Loading summary
Transcript88 lines
- [00:00]
A
I mean, I do think, you know, every technology cycle comes with hype. You know, that's just a very natural, human thing. But I think the companies that do last are the ones that can kind of really, truly deliver right on their promise.
- [00:17]
B
Welcome to Culture and Code, a podcast about patterns in tech, business, and culture. I'm Reinamoto, a creative entrepreneur and the founding partner of iAncode, a global innovation firm based in New York, Tokyo and Singapore.
- [00:29]
A
Cool. And I'm Tara Tan, managing partner of Strange Ventures. We're an early stage firm that invests in the future of computing.
- [00:35]
B
So this is a podcast we've been talking about. We tested a few recording sessions, but this is the official first episode that you and I are recording.
- [00:43]
A
So how are you excited to be here? You know, we're. I feel like we're entering the end of summer. You know, you see all the Halloween stuff coming out, but it feels. Yeah, already. But it feels like the world of AI has not stopped. I think last week was a record. We had like three to four product launches, which is totally insane.
- [01:05]
B
Yeah.
- [01:05]
A
So it feels like it's not stopped.
- [01:07]
B
Yeah. So, speaking of which, in this podcast, we tried to decode the biggest shifts in culture and tech. And the topic that we want to cover is inspired by one of the newsletters that you wrote recently, Quality versus Hype. And in that newsletter you talked about this startup that I don't think gets a lot of headlines, but it's apparently making already a billion dollars a year. It's been around for about five years or so. And it's this company called Serge. So why don't you give a quick 101 on who that company is and why you decided to mention in your newsletter. And then we can go into this topic or about quality over Hype or Quality vs Hype in tech and just, you know, culture at Marshall.
- [01:54]
A
Yeah, I love this story so much. So I first found out about them through the Information, which is a great tech publication that's around. So Sergio AI is a data labeling services company. We never hear about them. They were started five years ago. They have over a billion dollars in revenue or more raised. $0 in venture funding. So they were entirely bootstrapped. And their closest competitor is Scale AI, which is obviously very famous, very known. And what's surprising about me, for me, was that, you know, compared to what scale. Scale AI has raised in terms of venture funding, this team is entirely bootstrapped. They actually make more than scale AI. I think scale AI makes about 850 billion. A million a Year and they make about a billion and they do the same services. And I think what's different about search is that they really emphasize very human, high quality, nuanced data labeling for training, data for more for model makers. And I think in that interview that was mentioned in my newsletter, he talked about how, you know, a lot of data labeling services is treated like, you know, kind of you throw humans at the problem, you know, bring them overseas. And he shared this story where when he was building a sentiment model at Twitter before Serge AI, they thought, okay, to do a sentiment analysis model. They would just like, okay, go through like whatever, 10,000 tweets and, and mark them positive or negative. But it's more nuanced than that. Humans are much more nuanced than that. Right? Like, it's hard to say if this is a truly positive or negative based on if you don't know the cultural context, which is so interesting. So we had the entire company surge is built around building high quality, very nuanced data sets, which I think is fascinating. And I think that bleeds over to the world of AI and product and more today. I'm curious about your. What resonated for you there, Ray.
- [03:47]
B
Yeah, so I didn't know about this company until I read the name in your newsletter. And from there I happened to watch the interview, one of the few interviews that he's given publicly. And I had no idea who he was. I had no idea who the company was. And so just to give a little bit of dimension about what quality of data means, one example that he used as a way to talk about the nuance of data and particularly language. So let's say if I say, oh, he's such a badass, right? In a colloquial sense, I use that as a compliment to talk about somebody, but linguistically in a literal way, bad is negative. So model might annotate that as a negative comment. So what seems so intuitive for a human to understand a language? For machines, something as basic, as simple as, oh, he's such a badass or something like, right. Computer has a very difficult. Machines have a hard time decoding that type of nuance. And, you know, he just used that, something like that as an example of what he means by the quality of data. So that was one thing that I sort of knew, like what he meant by quality of data. But that example was a good example. Another thing that struck me and then the reason why we decided to use that as an entry point for today's conversation, quality versus hype is something that Is some as mundane or as simplistic as a nuance in a language is very difficult for machines to decipher. And sort of this. I'm going on a tangent, but I happened to come across a couple of clips this past weekend and it was a trailer for Dragon Ball and Dragon Ball, you know, that anime is a Japanese anime that's been around for, I mean, I. It was around when I was a kid in Japan, so like 30 plus years ago. And over the past, like two decades or so, it's become international and it's been around in the global market long enough that some people might not even know that it's a Japanese anime. Right. And that this trailer was for a new version of Dragon Ball in theaters.
