This Startup Thinks It Can Give AI Good Taste

[00:00:00]So, somebody just raised $18.5 million to try and give AI good taste. They want to end AI slop, and they're doing it in the corniest, most Silicon Valley tech bro way imaginable. But let's at least see what they have to say. This is called Taste Labs. Right off the bat, the logo is very strange. Um the corners here create a lot of visual tension because there's such a small amount of space uh where these two shapes almost intersect but not quite. But you know what? We're going to give them a fair shake.

[00:00:33]>> Let's be honest. AI has a taste problem.

[00:00:36]>> That's true. Okay.

[00:00:38]>> We're building the data and infrastructure layer to fix it.

[00:00:41]>> Oh no. Oh no. Oh god help us. All right, let's keep going.

[00:00:45]>> We're going to give her We're going to let her talk for more than 5 seconds.

[00:00:48]Okay. Tais. I'm the founder of Taste Labs and I'm excited to announce we raised $18.5 million to end AI swap. AI has made it easy to generate anything.

[00:00:58]But it still feels off.

[00:01:00]>> Notice the wording here. It still feels off. Feels very subjective term.

[00:01:06]>> Sameness, repetition, templates, no sense of individuality. We want a future where >> this is all true.

[00:01:12]>> AI understands what feels right. She's sort of accurately addressed the problem, but uh completely missed the way that one might be able to solve it, at least the way that she has framed the problem.

[00:01:25]>> What fits, what's beautiful, what feels like you, what is truly great. Today, those decisions live inside human judgment. They're difficult to explain, difficult to measure, which is why it's been so hard for AI to become great at these subjective domains. No, that's not why it's been so hard for AI to be great at these subjective domains. But let's continue.

[00:01:45]>> We're building the foundation to measure, judge, codify, and search over what great means in domains where there's not a clear right answer. And we're starting with design.

[00:01:57]>> Oh god. Okay. Um, the simplest way that I can put this is imagine trying to engineer or reverse engineer what makes something cool. Let's just keep going.

[00:02:08]We've already been working with the top frontier labs to improve their models, crafting the right post training data and our own environments. And we've also been working with top Genai and agent companies to craft the right context and verification tools for their agents to be able to generate better, more creative, more on brand designs. And we're so okay, I wrote a tweet in response to this and I basically said this highlights a fun fundamental misunderstanding in tech. Not everything can be codified and analyzed. Even if you can make AI imitate taste, whatever that means, because taste is a very, very nebulous term, which we'll get into in a little bit, it still won't mean anything. Taste emerges from craft, context, meaning, subjectivity, and genuine care.

[00:02:53]>> And we're just getting started. If >> Oh, God help us. We're just getting started. Oh, no. What's she going to say next?

[00:02:59]>> If you're as passionate about this mission of ending as we are, join us.

[00:03:03]We're hiring across research, engineering, product, and more.

[00:03:07]>> Taes, you should hire me. Unironically, I could help you a lot with this. I don't think that you would ever hire me, and I don't think that we would get along at all. I'm sure you're a very nice person, but I could tell you all the ways why this isn't going to work, and I could redirect your company to a better point.

[00:03:27]I've been making videos about the intersection of design and culture for several years. I have 780,000 subscribers on YouTube. I'm a professor of design at California College of the Arts, or at least I was a few years ago.

[00:03:39]I've been practicing design professionally for 15 years. You should hire me, Tais. I can help you and explain to you why this is not going to work. I can save you a lot of time and money. Anyway, I I shouldn't be so mean.

[00:03:50]Let's keep going.

[00:03:51]>> Apply at tastelabs.com today.

[00:03:54]>> Oh, let's apply. By the way, this is my second channel, Design Theory Workshop.

[00:03:58]My main channel is Design Theory. Make sure you're subscribed to both so you get updated when I post new videos.

[00:04:04]Okay, so right off the bat, like I said, the logo is a little bit strange. It feels like the the spacing of this really broad shape here and this little tiny thin line here creates a lot of visual tension and imbalance. Um, the logo doesn't mean anything or say anything, which is kind of hinting at the the fundamental issue here. Even if you can make AI imitate taste, it still won't mean anything. It's very interesting that the logo, at least for me, doesn't mean anything. Taste Labs, the name Taste Labs. Taste is a very nebulous loaded term. Now, I understand she's probably chasing trends. I get it.

[00:04:46]Taste is really trending big on Twitter right now in like the AI tech bro universe. I get it. I think it's a terrible word, which we'll get into later why why that why I think that's the case. But man, this is really really missing the point. Let's keep going. I think that this meme kind of encapsulates it. You can steal my swag, but you you will never replicate my lore. It's like it's kind of like what I was saying. It's like what what do you love about your partner or somebody that you care about?

