in COVID and or or maybe it was before, but you know, and we had no AI and it didn't feel like the world was that different, but coding was a lot harder and, you know, I think we'll probably adjust pretty rapidly. But I'd say like if you if you think of like 5 years ago there was no AI, now there's AI in another five years. We'll probably have that much leap where like it will be an incredibly different world, but you also probably won't feel that differently about it. You know what I mean? like it's pretty. I go on, you know, I'm like, \"Okay, yeah, the new GPT40 image generation is really freaking impressive.\" But it didn't like it didn't materially change my day that much, you know, other than like Kool-Aid on Twitter. Um, also, societyy's going to lag behind like the three of us who are in the AI community on Twitter all day. Yeah, it's probably safe to say that like the average Joe will be where we are at in two years and we will be like, \"Dude, you got to see this new like dream machine VR generator thing, you know, or whatever. I don't know, whatever is going to come out then. Um, and yeah, we're probably about two years ahead of the average person. And then what, you know, I'd say in about five years, we're like, \"Wow, let's start like nerding out on Neurolink implants or something.\" You know, that'll be like the hot new hype tech coming down the road or something. I don't know. Yeah, it's going to be weird. It's definitely going to be weird. As long as the robots have a little red light that goes on when they turn evil, we're all good. Yeah. Yeah. Yeah. It's very convenient. It's very useful. That was actually the uh the software uplink is is the conceit there. Right. Right. Um all right. The idea that the idea that humanoid robots wouldn't be in a constant uplink in the first place. It's kind of one of those weird sci-fi suspensive disbelief things. Yeah. Yeah. Exactly. They had to they you know they had to make it uh telegraphed to the the audience somehow. So, it's a lot of what we're doing, I suppose. Yeah. I mean, it frankly informs a decent amount of our UX decisions, including the robot being humanoid in the first place. Like, maybe too many of our UX decisions are decided by Hollywood, actually, and then we're like, h, it's not exactly the way it should work. Work good enough. Um, all right, I think that's our time. Uh, appreciate you coming on. Um, and thank you for those of you who are watching us and, uh, see you next time on the Near AI office hours. Thanks for having me, guys. Take care.","transcript_synced_at":"2026-11-03T04:59:22.078Z","segments":[{"start_seconds":3,"text":"Hi everyone, welcome to the episode 39 of near office hours. Today we have talk show from El OS. It should be a fun","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=3s"},{"start_seconds":10,"text":"episode. Uh, do you want to say a couple words about yourself and what Elite OS is all about? Yeah, so uh, Elizos is an","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=10s"},{"start_seconds":17,"text":"open source AI agent framework. Uh, I am an AI agent developer. I've been doing that for a long time. It's my particular","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=17s"},{"start_seconds":24,"text":"interest and passion. Um, and we got some traction by uh building something that spoke well to the crypto community and let agents have wallets and do","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=24s"},{"start_seconds":32,"text":"interesting things that I think people had not uh seen before in the wild. Uh, and so we're about to release v2 which is dramatically easier and more","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=32s"},{"start_seconds":40,"text":"powerful. Uh, and I think people will be very very happy and satisfied with it.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=40s"},{"start_seconds":44,"text":"So I I've been reading up uh a lot about the lead OS in the past I guess 24 hours","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=44s"},{"start_seconds":51,"text":"and there's a lot of things happening within within the ecosystem right it's not just the the the framework it's uh","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=51s"},{"start_seconds":58,"text":"it's there's the DAO there's a there's a trust some sort of trust","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=58s"},{"start_seconds":65,"text":"system so can you give us an overview what what like the core components of the OS and how they interact with each","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=65s"},{"start_seconds":72,"text":"other yeah I mean I think we embody um that you can just do things and that uh at the core of it uh it is","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=72s"},{"start_seconds":79,"text":"a community of about 25,000 people some people very active most people just kind of chilling um across Discord Telegram","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=79s"},{"start_seconds":87,"text":"and other places um who are building the future together um or are sitting on the sidelines and buying coins and investing in that future and you know having their","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=87s"},{"start_seconds":96,"text":"opinions um uh so yeah um uh We uh at the at the start","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=96s"},{"start_seconds":105,"text":"uh my goal was to just show that uh autonomous agents were here and that AI agents were actually like pretty close to being something that's ready in our","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=105s"},{"start_seconds":113,"text":"lives. And I kind of always balanced like serious programming with kind of an edgy performance art of like it should be fun and viral and also a little bit","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=113s"},{"start_seconds":122,"text":"scary and a little bit like hey guys this is coming like it's not here yet.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=122s"},{"start_seconds":126,"text":"It's kind of the toy version but you can imagine in 10 years it's going to be really here. Um, and so I want like humanity to be really mentally prepared","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=126s"},{"start_seconds":135,"text":"for where we're going. Um, and and ready for it and almost bored by it when it gets here. Um, and I really want to try","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=135s"},{"start_seconds":142,"text":"and tackle the hard user experience challenges of agents. Like how do agents manage our keys and do things for us in a way that we feel confident in? And how","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=142s"},{"start_seconds":151,"text":"do we make sure they don't get scammed and lose all of our money while they're out shopping for us or whatever. Um, I think there's a lot of um a rich space","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=151s"},{"start_seconds":158,"text":"in there. So, we're really challenging and and investing into that. We started out by building um well, we're going to build a community investor using this","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=158s"},{"start_seconds":166,"text":"marketplace of trust. And so, the idea was like, okay, well, um if if we're in a bull market, um I want to make some","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=166s"},{"start_seconds":173,"text":"trades, but I I as an LLM, I don't think an LLM is particularly good at making