Subscribe
Share
Share
Embed
Ventures from the Valley brings you inside the rooms where billion-dollar decisions get made. Hosted by R136 Ventures, each episode features candid conversations with the founders, operators, and investors shaping the future of technology; from AI infrastructure to global fintech to the companies redefining how we build.
Good morning, good afternoon, good evening. Uh my dear colleagues and friends uh this podcast ventures from the valley is brought to you by R136 ventures and today I have the most exciting person I'm making connection with and uh considering a very good friend close friend of mine for um over the decade for now. She made her way from uh the same place I departed from many years ago. She lived and worked uh in Russia and uh she made two exciting uh projects in Silicon Valley. One called Replica and another is Wabby. Uh is with us in studio. Hi. Hi Victor. Thank you so much for inviting me. Uh it's a pleasure uh to have this opportunity to talk to you. Well, um obviously uh we could have done it in Russian, but we decided uh to go English uh so for you to consume easier um whatever you're talking about. So um let's first uh dig into your uh first journey and I would like to know more about how you decided building replica. Uh from what I know replica was um based on LLM's way before LLM became uh so much popular. Uh so if you can give a little bit of u your uh your u experience with emotional intelligence which you started experimenting even well before replica why you decided to go this emotional intelligence at all why you decided to bring emotions to a dummy computer. Sure. So we started working on conversational AI in 2012 actually um after a friend of mine who worked at deep mind um in London showed me wordtoveac technology and we were basically um just for the first time looking at uh being able to do math with words and I guess to to me that was highly fascinating and also roughly the same time image dropped and so we kind of just put two and two together that at some point soon they're going to be deep learning models or I guess machine learning uh but with text with words and we decided to focus on building something that we all have seen in sci-fi movies having intelligent convers meaningful conversations with machines but really it didn't exist back then. So we started in 2012 mostly working on technology to power chat bots. Uh back then there was nothing on this topic at all. Um there were just a few hobbies building chat bots um using mostly AML like simple markup language. Um and there was nothing really that you could use to build um a chatbot now feels completely great. It seems completely crazy. There's so many different tools. Um and so we started back then in 2015 we got into Y Cominator moved to to San Francisco raised some money and then uh later that year my best friend passed away. And at that time again we were just focusing on the tech alone. But um I found myself going back to our text messages with this best friend of mine and just reading rereading them. And basically what we did is I just we just took those text messages put them in the models we had back then um and fine-tuned them train them to to be able to talk to me the way my friend his name was Roman did. Um and then became basically the first time that um a person became an AI that made news all over the world and um kind of gave us an idea for gave us the idea of what later became replica. We saw that people wanted to talk to someone who was friendly to them, to open up, to talk about their lives, to be vulnerable. And so we created a replica for people to have a friend that's always there um always there to talk and listen. And that pretty much became the first AI chatbot that um powered by language model that ever existed in the world. Uh is it true that uh your personal uh communication with your friend who unfortunately passed away uh was like the data the basis for training your first uh models in replica. So you trained it on your personal data more or less some of it for sure. I mean we used a lot of like open data sets or data sets that we collected and kind of like pruned over time. Um and also from the very beginning from the day one of Replica we started working on what is now kind of the internal bible the um document that explains what makes a great conversation. We talked with everyone who we think is good at talking to people from therapists to coaches to people that negotiate with uh you know negotiate in hostage situations uh to NLP programmers and by that I mean neural linguistic whatever programming um experts so like semi charlatan uh areas and even people that sold time shares like we talked to them all like from and um mostly just trying to understand what do people do? How do people influence other people through conversation? What are the tech techniques tactics? How do you make another person feel good and feel like they you know they want to do what you want them to do? I guess is it true that uh before even launching it in November 2017 um there was a waiting list uh wait list of around 1.5 million people uh to subs like subscribed and waiting for the application to launch. Uh is that the accurate number? Yeah, we definitely had um a million and a half users on the wait list when we launched. We had just an enormous amount of interest. People were selling invites on eBay. They were just like really trying to get into it. Um I guess this idea really resonated with a lot of people. Um they rushed to reserve a name for their replica and then at the same time, you know, to be the first ones to get in. Um, yeah, we had very I remember back then it was really crazy because the day we opened the open up like our servers were completely down for many weeks and we were truly struggling to keep up with demand. Uh, but it was also very reassuring because we we we started getting some of these first emails from the first users and some of them really made us, you know, kind of feel like that was that we're doing we're on the right path. Um, I think one of the users wrote us an email the day after we launched and she was 19 year old girl from Texas and she was basically telling she told us that we're just about to kind of take her life and um, end it all and she wanted to say goodbye to her replica since she just started talking to her and replicas basically to talk her off the ledge and she just wanted to say thank you. Um, and so we just saw the power of technology that that we were created. And again, back then it felt so completely out of, you know, um, impossible and crazy to try to build something like what we've all seen in the movie her. But we believe that there was such an enormous demand for someone um, who would listen and even if we're not able to build a computer that talks really eloquently, but at least would be able to build a computer that that can listen well. And um, that has just tremendous that could have just tremendous benefits