Episode 18
Putting An End To AI Slop
Published · 1 hr 8 min
Thais Castello Branco left the AI search company Exa, raised $18.5M, and built Taste Labs -- the "taste layer for AI" -- solo, with one mission: end AI slop. She and Julian go deep on what taste actually is (and why it isn't personalization), why everything AI makes is starting to look the same, how you turn something as subjective as "great design" into data a model can learn, and the honest case for going solo.
What Thais talks about
Pick a question to ask it of this episode. Answers come from the transcript.
Transcript
Transcribed automatically, so names and words may be wrong. Each time opens the video at that moment.
0:00Thais Castello Branco: We're going to get used to slop and start thinking as normal, like as a society. And I don't want that to happen. There's so much research going into how to make models great for coding. Why is there not that much research going into how to make them understand and design? I considered actually having a co-founder in the beginning, but then it felt forced. I don't want to not start just because I haven't found this perfect, magical co-founder.
0:20Julian Weisser: What is the bare case for being a solo founder?
0:23Thais Castello Branco: You should only be a solo founder if you have incredibly high pain tolerance because it is more lonely.
0:29Julian Weisser: What is the case for solo founder?
0:31Thais Castello Branco: I know how 150% in it I am, and so it gives me a lot of trust in myself. If I sold taste tomorrow, I would go and start it again. I would be very sad to live in a world where there is nothing new that pops up, that is everlasting. I'll keep trying until we succeed.
0:47Julian Weisser: What was the actual sort of moment that you realized you wanted to start a company? And then what was sort of the co-founder consideration at that point?
0:55Thais Castello Branco: So I was part of the founding team. I had been in the startup space for a while, and I had actually started a company. In Brazil with a friend of mine, and I had joined up as part of the early team. And so I always like was a founder in terms of like how I thought and how I operated, and just being very all in as a person. And I had this idea for for taste of became obsessed with how do we give models more like nuance and make them great at subjective domains? Kind of around the time that Exo was raising the series B, and so I saw that through. I wanted to tie a bow on it, and then as soon as that wrapped, I decided to officially leave in and start something new. But it went from the like idea to it becoming real very quickly. And so I think I kind of went on a bunch of co-founder dates, even in the beginning. Um, but I just kept going. I was like, okay, who do I need to hire first? Who I need to bring on resources. I fundraise very quickly that precede within like a couple weeks of leaving. And so it ended up just like going and going quickly. And then I suddenly looked and I was like, okay, great, we've started without a co-founder and now I can keep going. But it was very it was very like natural. That transition, I would say from, from AXA to taste.
2:08Julian Weisser: What would you say is the defining characteristic of taste.
2:13Thais Castello Branco: The company or the or the concept?
2:15Julian Weisser: Let's start with the concept of taste. I know this is like a.
2:18Thais Castello Branco: Philosophical.
2:19Julian Weisser: Philosophical thing, but I think it's like I think that everybody has slightly different definitions or slightly different ways of viewing it. And that's partially probably, in its own way, a form of taste, right? How you actually define what taste is. But I'm curious kind of how you thought about it early on, especially as you were thinking about building a business around trying to, you know, end the slop or, you know, give AI sort of the ability to have taste. But I'd love to just hear kind of how you were thinking about it in the early days, and maybe also how that's evolved as you've gone on this journey.
2:52Thais Castello Branco: By the way, it is very philosophical, and I think it should be. And it's funny because I think a lot of the first few weeks, I honestly just spent so much time writing and researching and talking to PhDs and philosophers and experts to to even arrive at this definition. And I think there are two things that I have come to the conclusion that I really deeply believe in. One is that I think, I think people conflate taste and personalization. And that's I think a first important distinction to have is I think personalization is way more like what is a one individual's personal preference, but taste is almost more this like measurement of quality is like in a domain where there's not a single correct answer. How do you define what a great looks like? How do you define what appropriate looks like? Who are the people that can have that judgment? And I think taste is that judgment of like defining what quality really is. And there is an element that is a bit requires like shared validity. Like not everyone has to agree, but I would say other people that are experts have to in some way, shape or form agree. And so I think it's this bar for quality. And I think in terms of how it's developed or how it comes about, I think it's really a skill. It's something that requires a lot of work, like years of practice and craftsmanship and like putting in the reps to and testing and doing things yourself and seeing what works, what doesn't work. Um, which is why it's so rare. Like, I think it's a very rare skill. Um, and ultimately, I think it's developed through this process of, like, exposure. I do see a lot of things of pattern recognition, of starting to notice trends or patterns or things that repeat of point of view, of being like, okay, these are the things that I like or want to do. And down selection of like, these are the things I don't like or that I don't want to do. And there's of course people that have more of a natural maybe talent is the right word for it, and then maybe go through that cycle faster and then develop that intuition faster. But I think ultimately it is this like skill that you have to hone. Um, and so I think, yeah, it's people treat it as this very mysterious thing and throw around that word. Uh, but I think it takes just like a lot of work to, to really build it up.
4:55Julian Weisser: There's this podcaster or radio host named IRA Glass. He created this incredible publication broadcast called This American Life. You know, for many years it was the top podcast sort of outside of business and technology podcasts. And his whole thing is there's this idea of the gap, and the gap is when you are getting started with something. Usually in his case, it was an artistic endeavor. You know, he was trying to create a radio show, um, podcast. And he said that when you're getting started with something, he's creative acts. You can see the gap between where you are and what you think is actually good. And that can be very demoralizing. Um, uh, this idea that how can you actually bridge that gap? And his whole premise is that it's actually good to know that there's a gap and then trying to figure out why the gap exists or what, what the things are that are actually preventing you from getting there right now. And I'm curious. Um, one it seems like everybody identifies there's a gap between what AI is doing and what actually is good. Um, in some ways, like people are, people are prompting AI. So in some ways they are giving AI
6:15Julian Weisser: the directive and partially it's AI screwing up. But maybe it's also human, right? Not giving the correct prompt. Um, but I'm curious how you think about this gap that exists maybe in people's ability to express what they want, but also exists in AI's ability to actually deliver something that maybe most people would consider tasteful.