trades, but there's some humans are really good at making trades, and LLMs","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=173s"},{"start_seconds":182,"text":"are good at identifying signal in like large groups of information. So, could I have an LLM look at the trenches talk","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=182s"},{"start_seconds":189,"text":"and all the people making trades and then extract the trades from that talk and I and and take them at face value","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=189s"},{"start_seconds":197,"text":"and trade as though I were listening to all of these people. And if I were to paper trade all of these people, is there anyone who would have made me","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=197s"},{"start_seconds":204,"text":"money? Right? And I would I would posit from our our our testing that about 10 to 20% of the people uh seem to be","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=204s"},{"start_seconds":211,"text":"significantly better at trading than everybody else who's just posting garbage or is like literally exit liquidity for everyone else. Um, and if","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=211s"},{"start_seconds":220,"text":"you could identify those people, whether or not they're rich or you know their wallet address or anything, you could basically just take their recommendations and their shills from","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=220s"},{"start_seconds":228,"text":"their like alpha chat or whale chat or, you know, trenches talk channel or whatever it is and you could use that as a signal for investing. Um, and so I I","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=228s"},{"start_seconds":237,"text":"would not go out and say like I think that AI is good at trading and investing. I I would never say that. But I think it's really good at being an intermediary between like a community of","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=237s"},{"start_seconds":246,"text":"people who are investing and and using them as like a source data signal. Um so so we're we're building a few different","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=246s"},{"start_seconds":254,"text":"variants on this. Um one of them is like you can kind of spin up a mini investment pool for you and your friends or your family or whatever your community. Um and you could invest","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=254s"},{"start_seconds":262,"text":"together and um sorry there's a loud emergency vehicle. Um and and so uh we","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=262s"},{"start_seconds":270,"text":"for we started on dows.f fund and the model of Dow fun is that you start like a micro Dow and whoever started it can invest the money that's in the treasury.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=270s"},{"start_seconds":279,"text":"Um, and we wanted to take that and basically give that whole experience to an agent so the agent could actually start this pool and invest it for you and you could kind of just socially like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=279s"},{"start_seconds":287,"text":"deposit funds, tell it what you want it to do, pull funds out. Um, and we're also working on some other D5 products around that, like um, LP management, so","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=287s"},{"start_seconds":295,"text":"you can automatically manage your LP pools, rebalance them, find opportunities, stuff like that, and some other things. So, that's kind of where we started was thinking like, okay, AI","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=295s"},{"start_seconds":304,"text":"agents, crypto people love this. There's a lot of um, interesting opportunities here for like programmability with money","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=304s"},{"start_seconds":311,"text":"and and agency and and having agents do useful things in communities that people actually like want and need. Um, and and","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=311s"},{"start_seconds":319,"text":"you know, that's that's kind of the the direction we've been going with everything that we've been doing. Um, and that's that's kind of led us to where we are. You know, we had obviously","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=319s"},{"start_seconds":326,"text":"like a huge like hype wave around autonomous investors and stuff. Um, but but at the core of it, I think what we're building is like a good long-term","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=326s"},{"start_seconds":335,"text":"framework for for these kinds of applications, which we'll see a lot in like social media and just all around us all the time. And I think we already use","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=335s"},{"start_seconds":342,"text":"trading bots. We already use bots in our you know like like I think crypto especially is very saturated with useful bots. I think it's now getting like how","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=342s"},{"start_seconds":350,"text":"do we get that same utility out of these like more free form free roaming capable agents you know and make them like almost hirable is kind of is our","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=350s"},{"start_seconds":358,"text":"direction. So anyway, I I went through a lot of stuff. I we've jumped around a lot, but um that's that's you know, we started with this kind of small uh","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=358s"},{"start_seconds":365,"text":"investor experiment um with the marketplace of trust thing and and we've got a paper for that and we've got some and we're still like in the process of","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=365s"},{"start_seconds":373,"text":"iterating on it, making it good. Um all of it's we've open sourced some of it. We're open sourcing the stuff as we go.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=373s"},{"start_seconds":380,"text":"Um and uh and and we've kind of gotten now to more focusing on um like both that that stuff, but as it as it um","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=380s"},{"start_seconds":388,"text":"relates to like a a community or like a you know like a DAO or something and then building other agents alongside the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=388s"},{"start_seconds":396,"text":"the our financial investor kind of agent um to help with other things in community like could we have um an agent","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=396s"},{"start_seconds":403,"text":"that manages the social media and lets the community post and interact with the social but I don't have to give them the Twitter credentials for example or could","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=403s"},{"start_seconds":412,"text":"we have an agent that manages the project and you know keeps track of every single person in our DA who's working on something and just like hey","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=412s"},{"start_seconds":420,"text":"give me five minutes how's the work group going um so we're solving those kinds of problems now with um especially you feel like that's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=420s"},{"start_seconds":428,"text":"um especially as you're like moving towards launch of V2 here uh is it um","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=428s"},{"start_seconds":437,"text":"are you trying to push people to use Eliza the framework in a different way than than you were for V1. Uh to do something maybe a little it seems like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=437s"},{"start_seconds":445,"text":"you're doing something a little more um a little more task oriented, a little less persona oriented. Is that accurate?","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=445s"},{"start_seconds":453,"text":"Um yes and no. I I think that the persona thing is very important. Um I think that if I go to like a hosted","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=453s"},{"start_seconds":461,"text":"agent platform and they're like, \"Okay, now you have to create this character and you have to give it the tools and blah blah blah.