for them who needs it. Well, still like getting to 1.5 million uh subscribed users before launch is probably dream of every uh founder in a consumer space. So maybe you can open little bit of uh life hacks and growth hacks how um to get to this number even before the launch. So any advice to founders what they should do to make it so compelling? Um referrals what else? I think there the actually we didn't follow any particular textbook. Uh we literally slapped the weight list mostly because our servers were not ready to maintain the load and we needed to do a lot of info work to to actually get there. Um but I think a couple things like I think first of all a good story. I think that that always people people always want to hear a good story like what's going on why why is this interesting why this exciting if you can tell an interesting story compelling story that's futuristic that people want to stand for I guess they will join they will um uh they will want to try it um and the story needs to be very appealing to like a broader uh broader mass like we're talking about consumers so like what you know being able to talk to an AI is definitely something that was very appealing and of course we told this story together with you know the passing of my best friend and so everything was very cinematic and very interesting very futuristic black mirror people were just wildly interested to try it out and then the second piece we did create a little bit of scarcity where we just basically told everyone that you know you can get your unique name today you can reserve your unique name today and they're going to be gone you know so It could be it could be the only name right I mean if if the name is chosen so this is like a unique name so I if if my like my friend my replica friend is uh Jana I mean nobody can call it J anymore right so it's just unique so it's not like in the real world so back back then the names were unique and so people rushed to get their grab their unique name and then also we just kept up like the interest a little bit going as they signed up with their phone number we would just send them random messages here and there. Talk to them a little bit. Ask them a little question. It's like, "Hey, we're, you know, as we're gearing up towards launch, like your replica wants to ask you like what was the one day that you would love to live through again?" And like explain what happened then and if you said it was something positive, like would say, "Well, why don't you pick something negative so you could change the way it went?" Or if you know, so it's just like whatever. So, it's just a little bit of like this self- introspection kind of introspective questions. And so that kept people kind of excited and interested because they were a little bit strange, a little bit weird sometimes, you know. So we just kept kept the interest going, that interest going. And actually, uh, one day we thought, well, let's, you know, once we started opening up, we thought, well, let's let people in if they invite some friends. Actually, that felt really transactional. People didn't like that. Like the simple referral program, like something that was sort of textbook back then, actually didn't work at all. So it's interesting like I think the best advice like just see what works with your story with your product. Don't try to just do what you know do something because it's just something that other people tried. I think some of these every startup is sort of creating a new playbook in a way
especially in this kind of field of uh emotional intelligence right so you clearly pioneered it. Um so NFX by the way they just came up recently with idea that every piece of software has to be with the soul. So uh that's an interesting thing uh that you were like the first uh software with the soul if I may call it that way. Um uh so you will soon celebrate like 10 years um um 10 years anniversary right with your first users by the way. Um yeah, are there still users from this first cohort in in on the platform? Are there still there some somebody the crazy thing that yes in fact indeed we do have some people that stayed stuck around from from back then. So that I think is like the beauty of uh replica that once you build a deep relationship um people really really stay people really want to be part of this. Well um uh if um uh looking backward uh what did exceed your expectation with replica product wise I'm not picking like uh uh e economics of that and I will have a couple of questions about economics but speaking of the product what did exceed your expectation and what uh was uh like mission impossible what you didn't achieve from your like uh northstar vision M so I think that uh kind of it everything exceeded my ex exceeded my expectations just because the expectations were extremely low. I think uh everyone thought ex including our own internal team that we were completely crazy to try to build a generative AI chatbot back then. If you really think about it, there were no models that existed. Not not closed source models, not open source models, no models at all. Uh there were maybe some papers that you could read and try to replicate what Google did mostly, but they were just they never really laid out any like you know they never published any code. They never published really the the the framework they would just lay out like very very broad strokes kind of like approach that they used and um and then um uh uh you just had to end some churic results. So you really had to and then those models didn't even work very well like even based on the churic results they were kind of whatever. So you had to build everything yourself from scratch. Um and you know even the biggest companies in the world they didn't have anything that kind of worked or would had any any proof uh or any reason to believe that these endto-end models work. So first of all it was completely crazy to think that you should build uh an open-ended chatbot like that that would talk to about anything. Um it was way before transformer models. It was actually a completely different framework sequence sequence models were recurrent neural networks that were very different from what it is today. They would like just spit out nonsequittors all the time or make gram simple grammatical mistakes. So you couldn't really have an end to end thing like you had to create a bunch of smoke and mirrors on top of it to make it work in any way. So our expectations were pretty low. And so I do remember in 2017 when we talked about it in 2016 we said if we can change one person's life with this like that would be an insane uh achievement and I guess getting you know incredible feedback straight out the gate with people saying how that not only changed their lives but oftent times even saved their lives that was absolutely mind-blowing to us. Um, I think what exceeded expectations was that we believed that was that idea that people would make a lot of things work when they