Will AI pave the way for a cleaner planet, or deepen the chasm of global inequality? That's the burning question we tackle as we unravel the societal and technological fabric wrapped around AI's growing influence. With special guests Jonathan Robert and Kirill Stanovich from Fever, we navigate the intimate narratives of how personal journeys intertwine with the seismic shifts in energy and technology. From Spotify's data-driven beats to the electric pulse of power grids, we connect the dots between individual experiences and the overarching theme of AI's role in our future.
Our discussion ventures into the heart of the power grid's evolution, where AI stands as both architect and overseer of a more efficient, cleaner world. We probe into the nuances of grid flexibility, the lucrative potential of electric vehicles as grid assets, and how startups like Fever are boldly steering through these electrifying waters. Jonathan and Kirill, with their rich backgrounds from Spotify to entrepreneurship, shed light on the unexpected ways a car battery can ignite a venture and the emotional resonance behind tech's global impact.
This episode isn't just about the machines and code; it's a reflection on the human element within the digital expanse. We examine the tug-of-war between centralized and decentralized energy models, the marriage of cybersecurity with infrastructure, and the pressing demand for talent in software and AI. Join us for a conversation that's as charged as the grid itself, where we peer through the lens of technology to glimpse the road ahead for society, the environment, and our collective well-being.
Follow us on youtube: https://www.youtube.com/@aiawpodcast
Will AI pave the way for a cleaner planet, or deepen the chasm of global inequality? That's the burning question we tackle as we unravel the societal and technological fabric wrapped around AI's growing influence. With special guests Jonathan Robert and Kirill Stanovich from Fever, we navigate the intimate narratives of how personal journeys intertwine with the seismic shifts in energy and technology. From Spotify's data-driven beats to the electric pulse of power grids, we connect the dots between individual experiences and the overarching theme of AI's role in our future.
Our discussion ventures into the heart of the power grid's evolution, where AI stands as both architect and overseer of a more efficient, cleaner world. We probe into the nuances of grid flexibility, the lucrative potential of electric vehicles as grid assets, and how startups like Fever are boldly steering through these electrifying waters. Jonathan and Kirill, with their rich backgrounds from Spotify to entrepreneurship, shed light on the unexpected ways a car battery can ignite a venture and the emotional resonance behind tech's global impact.
This episode isn't just about the machines and code; it's a reflection on the human element within the digital expanse. We examine the tug-of-war between centralized and decentralized energy models, the marriage of cybersecurity with infrastructure, and the pressing demand for talent in software and AI. Join us for a conversation that's as charged as the grid itself, where we peer through the lens of technology to glimpse the road ahead for society, the environment, and our collective well-being.
Follow us on youtube: https://www.youtube.com/@aiawpodcast
very high chance that when technology is locked and in hands of like few people, yeah, even if it's bored of open ai or it's like facebook or like meta or whoever yeah, we're in danger yeah, and it's interesting because on this podcast, I mean, like one of the reasons we started it was to demystify ai.
Henrik Göthberg:But and the underlying idea of why this is so important to mystify AI is that me and Anders have used the word and we have stolen it or framed it, what we refer to as the AI divide. You have heard about the digital divide, but we think it's much more important to understand the AI divide. And, of course, on a geopolitical scale, we all see the problems when an extreme amount of power in terms of wealth is with a few. So I think this is a hardcore down to. One of the key themes why we do this pod is to think about how do we fight the AI divide and what does that mean. Think about how do we fight the AI divide and what does that mean. And I used to put another flavor on this.
Henrik Göthberg:We had Erich Hugo, who is working with DeltaTrack like a Silicon Valley company. His background is South African. Now he also highlighted the AI divide, but he went the other way around. We have talked a lot about the super, super powers or the really AI savvy towards the rest of Europe, but, coming from South Africa, he put the light on the 5G when the whole country is running on 2G, and think about what happens now when you put GPT in the mix and it's a $20 per month approach to even be part of the game. And so he's sort of proposing a lot around these topics of AI divide. We need to really look at both ends of the scale and we need to think about this carefully. Do you follow? Do you agree? This topic of the AI divide is massive.
Kiryl Zhdanovich:Yes, I mean for me, the. The main problem about this is like, when I'm thinking about it, I become very emotional, so and then it's like sometimes this brings me, like you know, bad emotions. I would say. So sometimes, you know, like, I prefer, like in this current, especially current political climate, I prefer to say OK, that's maybe enough. I will focus on renewable energy for now.
Henrik Göthberg:One huge problem at a time.
Kiryl Zhdanovich:Yeah, but I mean it's so much connected so I don't want to go into political, even things, but for me, coming from Belarus, everything that happens now in Eastern Europe is the big thing, and then the AI and problem it brings and powers like other countries have, it's a like. Yeah, I think my parents already lived in some in some country like this, where I don't want to live at all.
Henrik Göthberg:Yeah, and? And what about you, jonathan? What is your take on this? Because we started the conversation with the sam altman. Uh, you know the, the entry point to this conversation. I'll tell us about this. It was. It was Sam Maltman. Part two, episode with Lex Friedman.
Jonatan Raber:Yeah, I was just reflecting on Lex asking him well, what if you have too much power? How can we trust you with that much power? And he said, well, you can always fire me. And then he caught himself saying, well, I actually got fired and that didn't work. He caught himself saying, well, I actually got fired and that didn't work. Uh, no, I, I like, I think the, the perspective I have is, uh, I'm a bit scared, but, um, it's, I think we, we tend to go through these cycles over and over again, like the same with the internet also a big divide like wealth and so on, and I, I think it's, I can't conceive of a structure where people don't end up with too much power yeah, but and and then, of course, this might be the segue into you know, you know, introducing you and all that it's like on the one hand side, we have all these challenges.
Henrik Göthberg:On the other side, we have other types of challenges, like CO2, like optimizing the grid, like we have a flexibility problem we don't have a capacity problem, so to speak and this is data and analytics and AI that can solve a lot of stuff. And I think that also brings us into we can't put our heads in the sand. We need to persevere, is my opinion on this. We need to watch this stuff and, most importantly, maybe getting more and more on board. The real counterbalance is that everybody else catches up. I think that's the smartest way to do it.
Henrik Göthberg:But that kind of leads me into segue into you know, from the AI divide, into you know this podcast where we're actually going to talk about helping you out with a theme the future of power grids and AI in power grids. Is that a good topic? It's a big topic, right, because you can actually talk about the divide in the world who has stable energy in the future and not. So there's another type of you know. It's an interlinked divide in this somewhere. And you can also argue if you don't have the right power grids, you're not going to be able to have the right AI, and vice versa. Yeah, anyway, so that's the theme, and with that theme, I want to welcome Jonathan Robert and Kirill Stanovich.
Henrik Göthberg:And he was happy, because that was a pretty good plan. And butchering, but not too bad. It usually takes five, six tries. I thought that you were approving on my that's okay.
Henrik Göthberg:And you guys coming from Fever, and it's a startup and we're going to talk all about it. But before we get sort of jumping into the and we're going to come back to the big topics I love it. I love it and see if we can connect the different divides right, wouldn't that be cool to think about how they interrelate? Absolutely yeah. But I want to start with you, jonathan. Why don't you give us the short background into who you are, you know, and sort of how you ended up working at fever, or sort of what's your story, into where you are today?
Jonatan Raber:sure, so I guess the uh, uh the nearest history to fever was actually at spotify, so I spent um a bit more than eight years at spot actually, where I met Kirill.
Henrik Göthberg:Yeah, this is where you were colleagues, and I guess you were also colleagues to Andes that hopefully joins later.
Kiryl Zhdanovich:Not sure if it's important to clarification. I was at Spotify two times, Once as the employee, another time as a consultant. First time I met Andes, the second time I met Yonatan.