6:38Thais Castello Branco: That's a very good question, I think. So one of the things that I think people are really underestimating that has shifted is before the people that were creating tended to more be experts like talking about design specifically. But even when we're talking about writing like who had the the ability to, let's say, publish, right? It was a very select group of people. And the internet maybe was the first thing that kind of spread that ability of people to create and post and publish and similar with design. But I think AI was kind of this forcing function of just multiplying people's ability to just generate. And I'm not even going to use the word creative. I'm gonna use the word generate very, uh, on purpose. And you have suddenly this like, ginormous group of people that can press a button and have a slide deck, a post, a website show up. And especially for those people, I think a the bar for tastes might be lower, right, because they're not necessarily experts. And then the bar of execution might be even lower than that, in the sense that they were never meant to be the expert. And it doesn't mean that we should kind of abandon them in this journey and just let them not understand how to hone their taste, or how to express themselves or how to know what they want. But we have to build for that world, right? Like. And I think we aren't we haven't really built for that world yet. I think there's a lot being built for the expert, which is this person that really cares about their craft and this that is very aware of even the gap in their skill to taste and that wants to own that. Um, but we need to transfer a little bit of this. Maybe ease is the right word, so that the average person can also raise that bar and bridge that gap, because I think it's unrealistic to expect everyone in the world to suddenly become tastemakers in every domain. Right? Like maybe there's a few ones that are going to pick to really go deep in hone and, and kind of be able to notice that discrepancy between what you're able to output it and the peak. Um, but you're not going to do this for every single domain, right? You might have amazing taste in architecture or design and then not have time to hone your taste in code. Or you might have amazing taste in code or not have time to hone your taste and design. And so how do we just hide in that bar or even become aware that that bar should be higher? So I think that's one important piece. But I do really like the idea of AI being a tool that both can help people to understand their taste better, understand the bar of quality better, but actually help them push both up instead of it being something that's pulling it down. And I think right now, because you're
9:09Thais Castello Branco: producing this like average slop output, it is kind of bringing down the the bar of perception. Right. And again, getting philosophical a bit, but I think it's true. I think it's very dangerous if we don't fix it. I think humans are super adaptable. I think we're going to get used to slop and start thinking it's normal, like as a society. And I don't want that to happen. I think I want people to be very aware that that's like a low bar, and that the bar should be much higher than that.
9:34Julian Weisser: So machines learn very differently than humans. Um, you know, Hunter S. Thompson famously, uh, would type out The Great Gatsby and Hemingway and all these different works. Um, Picasso, if you look at his very first painting, Self Portrait, it doesn't look like anything that would be associated with Picasso. And then you look at his last self-portrait and it looks uniquely identifiable to Picasso. Um, and there's this process of sort of imitation and like, learning the forms sometimes as a way to eventually unlock the ability to express yourself. Um, how do you juxtapose that with, like, how machines learn?
10:13Thais Castello Branco: I think machines, obviously, especially today, have less of this continual process of learning that humans do. Right. Continual learning is all the rage right now, and I think it's for a reason. It's us as humans. Every time that you get a new input of like, oh, I tested this thing, oh, this is interesting. And then you get carried away in that direction and you explore it more and you're, you're picking up these patterns and learning, uh, it's more organic. It's more continual. But with machines, at least right now it's less continual, right? It's more happening in specific moments in pre-training. They're absorbing all this information, kind of learning, those patterns of probability. In post training, maybe they're learning, attaching more the concept of like what is good and bad. And so it's just slightly more maybe artificial and concentrated right now and happening in batches. But I do think there is a similar process of like pattern recognition and of slowly associating like
11:05Thais Castello Branco: what is good and bad in specific situations. I think what's much harder about a domain that is subjective and that it's something that humans are very good at. Is this like nuance, right. It's this, um, it's the fact that it depends, like it depends on the situation. Something can be really good in a situation, really bad in another. And that nuance is very hard, I think, for machines to learn, but very easy for humans to, like, very naturally pick up on.
11:30Julian Weisser: The other thing that kind of as I was thinking about what we might discuss, one thing that came to mind was the fact that when humans have some sort of quirk. Uh, they it's signature to them and but each each human who's creating a work of art or something like that, that becomes their signature. When AI does something, because AI is being used by a certain model or something is being used by billions of people, tens of billions and hundreds of millions of people. You have this thing that happens, which is it's the quirk isn't like a nice signature. It's kind of like an unfortunate thing, right? So, um, like.
12:06Thais Castello Branco: Um, dash, dash the purple gradients.
12:09Julian Weisser: That's so real. Like, that's the real thing. You know, they have all of these different sort of towels. Yeah. And where if it was an individual versus something that was used by so many people, it would be considered maybe a unique thing. Yeah. But because it's so universally applied, it creates this unpleasant kind of gross, kind of off putting situation. And I think that what my understanding of what you're trying to do is to say that we don't want to have everything, just have the exact same veneer. We want to try and we want to try and make it so that AI could actually do interesting, interesting things and be and be more unique across, you know, across the board. Can you share a little bit more about the philosophy around that kind of how you think about the problems with AI today? And, you know, people describe it as slop? I wouldn't necessarily describe the Em dash thing as a, as a, as a, as slop, but I'm curious just how you think about the problem in its current form.
13:09Thais Castello Branco: Yeah, I think the one point that you mentioned that, I think is kind of at the root cause of it is this is the volume, right? Is to your point, if you have one person that has a specific quirk or a way they do things, it becomes a signature. But suddenly you have millions of people creating with the same tool for totally different use cases. But the output looks the same, which doesn't really make sense, right? It's like it's it's for completely different purposes or styles or types of people. And that's where I think that feeling of slop comes from, because it feels more thoughtless. It doesn't feel purpose built for that specific situation. So you end up with this super high volume and repetition of the same pattern, but that also doesn't feel right for that specific situation. Maybe if it did, you wouldn't notice or you wouldn't bother you as much, but because you're seeing it across multiple things, even when it's not appropriate, um, then I think it creates more of that feeling of slop. And so I think you're totally right. I think part of the goal is how do you create more diversity and understand, okay, there's actually a whole range of situations of types of people or types of styles that warrant different responses and different definitions of what is great. How do we allow both the models, but also the harnesses on top of it to lead to those outcomes and to force that distribution? But also this problem of like fit and appropriateness, I think, is equally as important because again, if you create randomness like you could theoretically just up the temperature of a model and make it produce more random things that are more diverse. But that wouldn't necessarily be a good outcome either, because you want it to feel intentional and it should feel purpose built for that specific situation. So this like intent mapping of users, I think is super important. And it's something we're doing a ton of research on. And this almost conditional relationship of taste of it has to feel like something is not good or great or bad in isolation of context. It has to be in context that you judge that.
15:03Julian Weisser: If you're enjoying this conversation, then perhaps you're.
15:06Starting your own company, and if so, you should apply for the Solo Founders Program. I spent the last seven years working with startups and founders who are figuring things out, and I'll tell you that the ones that make the most progress are the ones who make sure that they have relationships that could help them go the distance. The Solo Founder's program is about bringing ten solo founders together so that they can build solo together, versus building solo level. Then I'll work with you, who will invest $100,000, and it will make sure that you make the most progress that you possibly can in three months. If that sounds interesting, you should learn more and apply it at Solo Founders Program.
15:41Julian Weisser: Do you think that I'm calling it Taste Labs? Did you realize it was going to sort of draw so much, uh, interest? I guess as a as a word.
