Tobi: Hello friends, this is the Alphalist Podcast. I am your host Tobi. The goal of the Alphalist Podcast is to empower CTOs with the info and insight they need to make the best decisions for their company. We do this by hosting top thought leaders and picking their brains for insights into technical leadership and tech trends.
Tobi: If you believe in the power of accumulated knowledge to accelerate growth, Make sure to subscribe to this podcast. Plus, if you're an experienced CTO, you will laugh the discussion happening in our Slack space where over 600 CTOs are sharing insights or visit one of our events. Just go to alpha list.com to apply.
Tobi: Welcome to the Alpha List CTO podcast, where we explore the minds of some of the most fascinating technical leaders shaping the future of engineering. I'm your host Tobi and today with me is my old friend Datran, tuned in from Berlin, is that correct? Dat: Yeah, it's correct. From the beautiful city of Berlin, but it's not sunny today.
Tobi: Not sunny today. Then I actually have a plus here, Hamburg is sunny. So, um, that we thought of a few times already about doing a podcast. So it's like now finally the time, um, that you're, you're, you're one of the partners and, and the CTO at, at, at a consulting company called data nomic, which is quite fresh, right?
Tobi: Um, and you help many companies to navigate, um, through their AI journeys, right? And data journeys. And, and you did a lot already. So maybe, maybe you just, Say a few words, like what you already did, like you worked as a, as a head of, I mean, uh, head of AI at Axel Springer and then built your own companies, now consulted like diverse companies, et cetera. Tobi: Maybe you, you, you, you, you say a few words on that.
Dat: Uh, yeah, sure. So, um, yeah, as you also know, I've been in tech for, for quite some time now. Uh, uh, basically, um, started in some engineering then. Then moved on also do it. Of course, machine learning, right? This is the area that basically I settled in. And then, as you said, I actually moved to Iliado, which is the daughter company of Axel Springer, right?
Dat: And I co headed the data team there. Did a lot, also a lot of open source, which was quite new for a German company. I think at the time it was kind of weird because, you know, we, we just pushed out open source projects in here in Germany and it was like, Oh wow. Okay. You can do it. This is something quite new for, for, for many of them.
Dat: Um, and then, and then, um, over time, basically, since it was also a daughter company of Aksel Springer. Um, I got in touch with the board and I thought, okay, we're actually quite cool to do this also, you know, on a group level because obviously, um, is a small company, right? But how can you do it actually, um, on a, on a group level?
Dat: And, um, yeah, then my role changed quite a bit. So it was more basically, you know, um, strategy doing initiative. Um, but, uh, luckily I also had a, like, you know, like a research team that I could hire. So, uh, that was also pretty good because we worked a lot on. On, uh, text to speech or, you know, especially voice technology.
Dat: So a lot of things that you, you see in, in, in, in tools like Descript or, or, uh, yeah, things like that. Um, or in EvanLabs, right. So, um, and it was. Actually funny. I think, you know, at that time, I think we were one of the first companies that were using transformer technologies in production, because I remember at that time also, you know, hacking phase started.
Dat: It was really hard to write transformers because you had to write it from scratch. So we used TensorFlow at the time, and it was like, you know, TensorFlow under version one. So it was pretty funny still to write all the layers by yourself, you know, now. These days it's so easy. You just use transformer, the transformer library from Hackingface.
Dat: Um, and then, and then you plug it in. Um, yeah. And then as you said, I, uh, I founded a startup, uh, a price group together with Richard Schwenker. Um, and it was very cool in the beginning, so we gave some traction, but then after Ukraine war and so on, it was very, very challenging for us, especially in the, in the, uh, in the e commerce space, because we were doing price optimization, uh, that area, um, and then, yeah, um, uh, afterwards, basically, uh, I took some time to think about what I want to do, like, do I want to go back to startup life?
Dat: Um, Do I want to, you know, go into a corporate life, uh, or just do something new, uh, still being entrepreneur, right? But do something that, that I really liked. And, and what I really liked was basically, you know, like I said, going back to the roots of a little bit, uh, uh, because I did a lot of obviously CTO work, right?
Dat: So it was just not machine learning data, but it was, you know, software architecture, how do you, you know, hiring, how to manage basically, uh, infrastructure and so on. Um, and I wanted to basically go back to, to AI and data, right? Because I still believe this is. One of the, it is still right. One of the core technologies, um, that we need, and especially to win in Europe.
Dat: Um, and I see that, you know, many, many companies here are still struggling, uh, how to make use of it. And yeah, in some form. Tobi: That's a good mission. So I would, I would now put you in the bucket, um, to start off with a joke. I would put you in the bucket of the 20 percent of AI experts, uh, that actually knows what K means means.
Tobi: Um, I mean, in the, in the pre discussion we, we, we, we spoke about that, that like, there are like many. Different directions now, um, like after LLMs really like, or GPT came out, it kind of exploded, like also the, the ecosystem of experts, um, and, and the ones that, that really, I don't know, started early and now TensorFlow from version zero point something, or even, I don't know, Apache Mahout, um, uh, is, is, is, uh, like very rare, right? Tobi: Or relatively rare.
Dat: Yeah, I mean, I don't think it is an advantage or something like this, right, given like this, but obviously, if you have people like me who have been in the area for quite some time now, I've seen quite a lot, right? I've seen the changes. Of the different library changes, but I mean, you know, when, when, when I started basically, um, people were using even not TensorFlow, right?
Dat: So there were, there were many different libraries, uh, then basically TensorFlow established itself more as a framework. You had Keras as well, right? Which was, uh, very cool, actually. So it was the project from Francois Chollet. He actually Uh, then, uh, moved to Google, but he recently left Google and started to actually a very cool new project, right?
Dat: To, to, to, to, um, yeah, creating this, uh, evaluation, uh, metric based or basically task for, for AGI. Um, and, and then, yeah, over time, you know, PyTorch and everything became more and more and more popular, but, um, yeah, I don't think it's a super advantage. It's, it's more that you've seen much more and you've seen that. Dat: Things became much more easier right over time.
Tobi: Absolutely. Yeah. And some things you basically don't necessarily have to know. But, but I mean, it's, it's like with like software development, right? If you know them, like if you know the, the, the, the, the, the baseline stuff as well, it's kind of handy, uh, every now and then, right?
Tobi: Like, especially if you then really have to dig deeper. Um, that's how I see it with AI as well. Um, so, um, another warm up question, like, like everyone whom I showed your LinkedIn profile says, Hey, why does that guy have like 80k followers? How, how, how, how did that go? Like, can you And I did, did you buy them or what happened there?