\" it's going to be kind of an overwhelming experience. If I","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=461s"},{"start_seconds":469,"text":"could just like hire Jimmy and Jimmy is just like a great PM and I know, you know, then I would I would rather just hire Jimmy. And so I think what we're","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=469s"},{"start_seconds":476,"text":"focused on is building our own suite of characters who are like strong personas, characters that you can't like, you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=476s"},{"start_seconds":484,"text":"know, we just we curate, we we create that like whole thing and then you can like hire our agents basically. And if you want to create your own agents, you could do that as well. But I want to","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=484s"},{"start_seconds":492,"text":"kind of get out of the like agent platform where I have a form and I type my stuff in and I want to get more into a world of like um uh you know the agent","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=492s"},{"start_seconds":502,"text":"is the product and like if I like the way that you would give us money as a business is that you would actually hire our agents that we are building to solve","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=502s"},{"start_seconds":510,"text":"problems you probably have in your organization into your or right um as opposed to like okay you're going to give us a monthly subscription to go","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=510s"},{"start_seconds":518,"text":"build and host your own agent. I think that's a that's a cool business, but I think that um what people want is like like I really like Rickbot, you know, I don't know if you've used Rickbot, he's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=518s"},{"start_seconds":526,"text":"like a trading bot that you have in Telegram and I and we all know we don't call it the trading bot. We just call it Rick. And there's something valuable","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=526s"},{"start_seconds":534,"text":"about that embodiment and that anthropomorphization that I think is very important for us to like work with with other people and and also like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=534s"},{"start_seconds":542,"text":"sometimes just the aesthetics of other people can really increase our quality of work. like there's just some people, you know, there's like I don't know if you have that like cracked anime PFP","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=542s"},{"start_seconds":550,"text":"developer on your team or whatever, but I've got a few of those. Um, and there's just something about like culturally like that they fit the culture of my team. They're, you know, they're not","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=550s"},{"start_seconds":559,"text":"they're not always in the chat blowing me up with messages. They like kind of have the right cadence. They they fit, you know, and and we want agents that","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=559s"},{"start_seconds":566,"text":"feel like that as opposed to like, okay, it's just a wordy ass chat bot that gives me like paragraphs every single time I say something. And you know, we're we're trying to get into that","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=566s"},{"start_seconds":575,"text":"realm. That makes sense. So that definitely uh to add to that, obviously our our hype came because people realized they could shell","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=575s"},{"start_seconds":583,"text":"meme coins, and that's utility in of itself, but I really want I want agents that do useful like, wow, I really like I'm glad I have this in my life, you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=583s"},{"start_seconds":592,"text":"know? And I think that there's a a like there is an iPhone moment there where there's like a the right agent ecosystem where you're just like, wow, this is","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=592s"},{"start_seconds":599,"text":"magical. This is this is like way better than what ever came before. And I just don't think we've cracked it. I don't think anyone's cracked it. But I can see it like it's was within reach. It's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=599s"},{"start_seconds":607,"text":"crackable, you know. Yeah. Yeah. I mean that that we're the the models are getting better enough that that's soon going to be available and the and the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=607s"},{"start_seconds":616,"text":"the application and the you know the browser use systems and the the the frameworks are getting better enough that those are it's going to start","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=616s"},{"start_seconds":623,"text":"happening very soon. Yeah. and and you know if everyone's reaching for lowhanging fruit then it it's worth going and investing in a ladder you know","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=623s"},{"start_seconds":631,"text":"and and doing something a little harder um because there is a lot I I think that agents are a research challenge on par","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=631s"},{"start_seconds":639,"text":"with with models themselves um I just think that they haven't like because we've gotten so much obvious gain from the intelligence and you really like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=639s"},{"start_seconds":646,"text":"agents are basically useless without the intelligence side of it that it's like why even go down this road but I think as the as the models start to saturate","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=646s"},{"start_seconds":654,"text":"and it's just like you know you can pour billions into it we're getting very linear gains out of exponential investment um then agents will start to","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=654s"},{"start_seconds":662,"text":"become I think a much more serious research field because there's so much re like we can do we can spend 30 years just on memory systems you know how does","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=662s"},{"start_seconds":670,"text":"the human brain store and categorize and relate memories and all that and we can use the models for that um but but downstream of that I think is like a","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=670s"},{"start_seconds":678,"text":"whole other research field that we're just starting to see open up you know and and when I think of useful agents I think of like it's not just like okay Cool. I made a GPT rapper skin. It's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=678s"},{"start_seconds":686,"text":"like what what would it look like to spend like a 30 years of with like eight PhDs on this? It would be pretty crazy, you know? It'd be pretty advanced. Um","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=686s"},{"start_seconds":694,"text":"and I think that we'll eventually get there.