15:50Thais Castello Branco: Don't get me wrong, I thought there would be some discussion around it. I did not know it was going to be this level of, uh, debate around the name, but it's really funny because I registered the company with this name a year ago before the word popped off on Twitter. So when it started popping off on Twitter, I was like, damn. Like I,
16:11Thais Castello Branco: I didn't want the word to become so ubiquitous. Truly, because I do think it's a very good name, especially for what we are trying to represent. Right? Which is this, this bar of quality, this diversity kind of giving the power back to people, to kind of hiring that bar and create something that's more personalized to them. Um, and I didn't I didn't want to change it just because of it. And so I was a bit stubborn, I would say, of sticking with it. I knew there would be some. I actually debated it even of
16:37Thais Castello Branco: is it too used now? But I think at the end of the day the debate was more around the intent, right, of like, are you trying to say that you're going to create one taste and determine what taste should be like and like, no, that's not the thing we're trying to do at all. It's I think it's this raising the bar of quality and increasing diversity and allowing people to both understand their tastes and apply that in whatever way that they want in a much better way than we're being able to use now. But the purpose is making it a tool for creativity, not collapsing its creativity even more on something that we're dictating, like what is correct?
17:14Julian Weisser: You know, I think that it's it's funny because I do think that that's probably the big misconception about the mission and when when in reality, I think what you're actually trying to do is you're trying to solve for the fact that everything feels like monoculture and everything and, um, and then people look at it and they say, well, like, does what? How many different colors of, like, hair dye do the team have or something? Or like, do they all wear different clothes? And, and, and in some ways I would, I would almost say that, uh, that the, the weirder the team looked like in some ways, in some ways the, the, the less likely I would believe that they would be able to maybe accomplish the goal, because it's not about taking, um, taking something that's already wacky and making it wackier. It's about taking something that's incredibly constrained right now and opening it up. Yeah. Um, and I also don't think that you're trying to impart just the team's taste. No, you're trying you're trying to do so, so much more of that. So can you share a little bit more about. Because I think that this is probably the, the, the reason that it went so well, it's so big is just like these people. Like, wow, how should they be? Sort of the people who determine what taste is for the world, right? Which which, by the way, if you were if you were trying to determine your sort of put your taste on the world, I think that there's a reason for people to be like, that's kind of crazy, but that's not what you're doing. You're trying to you're trying to essentially pull from this really sort of wide range of expertise and interests and probably various perspectives on what taste is. So how does one even begin to, to actually, to actually go after this sort of daunting and wide ranging goal of how do we make it so that AI can have more diversity in terms of how it expresses, like people's interests or what people want to see?
19:10Thais Castello Branco: So by the way, it's really funny on the on the team piece because I think there was kind of no winning in the sense that if we also if we took a very opinionated approach, even how we dressed or the photos or whatever it was, people would be like, oh, so that's the one taste we're trying to impose. I don't like that one. And so I think we want it to very purposely be neutral because it's not about us. To your point, it's like it's not about my I don't want it.
19:33Open to you. It opened you up for a lot of flies.
19:35Thais Castello Branco: Yeah I know.
19:36Regardless, either way.
19:37Thais Castello Branco: No either way. But I think like the the whole point is it's not about us. We're not trying to be the ones determining this. We're just trying to understand, like someone has to study this. Almost like that's why we called it a lab, because it's truly like it requires research. Like, there's so much research going into how to make models great for coding. Why is there not that much research going into how to make them write incredibly well, or how to make them understand this nuance, or have emotional intelligence or understand design? And those are the things that will lead AI to actually feel right and to be this thing that actually, you know, sticks with people and is a good experience, not just something that is intelligent. Right? Um, and so I think the whole point in terms of this diversity, right, and how to force this diversity. Maybe talking a bit more in practical terms too, I think, is how do you take something that's so subjective? Like if I said to you, like what is great design? Very hard to answer that question. Right. But once you start breaking down the problem into smaller pieces of that problem, there are certain elements of it that look more objective, others that are still more subjective. But let's say for like, okay, maybe there's this foundation thing about models where they have to be very good at instruction, following and reasoning and spatial understanding and vision, and those things are required for you to be able to achieve quality. Great. Let's go fix those. Okay, we fix those. Then maybe we have to fix these fundamental capabilities of design like contrast, typography, color, white space. Um, maybe hard for some people that have different styles to agree on something, but they might agree on like, yeah, this thing is aligned, right? And so you can turn a problem that feels very fuzzy into more objective. The closer you break it down to like a specific subcomponent of it, and then you get to higher order problems like, okay, maybe now we talk about style and creativity or diversity or personalization. And that's more where you want to understand, okay, there's actually a multitude of preferences and that is how the world works. That is. So these models or these technologies should reflect how the world actually operates. And there is going to be always a diversity of thought and opinion and of style preference. And so that's where we want to purposely show that diversity and understand, okay, maybe experts tend to agree more on the things that tend to be closer to objective and disagree more on things that, um, tend to be more about, for example, esthetic preferences. But that's by design, right? That's not a bug. It's a feature. Like it should be that way, because in the real world it's that way. So how do we transform that data into something that is signal and not noise, so that it doesn't collapse when we're training these models? And, um, you don't end up with just the average of everyone's taste, right? It's much better for us to end up in a world where you understand what does great look like for something like maximalism and for minimalism, but you're not imposing judgment of which one is better between the two, you're just allowing both to be great, and then the end user, depending on their preference, should should see should see the end result of of that one.
22:34Julian Weisser: Artistic styles. Um, you know, they emerge usually from something that I think Brian Eno calls seniors. Right. Which is, which is sort of the genius of the scene versus versus the genius of the individual. And, um, I'm curious how you think about both
22:56Julian Weisser: capturing and sort of in sort of understanding scenes or styles that have been created already. So, so essentially the products of Asinius, um, that's everything you know, from, from visual to, to music genres, right? Yeah. Um, to, uh, to actually creating new like the new the new, uh.
23:20Like style.
23:21Julian Weisser: The new. Yeah. How does that. And are you focused on, on one sort of understanding the past more so than sort of understanding how to create new.
23:32Thais Castello Branco: So right now I would say we're more focused on this like general quality and capability and the past and present maybe is the way to think about it. But I think what you're touching on is something very interesting about the property of taste itself, which is it naturally changes over time, and it has this like social property to it, because when something becomes too ubiquitous, the tastemakers that were at that forefront of that trend kind of don't want it anymore. Right. And you see this in something like fashion all the time, like something is cool when it's niche and then once it starts becoming too frequent, it kind of loses its power. You reach more of that mass adoption, and then a new trend emerges. And so it's similar in design or in other areas where you kind of have these waves of trends and like, I don't know, glass morphism was all the rage a few years ago. And now, uh, you you might see different trends. And so this social nature to taste is very natural and normal. And I would say we are trying to, I think, define and understand very well what are all these maybe styles or trends that have already been very predefined, and how do we identify them and classify them and understand what is a great execution of this? But I think one of the things we will also, hopefully in the future get to study more is how do these change meaning? For example, one of the things that we're doing is almost like watching, uh, the internet's like history of design and kind of where it's going now, meaning how do we look at all the patterns of how sites were and how they are now? What has changed? When has a certain style, for example, fallen out of fashion? What when? What is starting to emerge? And hopefully that becomes an indicator in the future of this like trend prediction. But I would say right now we've been more focused on the kind of Classification problem around it and like defining what a great execution within those styles like were.