Dat: Yeah, of course. I bought them, you know, 20, 20, 20 K follow, you know, for, for 20 Euro. And, and, uh, I, I, I, I have, you know, you do remember this in China, they had this Uh, iPhone bots, you know, you, you bought, you, you, you , right? The click pots. You, you basically, uh, paid a certain sum then, uh, yeah. Then basically you have all the iPhones, mining or something like that, right? Dat: Or farms, right? And then, uh, they were doing some stuff with that
Dat: and no, uh, uh, jokes aside. Um, yeah, I don't know how it happens. I, I, Dat: I think everyone asked me like, did you have a strategy or, or, uh, did you really pursue to, to do this? I think it was just random, right? So, um, um, for me, you know, when I, when I, um, joined Udialo I didn't really have much followers or something like this, right?
Dat: So I really love blogging at the time already because I really love to write and share my knowledge. Um, and then, um, you know, when I, when I joined Indialo, one of my strategy was, Hmm, in order to hire, I actually need to be a brand, right? Especially when I want to, I want to have the best people around the world.
Dat: Right. Um, and then, and then obviously I just started to write on, on LinkedIn and, and, um, yeah, I just shared my knowledge, right. And somehow. And it became a little bit about a little bit like a diary, you know, like something like, you know, a notion page where you put your notes in there, right? And then, uh, for me, it was like, okay, when I write, I, I also try to, you know, obviously I read the paper or I read the project or I run the project, right?
Dat: It's for me also to, to, uh, stay up to date. Um, and somehow, I don't know, people liked it. And over time, you know, people just followed me and, and obviously I also didn't share only about tools, but also projects that I did, you know, with the dialogue with my team, um, through that, for example, we also got a lot of get up starts, right?
Dat: So like one of the project, um. I think we have now around 6, 000 stars, right? So it's not a lot like compared to the top popular libraries and some, but it was for us kind of a, yeah, you know, it was some internal tool that we could use within the ILO and we thought, Hey, wow, cool. Um, that thing, uh, or this library could also be used for, for other people, you know, in the wild.
Dat: And then we, we put it out. Um, and, uh, yeah, Lincoln was It's kind of a channel for me to, to help promote, you know, the, the team, the library, you know, and, and obviously at the end of the day, also myself. So it was a, basically a win win, uh, for everyone. Tobi: Time consuming win win, right? But uh, maybe you do that with AI agents these days.
Tobi: Let's let's. Chat about that later. Um, so yeah, um, today we wanted to talk about like the, like most recent, um, biggest development and, and, and hypes that are out there, right? Like there's, there's some waves like, uh, I mean, deep seek being the most popular one, then I don't know. I recently. Uh, unfollowed read Hoffman because he was like, uh, annoying me too much with, with agentic.
Tobi: Then I don't know, productivity for engineers. Um, like, uh, I don't know. I always like to refer as LMS as like the autopilots or exoskeletons for developers, right? Like how effective or what, what, what, what do you think about that? Um, then yeah, a bit about Europe as well. Right. Even if it's like a bit. Um, worn out, uh, and, but, but I, I still think like, um, we, we shouldn't give up on the, on the, on the, on the, on the participating or even like trying to win if it, if it, if it actually is a, a race, I'm not sure, like, happy to get your perspective on that and yeah, um, let's, let's, let's start, right?
Tobi: Uh, let's maybe, um, jump in, jump in a bit earlier, like, Why, why do you actually do what you do? Like, how did you learn about computers at first? Dat: Uh, uh, good question. And also funny thing. I think it was completely random. Like my dad, I think I was seven or eight or something. He saw a computer. And he was like, oh man, this, this, this could be the future.
Dat: And then he just bought me like a computer, uh, at home. Um, and then, uh, yeah, it was funny. I think it was Windows and t or Windows 95 or something like that. Um, and, and I don't know if you remember, I mean, you still had those, uh, dis, right? Like six disk, uh, and then one cd and then you had to put on the disk first to, to boot the machine and then you can install, uh, windows on that.
Dat: Right. So, uh, I somehow got very early in touch with computers, um, somehow, I don't know, uh, yeah, there was no particular reason because my, my family, um, actually didn't study or something like that. Right. So they, they, they, they don't come from an, um, academic background. Um, and then. Yeah, we're time. I, um, I don't know whether this is a stereotype for Asians, but, uh, I played a lot of games, you know, Starcraft and things like that.
Dat: So, uh, uh, got a lot of gaming. And then obviously, you know, when you, when you, uh, a little bit dirty on the side, you also start to program, right? So obviously I look and Oh, wow. What is coding? Uh, I started my own website as well when I was pretty young. Um, and that is actually was pretty interesting because this was a time when you didn't have those fancy frameworks, right?
Dat: Like today, like you didn't, you didn't have things like, uh, react next Jess or, or you, but you really had to build your own websites. And, and, uh, I was somehow also a lot into, um, Yeah. Like PHP boards. So, uh, like in, in Germany, I think it was a trend that everyone should have their own community board and I set up my own, Tobi: right, right, right.
Dat: Yeah. It was kind of weird. I set up my own page and I, I, I played around quite a bit, you know, to modify the board, uh, I know it's kind of weird, like thinking about this in the past and, um, yeah, and then over time, I actually, so as I said, I actually, I was a bit nerdy, but on the other hand, I also played soccer and so on, right?
Dat: So, um, at some point I kind of lost I was basically interested in that, because I also didn't know what to do with that, right? So, um, I think it was around 2006 or 2007, um, when I was like, hmm, okay, uh, what do I do with computers, right? At that, at that time, it was not like, uh, like a fancy job, uh, like it would be today, right?
Dat: At this time, Basically everyone went into investment banking or, or, or consultancy. Um, and, uh, no one has thought about, uh, doing tech and tech in Germany was always about becoming a project manager. If you're a project manager, you have an outsourced team and then basically you manage the outsourced team.
Dat: You just don't really code. Right. So, um, then, then basically I got. I got somehow sidetracked it to other areas, you know, um, obviously I also work for my, for my family business, uh, but then also was more into investment banking. But at that time, when I was in investment banking, um, it was also pretty funny.
Dat: I, I, I always coded. So I remember there was a night. When, uh, some of the, uh, VP gave me a task, um, to sort some certain lists, you know, obviously if you're, if you're a good, uh, or if you want to be a good intern, you just do it by hand, you know, and you do it until 2 a. m. or 3 a. m. Right, Tobi: right, you do it through, throughout the night, yeah.
Dat: So, so this, this is, this is how they, uh, call it a good intern, right? Uh, but, but obviously, uh, I don't know, I I, it wasn't vb, so I, I knew a bit about VBA, right? And I, uh, uh, uh, look at the documentation and wrote a VBAs script, and instead of like two hours what they expected, you know, it was done in four minutes, , and, and, and I always thought, Hmm, wow, this is kind of a monkey business sometimes what they do, right?
Dat: So if, when, if you, if you use computers, um, you can do, uh, many things. Faster, right? Uh, and, and, and, uh, solve problem faster. Tobi: And if you, if you know coding, especially, right? Um, like it always has been a superpower. Now, like the bar was quite a bit lowered from my perspective, um, in the last three years, right?