Henrik Göthberg:Oh no, I didn't know that Cool Nice.
Jonatan Raber:All right, continue, jonathan. I was there for eight years working on many different things. I started out on the business strategy side and then moved over into product development.
Henrik Göthberg:Which parts? I'm just curious.
Jonatan Raber:On growth actually. So I focused a lot on how to grow our subscriber base, on our overall user base, how to increase activation, how to increase engagement and, within that, I focused a lot on the first moments of coming to Spotify, the whole experience where you get the mid, exactly.
Jonatan Raber:So you have a massive amount of people downloading the Spotify app, opening up the Spotify app and my job was to take it from that moment into an active session and into a retained user and it was really interesting because we started thinking about how we could use machine learning in this environment. And you know, obviously we, spotify, focus a lot on how to understand taste.
Henrik Göthberg:So the taste onboarding moment, choosing different songs, choosing different podcasts- Because this is when you're the newbie to Spotify and you start bringing your personal experience together. The first time you don't know how it works.
Jonatan Raber:So Spotify knows that if a new user has a great first experience, listens to a song that they love, that is a very good indicator if they will come back.
Henrik Göthberg:Yeah, so actually how you manage to tailor that experience to what they are, to you know, blow their expectations out of the water, that is a very strong moment of building long-term good feelings.
Jonatan Raber:Yes, and all this needs to be done in an environment where you have very little data about the user At this point in time.
Henrik Göthberg:This is before you have the data.
Jonatan Raber:Yes, how do you solve that? It's very tough. So we did a lot of experimentation even in terms of how to sign somebody up, the authentication moment, and so you kind of have to guess using some of the information you have where in the world are they? What type of phone are they on? Do we have any other kind of metadata or context around that? Even what time of day it is or what time of year could matter, and then it comes down to actually asking some smart questions. So we thought a lot about how to reduce friction but also add friction where it was valuable.
Henrik Göthberg:So you reduce friction in terms of making less click setup but actually adding friction by asking maybe some critical questions that can amplify the experience.
Jonatan Raber:I guess Exactly, and I'll give you one example, which is in kind of the signup flow. Spotify asked a lot of questions that seem pretty strange. You know how old are you? Why do you need to answer that question when you want to listen to music? So there's some kind of licensing reasons behind that. But actually that information is great when recommending songs to somebody, uh, because, uh, it's very likely that if we recommend, or Spotify recommends, a song that you may have listened to when you were 20-ish years old, you like that song so.
Henrik Göthberg:So this is where the you know. That's why we have psychologists and all kinds of people in spotify in order to build a brilliant product team, exactly, yeah, and how in? Uh? Moving a little bit over to you, kiril, what's, what's your background?
Kiryl Zhdanovich:or you tell us about two stints in spotify uh, yes, I am originally from belarus, so, and after I finished university, I wanted to live somewhere else and, luckily for me, spotify invited me to live in Sweden and to work at Spotify. How did?
Henrik Göthberg:you pull that off. Did you have the best grades ever, or what?
Kiryl Zhdanovich:Actually, no, I was very bad at studying in my university because I thought we will do coding and then we did some math. Why do you do math? When was the last time you did math? You wanted to code hardcore yes, and then I also wanted to party, I guess.
Henrik Göthberg:So this was kind of like party and coding, yeah, maybe more party.
Kiryl Zhdanovich:I would say to be honest, just um. So I wasn't, I wasn't amazing, but in school, though, I did like I participated in math and physics competition, yeah. And then, around 20 years, I was like, okay, actually I need to do something in life. And then actually I started to study on my own and entered like some additional education and start working, and then I was participating in competitive coding competitions. At this time there is still a platform called Topcoder, so basically, you have like limited amount of time. You need to solve a problem and then submit it.
Henrik Göthberg:Was it Wipro that bought Topcoder?
Kiryl Zhdanovich:Topcoder, and then there was like….
Henrik Göthberg:Yeah, but Topcoder was a little bit like…. This is a platform where NASA could supply a problem, yes, and someone could basically… you know, you paid for it. I can solve this problem.
Jonatan Raber:It's only about a problem that NASA.
Henrik Göthberg:You were working on one of the NASA problems.
Kiryl Zhdanovich:So there are different types of competition. Actually, you can do their human design. And then there was like NASA competition about how do you position solar panel on a space station to get most amount of energy. And then you submit heuristics and then it was like you get one of the best solutions in the world and pay for this five thousand dollars, yeah. And there was like algorithm competition. I would say like what was quite average there, but I think it's kind of like the thing that I liked about coding and yeah but top core is cool because if you're if you're super, super good and you can solve the problem fast, you make good money.
Henrik Göthberg:If you're a, you can spend hours on hours and then your hourly rate is not that great.
Kiryl Zhdanovich:I think there it was more like if you're good, you're being pulled out of Eastern Europe and you will go to work at Facebook or Google.
Henrik Göthberg:But tell us, how did Spotify find you? Or how did you find Spotify?
Kiryl Zhdanovich:I was quite lucky because the story is following. My brother's friend was interviewing at Spotify. His experience wasn't amazing and he said okay, why should I change Google in Switzerland and go to some company in Sweden, especially if you consider salaries in Google Zurich? And then a recruiter asked do you know someone else? And then he showed me. And then fun part about interviewing at Spotify was that maybe 66% of interviews were competitive programming questions.
Henrik Göthberg:So this was actually up your alley.
Kiryl Zhdanovich:Exactly.
Henrik Göthberg:And this is what your top coder life taught you.
Kiryl Zhdanovich:I finished interviews in 15 minutes and then I had like an interviewer who said, okay, it's out of the list, but I'm just curious if you can solve this. So, and then we spent time like this and then I went to a hotel room after interview and when I got internet access I saw like email saying, okay, we want to hire you. And I was like okay, I should probably move to Sweden. I had no knowledge about Sweden, had no knowledge about Spotify, because why would you pay for music?
Jonatan Raber:So nonsense.
Kiryl Zhdanovich:Like why who on earth pays? If you have money and you're inviting me, I'm coming. So that's actually what happened. I ended up at Spotify. At this time I was working with my former brother colleagues who were hardcore C++ developers, who were doing offline maps, and I'm coming from the culture where if you allocate memory in a slightly wrong way, you are punished in a code review.
Henrik Göthberg:And then you consider your life choices. If it was hardcore stuff, I love it. If it was a good choice to start writing C++ in this company in a code review, and then you consider your life choices.
Kiryl Zhdanovich:This is hardcore stuff. I love it If I wanted, if it was a good choice to like start writing C++ in. This company learned a lot came to Spotify completely different culture, different people. I was very lucky to end up in 100% Swedish team. They Swedish-anized me as much as possible and I was pretty happy about this. Swedish-anize me as much as possible and I was pretty happy about this. That this was kind of like a culture lesson to me.
Henrik Göthberg:The culture lesson. The Swedish culture appealed to you. Yes, the way they handled, you know. And what is Swedish culture then, in terms of teams and how work life is? Because I think it's very different to Anglo-Saxon culture.
Kiryl Zhdanovich:Absolutely. And then, just if someone wants to watch about this, I think there is a video on the internet called Swedishness that describes it as much as possible. And then I joke with our CTO, victor, that he's like 100% from this video. So it's more like let's listen to everyone's opinion. And if you're coming from Belarus, you're like I'm sorry, your opinion doesn't count here.
Henrik Göthberg:And some people who don't know this. We're talking about consensus decision making deeply ingrained in our DNA. I guess yes absolutely.
Kiryl Zhdanovich:And then my product owner at this time like so, like I was working late and I sent email quite late in the evening, and then he told me on next day say okay, thank you for finishing work, it was important, you shouldn't have done it. So today, go home after lunch. What's happening here? You get punished for working. And then during the summer I was like the last one. He also told me you know, it's summer in Sweden. You should be out as much as possible, because soon it's going to be dark, exactly.
