Chad Reynolds: Some of the other frontier labs where they're continuously training models, we're doing the same. It's just our dataset we're working with are real people, and our ambition essentially is to build, you know, the largest computational engine of human behavior.Song: Built This Week, breaking it down. Built This Week, we show you how. A fresh idea, a clever tweak you locked in. You built this week. Hey, everyone,Sam Nadler: and welcome to Built This Week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder here at Rise Labs, and each and every week, I'm joined by my friend, business partner, and cohost, Metzner. How you doing today, Jordan?Jordan Metzner: Hey, Sam. How's it going? Happy to be back. Another fun episode. Lots going on in the world of AI.More new models out this week, like almost every other week that we meet. But, yeah, even more excited for, this week's guest.Sam Nadler: Thanks, Jordan. And with that, I'm super excited to introduce our guest Chad. But before I do, don't forget to like and subscribe. We've just crossed over 60,000 subscribers on YouTube. Please hit the little bell to get notified.And with that, Chad Reynolds, the CEO of Verve Labs is here with us. Chad, if you don't mind, give us a high level intro of who you are, and then we'll dive into your company and and this amazing solution you're building.Song: Yeah. It's great to meet you both.Chad Reynolds: So I'm a... I I would say, like, a designer by trade who fell in love with consumer insights and consumer science. And Purvey Labs is kind of a lifelong journey to bring kind of the consumer to life in essentially AI worlds that help you create... Essentially create with everybody in the world all at the same time. I've I've been collaborating with enterprise brands, retail brands, consumer brands, health care brands over the last two decades.And, this is kind of a manifestation of all the pain points that I'm sure you guys have experienced as well when you're trying to build ideas, and include different types of stakeholders in the decision making process. So that's essentially what we're building right now.Sam Nadler: Very cool. So my first question is, and and I hope you can explain to me via ELI five, like super simple. The the the unique kind of approach Verve is taking is that you're using kind of real consumer videos, real reactions to different products, and that's how you're training your models. Is that accurate?Chad Reynolds: That's accurate. Yeah. So if you think of language models are your reflection of all the data collected, kind of like the scraped Internet and all the the text and and video from the world, you know, those are how you and I might communicate through a blog or how we write things. But that's not necessarily the best representation of who you and I are. And so we've developed a human network of real people, and we continuously interview them through video, you know, around all different facets of their life.So think of almost like the world's largest, like, panel of of people where we're continuously interviewing through video, and that feeds what we call people models, which get into all the different dimensions of how you and I behave, how we see different things, how we feel different things. And it's very much so focused on the emotion side of how we make decisions. So less important of, like, predicting the next word. We're trying to look at human behavior.Sam Nadler: Got it. Could you run me through, like, an illustrative example? Maybe it's something like, I'm, you know, I'm one of your respondents. And maybe you're interviewing, like, I don't know. How do you feel about planning your next family vacation?Or is that is that like the type of question? Or how do you feel about, you know, if you were to get fired or something like that? I don't know. I'm just trying to like put a more concrete example for my own understanding.Chad Reynolds: Absolutely. So for like members that are part of our kind of our our member group, we're launching questions to your phone. You know, so they essentially pop up as video questions, and it could be around different dimensions of, you know, what's happening without, know, through your day, through your week, through your your month, your year. It could be big transformative moments like you just started on GLP ones and you're trying to figure out, you know, what this kind of new reality looks like. Or you just went through a breakup or you're, like, seeking out new types of music.It's around, like, all different types of categories that we we help support. But the questions are really focused less about, you know, do you like product a versus product b? That already exists on the Internet. What I really care is, like, who you are as a person. How do you make decisions?How do you think about things? And so all of that really valuable data goes into our models to help help develop these different facets around, you know, what does it mean to be this age? What does it mean to be, you know, using GLP ones? What does it mean to be going through a breakup? So that's really what we're focused on by having, like, real people participate in the kind of model making process.And so, you know, the use case for the customer, you there's a really, really cool use case we we put out around Magnum ice cream. So Magnum ice cream, you know, obviously, you're familiar with them and if you go into the freezer section and buy ice cream. And so they were looking to create a new a new product around a different kind of occasion. And so what we did was build a population of agents who were, we called healthy hedonists. Right?So these are people who were looking for, you know, looking to escape through, like, music and these different kinds of experiences, but also were working out the next morning. You know, they were being healthy, making all these different choices. And so we would interview, you know, populations of them and then transform them into agents. And then the brand could then collaborate with them and, you know, go with brainstorm ideas or test new product concepts or ways that they could talk about it from, like, the creative they would make to, you know, the words that they they would use around the benefits that they could deliver.Sam Nadler: Super interesting. So if I'm understanding this correctly, a company maybe has, a a type of person that they wanna engage with or market to or sell