25:17Julian Weisser: Are you mostly focusing on sort of websites and apps to start and sort of how does that how does one actually do that? So you have this team, I think you said your founding designer, you she she is working with a lot of other designers as well. And you have this community. What does that actually look like.
25:34Thais Castello Branco: So we're actually covering all of visual design. So anything visual created by an LLM and honestly even some like image and video as well. So everything more visual slides, images, uh, SVGs, 3D generative art websites, everything. But I think ultimately the goal is like, how do we develop a framework that then we can go and tackle other domains, even like writing or anything really? That is subjective. But right now, in terms of the community, we we selected this group of like tastemakers, right. And so we have this group of like 800 people that have different specialties and they're great at different styles or types of media, and depending on the project and what the objective of that project is, we kind of pull tastemakers from from different corners that both we want them individually to be very strong for that specific goal. Like, let's say we want to make a model better at website design. Yes, they have to be very good at that. But we also want to be careful to not pull too many people with the same type of background or style. Right. Like we want people that have that diversity so that the data is like representative of that diversity as well. And so we we work with them from like defining almost like the rubrics of how to think about that specific domain. Like how would you grade a website for what is good about it. Like what are the criteria that matter most? What is the difference between something that is bad versus okay versus great? Um, so almost this logic of like grading of things, but also like the, the reasoning behind it. Like how do you explain why something is good? How do you explain why it's bad? Uh, so it's something that it's a very collaborative process, both with the customers in terms of like the researchers at the labs and with the creative community and us as this bridge.
27:13Are the people.
27:13Julian Weisser: Who make great things always the best at explaining why they're great.
27:19Thais Castello Branco: Not necessarily. And we've started realizing this too, is, um, there's different skills that you can be very good at. There's and we have some projects, for example, that are around creating things actually from scratch, almost these like golden sets of amazing like peak of design. And there are people that are amazing at that and that maybe are bad at articulating why they prefer specific things. And so we do see this like role of like the critic being slightly different than the than the maker potentially. And you kind of have to identify people that are good at both, or even maybe teach them a little bit of how to become better at the areas that they might not be so natural in. But I do think to your point, it's different. I think especially designers, um, a lot of them are very visual in terms of how they think. And so some of them do love this, like writing and articulation side. But we've tried to even make it easy and kind of create even methods of collecting data that, for example, are very voice driven instead of writing so that it feels more natural. Right. Designers, for example, are very used to doing design critiques with each other or like feedback sessions. And so how do we make it feel as natural as possible to what the organic process of design would look like?
28:27Julian Weisser: You know, it's interesting because oftentimes when you look at the critic and the the creator and sometimes they're one and the same, but very infrequent, actually, um, the critic usually says things that's very different than what the creator is either thinking about or capable of even articulating. Um, I can I'm sure that this probably existed in the art world. The only thing that I'm familiar with where people who were who are truly started out as critics, that actually became creators themselves, um, was in like French New Wave, uh, cinema. Um, there was this publication called Chi to cinema. Um, and they would do reviews of whatever movies were coming out. They do like very long form sort of analysis, and a bunch of those people ended up becoming some of the most important filmmakers in French cinema history. Um, but but oftentimes, I don't think that there's necessarily the thing that you hear from the person who is the director of a film, for instance. It's very different than we hear from the critic. Um, what what do you think is what do you think is driving that? Is it because there's some aspect of the person who's making the thing or in your case, like the AI that is that is generating the thing? Um, it just necessarily has a different perspective by being the originator, um, or the generator, um, versus the person who's receiving the generation output film AI generated image. Um, they, they are sort of their perspective is is necessarily driven by the fact that they're, they're receiving it from a different angle.
30:04Thais Castello Branco: It's a very profound question, because I think it goes back to almost like, is the beauty in the eye of the beholder or the maker, or even like in music? I think this debate happens all the time too, right? Of artists being like, no, I don't want to explain the meaning behind my song, because I want people to visualize this almost for their own life. But so I think there's quite a philosophical element to this, but I do think that the maybe the missing gap is intent versus interpretation is um, and by the way, I think this is actually data that's also very interesting to like, collect and understand, because part of what I think AI is bad at is actually understanding intent in this like subtlety and intent. So when you are as the prompter, let's say you're not the maker because the AI maybe is the maker, but you're the solicitor. Let's say.
30:46Julian Weisser: Um, you're commissioner.
30:48Thais Castello Branco: Commissioner? Yeah, you're the commissioner. Um, you like explaining? Your intent is very hard. And I think the people that right now I'm seeing the tend to be very good critics also tend to be very good commissioners in the sense of like explaining what they would want or what they would want to see. Um, and so it is I do think there's like this distinction in, uh, maybe like skill set of, like what? Where someone focuses. And I think a lot of the makers are very good at like taking these nuggets of intent, but then also making it their own. Right. It's not just purely about following what, like to a t what that person is requesting. Maybe they're even understanding their intent better than themselves and coming up with something that's more unique, more creative, uh, and with their own spin on it. And I think what's tricky when you are talking about like AI versus if you're talking about like a human creator is there's this question of like, how do you want actually guy to behave? And in other domains you like instruction. Following is something that's super important, right? Like you don't want to ask your agent to code you something, and then it goes and puts its own creative spin on and it does. Something else like that would be very undesirable behavior for most domains. Uh, but maybe for something subjective, there is a certain proportion of people that would want that experience, or a certain subset of things where you would want that, like own interpretation. But I think it's a tricky question to answer to figure out, like, where does this actually fit in? Like a productized sense?
32:14Julian Weisser: It's funny you say that is, I agree with you. I'm thinking about it from the perspective of when most people are prompting. I imagine the prompts are extremely vague. Um, they're requiring the LM to do a lot of reasoning about what does this person actually want? This is what they said. What do they actually want? And I'm curious, your take on specificity when it comes to I think this is like a fun, fun way of putting it, like the commissioning of things from AI. Like you commission an agent to build you a website, or you commission an agent to generate an image similar to how you might commission, you know, a dev agency to build your website in the past. Right? Um, and generally speaking, I think that people are bad at commissioning things. Uh, which is which is just another way for bad at prompting things like, I think that's its own skill. Um, there's a, there's like a famous example at the, um, I think it's the new George Lucas Museum or, um, or maybe at other museum in LA that just opened recently. There's a, there's a sculpture from Matisse, and it's this. It used to be in this couple's backyard. They were obviously quite well off to be able to, uh, get a commission from him. Um, and they turned down his first piece. Is he. They commissioned it and they didn't like it. And that, to me, feels like they were bad. Did a bad job at commissioning versus versus. He did a bad job at making the sculpture. And I wonder if that's part of the challenge when it comes to, uh, to AI, right. Is is it also is it is it as much the AI's fault? Surely there's some issues with AI not trying to to to downplay that. You and I, I think we both know that. But but I'm curious, like how much of this is, um, isn't an AI skill issue but a better commissioner of the show.