Tobi: Um, uh, but, but, but, but still, I think it still is a superpower, right? Like, um, you, you still need to be able to, to, to handle, um, the, the, the output that an LLM spits you out, right? Um, and, and, and, and. Transform it, change it. Um, Dat: yeah, definitely. So Tobi: it still is. Yeah. Yeah. Um, what, what, what are your thoughts?
Tobi: about like recent developments in AI. Like, I mean, DeepSeek, um, was like a big splash. Like funny enough, it was like, there was this splash and you didn't hear anything for weeks and then, then, uh, the waves came. Right. Uh, how do you, how do you see that? Um, and, um, how do you see like this changing the game in longterm?
Tobi: Like also, I mean, there was, there was Lama before, right. The, which was open source and, and obviously like, um, a tremendous amount of others, um, but this was really like, um, a fundamental shift again. Like, what would you say, like, from your perspective, why? Yeah. Um, and how do you see the, the long-term effect, um, of, of open source in this world?
Dat: Yeah. Uh, I, I would say, uh, there were two components where, uh, basically I got impacted. The first one basically, I, I obviously had, uh, media positions, . That's, that's, that's how, how basically Deep seek was was more impacting than, than I thought Tobi: had had, or have. No, I still had, or still have, I still Dat: have, yeah.
Dat: You need a good stop loss and all this stuff. So it's, it's not, it's not a, it's not a big deal. Um, on this side, um, on the other side, um, I wouldn't say it was, uh, basically super, like no one would hype about this. If, if Nvidia would not fall too much, you know, like, because it impacted the stock market, uh, quite a bit.
Dat: Um, obviously I think one of the cool thing, what DeepSeek brought out is because I don't, I don't know if you tried it, right? Like I tried it. Especially they are one model. Um, I think it's super, super cool, you know, to, to have something with open weights, right. Um, so that you can deploy it yourself. Um, what you would actually, for example, have with open AI, right.
Dat: Um, obviously for open AI, you, you pay the price, right. To use it for, for deep sea, basically you can. You can put it, you know, in your own infrastructure, you know, you can control, um, your own data flowing into this, right? Um, and yeah, you know, some people are saying there's some Trojan or some security leak in the LM and it will send the data to some Chinese servers. Dat: I know people are sometimes, you know, they don't know how machines work, right? If you deploy it.
Tobi: Yeah, yeah, Tobi: yeah, I think that's, um, yeah, that's, that's obviously not realistic as we all know, um, but so you think this, this, this, um, local trend, um, is, is something which is, which is fundamental, um, or, I mean, at first it was like the training efficiency, et cetera, right, but, but, um, do you think also that People understood that this will lead to something way different, far bigger, which is not just like DeepSeek, but like, just, I don't know.
Dat: I mean, I mean, if you, if you ask me what the question, whether DeepSeek will lead to AGI, I don't believe it is right. It is if you, if you, if you look at the basically trajectory where DeepSeek is, right, it is basically on, on, on, on where we are, because if you. Look back in the last few years or last few months, we basically, you know, every week we had some company like Mr or, uh, I don't know, all the other open source providers, right?
Dat: Dropping a new, uh, model, right? The bigger, the better it was, right? Like, you know. like 100 billion, 200 billion, 340 billion and so on. Um, um, and it was a lot about, okay, how we keep more spending more and more and more right to basically build a good model, right? And, um, then people realize, okay, this is not a way to go, right?
Dat: Because you want to have efficient models. Um, and how do you find a way basically to, to train more efficient models, but, um, there's, there's this magic number that people would tell, right. That basically you'd see only use six million, uh, to train the R1 model. Um, yeah, you know, you have to look at a little bit of perspective, right?
Dat: So, uh, deep seek employs 150 people. 150 engineers, right? So just, just calculate the numbers in there. And, uh, I can definitely tell you that 6 million is not the number. Tobi: Not, not the correct number. Yeah. And then you have like a, like a shitload of hardware that you can just use, I guess. And, um, then like, yeah, the question is like, how do you calculate it?
Tobi: Like precisely, but, um, Obviously, it was way cheaper than, um, uh, what, what open AI did, right? Um, but also through like just being smarter and just building on top of something which was already there. Yes, this, this Dat: is, this, this is, this is what, what, uh, you, you hit the right point. It's about open science and open source, right?
Dat: Like I'm, I'm always a believer that open source will bring us more forward, right? With, with, with technology, with evolution and so on, with changes. Um, yeah, there's also a meme about open AI. Because it's more like closed AI, right? Because they, they don't disclose, uh, how they do certain things. Um, but DeepSeek, you know, one of the reasons why also, you know, they spent less is because other people did it before, you know, like the team at Meta, you know, people, people at other labs in the U.
Dat: S. who did open source. They, they, they obviously open source their, their research, right? On, and based on that, they, they basically could use, um, something. But there's obviously one thing that they discovered, uh, by themselves, right? So, um, um, especially for the reasoning part, because this is all, this is pretty.
Dat: Cool, actually. Um, because, um, um, open AI, you know, released this one model, you know, a while ago, and everyone was actually thinking, how did they manage to do this? Right? So, yeah. And, um, what DeepSea actually found out is actually quite, I mean, I don't think I don't know what OBI did, right? Because they don't disclose it completely, right?
Dat: But DeepSeek, you know, they found out, you know, the kind of approach, and I think there was a Twitter comment from one of the employees at OBI, and they actually confirmed, you know, that they also use this kind of technique, right? To basically train their one model, in this case, for OBI. So, this is pretty cool, and because of that, you know, we have more information How we can actually train more reasoning models, improve this kind of things, right?
Dat: Um, and this is actually the advantage of, uh, Open Science. Tobi: Absolutely. And, uh, but, but, um, really what, what, um, what's happening under the hood of, of, uh, 01, 03, Claude and others is like super closed, right? And you can't really compare it to what DeepSeek now did, right? Dat: Yeah, I mean, it will be challenging for me, right? Because I don't know what OpenAI and Anthrophobic is really doing.
Dat: I mean, there are obviously technical blog posts, right? Like a little bit, um Like showing a bit under the hood, right? But we don't know exactly, right? Uh, but if, if we, if we look what DeepSeek is doing, we know that they are using reinforcement learning technique, right? So there's a technique called, um, GRPO, like creating, um, I've forgotten the full name of it, but it's a policy, like a policy technique, right? Dat: That basically improves, um, the kind of learning in this case.
Tobi: And, um, do you think, like, if you compare, like, the results, um, of R1 to, like, the, the, the others out there, to, like, O3, Mini, et cetera, um, Cloud 3. 1, Sonnet, like, um, you feel that, um, there are any advantages of, uh, I don't know, still Uh, using Cloud 3.