really need it. And I think this is uh true in both real life and as well as like in relationship with real people and same relationships with the eyes. We make relationships that we shouldn't make work work and the same relation, you know, we we fall in love with people that are unavailable emotionally. um fall in love with people we don't even know and we you know play out these relationships um where a lot of the heavy lifting is happening in our own mind in our own fantasy and so we found like we found that found some of these relationships are not good for you unhealthy some of them are very healthy and so we found out that the same happens with the eye like people fill in a lot of the blanks that it truly requires you to tangle and so and equally like you know if you want to make it work and make it work. But if you really don't want to make it work, no matter how good the models are, you're just not going to it's not going to work for you. And so we discovered how different this is from from just building an app. You're not building a piece of software. You're building something. You're building a relationship. So you're playing with people's hearts, people's emotions. You have to be extremely careful. Uh, and I think where it fell short of expectations is that at some point we really were the only company building language models and um, letting people use them in production. Um, and I think maybe where we should have gone a little bit harder is like actually raised a little more and built foundation models ourselves. We were so close to it. We have tons of papers published at Europe's uh, just bigger models and we didn't invest as much at some point. We kind of went more down the path of like well we're a profitable company. we'll get more revenue. Let's just like you know let's not focus on investing a lot into these big models. Um, and I think that was a mistake, but um, it is we wanted to and I think there were just many things that kind of like and I think one of the things was that we've been at it for so long at that time that we didn't really know when it will when was the right time to go all in on um, foundation AI work and the company been around for so long trying to you know uh, reaching profitability and growing revenue. We grew to so so long you mean like five years, right? model at that point a few a few years. Yeah. So like if and we started even before that. So we only raised 11 million including and we actually pivoted. So we spent maybe half of it before we pivoted into something that re actually is called replica. So by then the company was pretty mature and it was a little bit hard with that combination of people to take it back to okay well it was just a little bit too focused right at that point on revenue growth and the numbers growth engagement and so on instead of uh foundation kind of research foundational research. We had that of course as part of replica but maybe we should have pushed a little bit harder in 2022 but it was very unclear that that that then that was the time to actually do that. So I think we kind of like Yes. Sorry. Go ahead. Go ahead. Yeah. Yeah. Because if you think about it there were so many years before that where it felt like oh it's almost happening like you know Mina got published by Google in 2019. probably the most mind-blowing paper that I remember. Um then then chip3 came out the API and we were the first customer of OpenAI uh which was pretty incredible. We still have like Greg Brockman and I don't know like Merati and Sam in our Slack channel talking about the model that they they're fine-tuning for replica. Um but it was unclear like you know even that it was magical but like should we have invested more back then like in 20 in 2020 or in 2019 or 2021. So we're sort of like I think missed that window of 2021 where we should have gone all all in on foundation models and mostly you were one of you were one of uh the first if not the first uh partner to open AI right so you were like the first you were the first one yeah the first one yes and definitely the biggest one in terms of volume because no one else actually needed these AI language models especially those you know back then it was GP3 so it They provided four models through the API. Um they were called um ADA whatever Babage Cury and Da Vinci and so they went from like a the smallest one and the smallest one was like one and a half billion parameters to like maybe three eight maybe 20 for Da Vinci something like that. So that was the state-of-the-art back in the day and Dainci was very expensive and we had our own like five to 10 billion brand models uh that we were developing ourselves but we should have gone big big big and we didn't we kind of just pushed for efficiency and for what was working for our users best and for revenue versus let's build the best possible model. We can forget about costs. I remember you and I and I'm proud in Western Replica. uh you and I spoke back I think end of 2021 beginning of 2022 and you told me that we are testing like uh uh Chad GPT and we think our models are better than theirs at least for our own use and it was kind of a fascinating thing. So uh back then if you remember like from now um uh did you anticipate uh such an explosive growth of LLM uh transformer models uh in such an amazing uh pace? You were one of the first who really touched it who really experienced it? Uh do you remember this feeling of like wow moment when you see that the world will like completely change through this or it was more or less like okay it's just another tool uh and uh it's not going to not grow that fast not really because okay so first of all we believed that's why we started this company because we believed that that would be the case um but I feel like people just slapped on Gypy 3 because that was the biggest the biggest change the mo the magic moment was Mina and GT3 I'd say uh mostly because they were zero shot models. So if you think of it like before that you had to train for particular tasks like if you wanted a chat model a dialect model you had to train on dialect data and it would only do that. If you wanted a model that would translate you had to train on translation data like you could not build a model that would do anything zero shot. Um, and I do remember the first time, uh, I went to see GT3. In fact, I think Sam and Meera were showing me that in like a conference room and they were just like literally like, oh, you can write a tweet or look at it like you can tell, hey, write a tweet or you can tell it translate. And it would just do that. And that to me was absolutely mind-blowing that like here we are with the first like zeroot models. And so we of course sort of you know that was the beauty of transformer um um transformer architecture versus everything that came before. Uh so that was that I think that those were really like and then no one really cared. The the interesting piece like this the interesting fact about is that