to. You go out and interview who you may think fits that profile. You get tons of data. You build these agents, and then the company can leverage, you know, those agents to like further refine their product offering their service to get kind of how this cohort would interact with whatever they're trying to build.Is that Absolutely.Chad Reynolds: Absolutely. Yeah. Yeah. I I... I'm I'm sure you guys have both been in the...In this kind of in this boat where you're... Like, as a start up, you're building you're building something for people. Right? And your ICP is not necessarily always accessible. Right?You can only ask so many questions. And so, you know, we we kind of fell into this trap where we would ask, you know, our AI tools to role play. Like, pretend that you're my ICP and pretend you're these things. And and time and time again, you can kinda see that the responses you get back are essentially the average. Right?So kind of the top of the bell curve. If you ask it to pick a random number, it's gonna pick seven. You know, it's like it's giving you the most probable thing. But you and I are different people, and also we make very erratic sometimes decisions or contradictory decisions. We don't always like behave how you think we're gonna behave.And so what we're really doing is, you know, making, you know, these ICPs or your customer segments more accessible. So now you can ask all different kinds of questions. You know, you can go like 200 questions deep. Right? You're not gonna make them tired.You can show hundreds of examples. And so it just gives, you know, you and I superpowers where we wish we could collaborate with our ICPs to see what is valuable and why is it valuable. And let me test this pricing model with you. And so, you know, you only get kind of one shot in the real world. And then in these worlds that we're building, you know, you have hundreds of iterations you can go through to kind of perfect, what you wanna put out.Jordan Metzner: Okay. So pretty awesome stuff. And I think, obviously, like, the edge of product innovation. I mean, you talked about ice cream, but, obviously, like, all types of product innovation. You know, usually, someone comes up with an idea, and then you...Like, to your point, they build it, and then they go to the market, and then they see if people buy it. Right? And, you know, Sam and I talk all the time about things like smoke testing or market research, obviously, like, but, this is, like, kinda simulated market research. So, talk to us a little bit more about kind of, you know, you know, for good or for bad, AI usually says, like, you're doing a good job and, you know, kind of is a pretty positive thing. Yeah.Yeah. Yeah. And never stops. Hey. I have a good idea.Oh, that's a great idea. Right? So talk to us through a little bit about kind of, you know, how how these, agentic workflows work to kinda give, like, realistic feedback and and comparison style feedback in order to kind of find what kind of cream rises to the top. And then, you know, maybe to your point about, you know, Magnum or maybe another example in which kind of... Then they've been able to implement that in the real world and find the success to prove, like, kind of the agentic models help them help them get there.Chad Reynolds: Absolutely. Yeah. So, you know, as you mentioned, like, LLMs and kind of the agent around them are designed to kind of be your your biggest cheerleader, and they're they're designed to do that. So kind of you're stuck in this, like, echo chamber. So we had to build people models, which essentially try to remove all of that...All of those positive reinforcements, you know, from that. And so as we started to build out all these different facets, they're really focused on combining psychological models, psychosocial models, social dynamics, perception, and to create this really unique model. And so what the user experience looks like is, you know, you kind of imagine coming into more of like a social network type platform where you can, you know, pick different populations you wanna share your ideas with, and then you can run simulations. So think of putting, you know, maybe you have two different pricing pages that you wanna get feedback on where you can upload both those pricing pages and they can sit on a a canvas. And then the population of agents can come in and start watching or reviewing each one of them and giving you feedback on which one they would pick.And more specifically, more importantly, I should say, is why they picked it. And so a a lot of what kind of the language models are doing is giving you the the kind of top of the bell curve. What we've really worked hard to do is to give you the edges because all the good lives in the edges. We just don't ever really get a chance to hear it. And so, you know, from a a platform experience, companies are...Our customers are building essentially their populations that are reflection of who their customer segments are. You know, and these also might be people who are really hard to recruit. Right? So for other companies like, a JPMorgan, it might be really difficult to get a hold of high net worth individuals, to sit down for an interview and talk to me about, like, all these different features you might be looking for. So where there's niche audiences or hard to reach audiences, this really excels, where you can essentially bring them to life, as these population of agents.Jordan Metzner: Okay. That's awesome. And then what... I guess, like, times is an interesting variable here. Like, how long does this take?Like, is this something that historically you would, like, maybe hire a market research company, procure people, then they would do this research, etcetera, etcetera? Does this take, like, overnight or over months? Or, like, how does the compute work to kind of find the answers or bubble them up or, you know, whatever what... Whatever the mechanism is to kind of, like, find truth, I guess?Chad Reynolds: Yeah. So we have preexisting populations just just like real people. We're continuously interviewing the world and then feeding the models and feeding the populations. Right? So it's just kind of like continuously training these populations.You can then, like, recruit just like you might for real people. You hear all the different dimensions that we're looking for to kinda create matches. And then you can show your movie trailer to them or you can, you