[00:05:16]It's the story and the history and the past that you have together. And in the same way, when it comes to a design or a product or anything that is made by human beings, it's really the story that carries it. That's what makes it valuable. Now, there are exceptions to this. There is a lot of like slop in the world. And I don't mean that as a porative or a negative term. Like McDonald's is sort of food slop. Um AI is is honestly slop in its own way. Um, reality TV is entertainment slop. I don't think that those things are categorically bad as and that's kind of a hot take. I don't think that slop is universally bad. I actually think that creating really good slop is is very hard to do. It requires craftsmanship.

[00:06:05]pop music for example, there are some incredibly skilled, talented musicians who make kind of, you know, like surface level superficial music, but the music is enjoyable and it takes a lot of craft to make that stuff. So, I don't think that slop is categorically bad. I think that it has its place in the world, but if you're trying to create something tasteful, that isn't going to work with AI. And we'll we'll keep going into why.

[00:06:33]I think that this is another key point here. The idea that taste is a problem with a technical solution is kind of hilarious. It's like trying to brute force coolness. Good luck. Yeah. Once again, it's like what makes something cool or what makes you love something?

[00:06:49]What makes you care about it? It just doesn't make any sense to try and codify it. If we use the example of the partner, of the romantic partner, you might say that there are certain characteristics that they have that make you care about them or love them. But it's not the characteristics themselves that make the thing. I could make a clone of, you know, your romantic partner in a lab, but it's not going to be them. It's your shared experiences.

[00:07:16]It's everything about your history together that makes it special. And it's the same thing with products and design and that kind of thing. Uh, on my Patreon, I got this really great comment from WGP. Uh, by the way, thank you WGP. Consider supporting me on Patreon if you want or don't. Doesn't matter. But you know, he's saying, I go to their website and look at their mission statement and I see there can be a world of creativity, beauty, and taste that requires cracking how to measure, classify, stir, search over, and codify subjective domains into data models can learn from, and tools agents can use so that models and agents can produce not just outputs that are correct, but that feel right.

[00:07:57]They want to standardize and measure something as intangible and subjective as taste. The outputs that feel right statement alone is in my opinion a paradox. And that's exactly right. And this is actually another comment here by Strawberry Puptart. His names are great.

[00:08:13]By the way, I guess tangent commentary.

[00:08:16]I've always thought of the term slop to be elitist snobbery. It's an arbitrary value judgment and usually meant derogatively.

[00:08:24]Soap operas, romance novels, and the we.

[00:08:26]They're not my thing. But it's also kind of being a a jerk to yuck someone else's yum. Yeah, I agree. But there's another thing, taste. I don't even think that taste is the right word to be using.

[00:08:38]It's a very loaded term. It has a lot of historical baggage. It's inherently elitist. It doesn't really mean anything. I think it's important to remember that taste was sort of formed by the European gentry or the ruling class and then that would sort of trickle down to you know the plebs essentially. So it's very very elitist even though it is entirely arbitrary taste. I mean I have some better words that you could use instead of taste.

[00:09:07]Judgment, discernment, care. Taste can mean class status. It can mean money. It can mean education. It can mean knowing the right references. It can mean having discernment. It can mean care. It can mean judgment. It can mean simply liking expensive objects. It can mean so many things. And then Silicon Valley shows up and says, "We're going to give AI taste." Okay, so what part of that sentence are they actually claiming to solve here? Trying to codify taste and individuality at the same time isn't even possible. They're opposites. Like if we read the full quote by Tidor or Tittor, um taste and individuality at the same time is not possible. They're opposites.

[00:09:54]Unique stuff is often disruptive and can be considered tasteless by some. Once again, sort of implying that context really matters. It's very subjective.

[00:10:03]He says, I think they can improve the quality of the AI generated stuff. They can give the AI a baseline to make some stuff more tidy and correct, but calling that taste is a stretch. It's a very cultural concept. Japanese UI and websites is awful from my point of view, but who am I to say it's tasteless when a whole country is into it. What will the A do what will the AI do in those cases? I think that this is a really really important point. There's another really great point by WGP here. The thing about taste is it's also tied to socioeconomic context of a person as well as the trends of the time. There was a time when a Van Go painting was considered trash. I imagine an LLM being trained with data of that time would output the exact convention of the time.

[00:10:45]Yeah. I mean, if you want to be a trends setter or tasteful, you kind of have to know when to break the rules, when to hang back. It's a very, very tricky thing. It's extremely nuanced. It requires a deep cultural understanding.

[00:10:59]And even if it does achieve that cultural understanding, which I don't think it will. I think that's impossible. Um, it's still not going to have the story attached to it. It seems to me like Taste Labs thinks taste can be learned like a language, whereas I see it more like growing up within the culture that speaks that language. This is very important. You can codify and understand words, but understanding the implications and the culture is really what's missing here. And I think Ryan is right here. He says it'll play well to executives nervous about the quality of their LLM output and looking to buy a solution who themselves also likely have no genuine taste. Yeah, they're just out of the loop. and it'll help people feel more secure and give them certainty in that regard. So, it's possible that this company might be successful, just not in the way that they're saying. So, here's something I could see Taste Labs being useful. Maybe if you're a company like Walmart or Target or Gap, they're sort of way downstream from the tip of the spear of culture. I could imagine it aggregating a bunch of data from past trends and then making inferences to do like low-end copies. And I don't mean that in like a peorative or negative way like this design layout and these shoes themselves and all of the patterns and trends here like that that could be useful uh that could be a good use case for taste labs where it's like they're they're just doing trend analysis to understand what has happened but that's very different from creating something that is tasteful or that is sort of setting the trend. Those are two very very different things. Like this is essentially a commercial imitation for making things more efficient.