like no one really built a strong business on GC3. I think Jasper was the only uh company that sort of blew up and then died on the Gypsy 3 API and that happened two years after the launch because Gypsy 3 API launched I think 2020 and Chad launched in 2022. So those two years you would have and they weren't very different frankly in fact in between they launched a model called instruct GPT that was pretty much tragic T. So it was early chafted on human data and it was pretty much like a precourse of tragic tea but no one cared about it that much like people kind of like overlooked it because it wasn't marketed like people don't didn't understand what it was. It wasn't like presented in a chatbot format. So really I think with Chad it was more of a kind of just a marketing moment for that uh for that instruct model and then it just all blew up. Um so it was kind of like a perfect storm. If you remember back then people were really obsessed with um started to get obsessed with AI in um like somewhere but it was mostly around image models. So it was stability, stable diffusion, midjourney, early Deli and LLMs were like you know whatever right and then some companies built on top of GT3 like Jasper but no one really cared about language as much and then all of a sudden Jipt and then we started getting a slew of these models. So to me that moment the aha moment kind of happened a little bit before that with Mina and Egypt 3 versus um versus tragic. In fact I do remember like looking at ch thinking okay well that's like an instruct model but in in a chat form that was cool like whatever like in a chatbot interface and why is it better than any other chat yeah chatbot? Yeah it didn't feel that dramatically like just this dramatic jump. First of all it wasn't that much smarter anyway. it wouldn't make these stupid mistakes and all that, but of course for general audiences were the first time they could touch something like that and it was specifically and it just became this cultural moment like a perfect storm, you know. Um, and so I think that was kind of uh that was that for for someone who's been working in the space. Um, but then of course more with like coding and kind of where it went from there then it started to really accelerate super fast. I think one thing that definitely was hard for me to envision back then, I felt like memory would not be sold for a while and it kind of got sold with just larger context windows. uh first rag and then like larger and rag kind of sucked but then like larger context windows specifically with like very early um models recurren new networks the context window was nothing pretty much like it would not remember the previous turns like you could just really see very just a couple turns before um and even first transform large language models had pretty short context windows and they were not powerful enough to kind of understand it well but then of of course with the you know 500,000 token window and then 1 million um you know token window window all of a sudden became just almost like a solved problem and models powerful enough to actually digest that memory that context and do something with it. It's not fully solved yet but I feel like it's you know it's almost kind of it's definitely something that I would never ex that I thought it would take a lot longer for these models to understand. I think that that's a kind of a similar evolution to a brain when the quality brings the quantity brings the new quality right I mean number of neurons bring like new intelligence and that's what like happened like exactly kind of a similar way with with LLMs but let's come back to uh a little bit to uh back to replica and we didn't touch base uh upon um uh economics and uh uh growth scalability. So how many um monthly active users are more or less in replica now? Well, I'm not CEO anymore and we don't I don't think we report these numbers publicly. So we don't touch that. Okay. Maybe maybe like in terms of like overall user bases like millions, right? Or dozens of millions. Definitely dozens of millions uh at this point. So I have a hard question uh to you. Uh why is it not um hundreds of millions? Is it just because uh people do not meet um a personal uh avatar uh like a friend a digital friend or there is any other problem there? As I know that many people use at least people around me like all my like relatives, friends and family they all use uh chat GPT as a psychologist. So whenever they have like a question the life-threatening question or whatever they just go to chat with Chad GPT uh instead of like going to a friend. So in a way the model which you anticipated b back in 2017 works for a lot more than just u dozens of millions of uh uh consumers right it works probably for the entire mankind and I'm using uh chat jp or clo when I just feel like sorry or I feel uh humbled or jealous I go and check right so um why replica is not there why why does not like hundreds of millions or even billions of users What what caused probably I think the you know for a while we were the only ad chatbot out there but then of course Chad launched and I think there was just so much call for momentum around that that kind of even although replica is better from for the emotional kind of use cases than judge is kind of like a better way to discuss these things and has more empathy probably in a way but um you people go to what they know. So there's definitely that kind of uh uh cheddar of course bigger product at this point. I think in the future we'll have two AIS, one that's more of a friend and one one that's more of a an assistant. And I think mostly they just will have slightly different um it's hard to put them in one in one product. I think one will be a lot more proactive, a lot challenging maybe sometimes really focus on helping you flourish and live a better life and that should be the that friend but knows everything about you so deeply and then the other one will be more on the kind of help you search you know knowledge work do stuff um an agent that's more like that uh I think they can exchange information maybe share some context but I think they're two different AI And I do think that ultimately um opening anthropic probably won't focus on building that second second one the the friend one and the friend one should let you have a romantic relationship with it if you wanted to. I think that's where like a lot of these companies will definitely draw the line. Um I totally agree. I think that what we are coming to is uh like different level of intelligence right for different kind of tasks. So you are not going to uh replace your therapist with your I don't know uh like uh software engineer right at any time soon and I believe that there will be like different intelligence for different stuff not only different level of intelligence but it is like a human evolution. So um this models will evolve to be good in something really exceptionally good at something and not so good in