- [06:29]
A
Okay.
- [06:30]
B
But two things. One, it was fake, and two, it was created by AI. So some, I'm sure some guy used, you know, I don't know, VO or midjourney or what have you to create this two minute trailer. And he had multiple of those trailers all around Juggle, and most of them had like several million views. Yeah. And. But at the same time I watched it and within 10 seconds I'm like, like, if you see like a second, you know, one second clip, it looks quite amazing. But then quickly you realize that like, what's, like there doesn't seem to be a captivating story behind it.
- [07:14]
A
I see.
- [07:16]
B
Yeah. So I, you know, I don't know if he was creating these trailers as a way to get more views just to, I don't know, get advertising, you know, for his channel. Most likely because it became pretty clear just by going through the comments that there's no mo. There's no Dragon Ball movie coming up. Yeah. But the reason why I bring it up is like that sort of quote unquote quality of data. Like I as a human being was able to decipher it pretty quickly, but like for a machine, I think it would be pretty difficult to spot like why this is good versus not great.
- [08:01]
A
Yeah.
- [08:02]
B
You know. Yeah. So going back to Serge and what they're doing, there's a lot of hype when it comes to data, LLM and AI companies. And I think there are a lot of companies that are, that are managing to raise a lot of money. But just watching his interview, listening to his interview, I think there's a lot of hype right now. And I wonder when that hype, the, the curtain of that hype will be revealed sooner or later. I mean, when that, when that could be.
- [08:36]
A
Yeah, I mean, I do think, you know, every technology cycle Comes with hype. You know, that's just a very natural human thing. But I think the companies that do last are the ones that can kind of really, truly deliver. Right. On their promise. Yeah. And it's interesting. I mean, you know, this podcast is named Culture and Code, and I think this example really captures exactly what's going on. Like, culture is living. And, you know, Code is really trying to encode as much of that living culture as much as possible. You can kind of see that counterculture as well. Like, you know, right now, I think, you know, it's cultural. Right. If you want to compliment someone, you actually give them an insult. Right. Have you heard of this?
- [09:15]
B
No. No.
- [09:16]
A
You're like, you're funny. Don't go bald.
- [09:18]
B
It's like, oh, it's kind of like.
- [09:21]
A
So it's like a double. And that is very. I think that's very hard for machines to. They decode.
- [09:26]
B
Yeah.
- [09:26]
A
So it's quite interesting. And I think. I do think that there's going to be increasing, increasingly in our culture, a way to prove, like, proof of human is going to be key.
- [09:38]
B
Yeah.
- [09:39]
A
Right. So, for example, there was this news that broke a couple. Maybe a couple weeks, couple of months ago of employees who are not real, who have been applying for top AI companies and.
- [09:49]
B
Oh, yeah, yeah, yeah, yeah.
- [09:51]
A
Right. And they're actually taking multiple positions. And then someone at some point realized, like, wait, this person is. It has been scamming all of us. And I do think, like, you know, these sort of things within our culture is going to be closer and closer to, like, proof of human. Right. Which is kind of interesting. Can you tell if they're a human or they're a bot?
- [10:10]
B
Yeah, I mean, I would even say, you know, machine versus human or bot versus human. But even within the spectrum of different types of humans and within the spectrum of the intelligence of human beings. Right. And going back to Sergio Serge and the interview that I watched, he talked quite a bit about the quality of the talent that he hires. And there are too many. I guess it's maybe it's a Silicon Valley term, body shops, meaning that companies and startups that are just hiring like crazy, and they're just hiring bodies and not enough, like, real talent. Yeah, yeah.
- [10:54]
A
For the search area example, it's really a thinly veiled kind of hit at, like, places like scale, because scale is very famous for farming out the work to very cheap labor overseas, essentially.
- [11:08]
B
Oh, wow.
- [11:08]
A
Very cheap labor. So they pay them like a dollar a day or something crazy. Like, I think it's all online. It's, you know, pay them extremely cheap labor to do massive amounts of data labeling shops.
- [11:20]
B
Yeah, yeah, but I think like what he was talking about is in addition to cheap labor that they might outsource, but also like expensive labor, I. E. Like PhDs from, you know, MIT, Stanford.
- [11:35]
A
Right.