Tobi: 5 Sunet for, for coding or something like that. Or do you think, or where does that come from? If that advantage is still there, does that come mostly from training data or because the, the, the, um, um, the, the, the use cases were like really like optimized and partly also the UIs were optimized to kind of really test the code, et cetera, or like, what do you think is like really like an advantage for you to use one or the other model these days? Tobi: Um,
Dat: I mean, I think the question that you're also like you were referring to is also about this leadership boards, right? So, uh, there's a lot of this LM arenas and leadership boards, uh, where every few weeks, you know, one beats someone in math, someone beats Tobi: the other and then you have an AGI. All of a sudden there comes like an AGI benchmark, which then someone beats, right? Tobi: And the world is afraid. And
Dat: yeah. And I mean, if you, if you know about the arena boards, um, then you, you, you have to take it, you know, with a grain of salt, um, because for the coding challenge, um, yeah, it's not very complicated in a way. Right. So it's, it's, it's mostly. Challenges that that is also easily solvable, you know, if you're, if you're an engineer, right?
Dat: Um, um, and, um, and then it's also basically, yeah, using very popular, uh, frameworks, you know, like in this case, for example, Python, um, if usually it's very hard to compare different LMS, you know, when you think about production code, right? Um, or basically maybe languages that are very exotic, right? Um, then then it also becomes harder because I realized that, for example, um, when I use one of the lamps for for to develop a Microsoft word edit, right?
Dat: I realized that some of them are doing worse than the other. And this is also because of data availability, right? So obviously some of the providers, they have more access to data, right? Because, because yeah, you know, we're probably in a, in a gray zone. Right. I don't know how they did it. Right. I don't think that all of this stuff is legally obtained by deals or so obviously not, obviously not.
Dat: Yeah. So, so, uh, I don't know, right? We're we're we're judged. We're basically just speculating, right? But probably obtaining by a scraping, you know, some of them might be legally obtained by deals, you know, and data access. Um, and some of them basically just have more access to data, right? Also, because Um, they are, they have more money and they have more partnerships, you know, with, with bigger ones, for example, um, yeah, I'm not sure if, if Microsoft, you know, would give, get up data to open AI, right, uh, for training, like maybe they do, maybe they're not, but if they do, but probably, you know, because I compared it, for example, open AI, um, With, um, with, with mistry, you know, in some cases only I was just doing better.
Dat: Right. And, and I, I had a feeling that was also because due to data availability, um, if you compare now, like thinking about, okay, now deep seekers out, should you use deep seek, um, you know, for your coding things? I think it's up to you in a way, right? Like, if you are happy already with CLAW 3. 5, you know, in your cursor and it's doing a good job already, why the change?
Dat: You know, like, it's also a lot about user experience. For example, for DeepSeek, Like, I'm not sure if, if Coursera already supports it, right. Uh, if not, you know, you maybe need to deploy it with Olama machine or you deploy it somewhere else in instance, but you know, people just don't think about it, but deployment is very costly.
Dat: You need someone to take care of it. You know, you need to upgrade versions, maybe something breaks. Right. So to me, it's sometimes it's also about convenience, right? You, you just pay for the provider because. You know, it's just easier because otherwise you should spend your entire time on, on, on basically, uh, DevOps stuff.
Tobi: Right. But I mean, on the other end, it's not super complex, right? I mean, that's what I think many people, I mean, apart from obviously all my listeners don't know, um, is yeah, just the fact that you can just deploy Olama, right? And you can just like download Docker, um, and install a Docker image and that's it.
Tobi: And then you, you have a ready made API. You can use it in your tools. You can use it in your, in your products actually. So the, like from, from my perspective, the, the, the, the barrier of entry. is super, super low in this, right? Um, and, and that makes it so fascinating. Like, um, I don't know how, how does that look like in two years? Tobi: Right. Dat: I Tobi: think for
Dat: tech people, definitely. Right. But, uh, I don't think that, that, uh, outside of our bubble, basically people, uh, use Olama or, or, uh, some other stuff. Right. Um, and, uh, but within our area, I definitely think, yeah, it is, it will be the goal, you know, that you have your own computer and computers, you know, like the.
Dat: The MacBooks, it's crazy how, how powerful they have become right when I think five years ago, I thought, you know, 16 gigabyte was quite okay. You know, it was super powerful over five years. If you don't have 48 gigabyte. It's nothing, right? You have so much stuff running on your, your, uh, docker container and so on.
Dat: Um, and then the question is, okay, how do you get access on on a GPU, right? Because, uh, make book. Yeah, I think they have some kind of this neural engine and so on, but obviously it's not an immediate graphic card, right? So most of the, uh, LMS are also optimized for this. Um, and, um, uh, yeah, I wonder where this will go, right?
Dat: Where, where they're basically. Personal computers, personal GPUs, for example, you know, the, the one that NVIDIA also announced, you know, with this like, well, personal desktop thing, I think around for around 3000 euros, something right, which you can put, you know, in your home Tobi: and Dat: you have basically a supercomputer at home, which is super cool, right?
Dat: So you can run stuff there and you can run your own Bye. Uh, speech to text and text to speech as well, right? Like a speech to speech model with an LM included in this. You don't really need, uh, basically, uh, yeah, Alexa and all of the stuff. You can basically build it all yourself. So I think it's fascinating.
Tobi: Yeah, I like that too. And I think it will basically, I mean, it will progress, right? And at a certain point, you'd have like, uh, the, the, the, the compressed knowledge, um, in quotes, um, of, of the world on, on a, on a box of a cigarette size, right? Um, that, that's, and, and that's, um, most likely. Dat: Yeah, definitely. I mean, this is the cool thing.
Dat: Technology always becomes more accessible over time, right? So, uh, I remember, as we discussed earlier, right, when I started, basically, machine learning, the AI was not really accessible, right? It was, uh, Very hard, probably before even me, it was even more harder because you had to write C code and, and, and, and program everything.
Dat: Uh, at my time, it was already more convenient, but obviously it was difficult because you had to, you know, change different, uh, uh, notes, you know, so, so that you have your network and then train it now, basically why it's so simple, you know, just plug in your data set. Just do whatever you want to do. Right.
Dat: And then, and then you get an output and then there's so many of this, uh, yeah, no code, no code tools, right. That makes AI so accessible. So, so that even, you know, business people can use it. Um, and, and, uh, yeah, another funny story. I. Uh, recently it was an event, um, where, uh, um, I, I talked to Hannah Schwer, I don't know if you know her, she, she's, uh, she's one of the, um, yeah, editor at Capital and, uh, and, uh, basically talking about, uh, tech, you know, uh, and she, she told me, yeah, wow, that, you know, I, I, uh, was able to program my own website on my own, you know, with an LM, um, Like, this is so fascinating because yeah, he's a journalist, probably five years ago, people would not have been able to do that.