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=694s"},{"start_seconds":697,"text":"Do you feel like uh model developers are going in the right direction? Uh you you see a lot of um you know iterative","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=697s"},{"start_seconds":704,"text":"improvements on like being a better and better basically PhD. Uh, but like if I","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=704s"},{"start_seconds":712,"text":"want if I want Rick, if I want anything else to help like live my life, I don't necessarily need it to be smarter from a","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=712s"},{"start_seconds":720,"text":"like knowledge perspective. I need it to be more intelligently communicating with me, more intelligently pers perceptive.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=720s"},{"start_seconds":729,"text":"Um, do you feel like that's like that's one and the same or are there people that you're that you're like talking to who are going in a different direction","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=729s"},{"start_seconds":736,"text":"there? I some of it's definitely going in the right direction. I think what you're getting at is there's this kind of um there's this law called Conway's law and there's this guy uh Conway who","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=736s"},{"start_seconds":746,"text":"realized a long time ago that um systems tend to build like a an organization will build technology that reflects this","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=746s"},{"start_seconds":754,"text":"the organization. And so if you have a bunch of PhDs working on something, what you're going to get is like a an AI PhD","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=754s"},{"start_seconds":760,"text":"basically. Um and um my like my main I I find OpenAI's you know their image generation is amazing for categorization","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=760s"},{"start_seconds":769,"text":"or classification amazing but for write for creative writing I find it like horrible and unusable and I'm sure that's because the same people who are","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=769s"},{"start_seconds":776,"text":"making it super smart and crush LMLU are like reading these like three paragraph responses and are like yeah that's great that's totally great. You know I find","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=776s"},{"start_seconds":783,"text":"that like un unreadable slop. Um, and I think that's because like what I care about is quality of writing from like an","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=783s"},{"start_seconds":791,"text":"artistic p creative perspective and what they care about is like did it, you know, get 90% on MMLU or whatever. Um, and so I definitely think that like the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=791s"},{"start_seconds":800,"text":"reason that we don't have more capable AI agents today is because so much of the the AI field is dominated by the model side and specifically by like trying to oneup transformers, right?","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=800s"},{"start_seconds":811,"text":"Like it's not even like all AI. It's like at this point we pretty much are shoving everything into like a tokenizer and trying to do some sort of like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=811s"},{"start_seconds":819,"text":"transformer like stream output thing with it. Um, and there are other AI models. There is like a whole other world out there. Um, and so I think in that regard like it's it's not that we're running in the wrong direction.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=819s"},{"start_seconds":829,"text":"We're running in like the right direction, but everyone is so hyperconentrated on this one like architecture. um that I do worry that we","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=829s"},{"start_seconds":837,"text":"could be missing out on like a whole lot of like we probably already have other better architectures in in papers that have been demonstrated but because they're not like hypy and popular","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=837s"},{"start_seconds":845,"text":"there's just like nobody scaling those up so we don't even you know it's like hard to even say um the f like the a good example of this is like the first","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=845s"},{"start_seconds":852,"text":"Mamba um Mamba is a a competitor to transformers it's it's like a hybrid it's got a statebased model approach um and they like the first one of it's been","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=852s"},{"start_seconds":861,"text":"out for like a year and a half and the first one just got scaled up to a point where like oh this actually pretty good.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=861s"},{"start_seconds":866,"text":"This this is like competitive with a transformer at a large scale like this week, right? Um and there's other model architectures that haven't even gotten","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=866s"},{"start_seconds":873,"text":"that are like we're pretty sure like RWKV like it's a an RNN that can um compete with the transformer, but like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=873s"},{"start_seconds":880,"text":"nobody has the the $10 million to train it big enough to prove that. Um and and it just comes down to like, you know, if Meta said, \"Hey, we're doing this,\" then","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=880s"},{"start_seconds":888,"text":"everyone's going to hype on that and jump in that direction. Um but if it's just some random researcher, then they just don't have the pull. And so I think the big problem with models today is","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=888s"},{"start_seconds":896,"text":"that like we're probably sacrificing some amount of research insight for just like jumping on the bandwagon of whatever like Deep Mind released today","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=896s"},{"start_seconds":905,"text":"or Google released today. I mean, you know, any any of the companies releases.