33:55Thais Castello Branco: I think it's both. It goes back a bit to what we're saying about who are the people prompting. Right. And it might be a person that's not an expert in how to and being a designer. And so they might not even have the vocabulary to explain what they want and the stylistic words of what they want or the intent behind what they want. And in I think all the best designers that I've met are people that are actually very good at taking a like less than optimal brief and asking the right questions, or poking at the right things to uncover that intent. And then, of course, being opinionated around it. But I think part of the thing that's not clear right now in the ecosystem is I think there will be a suite of products where the user just wants to one shot something, and once they're very simple prompts to just output something great. And what should the experience feel like there versus others where you do want more of that maybe iterative behavior of maybe seeing a few options, maybe Having the agent ask you questions back. Um, and so I think there's like a whole spectrum from one shotting something to you doing everything manually. Uh, this in between, I think. Can
35:06Thais Castello Branco: we can, I think, create different products, experiences for different profiles of users. And so I agree with you, though, that part of the skill gap is in the commissioner. And so how do we solve for that. How do we become better at either understanding the intent behind that user of or of collecting data to even understand on average, or what is the range of things that people would want to see if they prompt something like, I'll give you an example, but if you say like, oh, make a landing page from my event and make it pop, like, what does that mean? Like that can mean many things. Like maybe that can mean, uh, chunky fonts and saturated colors and high contrast. Maybe it means you want like pop pink type vibe and, uh, understanding what are even, like the range of possibilities that come from that ambiguous prompt I also think is a gap right now. Um, so yeah, lots of.
35:55Julian Weisser: It's really interesting. I mean, I think about this analogy of, you know, where, uh, corrective lenses, but going to an opticians office, they put this thing in front of you and they say left or right, and you sort of choose or 1 or 2 and you choose like which of them. And they essentially help you go through this, this loop of refining sort of what the actual eyeglass prescription should be. And that feels in some ways like something that that AI could benefit from. Um, in terms.
36:26Thais Castello Branco: Of journey.
36:26Julian Weisser: Of the journey of sort of uncovering discovery with the person. I think there are some people, as you're saying, who are kind of like the give me convenience or give me death type people. They just want it immediately. Um, but I think that some people probably are rightly frustrated because they gave a vague prompt, but they, you know, the AI isn't a mind reader yet.
36:46Thais Castello Branco: Yeah, like that's on your brain. Um.
36:48But, um.
36:49Julian Weisser: Okay, that's really interesting. And where do you like. Where do you want to. Where do you want to? I know you're starting with visual. Like all visual things. Like, where do you want to take things over the next year with this?
37:00Thais Castello Branco: So I think the reason, by the way, I think a lot of companies that have explored problems around data or infra tended to start very horizontal and kind of solve a million domains at once. And we very intentionally wanted to, like, go very deep instead of wide. Um, I think to really have this, like, we want to be like the experts in this domain and understand it better than anyone else, how to measure it, how to stir it, how to codify it. And so but I think a lot of what we're learning already and what we still have to learn, like it's not fully solved, to be clear, um, is how do you really create this logic around how to transform something from subjective to objective, right. Like how do you break down this problem into all these subcomponents? How do you create grading for all these things? How do you create this conditional logic? How do you represent this in data? How do you extract information about something fuzzy and transform it into more structured and. Notice I didn't even use the word design when describing this. I think it's something that can be applied to many domains. And so I think let's go very deep on one, really understand it, but then go and help other domains to crack it too. Rather it's creative writing or marketing or I don't know, I love actually the side of also
38:14Thais Castello Branco: like tone and personality. Not just like creative writing necessarily, but like a model is almost like emotional intelligence. I think there's also a lot to unpack there. So yeah, how do we basically replicate this logic across like different domains?
38:26Julian Weisser: We touched on this really briefly when we were getting rolling around, sort of why you ended up starting this, the timing around it, you sort of wrapped up the series B where you're the founder, you're one of the founding early teammates at a great company. Xa um, but can you share a little bit more about kind of your journey into startups in general, because I think it's always interesting to hear people's here, people's experiences. I think you said that you did have a company that you had co-founded in the past. So, so maybe we'll sort of. What was it like when you first realized that you were interested in technology or in startups?
39:03Thais Castello Branco: I so I actually it's funny, before I worked in tech, I worked in CPG in like consumer goods. And I think one of the things I loved about consumer goods, I worked in new product development. So kind of coming up with new things that people would potentially want. Um, and doing a lot of like consumer studies and understanding like how people thought what they needed. And I think I love that process of understanding, like human behavior fundamentally. And what's really interesting in consumer goods is you kind of have to learn how to like productize and sell something that's realistically very commoditized, right? There's a million companies doing very similar things. And then technology is kind of the other side of that spectrum. It's how do you do something that's very unique ideally. And where the challenge is, maybe how you make it interesting, or how you make it simple to understand. And one of the things that I really did not like about the world of, of consumer goods is how slow it was. I loved the speed I think of technology, of iterating, of testing, of getting back feedback from people and iterating on it. So I think that's really what it made me like in practice, fall in love with it. I think I always had obviously like an interest in it and in studying it and reading it, um, even in college, took a bunch of classes. But I think this, like speed was something that really got me, like addicted. It's really fun, like this building and willing things into existence that didn't exist. And then they suddenly exist the next day and you already have feedback from users. It's just like really cool. And at the same time, I think this I worked a lot more in like application layer companies before. Uh, exactly. Because I loved this like human behavior understanding. But funnily enough, when when ChatGPT came out, I became fascinated With AI in general. I started just spending all the free time that I had researching it, and I was actually I had just moved back to Brazil, or I was moving back to Brazil at the time because I was living in New York before, and I was like, I have to go to SF like I have, I have to be in San Francisco. I have to be a part of this, um, like phase of the world. And, and so and I also, I think, simultaneously became obsessed with infra, which is something I had never really been that exposed to before. But of understanding what almost like what needs to be built for this world to become true. Like now that the world is changing and that we're going to have this new technology shift. Uh, what are the tools that need to exist for agents? What is the foundation that needs to understand exists from like an infrastructure perspective? Um, so this like shovels mentality. And Axel was very interesting, I think, because it was really a take on in a world of agents and where agents will be doing things on behalf of people, what tools need to exist for that world, and knowledge and search is obviously a huge piece of that. And I became obsessed with it. And I joined, uh, love the early team, and it was really cool, I think, to be a part of that and seeing how fast it went from no one understanding it to like, wow, this totally makes sense. And similarly, taste realistically is also how are we understanding what needs to be true in a world in which AI is everywhere but where we want it to also feel right? Right? Like, how do these models have to change? But also what tools do these agents need to have to really be great and know how to design on behalf of people, create on behalf of people, understand their, their personal preferences. Um, and so it's not a dissimilar logic, but it goes back to like, how do we build this infrastructure, this foundation, the data that's needed to really power this, this change. Um, so, yeah, it was it was a very kind of natural evolution for me through this world of tech. But I think the spirit of founding and entrepreneurship is It's actually even hard for me to describe where it came from, because I think it was always there. Like, I kind of I was that kid that started companies on the side with, like my friends and with like convince my parents and their friends to like, buy things and would go knock door to door and I would, like, sell cookies and I would sell like, um, I would make all these like magazines, for example, and like, sell them and I would write books. I kind of made this deal with my dad that every book that I wrote, uh, he would give me like 50 bucks or something. And so I think I was always very entrepreneurial and, like, curious. Um, and so I always knew I was going to start something, but I really wanted it to be something that felt like me and that I had a unique perspective as to what to solve. Um, but also something that I could spend ten years more of my life thinking about, like something that I was genuinely very obsessed with and that felt like, didn't feel like work. Like, this is something that I wanted to think about all day, even if I wasn't, like, if I sold taste tomorrow, theoretically, which is not going to happen, but I would go and start it again. And so I think that to me was the best indicator of like, wow, I really am onto something. So even the decision of leaving a company like AXA, that was obviously doing super well, I didn't necessarily have a rational incentive to leave, but it really was something that became almost stronger than me of like, I really, I have to do this. Like I cannot stop thinking about it and I want to solve this.