Tobi: Yeah, Tobi: and many engineers, I think, are also afraid that they might lose their jobs in the near future. Um, but like looking at it, I don't know if this is true, right? Um, I think like, I mean. Generally, they're like, there's AJI, et cetera, like, no matter how far it is away, then like, most likely, like, everything will change, right, and everything also changes on the way.
Tobi: Um, but, but, um, I think this leads to another level of complexity because, uh, like tech will be basically everywhere and, and AI will be everywhere, um, and this needs to be somehow managed. So, right. Yeah. Yeah, I think, Dat: I think there's, there's, uh, always two sides on this one. Right. Um, I think, um, in the short term, probably with all this coding assistance and so on, um, and maybe also because of the, um, economic environment that we are in right now.
Dat: Right. Um, I think it, it will be very hard for juniors or, or basically people entering the industry, right. To find a job because, um, for now, you know, if you are a senior engineer. Why would you need someone like, like, because you had always a junior to manage certain stuff, right? Or write a Docker file because it's serious tasks, right?
Dat: Like, okay, please manage the, the develop stuff because it's mundane now. No, basically you don't need that. And right. So you have a coding assistant, just write, okay, please, uh, create me that Docker file for AWS and, uh, so that I can deploy it. Did you do it? Right. So. Obviously, now, basically, I think it can be tough for some engineering level, right?
Dat: But as you said, in the future, I don't think that, yeah, software will completely be rewritten by AI. I mean, I use coding assistance quite a lot in my tasks, right? And the quality is not so bad, but it doesn't solve Complexity, right? It doesn't solve dependencies, right? It doesn't solve architecture, uh, and and also business requirements in a way, right?
Dat: So you still need to ask the right question. You know, you need to put it in perspective. Um, and you still need to combine it with with the right technology, right? Uh, so, for example, with cursor. Yeah, you can add multiple files to it, you know, and it identifies basically, okay, the, the, the dependencies in there. Dat: Wow. How do you connect it to the database and everything else and so on, right? How do you deploy it and so on? You still need to do it.
Tobi: Yeah, well, and even if that is solved, well, then you hit the, the, the problem of the context window at a certain point as well, right? Yeah. Which is a bigger problem right now, which will be somehow.
Tobi: Uh, like solved, uh, at a certain point, or maybe, maybe, maybe there will be workarounds. Um, but, but, but, but then, yeah, um, it, like a long term management of something is still, um, I think of what you build is, is, is still a bigger problem. Right. And it's, it also got better. Um, I don't know, throwing in a code that you have written like years ago.
Tobi: And, and, uh, really refactoring it, um, really like building on top of it, uh, using cursor, et cetera, is, is, is way easier now than two years ago. But, um, yeah, I think the messages that you still have to stay, um, at the pulse of the time basically, right. As, as always, as this was always in the case, it just accelerated a bit from my perspective.
Dat: I mean, that's, that's a, that's the good point that you said, because Um, um, you know how engineers are sometimes, right? Engineers are always about, I love this programming language, right? I love, I love Rust, or I love Scala, or I love Java, or I love Python, Tobi: and so on. Dat: Um, with the whole LM game, it's not just about the programming language anymore, right?
Dat: Because, uh, the programming language. You know, you can easily get in by the language model, right? Uh, and, and basically, uh, write things very fast. It's more about what, what is the right technology that you can use? And if it's not the right technology, technology. How fast can you move with it? With it?
Dat: Right. How fast can you migrate? Hmm. Like, just look, just looking at the perspective we have, uh, in our industry, right? There's still so many old languages out there, like, like, uh, cobalt or, or, uh, , cobalt, you know, as, as a language and so on. I, I, I just still find it fascinating, right? And, uh, you know, I believe that, you know, with, with a language model, uh, basically train on, on the right language, right?
Dat: Um, you are able to basically do this migration projects much faster. Tobi: Yeah, migrating from A to B. Also, I don't know, migrating data, like if you have like, let's say, event loops, and you want to transform A to B, that's also like super helpful where LLMs really can, can, can, uh, can move you much faster. What do you think, like, about Um, like productivity of engineers. I mean, we just like briefly touched it, right?
Tobi: Uh, like you don't need juniors anymore necessarily. Uh, or maybe everyone's a senior straight away. If you learn programming, then maybe you're, you're a senior straight away, um, or on a good level. Um, to what degree do you think it's, uh, what, what, um, Or how significant from your perspective is the increase of productivity if you use, I don't know, Cursor, for example, did like, do you personally use Cursor and, um, how much does it help you?
Dat: A simple question. Yes, I use Cursor, but I also use just VS Code with Copilot, right? Depending on what I do and the mood with the editor that I'm in it. Um, you know, sometimes going here, um, but in terms of productivity. It's definitely makes you much faster, right? I already said it, right? Um, for me, um, it basically changed the game because, uh, obviously I'm, I'm not the super expert in front end, right?
Dat: I know pretty well about backend DevOps, you know, and machinery stuff. Uh, but because of LM, you know, I can easily, you know, do front end work and, and because I can read it, you know, it, it is fine and, uh, and LM basically supports that or so mundane tasks. You know, it is rare if you need to write a Docker file from scratch, right?
Dat: Okay, what do you do? You go to Google, you know, and you, or you go to Docker Hub and you search for the right one. And then you need to basically, you know, change certain stuff. How cool it is. You know, you just say, okay, I just need this base image. I want to do this and that. I want to copy this file to this.
Dat: And it's done basically in a couple of seconds, right? So, so basically, you know, if you just accumulate all those tasks that you would need, maybe 10 minutes, 15 minutes just to write a Docker file, right. Or, uh, from scratch, right? Because otherwise you would just copy and paste and then, uh, change certain stuff and so on.
Dat: Um, versus basically just telling the RM what to do. It's so, such a powerful game, right? Uh, similar, similar to, to that is also SQL. Yeah. You know, I, I, I, I have, I have always, you know, SQL is always. You know, you know, you know the basics, right? Select, uh, statement and, and, and, and, and filter and all of the stuff, right?
Dat: But when it comes to basically very complex stuff, you know, window function, I used to spend on window functions sometimes half a day, right? Because, uh, I had to think about this, you know, like how to do that. Uh, and then sometimes, uh, there was. It's something, you know, missing, or there was a bug in the code, it's done in one minute or less.
Dat: Right. So then I can just read through it. And then, ah, yeah, you know, that's right. You know, this is right when he came up and, uh, basically, you know, I saved half a day on certain things, you know, just think about this, how cool it is, right. Tobi: So the true full stack engineer with data capabilities is now born.
Dat: Yeah, kind of. I would, I would say it's not for everyone, right? Um, because I think you mentioned Can, can, can any engineer become, uh, like a bad engineer with an M? I think it can, but it depends on the quality of, of your level and your seniority, I had a feeling, right? Cause, uh, obviously I also interview a lot of people and, uh, yeah, sometimes it's, it's more about like.