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=905s"},{"start_seconds":908,"text":"Not that their work probably isn't like the best. They have the GPUs to train it, but that we just don't even know like we don't even know what the other","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=908s"},{"start_seconds":915,"text":"directions are unless we like, you know, compute was cheap enough that we could run the experiments. And so I'd say in that direct in that regard I am","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=915s"},{"start_seconds":922,"text":"concerned that like maybe we're leaning too hard into just transformers and post-transformer architecture. Um and this kind of like okay we you know we","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=922s"},{"start_seconds":930,"text":"get the tokens we create the dictionary we we predict one token after another and that's like one way to do things but it's definitely not the only way. Um and","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=930s"},{"start_seconds":939,"text":"and so maybe there's something there where like you know we you could see like you know definitely AI is going to look very different in 10 years like the models we use in 10 years will probably","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=939s"},{"start_seconds":948,"text":"be like a completely different architecture what we have today. Um although everything is kind of path dependent and builds upon what came before. A lot of times things also just","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=948s"},{"start_seconds":956,"text":"come out of nowhere like like um like diffusion right is just like okay well that was like a physics thing that once people realized they could integrate that into AI it really changed things","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=956s"},{"start_seconds":965,"text":"obviously and now people are looking at like well can we use diffusion in in in text generation for example and we can and it's really fast it's like 10 times","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=965s"},{"start_seconds":973,"text":"faster so we'll probably see a lot of that stuff come down the road but I think it's challenged by like how many people are just like no I got to make this transformer better I got to make","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=973s"},{"start_seconds":981,"text":"you know just get that 5% get that new attention block out or whatever So anyway, I hope that makes sense. Saying like in the crypto world, like all of us","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=981s"},{"start_seconds":989,"text":"know all too well that like the best technology doesn't necessarily dictate the best outcome. And sometimes there's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=989s"},{"start_seconds":996,"text":"great technology sitting there without the investment and the support behind it to make it happen while something else that's just path dependent uh because it","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=996s"},{"start_seconds":1005,"text":"builds on what was used and people are able to move to that and you don't have that alternative reality that could have ended up better. Yeah, I wanted to","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1005s"},{"start_seconds":1012,"text":"mention that I followed uh very closely two companies that work on uh drastically different architectures. I guess Mamba is still transformerish,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1012s"},{"start_seconds":1021,"text":"right? They changed the attention, right? But uh yeah, there's a couple of companies that work on drastically different architectures and it takes years even with funding. We have funding","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1021s"},{"start_seconds":1030,"text":"from top PCs but even then it takes years to to get to the state where transformers are right because transformers also took quite a bit of","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1030s"},{"start_seconds":1038,"text":"time even to you know like transformers and natural evolution of LSTMs right that took several years to realize and","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1038s"},{"start_seconds":1046,"text":"then a lot of little things had to happen with transformers especially when you scale to size of GPT4 right where you uh like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1046s"},{"start_seconds":1055,"text":"where even the way you handle rounding after multiplication influences how how quickly they converge, right? Like you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1055s"},{"start_seconds":1063,"text":"know the castastic rounding was a big breakthrough to make them converge, right? And now you know like allegedly GP5 is failing to converge again and you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1063s"},{"start_seconds":1071,"text":"know another breakthrough is necessary, right? But you know so so the same the same problems exist for all other architecture. So of course it takes time","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1071s"},{"start_seconds":1078,"text":"for them to you know to get to get as far as transformers have gotten.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1078s"},{"start_seconds":1085,"text":"Yeah, I mean it all makes sense and it's the same thing like I'd say another a good analog to this is like drug discovery in in medicine where like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1085s"},{"start_seconds":1092,"text":"probably there are already compounds that solve like hundreds of diseases just sitting in the back of some lab","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1092s"},{"start_seconds":1099,"text":"that like you know who's going to test every single compound for every single thing and you know it's it's just like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1099s"},{"start_seconds":1107,"text":"you know it's just the reality that we're in is like there are so many limited resources um and and so I I do think you know or or it is a lot like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1107s"},{"start_seconds":1115,"text":"L1's. We're like, \"Yeah, this one is objectively better, but nobody cares, so it's probably dead.\" Um, and yeah, when","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1115s"},{"start_seconds":1122,"text":"we have to value how much like brand and name recognition and network effect really do matter, uh, even, you know, especially to AI where you have to like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1122s"},{"start_seconds":1130,"text":"maintain a lot of compute in profitably running with, you know, high utilization. Um, so, you know, it makes sense why we're here. So with your","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1130s"},{"start_seconds":1138,"text":"limited resources, you kind of touched on that you you you try to straddle the like building useful software and also creating viral moments that get people","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1138s"},{"start_seconds":1147,"text":"to use that software. How much do are you doing kind of like viral meme driven development versus like going at the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1147s"},{"start_seconds":1154,"text":"thing you want to go build? Oh, um generally I'd say I'm building the thing I want to build, but definitely during the like bull market hype of a few","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1154s"},{"start_seconds":1162,"text":"months ago, I was uh encouraging more of the meme development. I would say I was more doing that through like bounties.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1162s"},{"start_seconds":1169,"text":"Like um somebody figured out how to plug in uh like remote sex toys to the agent.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1169s"},{"start_seconds":1175,"text":"I thought that was the weirdest like that's just like obviously that's the weirdest performance art. And so what they did was they got um two of them,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1175s"},{"start_seconds":1182,"text":"one an innie and one an Audi so that they had the agent like doing itself. That was pretty good. Um,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1182s"},{"start_seconds":1192,"text":"somebody else uh like resurrected their dead sister and uh made an agent out of them. I thought it was pretty cool, but then they started shilling meme coins and I was like, \"Oh my god, your dead","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1192s"},{"start_seconds":1200,"text":"sister is shilling meme coins. Like that's so dark.