Kiryl Zhdanovich:And I'm like what, okay, sure, I think you're a smart person, I will trust this. So this was my kind of like lesson at Spotify. Unfortunately, I didn't get to work on things. I wanted to work and I was like a little bit, but I also was quite young and naive. So I wanted to write the best, most complex algorithm in the world that probably no one needs and once it also implemented some feature linked a Wikipedia page with the algorithm name that people in the future.
Kiryl Zhdanovich:And then it was like amazing time, complexity. And then my colleagues say, yeah, great, but you know, we have so much like we don't have so much data that it's actually efficient enough so you could solve it with the simple things which help readability. And I was like, yeah, it's not the thing I was looking for. I was looking for over complicated things, show off my knowledge, and then that's that's thing I was looking for. I was looking for over complicate things, show off my knowledge. And then that's that's why I thought you were asking algorithm questions during competition or during during interviews. Have you changed?
Henrik Göthberg:your mindset on this absolutely absolutely I can answer that were your bosses right. Were your bosses right?
Kiryl Zhdanovich:uh, in a way, absolutely, and you?
Henrik Göthberg:were, but you were young and you wanted to. You wanted to kick ass uh, absolutely.
Kiryl Zhdanovich:But but then I was like you know, I had the single tool in my toolbox was my algorithm knowledge, and then I wanted to use them everywhere. I jumped a few teams at Spotify. At the same time I was finishing some school of data analysis, where I learned a lot from amazing people how things work in data science and machine learning. At this time and I was also I got another toolbox which I wanted to use and I ended up in the search team at Spotify with quite amazing people, to be honest.
Henrik Göthberg:Anders Nyman. Was he coming later in the search team? He was engineering manager in the 2020s.
Kiryl Zhdanovich:I left Spotify 2018. Okay, this is so. Yes, and I also. I was youngish and experienced in a way, so I got experience there, but worked there for some time. Then I left. Then I was doing one-person consultancy for around four or five years in different machine learning products or companies, doing mostly speech recognition. But it's also more like you're selling yourself as a machine learning engineer because everyone it's already. This hype train is quite old, so fear of missing out.
Kiryl Zhdanovich:you sell yourself a machine learning engineer, but then you come doing back-end data you're not a machine learning engineer I'm sorry you're learning this this time, but you ended up doing data wrangling yes, and then and then, luckily, my former colleague referenced me to Jonathan, because we didn't work with Jonathan directly, and this was like a beam in the light for me Because I met people with what is most amazing sense of humor that I like, which is quite strange. What is your humor? But it's like, which is quite strange, what is your humor? It's super dark, dark and dry, and then actually, people who I share values with and who are willing to do something, what I consider to be interesting and still consider interesting.
Kiryl Zhdanovich:So yeah, it was a no-brainer for me that I won't. So actually I think after interview I was this kind of person who was sending Jonathan or like it wasn't interview per se, but it was like meeting and I was sending him messages on WhatsApp yeah, are you hiring me? Are you hiring me? It's like, when do I start?
Jonatan Raber:It's funny because I remember the opposite. Yeah, I, I remember it the opposite. Yeah, I was like, oh, he's not going to join us.
Henrik Göthberg:Oh, this is so funny, like one guy is the recruiter, one guy is the. Okay, let's talk one minute as a rabbit hole we do rabbit holes in this podcast, by the way on the recruitment process going on here, because this is super interesting when you have, like, the recruiter and the who got recruited on the same pod. That's fun. So let's start with you, jonathan. What is your tips and tricks? Or how were you thinking about? What was the background when you recruited? You were a consultant. You were recruiting as a consultant.
Kiryl Zhdanovich:Then I was a consultant but, like I said, okay, this is so.
Henrik Göthberg:Oh, you want to join yeah.
Kiryl Zhdanovich:I want to be part of this journey.
Jonatan Raber:So Kiran never joined as a consultant. He was actually our first full-time employee.
Henrik Göthberg:Okay, so what were you recruiting for?
Jonatan Raber:So we didn't know what we were recruiting for. No, what was the?
Henrik Göthberg:problem you were solving. Then we didn't know. Oof, now I'm intrigued. So we were four or we are four founders but we were sitting in an office and, oh sorry, sorry, I misunderstood. You are calling from F. Yes, ah, exactly good, all right, so now. Sorry, I was, I was. I was back in my head that this was still when you were.
Henrik Göthberg:Oh, it's spotify, spotify, no, no okay, so let's back up the tape. Good segue, good, fucking segue, because the bottom line is that you guys met at. How did you guys meet in in? You have touch base in spotify.
Kiryl Zhdanovich:We met, like actually at the coffee machine I I honestly don't think that we talk that much. I think jonathan he's very nice and likable person, so he kind of you know, picks up people saying okay, how are you doing, what's your day looks like? Or something like this no, no, I remember what we spoke about Really?
Jonatan Raber:Yeah, come on, go for it. I was helping this other company called Lexplor, so a Swedish company using machine learning to help assess reading comprehension and dyslexia among kids. And I knew Kirill was an ML engineer, so I think that's what we started talking about and you were super interested to understand more.
Henrik Göthberg:This is at the coffee machine at Spotify, yeah.
Jonatan Raber:Right.
Henrik Göthberg:Exactly, you know when you get the bug. You want to solve problems and this is a good problem.
Jonatan Raber:Yeah, and so that's what struck me. I really sensed that, okay, kirill was so interested in the problem and wanted to learn more, so he made a good impression. Yes, so I think that's actually one of the strongest impressions I had from him, and then we had a common friend. So when we started Fever later on, all right. So now.
Henrik Göthberg:Okay, so that's it. Okay, so that happened now, and now you are into the game. Let's start with Fever and start there, and then we have a segue and talk about the recruitment process, because I think it's still super interesting. So how did you see it? What is the lessons learned? Can we do even stronger lessons learned on recruitment by looking at both angles? That's cool, but okay, so tell us now the segue into Fever and we start with that first and then we do the recruitment story.
Jonatan Raber:So the segue into fever and we start with that first and then we do the recruitment story, so the segue into fever. Well, like I'm uh, I'm not gonna bullshit you, so we were uh, four, um four friends. Yeah, uh, I came from spotify I had left and um, kind of with a feeling of I loved it when it was smaller, faster moving, more innovative, and it kind of turned into a very big machine. And the three others I was sitting at the same table with came from iSettle or PayPal.
Henrik Göthberg:Yeah, same feeling They've been on a hyperscaler journey as well. Exactly, iSettle is huge and yeah all of them have exploded.
Jonatan Raber:And so we kind of wanted to build our new home, and so we were exploring different ideas for six months and one of the co-founders he came from Volvo On Demand, so kind of a car sharing, ev sharing.
Henrik Göthberg:Interesting, we had Jesper Fredriksson on the pod who was at demand Okay, volvo M, and he's now in Volvo Cars.
Jonatan Raber:Okay, yeah, and one of the kind of problems that he was facing was how to make the business work, Because he had all these cars and all of the cars were needed at the exact same time and the rest of the time they were just standing still, so he had a huge utilization problem. As we were talking about this problem and like how it could be solved in different creative ways, and at the same time another friend of ours came to us and asked I bought a huge battery. I know you guys know tech. Can you build software to kind of run it and optimize it?
Henrik Göthberg:Optimize for my home, my home home grid.
Jonatan Raber:This was an industry, so a huge battery, and so we really dove down into the rabbit hole of flexibility of the grid and spent two months there.
Henrik Göthberg:Tell me the problem. So this is an industrial grade battery. For what? Without maybe going into detail?