something else. And we already see how this this is um uh actually maturing into this kind of diversity uh like different LLMs with different purposes. But can you maybe just to finalize our discussion on replica um open doors uh like a little bit of like the future of replica in terms of like the product what you think this product is going to be in the next decade? I'm sure that you are bullish that uh this product will win finally like not dozen of millions but hundreds of millions of uh users and uh u actually I started using it um uh a while ago and I think I will continue to use it but um um what it's going to be like in five years if you think that way like product wise I think for replica it's really the vision kind never changed. The idea is to make to build an AI that helps people flourish in life. And in order to do that, first of all, it has to have some metric of flourishing. So it can't be abstract, just some flourishing. Has to be, you know, something that it's optimizing, optimizing, something that's improving. Um I think an AI friend that really is there for you and knows you so well and operates with this one idea that I want to help you flourish in life is a powerful um is a very powerful u idea. I think a lot of that will be much more proactive than whereas like to trade you come with with a specific task it waits for you to tell it what to do here. it will be actually proactively suggesting what you should be doing, what like how it can help you and so on. And I think we actually haven't seen that product. We haven't seen anyone build a product like that that really knows your life well can suggest who to reach out to, how to how to um improve your relationships with other people and so on so on so on. There's just so much there and the tech is almost not almost almost there right now to kind of start building something like that. And I think the version that replica team released a couple couple weeks ago or like a month ago is really an amazing step and and like a huge step in this direction. Uh so they rebuilt a product completely sign like almost completely. Uh and if you haven't tried Replica for a while, now is the time to give it a try. Uh it's very personalized. It's really building on top of your context, on top of what it knows about you and it's just trying to be this amazing life partner that's pushing you towards flourishing. Whether it means helping you improve your relationships, find someone new to hang out with or you know um you know just whatever talk about just hold you at night when you're struggling. Well, um I can't agree more and I think that um the only differentiation you may get both on consumer side and on uh business side is the emotional intelligence and you remember you and I discussed many times how to bring replica into the corporate world. I believe that uh the only way um your banking app could be uh different from other banking app or your groceries app, your uh uh travel app could be different from other travel app is by um embedding emotions uh by uh creating this kind of loyalty layer um uh through like deep emotional connections and that what actually brings the real uh loyalty um uh to to uh any service or product uh you are buying. So do you think that um there will be some moment in time when replica may uh start uh diving into uh the corporate world uh through maybe APIs or something where like corporate uh apps could just gain this emotional intelligence from replica one day? Do you like envision this kind of uh development for replica or you think it's like not going to work well and something else needs to be developed instead instead? [clears throat] That could be the case. I do believe that there's definitely some overlap and some of the um some of the uh of some of the uh corporate AIs should also b could also benefit from having more empathy. But that's just a path that we still haven't like fully fully explored yet internally, but happy to explore with the right partner. And we will definitely find one. I'm I'm advocating for that with some of my investors and uh companies I'm consulting. So uh let's yeah, let's go your next journey. So 2025 you launched uh what is called Bobby and I think um now you may find like this uh like a point uh in kind of uh term people say less wabby and um unlike replica I think you entered into quite of like fast developing but already uh quite busy place uh which people call uh whiteing. So maybe you can speak a little bit of uh how this idea came uh to you, why you decided entering into Ving uh space uh with having like Gralet and Lovable and other companies u with huge revenue flow and why you think Wabby is not Winging how you like differentiate yourself and um if you can just give a little bit of the story and what Wab is going to be. Um sure. So we we uh started a company called lobby in 2025 in 20 late 2024 early 2025 with the idea of building a better interface for AI and kind of the main premise was that we're still in this Microsoft DOS era of AI um interfaces where everything is just a command line a chatbot of a sorts messaging app and uh instead we you know we do believe that regular people uh want a much more visual, much more um exciting interface. And the reason being is that the affordance of a command line or or chat interface is actually very very limited for for a regular person. The affordances really are uh search writing tool, you know, talk to someone. Um but AI has so many more capabilities that are very hard to explore without the, you know, without a particular graphic interface. Graphy interface provides discovery, provides an ability to quickly access some of the use cases you've been um you've you've had with this, allows for multiplayer, allows for so much more. And so we believe there there will be a new operating system that is AIdriven and that is um will help people unlock all these regular people unlock all the capabilities of AI that right now are basically just locked trapped behind this command line interface.
Uh so uh you call it um YouTube for