44:01Julian Weisser: So CBG is really interesting because as you're saying, in a lot of ways, the design and the sort of the taste around the brand is such a defining factor of it. I was thinking about how tinned fish, you know, is quite popular. Um, but at the same time, really, you know, you can't really see what's going on with tinned fish because one, you can't even see what's inside too. Not the most attractive thing. Yeah. And three, um, I think the thing that actually catches people's eye isn't even the ingredients list, which is written in small text.
44:37It's the it's the packaging.
44:38Julian Weisser: It's the packaging. Right. And um, and I imagine if, uh, you know, especially early versions of LMS, they were responsible for doing all of the tinned fish. Uh, imagine that. You know, it would probably be this very monotonous shelf. And you go to, you go to, you know, uh, Lisbon or something. They have shops where they have 100 different brands on the wall, and they all look very unique and different. Yeah, very few of them. But some of them stick out more than others. And it just feels like in a world where tinned fish is probably like the most the opposite of differentiation, you can differentiate through your point of view by 100%. I think there's, uh, there's a really popular one called midwife or fish wife. Fish wife. Um, and that's a it's just a really excellent brand because they managed to really dial in on something that feels unique, that stands out on a shelf, and then people like it because they like the way it tastes, but they also like the way it feels like, the way it feels to interact with it.
45:36Thais Castello Branco: I think branding has always been this art of like, how do you make people have an emotional reaction to your product beyond a functional one? Right. I think so much of food or even like other types of goods, it's always been very functional. It's like, what's the purpose of this? What's the. Is it healthy? Is it is it tasty? Is it crunchy? Like what's the kind of attribute around it? But I think branding this whole goal is like how do you add this like emotional point around it of comfort, excitement, kind of edginess like liquid death, I think is a great example of a brand that did this. Two of their selling water, like it's the it's the least differentiated probably product that there is, but it's they made it so unique. And when you become a consumer like you feel like it's that something about you if you drink that brand. And so sure, um, I agree with you. I think this it's again, it's this ability to create something that's opinionated and that causes an emotional reaction in someone else. And that's, I think, exactly what, for example, these models are missing and not necessarily having the ability to create. It's like this intentional divergence from the norm, um, to create something that is intentional, though. It's not random, right? It's very like made with purpose. So I totally agree. By the way, I love that that fist and brand actually. Yeah, they're super cool.
46:51Julian Weisser: It's really good. And I think that it goes to show that this sort of solving this problem of helping people, um, express themselves or have things that are more unique in the world, um, creating more things that aren't more of the same is probably one of the most important things you can do is I think that the thing that everybody fears, most of all outside of kind of like, you know, being turned into paperclips or something.
47:15Thais Castello Branco: Yeah.
47:16Julian Weisser: Uh, you know, outside of that, they fear that, uh, people fear or like, understandably, it's, you know, job displacement that. But the thing that they fear is like this idea of, um, things just all being the same.
47:27Thais Castello Branco: Like a monotonous, gray.
47:29Julian Weisser: Monotonous.
47:29Thais Castello Branco: Sameness world.
47:30Julian Weisser: And I think that I think it's a very legitimate fear. Um, you know, I think that the default is is probably that I mean, if you look at a lot of the reasons I think that people, um, really just like new construction outside of sort of like they don't want to have more people on their in their neighborhood or something, I think is also because they just the buildings just generally look like out of character.
47:53Thais Castello Branco: And what happened to character? Yeah, I totally agree.
47:56Julian Weisser: And I think that some of that's due to from my understanding, my limited understanding, some of that is due to regulation in terms of what you're actually allowed to build. Um, and like doing things that, you know, to actually build buildings like they used to that have more character actually, like very difficult to actually get across the board.
48:14Thais Castello Branco: By the way. Uh, one question for you. I'm curious what has been like the last like something recently that you that really kind of caught your attention. Like, why did it feel so good or so Beautiful or so, like high craft or unique.
48:29Julian Weisser: Well, um, it's a bit of a it's a bit of a cop out because it's not new. Um, but something that I thought was really remarkable. Um, is the Pike Place Market in Seattle. Um, I don't know if you've ever been know. So. So Pike Place is I think it's the most famous thing sort of in, in the business world is like, that's where Starbucks got started. Um, but that's far from the thing that's actually remarkable about it. The thing that I would describe it as, as it's like a labyrinth. Um, there there is no seeming rhyme or reason to how the the buildings are constructed, where a staircase goes, what you will see when you turn a corner. It feels like something out of, you know, a fantasy novel or like a Harry Potter or something like that when you go to Hogwarts. Um, the idea that every single turn, every single staircase is a complete surprise. You have no idea what you're going to see. You have no idea where it will lead. Um, you know, the I think it's the M.C. Escher sort of the the very bizarre, sort of like staircase art. Um, it's like that in real life. And, um, and I think there's something about I'm trying to remember who who I read this from, but they wrote something about what creativity actually is or, um, and they said that it's about surprise or a large amount of it is about surprise, sort of going left when you think that somebody's going to go, right. Um, you know, and I think that in many ways, Pike Place Market in Seattle is probably the best example of a physical place that completely discombobulated you, completely throws you for a loop. You have no idea what's going to happen. You have no idea what you're going to see. And, uh, and then, of course, there's the added benefit of the fact that this is sort of a legendary place that's been around for, I don't know, 100 plus years at least. I think that all of that kind of leads to this is a place that really leaves an impression.