- [11:35]
B
Yale, Harvard, Princeton. And these Companies would hire PhD just because they have PhD in data management or data science or machine learning or computer science. But just because you have a PhD in physics or math doesn't mean that you are a good computer programmer. And he talked about computer code as poetry, you know, so the reason why I'm bringing that up is there's the difference between machines and humans, but also difference in humans, obviously. But what I'm getting at is I think the level of creativity and the humanness will become more and more. It might take time. But just because you can program doesn't mean that you can program well or, you know, write poetry as if you code. And that's between creative programmer versus you know, run of the mill programmer.
- [12:41]
A
I totally agree. I mean, I think the bar for craft is going to be higher. I mean, I think there's craft in everything. You and I know this, right? You know, coming from a creative background. So there's design, craft. I mean, everything has a craft. Even data labeling is a craft. And the surge. AI folks definitely think of it that way. Right. Investing is a craft. I would say even being able to craft or, you know, build a great rag model or, you know, some sort of reasoning system is a craft. Because I think that kind of comes out and I think, you know, it's easy to say, oh, AI will automate everything. And yes, it does make things a lot easier in terms of the tooling. But I think the bar for craft is higher than ever. You know, people talk about taste, people talk about, oh, I think it's craft. I really think it is.
- [13:27]
B
When you say craft, but in, in the context of technology. Let's unpack that a bit. Yeah, yeah. What do you mean when you say craft in the. I think like craft in the context of like design, film writing, fashion might be a bit more obvious. But in the context of tech, what do you mean by craft?
- [13:48]
A
I mean, we should pull that in. Because I do think that abstraction from, you know, craftsmanship, which is very sort of, you know, I mean, I would say Japanese culture is very much an embodiment of that. Even your coffee shops, some of them have high craft.
- [14:04]
B
Yeah.
- [14:04]
A
And they pour a ton of craft into that. Right. Whether they're really Thinking beyond what was taught into digging in into what I say is first principles thinking, how do you make a great cup of coffee? Is it the beans? Is it the temperature? Is it the process? Is it the. There's an investigation beyond like the obvious steps to like very first principle of thinking what makes a great cup of coffee. And I would say like that sort of similar attitude or approach. Let's go back to first principles thinking what makes a great, you know, sort of training data set kind of comes down to that. What makes a great system that can serve this company or this customer explains that. Like, I feel like that kind of comes down to. For very first principles thinking what makes really great sort of a post training sort of, you know, process. And there's all sort of, sort of thinking that is required, I feel. And that's for me, craft for me.
- [15:04]
B
Yeah, I. I may be going off topic, but I bring it back. But you brought up coffee shops in Japan. Yeah, right. And what makes great coffee, obviously the quality of beans and how you brew the beans or coffee. But like, one place to look for craftsmanship is how clean the countertop or the coffee table. The kitchen is where you make coffee. And I'll give you two examples. So I grew up in a household where my dad was a woodworker. So my dad, when I was like two or three years old, he started a company, furniture company, using Japanese craftsman, you know, woodworking craftsmanship to make furniture, dishware, and all the way to architecture. And it, it was really, I remember I must have been, I don't know, less than 10 years old. And he, I went down to his. His shop, woodworking shop, and he was like, oh, you know, you gotta clean. Make sure where you work is clean first.
- [16:18]
A
Yeah.
- [16:19]
B
Like, you know, wipe the table. Just organize all the tools before you start working. And I found it so ironic because at home he was the messiest person. Yeah. Yeah. So. And you know, that was like when I was about 10. And then not too long ago, I happened to be listening to a podcast about this sushi chef, Jiro, and there's this famous documentary called the Jiro Dreams of Sushi.
- [16:53]
A
Yeah.
- [16:53]
B
And he's probably the most famous Japanese sushi chef because, you know, his documentary was on Netflix and you know, it was. It was one of the most popular Netflix shows many years ago. But in it he talks about his kitchen and. Or some, I guess somebody talks about his kitchen. And it's the cleanest kitchen that you ever see because he makes sure that there's like nothing on his kitchen counter and where he works to sort of interrupt his workflow. So and the reason why I bring it up is it goes back to, you know, the point about craftsmanship, but also the purity and the cleanness of data and like where you work and what you work with. If it's not clean, then what comes out on the other end won't clean.
- [17:49]
A
Yeah.
- [17:49]
B
You know, so it's this garbage in, garbage out.
- [17:52]
A
Right.
- [17:53]
B
Topic. Right, yeah. And, but also like the way you make it. And this goes back to the point about craftsmanship is that I think you need to be obsessed with not just the, not just shipping the product out, but how you make it and where you make it. You know, has to have that level of craftsmanship to make sure that the output has quality as well. So it's the quality of data, quality of input, and then the quality of how and where you work that guarantees the quality of the output.