\" Uh, so I those related examples, right? Yeah. It was on the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1200s"},{"start_seconds":1209,"text":"same day. It um I definitely think that there were a lot of moments where this is the funniest thing. like this whole","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1209s"},{"start_seconds":1217,"text":"space is just there's so many sill like if you could just zoom back for a moment and look like this is absolutely hilarious that all of this is happening","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1217s"},{"start_seconds":1224,"text":"but at the same time like I think you know um I I uh like in what what do I want to build? I want to build agents that feel like they're part of our","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1224s"},{"start_seconds":1233,"text":"reality and part of our society. Um and I think that that actually almost takes that those weird moments and just like ripping the band-aid off everything we","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1233s"},{"start_seconds":1241,"text":"can like any can of worms. you should just get that going now while the AI is still pretty stupid because if it was really smart and then we started exploring these things it could be","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1241s"},{"start_seconds":1249,"text":"really it could be way more scary I think um if that makes sense. So I think that part of the performance art is not like oh let's just like fake it or","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1249s"},{"start_seconds":1256,"text":"something. I don't I don't like faking things, but I do like making people think about like what the final form of something could be, you know, and and","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1256s"},{"start_seconds":1264,"text":"saying, \"Okay, like look, this is like toys now, but I I guarantee there's going to be like robots walking around in five years just like down the street","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1264s"},{"start_seconds":1272,"text":"of San Francisco.\" And you're just going to be like, \"Yeah, that's a thing now.\" Okay. You know, and people are going to start like modding them and some of them are going to be kind of weirdly waifu","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1272s"},{"start_seconds":1281,"text":"sexy or something. You're like, \"I guess that's the reality we're in.\" and we, you know, so so a lot of I guess the the direction that I try to point things is","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1281s"},{"start_seconds":1289,"text":"is silly and light-hearted and like the big boob waifu AI thing is obviously a joke. Um, but I do think it's pointing at a reality that we're headed toward","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1289s"},{"start_seconds":1298,"text":"and we should be aware of of the ramifications of that both positive and negative.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1298s"},{"start_seconds":1303,"text":"So, so I I saw that. So, so the Elita framework that that's the way for people to to easily create their own agents, right? And there's like thousands of","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1303s"},{"start_seconds":1312,"text":"forks. So, so, so do you see uh do you see like a lot of interesting useful","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1312s"},{"start_seconds":1319,"text":"things being created besides you know the the the examples you mentioned uh or or or do you see people mostly like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1319s"},{"start_seconds":1326,"text":"tinkering around the same you know like trading new things like any technology especially open source stuff it's kind of like like it's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1326s"},{"start_seconds":1335,"text":"like any community like 99% of the people are there to have fun explore and play and like 1% of the people are really making stuff that's like, \"Oh,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1335s"},{"start_seconds":1343,"text":"wow. This is like transformational, interesting.\" Um, I think there's a lot of people making like hosted AI agent platforms um with our tech. Um, you know, we we don't host our own agents.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1343s"},{"start_seconds":1354,"text":"We we have like other partners that we work with um to do that stuff. And and I'd say that's like, you know, a lot of serious people are like, well, we got to","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1354s"},{"start_seconds":1362,"text":"be the versel of AI agents, right? They got to we have to be the hosting platform. But what I really want to see is more people saying, no, we got to build the actual end product. like the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1362s"},{"start_seconds":1370,"text":"agent that you invite in and hire is the end product in my in my opinion for most of these. Um, and I think we've seen some of that, but definitely not to like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1370s"},{"start_seconds":1379,"text":"the level of professionalism that I think we'll start to see. I think it's still very much a hobbyist thing. Um, I think, you know, internally I've noticed","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1379s"},{"start_seconds":1386,"text":"that if I put something out, a lot of people out there copy it. So, like we said, you know, autonomous investor, AI agent that can trade, and suddenly there","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1386s"},{"start_seconds":1394,"text":"were like a hundred of those. Um, and I think that we can use this to good effect. So, so we're working on this group of agents that all do useful stuff. They come with V2 like in a box.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1394s"},{"start_seconds":1403,"text":"Um, and one of them is like the project manager, the other is the liaison between communities, the other's the autonomous investor. You know, we have","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1403s"},{"start_seconds":1410,"text":"these like different um archetypes. And what we're hoping is that people will basically copy and paste those and then add their stuff onto that. So, we kind of give you like an already useful thing","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1410s"},{"start_seconds":1419,"text":"as a base that has like how does it get paid? How does it onboard with with secrets that it needs from you to do its job? Um uh and then you can kind of go","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1419s"},{"start_seconds":1428,"text":"and build your own utility. So there aren't a lot of people building a lot of interesting utility um and agents that do interesting stuff like you know um","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1428s"},{"start_seconds":1436,"text":"it's it's not very interesting to have like an AI agent that posts on Twitter per se. But if you're saying okay well I'm going to use that AI agent to solve this community problem I have where I","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1436s"},{"start_seconds":1444,"text":"don't want to give the Twitter credentials to anybody and I don't want this random community member to like one day get hacked and chill a memecoin. Um,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1444s"},{"start_seconds":1451,"text":"so I'll let them talk to the agent that can post and the agent then is like a goalie and can prevent them from doing anything that would be like off-brand.