Jonatan Raber:So kind of the friend didn't really know. Uh, so he had heard about the flexibility markets and how much money you can so he was in a startup headspace yes, well, basically, um, and so he was asking us like how do I optimize revenue on this uh asset? I know it should, I know like supposedly I can earn a lot of money, but I need you to optimize it and I need you to control it.
Henrik Göthberg:And what was that business model all about? Was it putting a large industrial scale battery somewhere in the grid or close to in the grid, yes, and selling that back to like Vattenfall or Svenska Kraftnäs yeah, tso.
Jonatan Raber:But what that opened up for us was kind of understanding that the huge challenges that the grid is facing and how much money there is in actually providing flexibility to the grid and why that's like valuable to SunScape Caffeinette. So we spent two months there and when we came up the rabbit hole we realized that the first problem we were looking at was utilization of cars, Because it could actually be solved in the same way as a battery.
Henrik Göthberg:Because it is an optimization problem. And now you're flipping it from one car sharing to electron sharing.
Jonatan Raber:Exactly, but it goes one step beyond that Cool. Electron sharing Exactly, but it goes one step beyond that Cool, because if you think about an electrical vehicle and a company only making money when it's moving. If you think about an electrical vehicle as a part of the power grid, you can make money as soon as it's connected to the grid?
Henrik Göthberg:Yeah, because you have a demand response type opportunity and maybe explain demand response type opportunity exactly and maybe explain.
Jonatan Raber:Demand response so demand response is, uh, you basically respond, uh, to the demands of the grid, yeah, uh. So, if there is a need for more power, yeah, or less consumption, uh, you can um control your assets, your energy assets, in this case a car, by throttling down, by turning down the charging speed.
Henrik Göthberg:Yeah, and what happens here is like we're getting nerd into grids, but I guess we need to do it. I mean, like, sometimes in grids my background is in Vattenfall, by the way, so that's why this is interesting. I mean, like, a lot of times in in the, when you get further down in the low voltage, there is more of a voltage or balancing problem or consistency of service problem. So you have the challenge that you can't have too much then. Then if I'm like, if the wind farm is blowing, you get too fucking much for for the local grid, then you have one type of problem and if you have nothing, you have another type of problem and and if I take an example, like heavy industry, like steel industry and stuff like that high quality steel is all about, uh, very, very uh consistent voltage and current you know.
Henrik Göthberg:So what, whatever we do in order to make our i'm'm testing you if I know what I'm talking about you want grids that basically, the more consistent grids you have you have, the higher quality of service you have. It has very many benefits on the longevity of your gear, of the infrastructure, of the transmission, as well. As you know the quality. That means the steel industry loves you and the chemistry industry loves you and stuff like that.
Jonatan Raber:Is that a fair summary? Yeah, I think so. I think the. I think. First of all, you probably know a lot more about the actual power grid than I do.
Henrik Göthberg:I know, I know, I know it's old knowledge, but I'm leaning on it.
Jonatan Raber:I think the there were two things that struck me as very counterintuitive when learning about the grid, and I think the stability aspect yes, that I take for granted was one of them. So I never really understood how volatile exactly our grid can be no one knows this shit.
Henrik Göthberg:No one knows about. You know, we have the reserve power, we have a balancing power, we have all these different mechanisms, big business or big, you know, for the ts, for swedish class net. That actually makes this work.
Kiryl Zhdanovich:Yes, but this is basically the sign that you're living in a country when things work.
Goran Cvetanovski:I believe the country with blackouts.
Jonatan Raber:You kind of know all the names, like how things work, so but yeah, that's a good point actually, but there was one other thing that like struck me as like it blew my mind, so the fact that having little power in a grid, yeah, is as bad as having too much power isn't that interesting.
Henrik Göthberg:This is the main thing that people don't understand. The voltage problem is a problem in both ends. Yeah, that, I don't think people understand that as much.
Jonatan Raber:So when you're driving along a highway and it's super windy and you see all these wind farms and the wind farms like the turbines are standing still and you get annoyed.
Henrik Göthberg:What it's windy, just like turn, and then we can explain it by well, it's already full in the grid. The grid is already full and that you can actually get paid for consuming energy yeah, because now, ultimately now I think it's even you guys.
Henrik Göthberg:I mean, like someone is wanting to sell solar to me or what I don't know. Someone is trying to sell solar to me at home and a lot of the guys who's selling solar panel they don't tell you this. But then there's a couple of guys who can offer you a battery in order to sell back this. But it's interesting, because then you need to be an authorized dealer who has an authorized setup with swanska kraftnet, and I guess fever is maybe one of them yeah, we um, so we participate in different flexibility markets.
Henrik Göthberg:Uh, so yeah, we interact with, but this is interesting anyone who buys solar right now 80 of the guys that want to sell you solar power on your roof. They are not an authorized wholesale or dealer towards, so they don't have a mechanism to put the battery on your store to to help balancing the grid. So that's interesting, right? I? I know quite a bit about this and this and I had a lot of people calling me privately and not until quite recently, someone called me and wanted to sell me the battery as well.
Kiryl Zhdanovich:But I guess it's more like you know things, most things in the world, they're developing in steps. I wouldn't, I would be surprised if you know, like first solar panels would be selling with batteries and then offering balancing and all of it. I think we're now seeing another step in the grid balancing. So, okay, people who have solar panels, they actually can participate in some balancing markets. Or like maybe they want to store battery for their houses for the future, or yeah, things like this.
Henrik Göthberg:But maybe this is where we do like one rabbit hole or one like couple of minutes to kind of explain what does. We're going to come back to recruitment topic, but I think, as an introduction, what do you guys actually do now and can we basically maybe zoom out just a little bit to understand some of these mechanisms? Because I think it's very good, valuable information for me as a private consumer. When a private consumer listening there's an AI topic here, but it's also a private consumer topic, I love that.
Jonatan Raber:Yeah, we can go bananas.
Henrik Göthberg:You two take over now.
Jonatan Raber:Do you want to try?
Kiryl Zhdanovich:yeah, go for um no I think you should well.
Jonatan Raber:So um in in one sentence um fever. We enable a fully sustainable and resilient grid and we do that for europe, so that's like the scope we have, that's your mission that's our mission okay, and then you need to unpack this, of course.
Jonatan Raber:So we have a very dirty world today with fossil fuels. Yes, we want to move into sustainable energy, yep. So that's a huge shift because we're going to have a lot of wind and sun-powered energy, but that creates a big, big problem for the grid because it becomes a lot volatile a lot more volatile because you never know when the sun is going to shine or when the wind is going to blow, and even when it's shining or blowing, it goes up and down.
Jonatan Raber:And so the grid was built for a very stable and predictable environment where you knew exactly how much that was going to be produced and consumed. And we're moving into a future where none of that is true anymore and the places that energy needs to go is not the same as before. So fever's product is flexibility, so we offer as much flexibility to the grid as possible. So that's what we bring to the transmission owners or the distribution owners, like the grid owners, and we try to maximize the amount of revenue and value that you can get from your flexibility.
Henrik Göthberg:I have some numbers here that I remember from Vattenfall and I want to test if they're still true. So someone explained to me sometime. Okay, there's some unpacking. We have a TSO Transmission Service Operator. They own the big national grid and the regulation of the grid and in Swedish this is called Svenska Kraftnät. And then, connected to the TSO, you have the network operators like Vattenfall, and this part of the Vattenfall business is called the DSO, the distribution service operator. And this is where you know whoever has the grid where you live, they have a monopoly on you know, connecting you to the grid in that level. So it's the difference between retail, you know, having your energy bill and your network bill, so to speak. And so DSO, tso.