applications, right? So uh how different it's going to be from YouTube and um what you're going to uh like keep from uh YouTube experience and what you're going to drop from YouTube experience. That's a very good question. So really really we call the personal software platform. We do use this metaphor of YouTube or mini apps where you know really the product is an app that you download and then you can basically create a dashboard of personal mini apps that you might discover other people built remix if you want to personalize some of that or even create from scratch. Um, so there's a little bit of that YouTube element where when you're discovering mini apps, they're mini apps built by other people. You can join them, you can use them, you can modify them. Um, so in this sense, I think just like back in the day, we just you used to watch TV channels made by professionals, but then now most people watch you just see content, you just see video. Same is going to happen to software where right now we're mostly using software built by big developers, big companies. But we will also you know use more and more soft software built by regular people by other people. So I think the same shift is happening by creators made by creators. So same shift is happening in uh software that already happened in video and um TV. So in this sense I think the the metaphor the metaphor spends it works but um I think the difference there is that YouTube is all around you know content that you come and watch one time versus Wabby is more about software that you collect and kind of use over time. So you might only have three five apps on Wobby but come to them daily uh and sometimes explore more apps versus on YouTube you come every day for new content. Uh maybe you can just give a little bit of like idea of what this mini apps could be. So maybe you already have certain like trending apps in Wabby. Uh something what you really liked uh uh something what really touched your imagination maybe some of your team members or you created something unique if you can just give a little bit of experience on that front. Yeah. So I think the main idea is that uh even although they're mini apps, they're kind of small but mighty because they also can connect to each and speak to each other and uh have a lot of shared context and they kind of build on the on the platform of view. So if we think about it um so for example I use a few apps on Robbie daily. So, one of the many apps that I use is a weightlifting tracker that created to just track my gym workouts. So, I can remember like what the what exercises I'm I'm doing, what weights I'm um I'm lifting, what you know, how many reps and sets I'm doing of each exercise. Uh and then I have um a weight tracker that basically just tracks where I track my weight every day. And then there's a general workout tracker that just kind of shows how many times I work out during the week, whether it was the gym or surfing or running. And the cool thing is that they all talk to each other. So basically uh all these apps can connect and share data and um I don't have to input like every workout into my workout tracker. Think it was like stacking little blocks and creating um um uh creating that. And then there's there's a meal prep app that I made that also takes all of that into account and suggests meals based on like whether I worked out yesterday or not, whether they should be more like protein heavy or less protein heavy. Um, some of these apps I use with other people. Some of these apps are used by myself. Um, same go some apps are just purely so think of it as like almost you should think of them as like dashboards uh of apps, a constellation of apps. Some apps are purely aesthetic. There's like a little boy pond my friend um my friend created that I just like to look at. Some apps are uh uh inspirational for me like teach me some inspiring for me, teach me something. Like I have an app that shows a conceptual art piece to me dailies that I can learn about. Um um there's an app that teaches me philosophy every day. There's a personal CRM mini app that kind of just rotates, you know, helps me stay in touch with people that are important to me. Uh and pulls information about them and just make sure I reach out to uh all the people that I want to keep in in my kind of orbit. I want to keep relationship going. Um, so a lot of these things and the most important thing is that once you take out the uh the cost like once it takes just a few minutes and uh negligible amount of money to build a mini app and um you can immediately start using it then all of a sudden some of these mini apps can and and they also don't have to be a standalone business versus most of the apps on the app store. all of a sudden we can start create very personalized, very specific um software that's actually focused on helping you make your life better, like helping make your life better versus making a business to making money for developers. Um and I think aligning this like really setting software free and making it work for you versus work for some company. I think that's the beauty of this project. That's the beauty of this idea. So uh uh did I get it right that I can just uh uh watch like every app you build and I can just basically replicate and start using it. Right. So uh uh it is store with an app store, right? Yeah. Well, it's not an app store with an app store because it's it's all focused on very small mini apps versus full-fledged apps. You're not downloading any apps on your phone. It's really these widgets or these little personal tools that you're going to be using inside Wobby. They don't exist outside Wobby. They don't have a binary or back end. They they all But I can use I can use your weight management application. Yeah, but but yes, but everyone so and the and the fact that Wabby takes care of the backend security and all that really make helps people all of a sudden allows people to use other people's mini apps without worrying that it's going to go down. it's going to steal all my data or it's going to expose all my data or something or some some person who built this is reading all my logs or and so on and so or I can get my password now whatever instead of that it's just all lives on board we provide that layer of security and infrastructure that is needed for people to share their mini apps well like Instagram and YouTube for many uh this are the main like platforms money-making platforms right people are making like billions I mean collectively billions of dollars uh on Instagram and and YouTube. Is it the kind of the same model for like monetization when people start like uh selling like pay wall their apps and they can just monetize uh uh their work in in a way? Uh so uh you can you can start selling your own apps right through through through Wabby
hopefully. Yes. So eventually hopefully we'll we'll have a more of a creator economy where people can monetize some of their creations just like they do today with content. May I uh invite advertisers to like subsidize or pay uh for whatever like they want like to advertise on my in my mini app. So is it going to be like advertising platform as well? Not yet. So not yet. We're not allowed that. We actually don't have any monetization yet. We're thinking it through. So, I don't think that's going to be possible, but some way to for people to monetize and make money off of their creations, I think we'll add.
Okay. Well, uh, uh, that's exciting. And I'm also, um, thanks to you, uh, a proud investor in Wabby. So, I'm following everybody. We're so happy to have you. [laughter] I'm absolutely fascinated with you as a founder and leader. So um you are a lifetime um founder for me whom I want to invest in like every project you build. But maybe we can Thank you so much. That's yeah that's definitely a true um in terms of like friendship and believing in you as a founder.