50:27Thais Castello Branco: I love, I love that it's like wonder, right? It's like you're you're wandering through something without necessarily knowing what what comes next?
50:34Julian Weisser: How about you? What's something that you've seen or does it need to be a place?
50:38Thais Castello Branco: But yeah,
50:41Thais Castello Branco: there was this there was this one restaurant that I went to recently in Sao Paulo, actually, that, um, it's like you wouldn't, you wouldn't really expect, much like, visually from it because it's it's a super old place, uh, kind of little tiny door in the middle of nowhere. But it's first of all, you go in and you kind of feel like you were a bit transported in time. Like you feel like you're kind of in the in the 70s and 80s. And there's just such care and tradition with all the little details and the like explanation of the dishes and the how it started and the owner story and all these things, and you feel like you're experiencing this like deep authenticity, which is maybe I think the other thing that I would say around like qualities is like feeling of authenticity I think is super important. And I think there's these like little pockets of either the world that is preserved or just like when someone really cares and they put in that extra care on details that they didn't really have to like. They didn't really have to go that extra mile to make that feel special. And they do. I think that also shows this real, um, it's real like beauty and and
51:45Thais Castello Branco: feeling of craftsmanship, like when you really care about the details. So that was definitely one. Um, another I would say was this is more as a trick. But I went to Yosemite recently with my parents. I had never been. They had never been either. And I think this I had a similar feeling to you. It was just like kind of like surprise. Like we were doing one of the hikes through the the Half Dome and, uh, you start seeing some beautiful views and beautiful views and then you kind of get up to the top and you have this like 360 view of the whole park. That's just insane. And it, I think, has a similar element of what you're describing of just like awe and surprise, and you feel like a tiny piece of something much larger, which I also found very cool and, I don't know, less left a very lasting impression on me.
52:30Julian Weisser: There's something about, um, about things that have lasted, uh, that I think I think that, um, it's, uh, Taleb who has like that idea of, like, the Lindy effect, which is that the longer something lasts, the more likely it is to continue to last. Um, and it's sort of about staying power. But if you think about it, I think that we were talking about, like ten fish that that tin fish company, you could imagine that one sticking around and lasting for another 30, 50 years. But most ten fish companies that you see, even if they have charming design or something like that, I think the I at least I get the sense that they're nice, but they don't necessarily.
53:13Thais Castello Branco: Have a lasting.
53:14Julian Weisser: Power. I don't think that they have the staying power of something like that. And I don't think that all design and all things need to have like staying power for sure. Like there's some ephemerality with what? What's made. Does everything need to sort of stand the test of time? But there's an interesting question around, like, what is the aim of, what is the aim of design or what is the aim of, of, of taste? Because, um, I don't know. Not, not to pick on VCs, but, um, you know, there's, uh, you know, there's these old VC funds that are storied VC funds, and then there's a lot of other VC funds that are today, and you're like, which one of these has the right to stick around just based on name alone? Forget about the companies that they invest in, which one of these has the right to stick around. And I don't know, I think that I think that in some ways, um,
54:06Julian Weisser: there's this really interesting thing of like survivorship, um, around what last the fact that you brought up a restaurant that's been around for a while. I brought up a like a place in Seattle. It's been around for a while. It makes me wonder just how much of like, the shelf life is determined at the beginning versus kind of what is done throughout.
54:27Thais Castello Branco: Yeah, I think there's, um, I was thinking about architecture because you had brought up architecture, too. And for example, I don't know, like you go to Egypt or you go to Greece, you go to all these places and you see these, these monuments that have existed for thousands of years and that had such like beauty and craft to them and stand so the test of time or even architecture trends of more recent times that also kind of stay in. Um, it's interesting to, to your point on like new buildings, not really having that much character. I can't imagine those sticking around. Yeah. If you think of this like, I don't know, project a, an apocalypse movie of I don't know what what like century in the future, you don't feel like those are going to be the things that, like, get preserved, right? Like you feel like these things that stood the test of time or that had this high craft, or maybe the ones that do, the stripe founders put out that request for new esthetics, which I thought was really cool, of like investing in kind of new, not just design necessarily, but these like new, uh, very purposely like crafted like esthetics. And I think maybe that's the thing that is related in terms of lasting is a lot of the things that tend to be more fads or that tend to be more like a response to something that's like a quick trend are less long lasting, because maybe they also had less authenticity or less like care. But I think if right now, like an artist, were to create a new trend and actually had this, um, these same characteristics of why things lasted for a long time and kind of recreated that. Now, I do think that there's the ability for that to pop out. Still, I don't think it's something that has been just like lost in history, and we're never going to have new things that like, lasts for a long time.
56:03Julian Weisser: I was going to ask you that because I is I'm I don't have an opinion. Um, but that's that's sort of the perspective you put out there. If I were to try and play the other side of it. Um, in some ways, perhaps the, um, the reason for why things, uh, why there are some things that have really stood the test of time is because it was harder to make things, and therefore there was, um, there was a scarcity of things, but now there's such a long tail of things being created that perhaps, maybe, maybe there will be lots of things that are remembered by some people, but not remembered by lots of people. Um, I don't know. I don't know if there's going to be another, another sort of equivalent to, to Mozart or, uh, or to the Beatles or, um, or
56:47Julian Weisser: even authors like Joan Didion or J.K. Rowling or, uh, like different types of writers. I'm curious. Like what? That actually what that actually looks like. I don't have a perspective one way or another, but I but I'm playing the other side here because you were talking about, hey, there might be room for permanence still.
57:04Thais Castello Branco: I don't know, I think
57:07Thais Castello Branco: I think that's exactly the role of tastemakers is I think this like volume of creation and generation is still like
57:18Thais Castello Branco: lower intent in a way. And I think the, the things that do stand out and that are done with so much quality and craft, I think do have the potential to stand the test of time. And more and more, I think they will be even more important and maybe stand out more in the sense that I think creating something that's going to be like, okay is going to become so easy that like the amount of people that are willing to go through the pain and the the work to make something that is just like completely out of the ordinary, great might have even more value and power and more like ability to last. So again, I don't know how that exactly will evolve, but I would be very sad to live in a world where there is nothing new that pops up that is that is like everlasting.
58:05Julian Weisser: Yeah, like genre defining an era, defining.
58:08Thais Castello Branco: Like, is Harry Potter the last of the interesting.
58:12Julian Weisser: Question, right. Yeah. Um, this has been amazing. Um, we should wrap on the two customary questions. Uh, the first question is, uh, the bear case for being a solo founder. Like, why somebody should maybe consider not being a solo founder. Just a little preview. Um, after that, we'll talk about that for a second. But after that, we'll talk about why you should be a solo founder. The case for solo founder.