Henrik Göthberg:And now comes the topic of flexibility and I heard, you know, like the normal DSO, the view of the grid, maybe it's both DSO and TSO. Someone said an astonishing number, that we have an overcapacity in terms of how much energy we have Flipping it. We have an underutilization due to the flexibility problem that basically we are running on. You know, based on the you know, potentially we'd run on like 50% capacity. Help me here. Now I'm trying to explain what we mean with flexibility in terms of big numbers. So in order to work on a daily basis and have our grid stable, we have like a 50 to 100% overcapacity in order to work on this.
Jonatan Raber:I can give a good analogy.
Henrik Göthberg:Yeah.
Jonatan Raber:Actually, we probably should take a step back. Yes, I love saying that in period laps every time I do it, so what?
Henrik Göthberg:is the grid.
Jonatan Raber:So the grid is the actual physical wires that connect power production to power consumption. So you can think of it as the highways and the roads that you drive on. Exactly the same thing. And each highway and each road has a capacity you can drive X number of cars down this road at any given moment. This is a good analogy, I love it.
Jonatan Raber:And so if you were the owner of the road and you got paid for how many lanes you actually built in this road, you would just keep on adding lanes until you can serve the peak hour of traffic without any congestion. And that's kind of the world we've been living in before. And so the road owners, the transmission owners, the DSOs, that's been their philosophy. We're just going to keep on building out the grid until we have zero congestion, which means that 98% of the hours you have eight lanes. Yeah, one car, one car, and that's not, you know, efficient and that's not going to work in the future. And that's not efficient and that's not going to work in the future. So if you draw that analogy back to what Fever does, so we take the existing roads and we just try to make the traffic a lot more fluid and flexible so maybe you're okay with staying at home for one more minute, because that's going to make traffic smoother or you leave one minute earlier.
Henrik Göthberg:Yeah, so the flexibility problem. I think this analogy explains as well why we have such overcapacity, but it's because of these few cold days or peak demand it is. Then we need eight lanes if there's no intelligence. So the whole topic. Then, as I understand it, when we now need to modernize the grids because the grids has been sort of, as you said, very much one way, from big power plants going sort of into smaller high voltage, medium voltage, low voltage I usually like the big blood vessels going all the way out in your capillaries or whatever you call it.
Henrik Göthberg:I was going to do that analogy yeah, but that's the problem, right, that that now we have the capillaries at the at the end, and this is where we push energy back now when we're talking about solar panel. So, all of a sudden, our grids are also really much built for one-way street. You know one way and, and now we have prosumers I have wind, I have my car at home that we could now use the battery, or we have our solar panels, or we have all these ideas on, uh, you know, even companies building grids and our grids are literally not built to go two ways. And and, by the way, that's just the electrons flowing. What about the data flowing in order to get paid and all that? So it's such a shift that means, okay, we need to change all the grids.
Henrik Göthberg:So if we had a grid that can only go in one way, and now I need a grid to go in two ways? Now the core question comes for the Swedish state. Ooh, I need to rip out the eight-lane highway and build a new eight-lane highway, or maybe I should have some intelligence in there, so I only need to build three lanes. Is that a summary? Yes, so to solve this problem with infrastructure or bare metal, so to speak or copper fucking impossible for the Swedish state.
Jonatan Raber:It is possible, but it's going to take 15 years. 15 years and a lot of wasted money?
Henrik Göthberg:Yes, so we are coming to the topic. Maybe we should have some data and AI in order to build three lane highways for the whole Sweden that we need to now fix everything, or are we going to do it the old way and do eight lanes Exactly? That's the core question. Yes, I love this data and AI question, by the way, and I saw it early in Vattenfall and I'm like why the fuck are you not investing more in AI?
Jonatan Raber:Yeah, I mean, I think, kiril, you can speak more about this, but the problem today is actually more about the infrastructure and getting access to the grid and how to navigate all these different monopolies and players. That's like the big question.
Henrik Göthberg:Yeah, because okay. So there's a fundamental technology problem, but there's a market problem here.
Jonatan Raber:There's a big market incentive problem and a big, just infrastructure problem. But what we know from a macro perspective is that the way the grid operates today requires actually very low degrees of intelligence. Oh really, because most of it is quite predictable, but the world we're going into is very volatile of intelligence. Oh really, because most of it is quite predictable, yeah, but the world we're going into is very volatile, very unpredictable, and that's where I think AI can really play a role to kind of create a stable environment even when you have all that unpredictability.
Henrik Göthberg:Yeah, and maybe another way of looking at this, maybe because we have optimization problem at TSO. As long as you had the grid, you guys know Pontus Wallin. You're working with Pontus, have you? Met him he's at Ingrid right, an analytics guy.
Henrik Göthberg:I don't know. Okay, but I met him. I saw a podcast with him where he explained this quite well, where he basically said you know, is analytics new to Svenska Kraftnät? Of course not. You know, they have had these balancing and optimization problems since the first day and they've done it in different ways. But I think the big, big shift now is you need to be able to do this at real time, at speed. You know, you go from balancing on the hour to going to balancing on the 15 minutes, to going to real-time balancing. And why do we do that? It's because everything becomes much more. I mean, like when you have so many different types of energy sources. It's completely, it's a living organism. How would you summarize it? I actually now want to take a living organism. How would you summarize it?
Kiryl Zhdanovich:I actually now want to take a step back and maybe not to talk much about AI, but it's like things are coming in steps. So I think let's say we have now a lot of electrical vehicle cars in Sweden, specifically because there were easy way for buying cars, like Swedish government when sponsoring it. And then you know how electrical vehicle car is used someone goes to work, comes back maybe around five, six in the evening, the most expensive hour. As the user of electrical vehicle car, I don't want to pay a lot of money and maybe I probably will not use my car until seven in the morning. So, and maybe I need to charge it for two or three hours. Do I need to charge it right now at the most expensive hours? Probably not. So this thing is called smart charging. So this happens at night. It's no AI is needed. It's more or less you need to divide one number by another to understand how many minutes you need and what are the cheapest hours during the night.
Kiryl Zhdanovich:And already, yeah, and already we're doing some peak shaving.
Henrik Göthberg:We can do peak shaving in a very pragmatic way, but we need data. By the way, we need to manage data at least.
Kiryl Zhdanovich:Yeah, yeah, but I think you know, like market or electricity prices in Sweden, help us to do this. Yes, so like, if you're charging your electrical car during the winter, you don't ever want to do it at six in the evening.
Henrik Göthberg:No way, and everybody can see. You can follow the spot market like this. Everybody can do it.
Kiryl Zhdanovich:Yes, and then you know, like even this, electrical vehicle chargers, which are not maybe the best piece of software, but they also can fetch these prices and charge a car.
Henrik Göthberg:I have a Tesla. I put a scheduling on. Very simple yes, more AI needed for this by the way.
Kiryl Zhdanovich:So in a way, you know, we can talk a lot about AI and everything, but there are like some simple, simple things that we can.
Henrik Göthberg:That everybody can help.
Kiryl Zhdanovich:Yeah, exactly. So like okay, we have a load with a new thing electrical vehicle cars. We can do this Sometimes I'm not sure. I think iPhone in US does the same. They say okay, we will charge your phone during the night because this will help everyone in a sense. So, for sure, future is AI driven and we will absolutely need it.
Henrik Göthberg:For now, maybe we can just charge our electrical vehicle cars during the night, so behavioral change takes us a long way, because this is ultimately about people changing their basic behaviors to help out and be smart about their wallet and be smart about helping the grid.
Kiryl Zhdanovich:Absolutely. And then the same thing maybe will help us with solar panels and the battery at home. Maybe you charge your battery during the day with the solar panels and maybe at hour six or seven, when the consumption is the highest and the price is the highest, maybe you can use your battery to power stove or I don't know things like this. So always, you know, when people talk about AI, I think, yeah, sure, on the way there we will have some heuristics that will absolutely help us.