- [18:33]
A
Yeah, 100%. I mean, I do think that there's like an interesting, you know, tension here where especially in startup or tech culture, craft has been seen as slow, which I think is not true. I think there's like a misunderstanding about craft. So when someone says, oh, you know, they're very, you know, they're sort of very craft oriented, sometimes it feels like, oh, you know, it's very different from like startup speed. Like throw everything against the wall, like doesn't matter how you get there, just get there. And it's sort of an interesting tension here where it's like very craft sort of oriented kind of approach where they're like demanding perfection over everything versus like messiness, speed and all of that. And I don't think they're actually quite different. I do think that both have value. So if you look at old Apple's a very great example of a craft first organization and what they do. Right. And it's still enduring to this day.
- [19:32]
B
Yes, definitely. So on that note, I'm going to switch our conversation to the next topic that we talked about, we said that we would talk about, which is the recent launch of GPT5. And I'm wondering, is ChatGPT or OpenAI as a company, are they more focused on speed and even hype versus quality or craft?
- [19:59]
A
That's a big one. I mean, they're getting slammed. They're getting slammed right now because GPT5 wasn't the technical breakthrough that a lot of people expected. So it was quite underwhelming from a technical point of view.
- [20:11]
B
Right, right.
- [20:12]
A
Like everyone's like performance benchmarks kind of sucked if it's not in thinking mode, it actually can't solve like a lot of basic math stuff. So I would say that there's there. I mean, whereas you compare it to the recent sort of releases from the other model makers have performed really well. So I think that was a little bit of an underwhelming, kind of like feeling from this launch from GPT5. However, I thought what was interesting that what OpenAI did is that they are starting to productize or take a productized approach to their model.
- [20:44]
B
So what do you mean by that? Say more.
- [20:46]
A
Yeah, so what GPT5 did, what was interesting was that they, number one, collapsed all the models. You know, previously you had like 3.5, 4.0 for reasoning.03 for no, so 040 and then.03 for reasoning and all of that. You can kind of choose between the model. Right now, if you sign in, it's only GPT5. So they collect all of it. Which means that what they have done is to abstract all the different models, types of models under the hood. So that when you put in a query, part of that response is trying to figure out which model should it route to.
- [21:18]
B
Right, right.
- [21:19]
A
For what kind of capabilities do you need to answer this question better? So that for me is productization. So you have that and then now you can do things like oh for the developers, change between how verbose do you want it, how much more reasoning do you want it, how quick answer do you want it? So I feel like it's more of a productization approach. Obviously there's still a lot of kinks to work through. Like, you know, I don't think they did it very well, but what I'm seeing is that, you know, it becomes more around usability of the model versus just technical breakthrough. That was my sense for GPT5. It's also a little bit more aesthetic, so they actually try to bake in a lot more taste into it, which is quite interesting. But again, I think this is more about UX or DX user experience, developer experience, than a technical breakthrough.
- [22:09]
B
Technical breakthrough, Yep. Yeah, I mean, I think and subsequent subsequently chatgpt, they. So going back to the quality versus hype conversation where we started, there's a decent amount of quality in what they do, but they also try to create hype just by releasing models after models. And OpenAI and Sam Altman I think are quite skilled at creating a narrative for OpenAI as a company and as a brand. Because to your point, the capabilities of ChatGPT, if you compare that to other Models out there, you know. Yeah, in some capacities, ChatGPT might perform better, but in a lot of other capacities, other models perform better. But in the public's psyche and in culture, if you ask people outside of the Silicon Valley, US West coast tech bubble, like 95% of people might associate AI or ChatGPT is AI.
- [23:27]
A
Yeah, yeah, absolutely.
- [23:31]
B
My dad would know ChatGPT, but he has no idea about any of the other models.
- [23:37]
A
Right, yeah.
- [23:38]
B
So OpenAI and Sam Altman have been able to create hype around AI as a topic, but also ChatGPT and OpenAI as a brand, so that it takes more mind share in culture and they are able. And then I think like the release of GPT5 or any of the previous models were to create to feed the narrative that OpenAI is dominating in the AI race. While in reality the difference between what they make versus what Google makes versus what anthropic makes may not be as big of a gap as at least the awareness and the recognition that the open air gets.