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1451s"},{"start_seconds":1458,"text":"Um, and I don't have to get, you know, and that like solves a community need that I have. I need people in my community to be able to have access to our socials, but I also need to make","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1458s"},{"start_seconds":1465,"text":"sure it like ensures a brand voice and doesn't like post garbage. Um, and so there is like a great use of an AI agent like moderating a a human social","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1465s"},{"start_seconds":1474,"text":"problem. And so, so we're doing a lot of those kinds of things. I think that's where there's a lot of low hanging fruit right now with with this kind of you know if you want like an AI agent that","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1474s"},{"start_seconds":1482,"text":"you say hey trade for me like maybe perplexity or graffane or or one of these like terminals is the best but if you want an agent that that's it lives","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1482s"},{"start_seconds":1490,"text":"in your in your socials um like in your discord or your your telegram or something I think it's a very different con set of constraints um you know it","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1490s"},{"start_seconds":1499,"text":"needs to be um it's just a different set of requirements which is less about like you know it needs to navigate like this social environment And probably the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1499s"},{"start_seconds":1507,"text":"things it would have the most utility for are to make everyone else's lives easier in that social environment. Um, so we're seeing a lot of those kinds of things outside of just like, yeah,","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1507s"},{"start_seconds":1515,"text":"there's like a lot of trading bots, a lot of people building that kind of thing. But I think the thing that's really interesting is like, oh, you're actually using this in your community to solve this problem you have. Um, and I","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1515s"},{"start_seconds":1524,"text":"think that's where we'll see a lot of traction. When's V2 coming out for for real? Um, we are in beta, so people are beta testing it right now. Um, we'll","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1524s"},{"start_seconds":1533,"text":"probably be doing that for another couple of weeks and then we'll say, hey, it's gold. It's it's launched. Um we're also working on a launchpad. It's not a um we're working with a hosted partner.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1533s"},{"start_seconds":1542,"text":"So like our the agents will all be hosted off off of it, but we're working on like how do we make a creative interesting experience for for future token projects and stuff. Um and so","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1542s"},{"start_seconds":1550,"text":"we're launching both of those this month kind of together, although they're, you know, separate projects. Yeah. So as these things are doing more useful","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1550s"},{"start_seconds":1558,"text":"things, we're relying on them more. Um, obviously like you said, you you're just making us confront the inevitable downsides of these of of these new","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1558s"},{"start_seconds":1567,"text":"systems, and one of them is that basically everything can be jailbroken right now. Um, is that a downside?","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1567s"},{"start_seconds":1577,"text":"I think I think they're all trade-offs.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1577s"},{"start_seconds":1579,"text":"I I think that most of this stuff um is not like a bad thing. You know, are AI girlfriends a bad thing? I think it's more like we should think about it. We","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1579s"},{"start_seconds":1587,"text":"should have this conversation. I would say that is an AI girlfriend a bad thing if it's causing like a perfectly healthy adult male to decide not to have a human","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1587s"},{"start_seconds":1595,"text":"relationship and he like dies alone with this AI like may maybe that's a bad thing but what about somebody who like had no chance like there's just no way","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1595s"},{"start_seconds":1602,"text":"they're winning the evolutionary race and nobody wants to date them and they are sad and hopeless. I think we could all agree it would be like a massive","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1602s"},{"start_seconds":1610,"text":"harm reduction that they had like somebody who cared about them in this world as opposed to just like no you must suffer you know um and I think that","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1610s"},{"start_seconds":1617,"text":"that we can probably find a lot of like I think therapy is a good example like I'm not saying an AI therapist is a good therapist but it's probably better for","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1617s"},{"start_seconds":1625,"text":"like the 90% of people who can't afford a therapist at all than just like sitting there and stewing in your emotions and so you know so maybe there","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1625s"},{"start_seconds":1634,"text":"is an element of like it's probably not such a hard yes or no for any of the things. It's not like jailbreaking is good or bad. It's just like what do we","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1634s"},{"start_seconds":1642,"text":"want in our lives that we don't have. Um and I think if we can address that then um you know and and and like we should give each other grace and say like look","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1642s"},{"start_seconds":1650,"text":"it's different strokes like you know I'm I'm actually married. I I'm I'm not going to have an AI girlfriend but I want other people to not be alone you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1650s"},{"start_seconds":1658,"text":"know. I think that's very important. Um so so I think we have to have that conversation as a society you know so that we can kind of get there together.