[00:12:41]That's not a system for creating culture or at least not a culture that I want to be a part of. I guess it is its own it it is culture in its own way. I'm not trying to be elitist here, but those are not the same things. Taste Labs is positioning themselves as sort of setting trends and and allowing for personalization, but I don't think that's going to happen. So, let's go back to this picture. Interestingly, these guys are trying to position themselves as tasteful, right? The tasteful team. Um, this just looks like a nightmare blunt rotation to me. Maybe they're really nice. I bet you they are really, really nice people. Please don't make fun of these people. Please be nice to them. Like, we don't know them. We're just We're just looking at this picture.

[00:13:22]Let's analyze the image a little bit.

[00:13:24]Right? It's like, okay, so they have this sort of uh beautiful background.

[00:13:28]They're in a very modern space. They have the Barcelona chair, which is a signal of, you know, an understanding of taste, at least in terms of the, you know, the Silicon Valley techbro elite.

[00:13:39]It's like, oh, the Barcelona chair. Oh, the mid-century modern chair here, which I can't remember the name of right now that she is sitting on. You got the, you got the beautiful backdrop behind you.

[00:13:49]You have, you know, the dark clothes.

[00:13:52]All of this stuff sort of signals certain things.

[00:13:56]But in another context, these are just cringey nerds. Like there are all these little signals but they communicate very very different things depending on the context. This is another really important point here. Most of what is considered taste is the real realm of zero sum signaling games. Yep. It's all about signaling. Your taste slop will just just be the next AI slop and then your anti-taste slop will become the next taste slop. Taste is defined in terms of slop and therefore can never transcend slop. What he's saying here is that Taste Labs is always going to be behind the eightball almost by definition because they are only training data rather than creating something new. That's key here. That's really important. Here's the thing.

[00:14:39]Something can be dismissed as bad, tacky, ugly, wrong, or trash and then become the thing that everyone later calls important. It's like the Van Go paintings. I mean, he was a nobody for his entire life, and now his paintings are incredibly significant. That's why this idea of training on existing judgment feels really strange to me.

[00:15:00]Training data gives you the convention.

[00:15:02]It gives you the thing after people have already decided how to see it. It's not about emerging ideas. Someone in my Patreon actually recommended this article uh in the New York Times called what Silicon Valley is coming for next.

[00:15:18]And it's basically about taste. You know, it says taste come comes from outside of you. It surprises you. It's not what you guessed it was. It's something that comes up and brings you somewhere new. I mean, here's how you can think about it. I already have access to my brain all day. I don't need an infinitely personalized machine to make increasingly precise versions of what I already think I want. I don't need that. Part of the point of art, books, music, culture is that someone else made it. Someone else had an imagination. Someone else had a worldview. Someone else cared about something you may not have known to ask for. And this allows you to be brought into their world. It allows you to empathize. That's a really important aspect of taste. The communal element of it, which is, I think, going to be missing from this company. What is inspiring is commitment to craft. you know, this endless practice, the drilling, the adapting, the failing, the adjusting, you know, trusting the people that you work with and loving something enough to really suffer for it. That last part is key. It's about loving something so much that you just obsess over it and aspire to greatness. That's really what makes something special. So taste isn't really the word in my opinion. I don't think that's the right word. I think it's care. It's about craft, effort, attention, and time. It's about sort of being willing to obsess over something and fail at it and just go back over and over again and keep going and it doesn't feel like it's this toiling awful thing. It's it's something that you really care about. And I think that's the key here. AI will probably be able to copy the the elements and deconstruct a lot of this stuff. And that will be useful to some companies.

[00:17:10]There already are trend research labs out there and this would probably just accelerate that. But the story is what's going to really be significant moving forward. Now look, I have my biases, right? I literally tell stories on YouTube, my other channel, Design Theory. It's literally my job to tell compelling stories about design and culture. And I do think that AI will increasingly uh sort of consume a lot of things that are more focused on convenience. But I think there's always going to be a place for the human touch.

[00:17:48]And I think the fact that people have reacted so violently and negatively to AI in ways that most most of the time technology is appreciated and respected and embraced by the youth. But that just isn't the case with AI. Even Gen Z, even maybe Gen Alpha, like AI is equated with slop. It's equated with low quality.

[00:18:11]It's uncool. I think that has a lot of significance and I think it's because it will never have that human element, that human touch. It will always be considered loweffort.

[00:18:24]Anyway, I hope you enjoyed this video.

[00:18:27]Consider subscribing or supporting me on Patreon. I really appreciate it. I hope you learned something and have a great