know, test these new packaging design. So it happens instantaneously for you to do that. In the background, you know, the...Our team's working to continuously train models. So similar to, you know, some of the other frontier labs where they're continuously training models, we're doing the same. It's just our dataset we're working with are real people, and our ambition essentially is to build, you know, the largest computational engine of human behavior. And so, you know, our our ambition here is to really, you know, make... Reduce the cost of asking.Right? So that, you know, you and I, as we're building ideas, we don't we don't really benefit from someone to saying, hey, this is really cool. This is really good. It's really giving us like pointed feedback into into something we maybe hadn't heard before or a different way to look at the problem. And that's really how we improve and and iterate in a more successful way.Jordan Metzner: Okay. Pretty awesome. How do, how do folks find out about you, get in contact with you, learn more kind of... And, you know, who would... Who's kind of optimal, you know, kind of customers that you think would be great?Chad Reynolds: Yeah. So right now, we're partnering with companies within CPG retail brands. I was thinking about the athlete... Athletic shoe brands that you might wear, those types of customers. We've also expanded into hospitality, media and entertainment.So you can test, like, creative and things like that. And then also in life science where we're building patient populations and, you know, HCP populations. If you go to our site, vervet.com, you can sign up to start your own world. And then we help construct kind of the population you want to be reflected in that world.Jordan Metzner: That's awesome. Rise Labs builds production grade AI for the world's most ambitious companies.Song: Supporting engineers.Jordan Metzner: Data scientists. AI architects.Chad Reynolds: All working togetherSong: on your problem.Jordan Metzner: While others talk about AI. RiceLabs is already implementing it. Alright, Sam. Any more questions? I I think it's super cool stuff and, yeah, I don't know.I mean, hopefully, we'll get some some new cool food objects out of it.Sam Nadler: Yeah. My last question, then we'll move on just to a quick news story is, is it is it also to some degree an income earning opportunity? Like, do your survey respondents make money? Are there... Is there opportunities for people to to join and become members?Chad Reynolds: Absolutely. Yeah. So you can sign up to become a member, and then we launch these video surveys to you that we compensate you for. So, you know, our our goal in doing this was to create essentially a consented pipeline for unique data training. And so because we hide all of the PII, it gives you a way to contribute who you are.And for us to all build models that are a better reflection of you and I compared to how kind of language is scrapes from us. You know, how I write a blog isn't necessarily who I am as a person. And so, you know, I think the I think the worthwhile endeavor we're on is to build... You know, we understand, like, the intelligence or super intelligence that's coming through and what these, you know, the capabilities of language models are. But, you know, when you can start to pair that with, more of the human understanding or human behavior, that's really where you're gonna have some breakthroughs.Amazing.Sam Nadler: Alright, Chad. Because you were a former designer, I wanna talk about this article. So Adobe brings AI agents and creative tools to Photoshop, Premiere, Firefly, etcetera. But I guess my question to both of you is, you know, with these... Obviously, you know, I'm sure that that some of these tools help designers move faster, but is Adobe gonna have the same relevance three, five, six years out when a lot of these things you can do, you know, directly in Cloud Code or in Codecs and, you know, kind of forego using these tools?Chad Reynolds: Yeah. I mean, you're obviously witness you're obviously witnessing, like, the battle for not only our attention, but the surfaces that we, you know, that we make things. And so, you know, whether it's putting agents inside of Adobe or it's Claude building, essentially building image... Imagery tools or meeting tools or... It's really a battle for the surfaces that you and I make things on.And so it's gonna be really interesting to see where people... Like, you and I choose to go first when we're gonna create something. Jordan, I know you've beenSam Nadler: a big fan of editing videos in Cloud Code. What are your thoughts?Jordan Metzner: Yeah. A Cloud Code... You know, it's funny with, coding. You know, we went to agents and we kind of removed the visualization of looking at the code anymore. And now we're editing with Club Code, and we've removed the visualization of editing anymore.So it seems like abstraction is the future. I don't I don't know. I think Adobe's in a tough spot. You know? I mean, as you know, Sam, we have clients that use Adobe heavily in highly technical areas.And I'm sure there's still a market for that at the edge, edge, edge. But, you know, like we've seen with, you know, the Canva news and the layoffs happening there, like, just users can go on ChatGPT and make a birthday invitation by telling it what it wants and it looks great and it... You know, it's a style they want and they don't need any tools. Right? And so, you know, as that comes closer and closer to the consumer, right, imagine that on your Apple phone or on your Android device and as it gets closer and closer to the user, I think it's it's definitely gonna hurt them.Know, same with photo editing and same with, like, their entire stack. I don't know where they're gonna be able to continue to maintain the edge. Maybe, like, again, maybe at the niche niche niche of experts, but, you know, we probably will see new innovation there as well. So, I definitely think AI is threatening them more than we've ever seen it before. And, I bet they're scared.So I don't know. That's my thought.Sam Nadler: Cool. Well, with that guys, Chad, thank you for sharing your your company with us. And if anyone's... Please, you know, go to vervet.com. And, yeah, great episode everyone.Thank you.Chad Reynolds: Yeah. Thanks guys. Really appreciate it.Jordan Metzner: Awesome. Thanks so much. Alright. Thanks, Sam. See you soon.Chad Reynolds: Bye, everyone.