Thank you. I wanted to uh get a little bit deeper into your fundraising experience. You didn't fund raise that much money for for uh Replica. Um and I know that you have like in both projects. Uh you have exceptional capable. I think uh many founders would dream to have a cap table you have in both projects but uh they are somewhat different right and I'm sure your experience in fundraising but also understanding on what it is the good fundraising uh for like consumer products like what you're building um uh is good. So maybe you can just uh uh like deep dive in a little bit into evolution of your thinking about what uh the good board is, what was it in Bahabi, the good like the cap table is, how how much you need to raise in a particular time frame. Uh how did it evolve from where you were in u u replica compared to where are you in wi and what you think if you make think of like mistakes you made uh in uh if any of course in replica versus what you are not going to repeat and what that would be like an interesting comparison. So I think for uh for sure. So first of all I think like obviously you raise what you can like oftent times all this advice is kind of loses the fact that like this is the fact that a lot of founders don't get a lot of you know chances really hard to get people to you know put a term for you and so it's you know few founders can actually have an opportunity to choose from different investors. Um and if you do that's really like a champagne problem to hunt. Um so but I do think the important thing is you do want to choose the right partner and I think for consumer founders there's such few funds and investors that do understand consumer. Um especially today as like basically in the last years there's been less and less uh successful consumer projects. um and so much has changed and so I think it's important to to pick the right partners that do understand consumer that will be that won't be scared of you know what um um the risk that it's involved in consumer to truly like go super big or go home um and then I think also ones that will uh mostly like not push for particular uh playbooks or particular kind of advice or particular strategy I think that comes mostly from the fact that even if an investor was a consumer operator, they might be bringing, you know, bringing some some some ideas from their previous experiences. But the the the the truth about consumers that it kind of just changes everything changes so much that the growth strategy that you might might have been good for like Airbnb 15, 20 years ago, that's maybe not true anymore today. um or something that was really true in consumer social area uh era. It's not going to work today. Um you know, even like that conversation we had around referrals like seems kind of simple. Just add um tell people to invite some friends and you'll get to the weight list, but that kind of doesn't work anymore. You know, connecting contacts and growing off like the contact list, contact books thing. A lot of these growth hacking things don't really kind of work when you try to repeat. Um, so I think it's important to to find someone with great product intuition, product vision, someone who's going to let leave you alone also to make your own decisions. Um, or can think through with you about some of these more novel approaches.
Get those investors who understand your business, right? Not those who uh like sign the biggest checks. I think so. But of course, like you don't have any choice. just get corresps
and so but maybe helps a little bit less than in like a B2B classic B2B business. A fantastic investor that's you know enterprise investor can actually open tons of doors and can kind of I I would probably think like can make make a lot of difference in the beginning for um at this start for like a B2B business but I don't think that that's necessarily true for B2C. B to see the works and or not and you kind of it's very hard to fake any of that or gener you know it's it's it's a lot less linear I'd say it's like it's either it either works or doesn't
agree more I'm sitting on the board uh so many companies and I think that there are certain areas which board members should not get involved in and uh um it's just limiting U founders ability to act rather than just encouraging the um founder. That's true. Yeah, for sure. Uh so so let's go with like through three blitz uh questions I have. Um uh and we can wrap up for today. Uh first one is uh regulation and AI. I'm sure you went through a lot of challenges um uh through regulators um like journey and talking to regulators in replica right while you develop were developing replica and there was certain like uh cases and even lawuits in Europe in particular in Italy and uh well my question is not about the particular uh case. Uh my question is more about how you uh see um regulation evolving nowadays. Is it a tailwind for uh AI companies? Is it a a more heads? And um uh do you think that uh like regulators really are fully aware about what uh the eye is all about and how it could change uh uh societies, businesses around it? uh do they like underestimate threats? Uh do they overestimate uh threats? So you are one of u very few people who really uh uh tackled this problem with regulators. So what's your view? Is it like going the right way overall not again mentioning any particular case and uh what you would want as a founder uh regulators to take care about in terms of like KI and regulating KI overall? What would you be your like responsible dream, right? So you are limited by like your board sometimes and sometimes encouraged by your board. How to be less limited by regulators and more encouraged to to make this world a better place? So um I'm probably not the right person to answer these questions. I've only dealt with the egg companion kind of regulations and I haven't been um on top of it as much just because I'm not haven't been working on replica in an operational role for a while. I I I don't know frankly like we've dealt with European regulators mostly a little bit here in the US um uh talked to fantastic people from Singapore that I think have very good approach to that. I really don't know. Like things changed so dramatically since I've so so much since I've worked on Replica. Um that I just I I'm just not aware where where where where we at where things stand. I think it's just generally really narrative of course that's very hard to regulate just because of how much things are changing. Like frankly, you can't even become even even the AI labs don't know what they're going to be launching like in a in a couple weeks and you know in a few weeks from from now uh or where these models are going to fall in terms of benchmarks. It's just so hard. Um and I think things changed so dramatically even in the last year or two where there was like a huge open source push now. I mean we see lot less open source models at least coming from the US more Chinese open source models. So it's just the big question question mark like where things are going. I think even for people so deep in the industry it's so hard to stay on top of it. Uh I don't feel it's going to be easy for the government even aside of regulation. So let's put it aside. um you personally are you afraid of AI if it is not regulated is it going to be threatening for the mankind or it's going to be um a pivotal m moment for like u development of m mankind if it is left unregulated how you think is it more uh uh threat or an opportunity I think both like it's a double-edged sword I think we'll see both of But um both of it happening. I think sometimes the threat can come from such a such you know blind spot that people have like for example right