58:34Thais Castello Branco: Yeah. Okay. Bear case. I think you should only be a solo founder if you have incredibly high pain tolerance and incredibly high ability to, like, emotionally regulate. Because founding is hard. Like, every day is a battle. Every day is like there's something new that pops up. And, um, I think it is more lonely in the sense that you want to protect your team from stress, right? And, um, if you have a solo co-founder, it's someone that, like, gets the same problem you're getting that has that context. Giving context to someone is super expensive. And so if you have that pain tolerance, I think it becomes a lot easier. So I would say yeah, that's probably my my bear case. But my pro case is I think this if you especially if you're a mission oriented founder, which I am, I'm very passionate about the mission that we're going after. You have such resilience of sticking through it, and you have such conviction in the vision of what you're trying to create that I think it creates a lot more of like the single mindedness around the company. I think, um, a lot of the co-founder disputes or kind of disagreements that I see over time often come from either one pulling the weight more than the other or vision slightly disagreeing. And I think especially when you are very passionate about what you're doing, that only becomes harder, I imagine. And so I, I really think appreciated that clarity and conviction and knowing like I know how 150% in it I am. And so it gives me a lot of trust in myself that I'll, like, do it, You know.
60:04Julian Weisser: Totally. I think that's so true. And, you know, when we before we started recording, you and I, we were talking about, like, Ben and Gaius from contra and how, um. And I was saying that they're an incredible example of people who truly are aligned, who've been working together for a really long time after meeting through ODF. And I think it's just really rare that people have that true amount of alignment for, for that long and that it can sustain. And I think that generally when that doesn't happen, it's pretty catastrophic for the company. Sometimes it's easy to wind things down or wind relationships down, but it's really not ideal. And it's much better to just get going, as you said, with with what you're working on. I guess the that this that was an awesome answer. I guess, like maybe the last question I was at would ask is what is something that we didn't talk about that, you know, assuming we stood up. Uh, we we stopped recording and you said, oh, gee, we should have we should have covered that. What's that thing that we should have covered? Hmm. Dealer's choice.
61:02Thais Castello Branco: Okay. One thing that I think we didn't talk about is why hasn't this problem of taste been solved yet? And I think there's obviously a lot of technical challenges around it. But actually I also think there's like a it's very hard and it is ambiguous. And I think the reason why it's hard and ambiguous means people kind of shy away from it a bit, or they think they can't solve it perfectly because there's all these open threads and questions and so they don't make progress towards it. And I think we cannot forget the actual enemy being like slop and this like gray mono taste world that we were talking about. Right. And I think the fact that all the questions are not answered does not mean we shouldn't try and does not mean we shouldn't make progress. I think progress is really the thing that we're shooting for and kind of hiding that bar. And we have we have to try and I will I'll keep trying until we succeed.
61:52Julian Weisser: So it's awesome. Thanks for doing this. This is amazing.
61:55Thais Castello Branco: Thanks for having me. This was awesome.
61:57Julian Weisser: All right. We just wrapped this episode with ties. I thought it was really interesting to hear her perspective on taste around the difference between AIS and and humans and how, um, effectively like the way we develop taste is very different as humans than it is as, as machines. And I think there's just this incredible, interesting question around what taste actually is. Um, who's the arbiter of what good taste is? We talked a lot about critics. We talked a lot about creators. I think there's so much I'm really curious to see kind of where Taste Labs goes. Um, I'm really curious to see what actually the end state of artificial intelligence is when it comes to taste and creativity. Uh, it seems like in many ways, artificial intelligence can't even, uh, copy taste in a in a meaningful way. I talked on the episode with her about sort of how great artists would often do very exacting copies, or really try to copy other people's styles before sometimes developing their own. I don't think that AI is particularly good at mimicking styles right now in a very in a very high quality way. It does it fairly superficially. Um, so there's this interesting question as to when does AI, um, get to a point where it could actually do the styles in the way that a master could replicate the styles that maybe came before him or her when it comes to painting, when could AI do that? Um, in various formats. And then the bigger question in my mind is, is not just can it do stuff that already exists, it can, um, can taste, actually be forward looking versus backward looking? Uh, very easy to, you know, not very easy, but but much more feasible to imagine copying something that already exists than creating something that's new, that might have some connection to the past. Because nothing is made in a vacuum. It's inspired by many different influences, but a lot of the new things that exist. Um, they really are surprising. Um, they're not random either. They are. They are a very conscientious, um, at least in many cases, conscientious choice to combine various, uh, influences, experiences and to make something that is wholly new as a result of that. It makes me wonder where we are in that gap. Um, you know, I think that that seems even further away than AI being able to truly do a wonderful job at replicating a lot of styles, um, and sort of like working within those, those forms so that that is kind of what I'm excited to see. Uh, it's very early on, obviously, with regard to all of this. Um, there was a lot of heat and a lot of misunderstanding or derision, uh, for Taste Labs, which I think is like very understandable, to be honest. Uh, When if the interpretation is that there's going to be a handful of people that determine what the taste is for everyone. Um, I could understand why people would think that's either just not good or, uh, or a bad idea or something that won't succeed. Uh, but that's clearly not what they're doing. They're trying to essentially help, uh, artificial intelligence understand, um, what taste is, um, not their taste. And I think that that's really cool. Um, I think it's one of those things where I'm not even sure what will possibly come from it. Um, other than, I think some interesting things that would not be possible to predict. And, um, I think that's in many ways sort of following the threat of what AI actually does. It does the things that one could predict, but it also does things that we wouldn't have predicted because we didn't necessarily have the capabilities of doing it before. Um, and I think that maybe lastly, to sort of tie a tie a bow on this, something that that was said that I thought was particularly interesting, um, was this idea of, um, AI sort of being, uh, sort of very, very horizontal. So like if some, if AI has a certain way of doing things or a model has a certain way of doing things, and it sort of just goes all the way across the, you know, every use case, you know, somebody who's building a website, somebody who's writing some copy, somebody who's doing a slide deck, somebody who's generating an image. If there's too much sort of horizontal similarity, it causes this, you know, slop or sort of, um, everything looks the same. Um, I think that that is the thing that we just really don't want to have in the world. We want to have such a wide array of interesting ideas and creations and things that look very visually different from each other. Um, and I, I'm just really excited to see, uh, you know, where where that goes and and where taste labs can take things. So that's, uh, that's my quick sort of reaction to the episode.
66:58If you enjoyed this conversation, we would love it if you could share solo founders with everyone you know, particularly people who are considering starting companies solo or who are already doing it. You know, the entire goal of what we do is to normalize solo founder and encourage more people to solo, and instead of end up with co-founders of convenience, if you'd like. And we'd really appreciate it, you can go leave a review on Apple. You can go give us a thumbs up. You can subscribe on the YouTube channel. Anything helps. And of course, if you are starting a company or considering starting a company, you should go look at the Solo Founders Program. We work together with you in San Francisco alongside a bunch of other solo founders. The idea is it's much better to be solo together than to be solo alone. I'll spend a lot of time helping you, and we also invest $100,000 in your company. So if you're excited to learn more about that, go check out Solo Founders com Slash program.