Dat: They just don't understand anymore what, what, what the basics is. Right. So, so as I, as I said, told you, yeah, you know, I asked the LM about SQL and, uh, and for a complex stuff. Right. But then, you know, I read through it and then I, okay, this is how I also had this in my mind already. Right. But for many people, it's just like, okay, just copy and paste the results, paste it in. Dat: Right. And I just don't understand what the code is doing anymore.
Tobi: And then we come to AI agents, the best topic in the last weeks, I think, um, and I think every, I don't know, even an executive kind of jumped on it, uh, like really thinking about like, Hey, can we replace big, a big portion of our workforce with AI agents? Tobi: Um, what's your take on that?
Dat: Yeah. I mean, um, I think it's, it's a really cool area that we're getting into. Right. Um, but, um, in our pre discussion, I always said, I just, I see agents as a, as a follower or, you know, as the next, uh, RPA 20 RPA, like, um, what is again, um, um, robotic process automation Tobi: process Dat: animation.
Dat: Um, this was this entire area where basically some companies built on that, right. Where basically, Okay. You have some processes, right? Uh, from A, B, C, to D, right? And, and you had this, this kind of workflow builder where you did it in code, right? Um, and you had to encode this with a lot of rules. Um, agents make these things a bit easier, right?
Dat: Because at the end of the day, it is a, it is an LN. This is how I call it. That's some function calling, right? So, so, so this LM can, can do some functions, right? Uh, and call certain APIs or execute certain code, uh, based on what you give them, right? Um, and this is obviously completely different from, from RPA because in RPA you run in a certain environment, you code the rules and you don't leave the, the environment, right?
Dat: Whereas as an agent. If you don't specify, you know, what kind of permissions this agent can have, right? So if you, for example, give the agent, let's say, all the permissions to delete your production database, right? Things can go wrong very much. And right now We are, I would say we're in a super early stage of agents, right?
Dat: There's a lot of use cases in place, you know, where, where people, you know, uh, the prototypes, um, there's also, um, no code tools like in an end. Right. So you probably know it. It's like this, this, uh, this flow. Tobi: Like a make. com or Zepi. Dat: Zepi is the same, right? You, you can combine these flows. You can call the M based on this output.
Dat: Right. Um, You can basically, um, build like an, like, this is what we call like an agendic workflow, right, based on this. Um, I think, I think this is pretty cool if you, if you can do this. Um, when I think about, you know, if I would implement that, I think about evaluation, right? I think about monitoring, uh, I think about, is, is the agent really doing what I'm, what he's supposed to do? Dat: Yeah,
Tobi: right, right, right. It gets like super complex, right? If, because, um, uh, like, like typically those, those automations in, in Make or Zapier, they have like a, Defined start and a defined end and, and, and also defined steps in between. Right. And with, with agents, this can change, right? Like you have capabilities and then you throw something in and then, or you turn it on and then, um, like, I don't know, A talks to B and B to Z and et cetera, et cetera.
Tobi: So it becomes, um, more than, uh, just, uh, just, uh, just a funnel that stuff flows through. Right. Dat: Yeah. This is, this is, I mean, this is what I also say that. LM has changed the soft engineering, uh, interface quite a bit, right? Before that, um, for a lot of the stuff we would build rules on top of this. Now, I think for most of the things we can use an LM and we can build these things very fast, right?
Dat: Like we can build this in a, in a And I would say, uh, 20 percent of the time that you would build on top of rules because words are complex. You really need to adapt to everything. And then you need to write test cases along the side with an LM. This is much faster. But then the problem is that the rest of the time you spend much more time on building the evaluation metric on top of this.
Dat: Right. Like is, um, uh, you know, doing what it's supposed to do. Do you, do you, uh, put on the right permission and restrictions to this? Right. Um, and then, uh, if, because, because we know, I know that language models are generative models, right? So. There's no deterministic, uh, component here, right? Obviously, you can, you know, work around with temperature a bit and so on, right?
Dat: But at the end of the day, every result might be different, right? And maybe it works for 100 test runs, but the one of the first test runs, how do you cover that, right? And if it's a critical workflow, for example, you know, maybe in sales, if you need to write Back to your most important customers, and it failed at that moment of time. Dat: Yeah. You know, can you really trust this?
Tobi: Right. Right. Like if money flows in, things are okay, right? It would be a good evaluation, but if it stops, it could be bad. And like, what, how would you like, in what cases would you, would you actually use? agents these days. I mean, they're like, there's like, it's, as you said, it's like quite early still.
Tobi: There's like, uh, well known services like crew AI, relevance AI, and, and a few others. Um, uh, and, and you can actually. meaningful things with them. But, um, I also feel like whenever I touch it, like it's, it's still quite early and, uh, like half of the stuff doesn't work, like where would you use it, um, especially in a, in a business critical set up that, that actually makes sense, like any, any, any learnings on your way
Dat: so far, as I said, it's more proof of concept that I've seen so far. Dat: Right. So, um, I know that. That some people are using this, um, for example, for, um, CRM or advertisement, right? So that, that, for example, if, if, um, if, uh, um, someone sends you an email, right? Uh, then, then basically, uh, based on the, the input of, of the text or the reply of, uh, or, um, of, of, of the email, you know, you decide, okay, based on this, you know, send this back and so on, right?
Dat: So this is, this is possible. And, um, Um, but as I said, when I, when I talked to one of the person where she did that, yours also, yeah, everything was about evaluation. Right. So we had to evaluate this, this quite manually. And, and, um, yeah, you know, there, there always, I think there will always be use cases where you can work it or work with this.
Dat: Right. So there can be also use cases, for example. In finance, you know, if you do a booking, you know, uh, double booking or something like that. Right. I definitely think that if you are able to, to work with data or this, uh, financial tours, right, you can, you can build an agent to read from a, from a invoice. Dat: Right. Um, and then based on this invoice, do some bookings into your systems automatically, right?
Tobi: Fill it into your systems. Yeah, that's that's that for sure. That's for sure. Like it's, it's, it's also again, our classification, right? Um, Dat: yeah. And, and, and this is where I think will, uh, will, will be very useful where you, where you can have a bit of, um, I would say verification that you could do, right?
Dat: Because, because, because for example, um, if you do an invoice, You know, obviously the agent will not do completely everything, but you will basically extract all the information and then the agent will, okay, after I had got all the information, I put it probably, I don't know, into, uh, like a CSV, the CSV will go to the system and then it will go into our, our booking system and then you can do the check, right?
Dat: You can check Is this the right sum that, that, uh, that, uh, the, the lm uh, transported right to the system, right? Mm-hmm. And then you can check both sides, right? Um, I wouldn't trust it if I don't really have, yeah. If I don't have an easy way to validate that, right? So even, even the CM case that I heard, Hmm, uh, I would be doubtful if, if I, if I would let an LM generate responses completely.