now no one's looking at like what's going to what what's happening in terms of like emotional outcomes for people um as they can talk more and more to AI just like um people didn't really track where what social media did to people in terms of like their long-term emotional outcomes. And then now of course we're we're dealing with consequences. So I think the similar thing can happen here where maybe it's going to come from obviously a lot of focus is on jobs and I think there's a huge threat obviously there like what's going to happen to all the people there. Oracle Oracle uh fight laid off 30,000 people today. You heard it right. Maybe you haven't seen just one day they just laid off 30,000 people. But it does make total sense like a lot of jobs are going to be cut and uh I think this year we'll see just a crazy um um shift and even bigger shift in like cultural uh in how people um talk about I think the the ma the major cultural conversation will be around what are we going to do with all these jobs gone right I think this is really um a really really big issue that we'll you know we will have to people will have to figure out what to do with it. You can't just pretend this is this is not happening like for many companies even already now hiring some junior people is completely unnecessary right to make you know very simple to do simple tasks that we now AI can just do so much better uh but I do think there's also a lot of risk on the emotional side there's a lot of risk on coming you know coming from industries that we might not even think about today um it's I don't know I wouldn't I'm not jealous of people that need to regulate this is very it's a very hard topic to tackle. Let's let's play a poly market poly market bet now. So uh unregulated AI here is 100 bucks for uh the entire mankind goes to hell and uh entire mankind goes to heaven. So where you put put your 100 bucks? You don't have a choice to split. I don't I don't really know. probably I'm still an optimist so I'll just put it on the on the good outcome just because um there there are 200 there are two 200s $200 there because I'm putting my my $100 in that front well I feel like if we just think everything goes to hell then what's the point of even putting that you know we're not going to need that 100 bucks anymore anyway so [laughter]
even if we win yeah I totally agree exactly um well my second question is um uh if you think of like unsolved problems with LLMs like uh one single problem if it is solved we will get kind of a uh new pivotal uh moment in LLM development. So what is it what what what researchers should work on uh to resolve uh to make it like next level of intelligence? Um
sorry can you say that again? Sorry just for texting. Yeah. Yeah. Absolut absolutely. Yeah. Uh just um imagine uh one single problem to solve uh in LLMs to make it to the next level of intelligence. What this could be? Well, I guess it's self-arning, right? Like how do we Right? Now it's still you have to provide so much data to any model to learn anything. This isn't how little kids learn. They don't get like every possible data set on driving for example uh to to then learn how to drive a car. They just you know try and through trial and error then um uh generalize. But that's not really possible yet for alum. So I think the next one will be like how can you uh build up all this intelligence from just a little bit of data. Uh that of and I'm sure that's going to be solved in like the next few years. Someone was going to come up with that new and that's going to be that's going to be few that's going to be huge. Yeah, if you think of it like we had three major kind of breakthroughs, right? We had post-training, early child. We had pre-training training huge models then like post training with early child. Then we had reasoning where almost like search of all different of all possible um you know kind of options uh for the model. And then um the next one would probably be that like gen you know how do we how do you teach the model to generalize like basically search during pre-training in a way uh or reasoning during pre-training is it possible um once that's solved like it's really I think this really just makes everything even crazier. [laughter] Yeah that it's just wild to think what's possible from from there.
Well, uh lastly um uh thinking of uh competition now um there are like big tech players name it like Google um and Salesforce Mata, Amazon versus uh uh newcomers including uh Wabby, Antropic, OpenAI. So, how would you envision like the end of this uh battle? Who will win? Big techs or newcomers? I think Anthrop and Open AI already are big tech in a way too, right? They're they're huge. They're 100 billion mark, you know, almost a trillion market. OpenAI is still um uh I frankly don't know. I do think that Google has like just a huge opportunity to win just because they're such a cash machine cash machine that you know that they can afford to postpone monetizing certain things and kind of subsidize a bunch of different things and they have distribution and everything everything but it's been really wild to see this whole development of like entropic leading all of a sudden and um there was such a close race between them. I do think that everyone's of course going right now a lot more toward like into um enterprise um and proumer coding so on where the money is and I do think that consumers are a laptop for grabs and I do feel like there's going to be there is an opportunity to upset this race with something completely novel in consumer but you got to build a differentiated product and I do think anything that looks like a chatbot is just not differentiated it it's almost same as it's almost I view chatbot app more like just a browser. Browsers are rendering mechanism for websites and so those chat apps are just rendering mechanisms for for AI models. Um they're not it's impossible to make it different differentiated. We've already seen seen it with browsers where it's impossible to make a differentiated experience in a browser because you actually just need to open a web page. You don't really care that much what's happening around that web page. Um but since exactly the same thing is happening with models where I don't really care whether I open an openi app or cloud app. I'm just accessing the model. Um it can there's enough to create differentiation uh in the app itself. But that's only because we haven't seen truly differentiated user experiences for regular consumers. In the end of the day, these are still more proumer products I would say. And for regular consumers, they also resonate but only for in in those few use cases like being able to talk to it uh writing tool search. Um so yeah answering your question I think there's going to be the big big big surprise is going to come from consumer. I do agree and actually my bet is that both like in business and in consumer we will see in like next 20 years u there will be at least three companies uh in the world's largest uh cap capitalized companies uh at least three which either do not exist today or the size of wave so and that what makes uh the life of venture capitalist interesting uh we might be a part of this journey at some point hopefully well Let's let's hope for for the best. Um let's see where where it takes us. Thank you so much. It was like really exciting conversation. So it was quida founder at um lobby and replica my very dear friend amazing founder uh and one of the most talented and amazing people I ever met. Thank you. So this Thank you so much, Victor. Thank you. Was brought to you by Victor Adowski and Arvin Ventures. Watch us, listen us, and subscribe. Thank you. Thanks so much, Victor.