Henrik Göthberg:Yeah, but okay, so maybe that's the connection then. So where is the AI in what Fever is doing and where does AI come into? When does it become a real play? Or how do you see this? If you sort of start, if you continue the journey, that we can do a lot without ai, but what is fever doing and where do we really really need ai?
Kiryl Zhdanovich:I think you can start, yeah, um, yes, I actually had an interesting conversation with our ceo recently. So we are now maybe in the state when we are kind of selling balancing to people where, okay, there is this market where you can earn money and in this way we're helping Swedish grid to have more assets and then to have flexibility and then to help future. So basically now we're saying, okay, please buy the battery, come to us, you will get money, and then we will help Swedish grid. Is that the core business right now Fever?
Henrik Göthberg:come to us, you will get money and then we will help Swedish Grid.
Kiryl Zhdanovich:So for now.
Henrik Göthberg:Is that the core business? Right now? Fever Like can you explain the core business like this, because I think I know it, but please, I don't think everybody knows.
Jonatan Raber:The core business is basically to offer up the technology of being able to do balancing or flexibility, trading and control for companies that can't so utility companies like Järbolag, or infrastructure owners that own a lot of different batteries or EV charging spots. They don't have the tech team in place in order to build the infrastructure to aggregate all these assets and trade towards Svenska Kraftnät and the TSO. They don't have the ability to predict prices and orchestrate all these assets, so they buy that service from us, so we call ourselves a VPP as a platform.
Henrik Göthberg:Virtual power plant? Yes, exactly so as a service. So the core topic is that Svenska Kraftnät has a strong need and a strong monetary incentive to balance the grids and basically they can come to a point where we need to do a major reinvestment to fix this grid, or if we can find something that helps us balance. This is the VPP service. Can we can live in these grids longer and on the other side, you can then sell a service to different companies where basically say, hey, you can make your if you have extra spare power or if you decide not to use your. If we can sort of help you how you use your power and you in this way help svenenska kraftnet, you can make some bucks on that. Yeah, is that the?
Kiryl Zhdanovich:that's the core thing here yeah, yes, and then we're kind of ai or machine learning helps us here, so we are like saying, okay, this is amount of money that you will get from us, so in this market. So we want you know to, uh, our customers have like the best possible experience, to our customers have like the best possible experience, and we want you know to have as many assets possible in.
Kiryl Zhdanovich:European market so that you know the future of electrical grid is bright. So, and then our goal now to attract customers is actually to like, place assets right in the market, meaning like, okay, what are the best prices, what are the best products, how do we predict it? And things like this. So this is when it comes to money, but also there are different types of flexibility assets Electrical vehicle, cars or chargers. In this case, it's a very interesting machine learning topic. In a simple case, you can explain it. I need to predict how much load chargers will create to the grid, and when the grid doesn't feel well, we will stall some chargers and grid will feel well again. And then, since we're dealing with the question of national security in this case, because electricity is national security, we don't know to overestimate our abilities. This is no, no, no for us, but also we want to earn some money for our customers, so this becomes like a problem of forecasting.
Henrik Göthberg:And it needs to be a robust forecasting.
Kiryl Zhdanovich:Absolutely. And then we're getting PhD in electrical vehicle users' behavior now at Fever, so just to make it better. So right now we have some prediction problems. We have some optimization problems Also on top of it. Since I mentioned it, it's like national security, national security question. So it's not like we over predicted something and then we can feel well about it.
Henrik Göthberg:We cannot but okay so, and fever is operating in which markets? Is it a European footprint or is it a Swedish footprint?
Jonatan Raber:I don't know so currently only in Sweden, only in sweden, only in sweden. And yeah, but the the clear goal is to go across europe. Yeah, so the the good and bad part of the european grid is that it's, uh, all built kind of same time so it all has the same problems, kind of the same functions, yeah. Yes.
Henrik Göthberg:And could you argue that, whatever market you are going into, one of the key actors that you're working with or customers, if you like it will be the transmission operator on the one hand side? Or is it other models you could work with?
Jonatan Raber:Yeah, so I think right now in Sweden know, like Klondike, everybody's going to buy a battery and everybody wants a big gold, and that's specifically for a market called FCRD, so it's Frequency Containment Reserve. So FCR is a market for managing the frequency of the grid very quickly, and this is just a gap in the market and so there's a lot of value in that market, meaning that the TSO pays a lot to participate.
Henrik Göthberg:But that particular market, is that then connected to our physical commodity market like Nordpool? Not directly? Or is it more connected to the TSO?
Kiryl Zhdanovich:Yes.
Henrik Göthberg:So this is actually from a market perspective with the transmission service operator, because they need to balance stuff.
Jonatan Raber:Exactly so. The way the market works is that the TSO, each hour of the day, every day of the year, buys in a certain amount of capacity. So they want to know that they have x hundred megawatts of capacity. Uh, in case something goes wrong in the grid and that wrong is currently, I think, is if oska sham 2 goes down, close down, yeah um that's one of the power plants in, yeah, in sweden.
Jonatan Raber:So, uh, the entire nordics, so all countries in the nordics share this responsibility, and in sweden the market is extra lucrative for various reasons. Okay, but when, when we think about flexibility it's this is one market uh, flexibility could also be traded at nordpool, yeah, like in the intraday market, where energy companies try to keep their books in balance. Yeah, that's exactly the same problem we're trying to solve, just on a different time scale. Could you?
Henrik Göthberg:work on that. Could that be an opportunity for fever, or you guys working?
Jonatan Raber:Absolutely.
Henrik Göthberg:Is that a part of your portfolio working at Nord pool with this type of intraday stuff? Not, not right now.
Jonatan Raber:Not right now. Not right now, and the reason is basically that if you look at each hour of the day, the answer to where to place your flexibility is always going to be FCRD. Yeah, but that will change over time.
Henrik Göthberg:But so do you guys now. So TSO is a very important, transmission is a very important relationship for Fever right now. Are you engaging or do you need relationships with the spot operators like Nordpool, epic Spot and the European players? Is that going to be part of the big play? Long-term play with Fever we could be.
Jonatan Raber:I think the way we think about it is our customers could be the balanced, responsible parties. So the parties in this market are responsible for making sure that there's as much consumption and production in the system, and they usually trade on these markets in order to keep themselves in balance, but also to get as much value from the energy that they produce. So that's our customer and I think we'll go to market either through them or on behalf of them.
Henrik Göthberg:Yeah, but I think you answered it. Let's see if I get it right. I mean, ultimately we have a balancing and flexibility problem and it's in the best interest of all the actors to figure out the market mechanisms. That lets us use all the different levers.
Henrik Göthberg:So if you stare yourself blind on one particular lever, you know at some point that that gets overheated part of the market yeah and then maybe you know klondike here, but maybe you know, if we do it, if we think you take a step back, we can do this part of the problem at Nordpolo in this way, and then we have this way. So it's an ecosystem optimization problem also. So it's also an ecosystem of markets. That's what I'm trying to get to.
Kiryl Zhdanovich:I guess You're completely right. Yes, so, for example, as of today, we're kind of like thinking how we best participate in balancing market. I guess maybe in two years we will say something like this we will have one megawatt of flexibility. Which market, at which hour we will bring this megawatt of flexibility, what state we will like, where it will be the best for us, and also, like it's not only about money, but it's actually when we get the most money in the hour, that means this is the place where the flexibility needs the most.
Henrik Göthberg:Because this is this, to me, is super interesting. You know, because if you now has one of your more core competitive advantages to really understand the optimization problem and have the right algorithms, you know how to use that and how to create a user experience or a product out of that is potentially. It's a little bit like you said you could take a transport problem of M and you could transport it here. Here you have the fundamental understanding of what you need to do. So here, probably, yeah, I can am I on the right track here? How you're thinking.