Dat: Um, but maybe if you do it just for the routing, so, you know, like you have an agent that just decides which template the, the, the, the, the email, you know, you have, right? And then which email needs to be picked so, so that you can reply back to your user, right? Tobi: Yeah. Or you just, I don't know, classifying your emails, right?
Tobi: That's again, classification, right? Um, like just like having my emails labeled, um, in the morning is. I mean, also nothing really new, right? Uh, but, but something I would, I would trust it to do. Dat: And you can, and, and, uh, if you know how to use it, you can do it yourself, right? I mean, this is, this is how, what some people already built, for example, with NNN or make, right?
Dat: Like, uh, read from, from Gmail. Uh, classify that, for example, as, as business or non business, right? Uh, and if this is a business and, and if it has a, uh, calendar entry, right? I put it into my calendar, block my calendar with that, right? So this is a valuable, easy use case that people can always build.
Tobi: Absolutely. Absolutely. Yeah. Yeah. Thanks for, um, like, like your perspective on that because like if you look at LinkedIn, if you open LinkedIn these days and you follow a few, um, uh, a few AI influencers, then . You, you can, you can easily get afraid. Right? Easily become afraid and, yeah. Yeah. You know, we know Dat: we always need a, we always need a hype. Dat: Last year it was about, uh. Right. And this year is always about agents.
Tobi: Right. Right. Right. Um, and let's see what, what comes next. Right. Um, also how those terms like quickly then shape, right? Like, I mean, like, uh, I would say like half a year ago, like almost no one knew about agents and then like it quickly shapes up because someone basically said, Oh, that's an agent. Tobi: Yeah. Yeah. Yeah. Fascinating. Um, do you already see like, uh, the next big thing or? Yeah, slowly,
Dat: I don't, I don't think, I think right now for this year, we're, we're still, you know, have this agent topic, right? Um, and, um, I think it's also good because for example, um, even if you think about this rock topic from last year.
Dat: Um, not everyone has implemented or basically knows what it is, right? So, so obviously everyone knows how to build chatbots these days because, uh, yeah, that's, that's a cool topic. And everyone, everyone wants to have a chatty BT interface and that's why that has become such a topic. It's also an issue. No, not everyone needs a JDBT interface. Dat: Not every business needs a JDBT interface, right? Yeah.
Tobi: It's first sounds logic, but then I like often people do, yeah. Or it's, it's just don't need it. Yes. Dat: And, and, and especially, you know, if you are. German miller stand, you know, and you produce tires or something like that, right? Like how, like, uh, why do you need a chatty PT interface on a tire? Dat: It doesn't make sense. And, uh, yeah, Tobi: chat with your tires.
Dat: Yeah. And, and yeah, you know, the gigantic space will still be a place. Um, obviously A trend or I don't know whether the trend but people are still working on that, right? So, um, we still want to have like, um, smaller models, right? So, uh, the models are still very big, um, running it, um, requires a lot of energy, you know, and also, um, infrastructure, right?
Dat: So this is obviously still. What people will continue working on, um, different models, right? So basically, transformers is used for all of these language models, basically, but obviously like the main one, right? Um, there are other, um, I would say architecture, right? Um, that is developed at the moment. You know, you have, you have, uh, for example, stuff like a mamba liquid net, you know, you have XSTM, um, that, that is built, you know, and, um, this is also a direction that, yeah, how can we make basically transformer much more like all different architecture, right?
Dat: How can we learn more? More efficient in a way, right? This is, this is a topic that is very important and still going on. Tobi: So, um, uh, like briefly touching, um, Europe versus us versus China. Like if you were like the Mistral founder these days, what would you do? Dat: Well, yeah, tough question. Uh, I mean, I think, Mistral guys are doing a great job, right?
Dat: So, so I think they, they took the right time, the right spot, got a good funding and so on. Um, but obviously if you, if you think about the customer base and what they do, I don't know whether it's a European company in a way, right? It's based in Europe. The founders are European, but obviously for me, I am for, for, uh, and through and drop it.
Dat: There's also many European engineers, right? So, so, I don't know if the question is correct in the way, Hmm, is it a war against Europe and US, right? Because if you just think about the Western Front, Um, um, it goes one hand in hand, right? Um, and when you think about MISRA in general, Um, why I think they're doing a good job is when I look at the customers, most of the customers sits in the U.
Dat: S. Right? So probably the question that I would ask myself, would it make sense to stay in Europe, right? If most of your customers actually sits in the U. S. Um, I'm still happy that they are here in the way. So they are basically holding the The European, uh, uh, flag kind of, right, like, uh, it's one of the few who is actually doing state of the art, uh, uh, research in this area.
Tobi: A few companies that are based in, out of Europe, right? I mean, that's, um, I think like a thing to always think about, like, I mean, if you look at Uh, Zurich, for example, or TU Munich, right, ETH or TU Munich, then there's a lot, a lot of, a lot happening there, right? Um, and a lot happening, uh, like four companies that are actually US based, um, like being, being, being down there.
Tobi: So like, does location matter is the question and does like origin matter, um, ultimately. If you now see open source, like from my perspective, it does not so much. Yeah. Dat: That's what I'm saying. You know, like look at Google, they have offices in Munich, you know, they have offices in Zurich and so on. Um, obviously if you think about, uh, from a, from a German or European perspective, um, it's not a European company, right?
Dat: So they pay tax in, in Ireland in a certain way, right? It's not, it's not the full contribution. Uh, and obviously if the U. S. decides at some point, Okay, we don't like Europe, right? Then, then we're losing all the services that we are used to, to have, um, and, and that would be obviously sad for, for, for a European, uh, customer, right? Tobi: But they could also use, uh, or lose very good engineers, uh, working, working in Zurich, right? Uh, so, I mean, it's a two sided risk.
Dat: Yeah, I know what you mean. But in, in, in times like that, you know, where terrorists are easily, uh, kind of, uh, you know, where you, where you can put terrorists easily, it's, it's not so trivial to, to think about this. Dat: Right. Um, Dat: I think, I think
Dat: the topic of sovereignty, um, is still, uh, is, should be important for you. Right. Um, I think your, um, needs needs kind of a change and it's not just about AI and, uh, in general, it's about innovation. Right. But risk taking, you know, um, and the question is, do we have enough wake up calls?
Dat: I don't know, right? I don't know, uh, like, like, uh, how, how people will react to this, um, that we are losing ground, you know, I mean, you know, all the stuff about saying about Germany, you know, that in the last next decade, we will be zero growth and so on. Um, and that can happen, right? We have a lot of issues.