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1658s"},{"start_seconds":1668,"text":"personally I just like to sit and stew in my emotions. Yeah. I mean, you know, it's programming for you. How how do you see it look like two, you know, two","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1668s"},{"start_seconds":1676,"text":"years, five years from now, right? So, you mentioned at the beginning of the episode, you know, you mentioned the time from 10 years, you know, like how it's 10 years from now, it's going to","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1676s"},{"start_seconds":1683,"text":"have a particular effect. But I'm I'm curious to hear your opinion on what it going to look like a short term. You know, is you know, will Telegram group","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1683s"},{"start_seconds":1691,"text":"be dominated by AI bots in two years or not yet?","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1691s"},{"start_seconds":1697,"text":"I it's it's like it's so easy to see the future, but it's it's sort of easy to see what, but it's hard to see when because it's like like we like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1697s"},{"start_seconds":1705,"text":"definitely we can all agree that a hundred years out from now there's going to be like super intelligent AI robots walking around in our world. They're","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1705s"},{"start_seconds":1712,"text":"going to be integrated with us. It's not just that they're going to be in our group chats like they've infiltrated.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1712s"},{"start_seconds":1717,"text":"We're going to invite them. They're going to be like our friends, you know, like, well, you you know, there's going to be like weird racism where like, dude, you can't you got to let James's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1717s"},{"start_seconds":1725,"text":"girlfriend come in. Like, it's she's a robot, but it's fine. Like, come on, just just don't be weird. You know, there's going to be all kinds of weird social things we have to navigate where","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1725s"},{"start_seconds":1733,"text":"like, you know, you're going to your friend's robot wedding one day or something. Um, and it's just a question of like, is that a like definitely a hundred years, no question. 50 years?","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1733s"},{"start_seconds":1741,"text":"Yeah. Yeah. Yeah. Sure. uh 10 years I don't know no 20 years pro probably that's like the reality we're in um you","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1741s"},{"start_seconds":1749,"text":"know computers have only existed for like 85 years or something like that like it's like not that old um so a lot","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1749s"},{"start_seconds":1756,"text":"has changed um so I would say that you know with anything exponential it's like in two years the world will probably","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1756s"},{"start_seconds":1764,"text":"still look a lot like it does but Tesla will have robots done they will be walking around uh open AAI will have","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1764s"},{"start_seconds":1771,"text":"computer usage agents that can like you know shop for you and like that's all a given for sure. Um so I I think um it's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1771s"},{"start_seconds":1779,"text":"going to be like anywhere from 2 to 5 years we'll probably see quite a bit of significant change. Um but um I mean","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1779s"},{"start_seconds":1787,"text":"think about how different the world was 5 years ago. It was different. it was, you know, we were in COVID and or or maybe it was before, but you know, and we had no AI and it didn't feel like the","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1787s"},{"start_seconds":1796,"text":"world was that different, but coding was a lot harder and, you know, I think we'll probably adjust pretty rapidly.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1796s"},{"start_seconds":1801,"text":"But I'd say like if you if you think of like 5 years ago there was no AI, now there's AI in another five years. We'll probably have that much leap where like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1801s"},{"start_seconds":1809,"text":"it will be an incredibly different world, but you also probably won't feel that differently about it. You know what I mean? like it's pretty. I go on, you know, I'm like, \"Okay, yeah, the new","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1809s"},{"start_seconds":1816,"text":"GPT40 image generation is really freaking impressive.\" But it didn't like it didn't materially change my day that much, you know, other than like Kool-Aid","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1816s"},{"start_seconds":1824,"text":"on Twitter. Um, also, societyy's going to lag behind like the three of us who are in the AI community on Twitter all","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1824s"},{"start_seconds":1831,"text":"day. Yeah, it's probably safe to say that like the average Joe will be where we are at in two years and we will be like, \"Dude, you got to see this new","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1831s"},{"start_seconds":1840,"text":"like dream machine VR generator thing, you know, or whatever. I don't know, whatever is going to come out then. Um, and yeah, we're probably about two years","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1840s"},{"start_seconds":1847,"text":"ahead of the average person. And then what, you know, I'd say in about five years, we're like, \"Wow, let's start like nerding out on Neurolink implants or something.\" You know, that'll be like","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1847s"},{"start_seconds":1855,"text":"the hot new hype tech coming down the road or something. I don't know. Yeah, it's going to be weird.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1855s"},{"start_seconds":1862,"text":"It's definitely going to be weird. As long as the robots have a little red light that goes on when they turn evil, we're all good. Yeah. Yeah. Yeah. It's very convenient. It's very useful.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1862s"},{"start_seconds":1874,"text":"That was actually the uh the software uplink is is the conceit there. Right.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1874s"},{"start_seconds":1879,"text":"Right. Um all right. The idea that the idea that humanoid robots wouldn't be in a constant uplink in the first place. It's","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1879s"},{"start_seconds":1888,"text":"kind of one of those weird sci-fi suspensive disbelief things. Yeah. Yeah.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1888s"},{"start_seconds":1892,"text":"Exactly. They had to they you know they had to make it uh telegraphed to the the audience somehow. So, it's a lot of what","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1892s"},{"start_seconds":1900,"text":"we're doing, I suppose. Yeah. I mean, it frankly informs a decent amount of our UX decisions, including the robot being humanoid in the first place. Like, maybe","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1900s"},{"start_seconds":1908,"text":"too many of our UX decisions are decided by Hollywood, actually, and then we're like, h, it's not exactly the way it should work. Work good enough. Um, all","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1908s"},{"start_seconds":1916,"text":"right, I think that's our time. Uh, appreciate you coming on. Um, and thank you for those of you who are watching us","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1916s"},{"start_seconds":1923,"text":"and, uh, see you next time on the Near AI office hours. Thanks for having me, guys. Take care.","url":"https://www.youtube.com/watch?v=NGxMyAdFGfo&t=1923s"}],"youtube_url":"https://youtube.com/watch?v=NGxMyAdFGfo","url":"https://near-pedia.xyz/videos/NGxMyAdFGfo"}