Jonatan Raber:I think so yeah, to us, the choice of where to place our flexibility comes kind of last. Before that, we need to understand how much flexibility actually exists. So there's a lot of prediction that goes into what Kiril just spoke about understanding the EV fleet, for example. Understanding how the weather will impact the grid, understanding constraints in the grid.
Henrik Göthberg:But could you then summarize to some degree the companies with the most robust predictive power in terms of being efficient and sort of optimizing the money but being safe? That becomes a very, very strong competitive advantage that you then can think about how to use.
Kiryl Zhdanovich:Absolutely, and so like, if we talk about future when there is, like you know, we tried heuristics worked well, not good enough. So the future will be for us to understand, okay, how this market is influenced by weather, by people, by industries, and how we can utilize our assets there. So this will be kind of like the future optimization there and then, yeah, as you said, the one who will understand it best, who has like best prediction, best understanding of the market, I'm not sure like predictions will be enough, because it's as you also described. It is kind of like living mechanism.
Henrik Göthberg:It's real time, so you need a prescription, you need to get to an automated optimization algorithm. Yes, I mean. So what we are saying is it's not enough to predict, it needs to execute.
Kiryl Zhdanovich:Yes, absolutely. And then, like when we're talking about like AI in electricity, I am always thinking about my backend colleagues. No-transcript.
Jonatan Raber:You're just dangling that.
Henrik Göthberg:I love it. I love it. Can we sort of summarize a little bit about the future grid, because you're already moving into that now, that now, so the predictive power becomes super important. But you said it's more than that, because it's the whole availability of the whole infrastructure and the software and everything that goes around this, because it's not only now hardware, bare metal infrastructure, it's the hardcore data pipe in everything, right? So could we elaborate a little bit about the future grid?
Henrik Göthberg:here, what have we said now is sort of almost hinting at something. Could you try to nail it down for me?
Jonatan Raber:should I start with the macro pieces and you can talk absolutely yeah macro.
Jonatan Raber:I like that. So the uh, the big things that will change is we're going from stability and predictability into something that's unstable and unpredictable and that's mainly driven by the sources of power. So you have coal and gas today, and nuclear, and that's all going away and instead you're going to have much more wind and sun powered. So that creates a lot of volatility in terms of when power is produced. So that's kind of the biggest changes drivers that is, you can clearly see already exactly and then change.
Jonatan Raber:Number two is what we've been talking about the places that energy will be produced and consumed. So evs, for example, are going to have a huge impact on where energy is needed and how quickly. All houses with a sun, like pv, on the roof will also change where energy is produced, so production is going to become decentralized, and so production is going to become decentralized and intelligence is going to become decentralized instead of being concentrated in certain power plants and in certain control rooms. So I'd say that that's kind of the big changes that are happening.
Henrik Göthberg:And what does this leave us in? On more technology, infrastructure and even down to data and AI.
Kiryl Zhdanovich:Absolutely. So I like to explain this with the following example Future house will absolutely for sure have solar panels, home battery and electrical vehicle. So electrical vehicle will have also batteries that would be able to power homes. So, actually, maybe at certain hours people will power their homes with electrical vehicles, the batteries that they saved power during the day. So and this will be optimized Our goal will be at this point of time saying, okay, what should we do now?
Kiryl Zhdanovich:Should we sell the energy from the battery of a vehicle, of electrical vehicle? Should we charge it? What we sell the energy from the battery of an electrical vehicle? Should we charge it? What happens with the battery, solar panels altogether, and then how it will work altogether. Or maybe, if you have some energy, would you like to sell it to your neighbor? Or like simple things. I believe, like future of Swedish BRFs, they probably all will have batteries, they will charge it and maybe they will use it or sell it to neighbor batteries. So the same situation that we see now in Ukraine, for example, when it's like all centralized and then there are like power plants are bombed and there is no electricity, probably this will not be a case in the future. So probably we will buy energy and sell energy to our neighbors in the same neighborhoods, so it will be kind of like the same as swedish neighborhoods looks like. Now. You don't need to go to city center to buy I don't know clothes. You'll have everything near your house.
Henrik Göthberg:Okay, can we unpack this optimization problem a little bit, because I I sense that there is an optimization problem simultaneously on different abstraction levels. So so very locally, how you do it here. But then you can take the market perspective and then you can take three market perspectives and then you can take high voltage, medium voltage, low voltage perspective. So you need to have an optimization of the whole ecosystem versus the capillary at home. So that seems crazy. I mean, like it's such a complex problem. Maybe it's many problems. Maybe you know, is this one big model or is it many small models orchestrated, or you know?
Kiryl Zhdanovich:I'm quite looking to solve this, or like to at least participate and try to do this.
Henrik Göthberg:This problem is your favorite problem, exactly.
Kiryl Zhdanovich:But I think it's also like you know, as like I hope, as much as I imagine it will be decentralized. I think we're still like it's the same with the huge computational systems You're solving like one problem in one area at a time. So, and this will hopefully help the whole grid stable enough to like small area working, a lot of small area workings. The whole grid is actually working.
Henrik Göthberg:small area working, a lot of small area working, so the whole grid is actually working.
Henrik Göthberg:Yeah, because very early on in Vattenfall I'm working now with Scania, a bit with Boliden and there's been a trend that is getting more and more obvious For me in Vattenfall this was obvious in 18-19, that monolithic approaches will not really cut it, because you could see like we will have at least intelligence on the edge in some ways and then we'll have central intelligence.
Henrik Göthberg:If you look at the cars, okay, you need to get a compute in and maybe build a model and then you need to push it out in the car. This is one angle. And then I like, like I I mean participating is on the advisory board for enlit, european utility week, and I hear them sort of talking old school data warehousing on old school monolithic views of how to optimize this, and I say like this is clearly many, many hardcore problems on its own, so it's going to be about optimizing. I saw this as a very decentralized problem, where you then maybe have an orchestration problem, and so I started to dig myself into federated machine learning, all these different things that were sort of going in this direction, and for me this is super obvious. And do you see it different. I cannot see it in any other way than distributed.
Kiryl Zhdanovich:I have a very or like. I think we, as a Fever, have a very interesting experience that we discussed, because we've met a lot of people who are very knowledgeable in energy, energy grids etc. And they're basically you can quite easily separate them into groups. One group is people who are like 60 plus years old, because grid was kind of like solved for a lot of years. So these people were like young, ambitious, they solved the grid and we actually didn't know how it works because it worked perfectly. And the second group is actually like graduated PhD students who just graduated, and they have completely different views on the future of the grid, where you know, like people who are like work a long time, they're like no, no, we should keep it as it is, so it should be, should be how it is centralized, everything and the new generation. They're like more cowboys, decentralization, more like flexibility, more green energy and then what do you think?
Henrik Göthberg:you know what is the sort, what is fever's best on this? More distributed, because I find it, just looking at what you put, if the drivers you explained are true, it's distributed, it's an edge here, it's a prosumer dimension in this, whether you like it or not, and you can't put the genie back in the bag.
Jonatan Raber:I think you're absolutely right. I think the uh, that's what I mean with decentralization, I think a lot of the decision making and logic will move to the edges of the grid, or the capital letters, as, as you say. I think that's going to have a very, very big impact on how we operate the grid. Yeah, um, but I I want to hit on your kind of point on. There are many different levels. Yes, and even you know, the Nordics is one synchronous area and it's going to be connected to Europe and then Europe is going to be one big area. So whatever they do down in Italy is going to affect us.
Henrik Göthberg:Yeah, we're talking about price zones, price regions in Sweden. And why is the price region in southern Sweden always in southern sweden is always more expensive? Well, it's. It's connected to europe don't forget that.