Dat: We have a lot of problems. Um, we made a lot of mistakes. We are not engineering driven, right? This is really sad, actually, kind of, because, uh, Germany was always about, you know, that I loved it. led by engineers and so on, right? These days, companies are, are led not by engineers, right? And, and, and, uh, and, and, and, and don't get me wrong.
Dat: You know, you, you always need business as well, right? Engineers need business accruement in this case, but, um, we made many mistakes. Yeah, not having right tech. You know, we, many managers don't understand tech, right? They don't understand what it is. It's, it's more about number. It's a number game. Um, and, uh, yeah, the time is kind of over, right?
Dat: If you think about that, maybe in some areas, for example, like automotive, I just give Germany maybe 10 years and then we're dead with industry, right? Obviously it will not die completely, but, uh, it will be not the, it will not be able. to, to, to power our economy and then the question that I always have to ask myself, what can power our economy next, right? Dat: If it's not automotive.
Tobi: Yeah. Yeah. Super unclear, right? Super unclear, unfortunately. Um, let's see. I'd also be happy if it's, yeah, go ahead. I think, I Dat: think, um, I mean, it was, it was a bit of, um, Like, I think it's not completely negative tone, but, um, it's more sad in a way, but I always say how, what can I do to, to help, you know, or what can I do to, to, to make a change, right?
Dat: And I think everyone maybe who listens to this or isn't in a certain position should ask themselves, how can I wake up tomorrow, right? And do the change, right? For you, right? And this, this, this, this is what I, what I would like people to have more, you know, being more. More positive about this and it's not over, right? Dat: We are we have very we are still the third strongest economy in the world, right? You know, I grew up in Vietnam and you know in both worlds, Tobi: right?
Dat: Thinking about Vietnam, Vietnam has such a crazy spirit, you know, they're super entrepreneurial and so on and this is What we need to learn, you know, we need to look we need to look at Asia. Dat: We need to look How fast the Chinese are. We need to look how fast the Vietnamese are, right? We need to copy this, right? We shouldn't be afraid that they copy us. We're far behind. We need to, we have, we need to copy Asia.
Tobi: We need to catch up again, basically. And, um, what, what do you do to change this? Like personally?
Dat: Well, personally, you know, obviously, uh, since I, I have a, I have a network, you know, um, and, uh, I can reach out to people, I obviously talk to them, right? And obviously, you know, try to change as much as people as possible. Um, and for me personally, you know, um, that's why I, I, I decided to, to also, um, be, you know, more like as an external consultant because, you know, when I was at Axel Springer, I thought, wow, this is super cool already, but you can only change one company.
Dat: Right? So if you are more free, how can you actually, you know, talk to more companies, you know, uh, work with more people and do the change, right? Obviously there's also a lot of initiatives that, that, that I could support at some point, right? But I would always say you need to choose your battle.
Tobi: Thank you. Um, so, um, we, I just realized we only have three minutes, um, like I, I have a few questions left. And I mean, obviously we could chat for three more hours, uh, as I just realized, right. Like, um, looking at all the questions I had, uh, and what, what we came up with, but I really like, it's like, like this stuff.
Tobi: It's really beefy. Um, but maybe like slowly coming to the end, like what, um, like you post a lot about tools. Um, anything like super, super cool. I saw you yesterday posting about web UI. Anything super, super cool you recently discovered that you'd like really using every day and that you think like everyone should know about?
Dat: Um, I mean, we already talked some of the stuff that, that, uh, I use every day. Right. So I, I use a cursor a lot, right? So I find it super amazing. It's not, it's not just because of the language model, it's a whole user experience, you know, that you can, uh, paste code and change, accept and reject. Uh, on this.
Dat: So. They've done quite a good job on this one, right? Because, you know, other than that, if you, you would have just placed code, you know, on, on your, on, on chattybt, and then you need to look at the divs, right? This is super annoying. Now, basically, with, with Coder, uh, everything is quite good. Other than that, um, yeah, uh, you, you, you said it yesterday, right?
Dat: Like, um, I'm, I'm very, uh, even though Agents and Identic Workflows and everything else is, like, super hyped, Um, I think the direction is really cool where we're going, right? Like, like, for example, with browser UI, web UI, um, that, that I, that I posted. So I think, um, the, the future is how do we actually, you know, interact with, with browsers, with operating systems, right?
Dat: You know, in a certain ways. Um, and how can we control this more, right? I mean, OpenAI released this operator. Uh, it also goes into this direction quite a bit, right? And I think this will be the future. Tobi: Did that before. Yeah. Absolutely. Like, uh, looking forward to someone finally controlling my mouse and LLM finally.
Tobi: Yeah. Yeah. That would be nice. Um, so, uh, final and outro question, um, which is a little surprise, um, I mean, you, we spoke about Idealo and your early days there, um, and I actually know Martin Zinner, the founder of Idealo. I guess you know, know him quite well as well. And, and, and he told me about like an early thing you implemented, um, as an Easter egg into, into the Idealo search, um, back, back in the days.
Tobi: Um, and it is called the time picker, AKA time machine. And it actually lets you physically travel back in time. Uh, so you were super crazy inventing that back then. Um, and now like we hit the year 2015, um, And travel back in time, like imagine that. Uh, and we see yourself like just having started as head of AI at Axel Springer. Tobi: Um, what would you, if you now had the chance, whisper into your young self's ears? And, uh, let's, let's say we exclude buy more NVIDIA shares.
Dat: Wow, that's a tough question. Um, yeah, I don't know. I think I wouldn't change anything. So, uh, even, even The ups and downs, uh, uh, taught me, or basically, uh, taught me a lot of lessons, right, to who I am today. Dat: Um, and, uh, yeah. Obviously, I wouldn't actually buy more shares, because, yeah, that's, that's, that's unfortunately that you excluded those questions. Um, if not, then I would buy more Palantir shares then.
Tobi: Palantir. Okay, okay. So, uh, let's, let's all look at the Palantir stock now. Thanks a lot. That's really fun talking to you. Tobi: Yeah, we should, we should just record another episode next week. So let's do that. Dat: At some point we can, I mean, if there's a new topic, I'm, I'm always, uh, you know, uh, open. Tobi: A, a, I update with that and, and Toby, let's do that. Thanks a lot. Have a good day. Dat: Bye. See you. Tobi: Thank you for listening to the Autholist podcast.
Tobi: If you liked this episode, share it with friends. I'm sure they love it too. Make sure to subscribe so you can hear deep insights into technical leadership and technology trends. as they become available. Also, please tell us if there is a topic you would like to hear more about or a technical leader whose brain you would like us to pick.
Tobi: Alphalist is all about helping CTOs getting access to the insights they need to make the best decisions for their company. Please send us suggestions to cto at alphalist. com Send me a message on LinkedIn or Twitter. After all, the more knowledge we bring to CTOs, the more growth we see in tech. Or, as we say on Alphalist, accumulated knowledge to accelerate growth. Tobi: See you in the next episode.