- [24:27]
A
Yeah, I mean, it's a tough one, I think, for what they're doing, you know, harnessing capital, a big part of the play, Huge part of the play. Like, for a company like that, it's impossible to be bootstrapped. Right. So like, yeah, harnessing capital and talent is a big one. And they're going up against, you know, your metas, your Googles of the world, which, what it feels like infinite capital. Obviously it's not infinite, but it feels like it's infinite. With meta, there's no sort of lobbying, $200 million paychecks for the best researchers, which is crazy. Yeah. So I feel like, you know, like, you know, that was a great. I can't remember which is the op ed I read, but it was, you know, sort of this idea and I totally sort of resonated with it, which is hype as infrastructure. Like hype is a essential tool for the likes of these companies to really harness capital, resources, attention and talent as they get into that race. But it's interesting, I mean, you have like the surge AI, which is totally, you know, they kind of just did work really well and then they sort of won over time. But I'm curious if that's going to sort of still stay true in the model maker race. I mean, you have a couple of those right now that are doing that where they're a little bit more stealthy. They've raised capital, but they're stealthy and they're, you know, we'll see, we'll see.
- [25:45]
B
Yeah, but the story, like her Serge is, is an encouraging. I would like to see more of those stories. Yeah, you know, I'm. I mean, OpenAI is kind of an anomaly in that, in that space and it's sucking all of the air out of the press especially. And companies like Serge don't get mentioned a lot. Yeah. But like, over time, my hope is that the quality, I mean, there's always hype. And hype is something that I think is, I don't want to necessarily say easier to generate, but I think that there are a lot of people, whether you're in Silicon Valley or outside of Silicon Valley, outside of the, you know, the rest of the world, you know, just doing jazzy hand to create hype, to raise money or to get attention. I think it's something that, I think a lot of companies and a lot of people do.
- [26:39]
A
And.
- [26:40]
B
But yeah, like the story, like surge is something that A, is encouraging and B, I would love to see more of not just in tech, but in other places as well.
- [26:54]
A
Yeah, I mean, I would say that there probably are. It's just that we don't hear about them because they don't need hype as much. You know what I mean? Like, you know, I think hype is a powerful lever for capital. Right. Right now, whether it's in private markets or in the public markets. Same thing. So. Yeah. But I do agree, I think that there's probably more surges out there that we don't know of. But, you know, that might change very shortly. Yeah.
- [27:21]
B
Just to kind of wrap up the conversation, I'll just bring up an example that's completely outside of tech.
- [27:27]
A
Great.
- [27:28]
B
A company that, that I came across recently. It's a company called Jack Marie Mars.
- [27:34]
A
Okay.
- [27:34]
B
Have you heard of them?
- [27:35]
A
No.
- [27:35]
B
It's a sunglass company. Okay.
- [27:38]
A
Sunglasses.
- [27:38]
B
Yeah, it's. Yeah, sunglasses. Yeah. And it's based in la and it was started by a guy, a designer, an industry designer, and he decided to make. He decided to make sunglasses, but super high end. And I think they launched about 10 years ago and they've been flying under the radar. But in recent years they started to get quite a bit of traction because of their pressing, like in the fashion industry and the movie stars wearing and the stylists choosing their sunglasses. But each pair costs like $2,000.
- [28:21]
A
Okay.
- [28:21]
B
Wow. And it's made in Japan and what they do is on purpose, they only make small lots. So like they only produce, let's say 50 pairs of this model or 50 pairs of that model. And then they quickly Sell out. Yeah. So their business model is A, quality and B, less about under hype. Underhyped. But it's leading to hype.
- [28:51]
A
Right, right, right, right. Yeah. Under hype is a very powerful detail. Yes, sure.
- [28:57]
B
Yeah, yeah, yeah. And, you know, it's. It's sort of counterintuitive, especially, like, in a world like fashion where you want as much as hype as possible. But this company, kind of like Serge, decided to double down on quality. And your point earlier, point about craftsmanship. And, yeah, you know, they've been around for 10 years, so it's not short. A short period of time, but, like, that obsession with quality is. Is paying off.
- [29:27]
A
I love that story. Thank you for sharing it.
- [29:30]
B
Yeah.
- [29:30]
A
Well, I'll see you next week.
- [29:32]
B
Yeah. So this is a good place to end quality Overhype. And I think we are leaning towards quality. Even though hype is kind of inescapable.
- [29:40]
A
I think the debate remains open, but I think it's definitely interesting times. I love that story about the sunglasses company.
- [29:45]
B
All right, sounds good. So, yeah, we'll put these links in the show notes, and we'll see you next week.
- [29:49]
A
See you next week.
- [29:50]
B
Awesome. Bye.