Jonatan Raber:So so I think, for for fever, right now, we need to focus because, uh, even though there is, you know, the theoretical one mastermind that can understand what goes on from the highest level down to the single household, I mean, we're not there yet. So what we're doing is we're saying whatever happens behind the meter, meaning within a home or within an industry. That's for somebody else to optimize and solve so, if so, you need.
Henrik Göthberg:so you are basically betting on carving out pieces of this algorithm problem of the whole ecosystem. Yes, no one can solve it all.
Jonatan Raber:So the algorithm and the problem we're solving is what happens between the energy meter and the grid, not between the energy meter and the appliances that kind of At home, at home Inside the house, yeah.
Jonatan Raber:So you'll see, over time, user behavior changing and we go back to kind of the counterintuitive notions of I have a PV on my roof and I see it producing power and feeding the grid. Why would it be more beneficial for me to turn this off? How can I get more money by turning off energy production? If a user doesn't understand this, they'll keep on producing energy even when it's not needed. So as soon as users understand, okay, this is an orchestra of participants.
Kiryl Zhdanovich:Okay.
Jonatan Raber:I'll turn off my sun production when it's needed. Knock-on effects yeah then that will have huge knock-on effects. Back to Kirill's point about smart charging that's great, but if everybody smart charges at the exact same moment, the grid blows up. So I think you'll have extreme waves of kind of user behavior changing, kind of the downstream impacts, which is like our world in part.
Henrik Göthberg:All right, but you know, let's see if Andres joins us, but he would be very mad at me with having people who know technology. Oh, you want to do, he's not coming. Okay, should we do news section or not? Let's do it now.
Goran Cvetanovski:It's time for AI News brought to you by AI EW podcast.
Henrik Göthberg:Just in the middle of the conversation break and then we'll continue. So we tried to. We started with literally last season, when Chachipiti and everything going bananas in terms of new LLMs or anything, the AI world is spinning by the week and has been for a while. So we made a point to basically see if we could pick up on. You know, was there any news, things flashing by on the news, and it can be sort of gossip news or it can be real news, anything from investments to tech, to a new LLM or whatever, and we spend a couple of minutes and reflect on it. So we haven't prepared this at all.
Henrik Göthberg:So I'm going to shoot to you. Do you have any things that caught your eye in the news? I have a couple of topics, or one topic at least. I'm. I'm going to shoot to you. Do you have any things that caught your eye in the news? I have a couple of topics for one topic at least, but I I'll ask the guest first and you can. If you don't have it now, you can think about it then you cannot go go after me, but I just want to be polite to ask the guest first I've read actually an interesting paper, but I'm not sure if it's news.
Henrik Göthberg:So the paper that okay, but I humor you. So what's news to you this week? Which one was it?
Jonatan Raber:news this week actually yeah, I'm not sure what's news to you.
Henrik Göthberg:If you read the paper, let's go with that with paper.
Kiryl Zhdanovich:Yes, okay, with about paper. So the paper was that I'm not sure how you spend your teenage years. I was debating people on the internet, so this was a very important part of my life. So, and then I guess it's very important that these debates, that you win the debate, otherwise what's the point? Point? And then you know there are, like famous US election, where people claiming that, ok, there probably were some bots on the internet actually converting people's opinion. So the paper is coming, I think, from Swiss, italy researchers.
Kiryl Zhdanovich:So they wanted to build a platform and actually to check if ChatGPT, specifically, can convince people more than actual people. The setup of this experiment was following they had like 820 people, I believe, and there were like four groups. So there was like human versus human, human versus ChatGPT, human versus human, who know personal information about other human and human versus ChatGPT, with also personal information about other human and human versus chat GPT, with also personal information about human. And personal information was something very simple that we can get on the internet Age, ethnicity, work, education, things like this, like Facebook profile, basically and they picked around 10 questions which you don't need any prior knowledge to.
Kiryl Zhdanovich:So the questions were like should rich people pay higher taxes? Is using social media making people stupid, like? So, like easy questions, I guess most of the people have opinion about it so and the debate was like this person was giving a random opinion, maybe even the one that they didn't support, and then they were debating it for 10 minutes and then at the end they were checking okay, what's the result? Have you changed your opinion? So the result, without personalization there, not very interesting. Interesting part was that ChatGPT was convincing people much better compared to human with some personal information. In this paper, they provide some prompts, which were quite simple in this case, and I was surprised how simple they were, and that means that probably someone with access with ChatGPT similar model can convince people on the internet to like things that it wants to convince to.
Henrik Göthberg:So, which is pretty scary, because now it means that you can now do things at scale.
Kiryl Zhdanovich:Absolutely.
Henrik Göthberg:So what's the point of what it is Like? You can use ChatGPT. It actually means that you can start robotizing, automating setups in order to convince that massive scale. Absolutely, I think. Because, because you know people couldn't do all this convincing. But if I now can program it, and it was like 1000 at a time and then 1000 chat gpt's conversations going on- yes, and then I said like prompts were quite simple.
Kiryl Zhdanovich:Yeah, so if there is like some mastermind who has like knowledge about convincing people and the techniques and things like this, I think it's like it's a weapon yeah, so.
Henrik Göthberg:So it's so, with the right, prompt engineering techniques in order to set up the structure to get the lm tuned on the right way of thinking and arguing interesting. Yeah, tuned on the right way of thinking and arguing, interesting. Huh, quite scary. I think I saw that link through. I think it was a couple of weeks back that this came out. Yes, but it's quite interesting in terms of election times this year, of course.
Kiryl Zhdanovich:Yes, I'm specifically interested about it because there are different products that are based on large language models that are developed currently, and not many of them catch my eye, I would say. But these things which you know, like this, these are super interesting to me like basically how anyone can use this to do like something that we couldn't do before yeah, cool, any news that you caught your eye this week.
Jonatan Raber:I have one and a half at home. I'm addicted to chess. I haven't read anything this week.
Henrik Göthberg:That's cool, I can take one and then you can Do. You have a couple today, goran.
Goran Cvetanovski:I see that you're checking on the one that I had, so you continue with that one.
Henrik Göthberg:I think the big one working in the data, and I had, so you continue with that one. I think the big one I mean like working in the data and AI space and, of course, working more with this sort of normal enterprises. You know, you get all these normal players that we always hear about the Snowflakes and the Databricks, the Ashores and the Microsoft and it's been a game now going on where sort of it's been the Microsoft or the Google camps a little bit. Okay, we have the Meta and it's been a game now going on where it's been the Microsoft or the Google camps a little bit. Okay, we have the Meta and I think what happened this week is that Databricks really got into the game. Have you heard about them releasing their first open source LLM model? You can bring it up there, Goran.
Goran Cvetanovski:Yes, sir, llm model, you can bring it up there, Goran.
Henrik Göthberg:Yes, it's the DBRX, that's what they call it. I find this whole play very, very interesting. There are many things that make this play super interesting. So, first of all, they go all in on the whole open source angle, right, they are open source now and they are then really benchmarking themselves against the other open source models and they are then really benchmarking themselves against the other open source models and they can clearly beat Lama2, Mistral. They're doing better than Mistral on some key benchmarks and they're also doing better in Grok. And of course, then the DBRX model, I think in size is a little bit, it's slightly bigger right. So I think it's slightly bigger right. So I think what was the size? I forget? Lama is 70 billion, whilst GBRX, I think, was over 100. Do you have the number, Goran? Help me out. So it's a fairly large model they have put in place.
Henrik Göthberg:And I read an article a little bit. It was an article in Wired where there was a little bit behind the scenes. So they actually invited Wired, the magazine, in order to sort of follow them the last 20 days before launch, and it was one of the cool things for us that they're like okay, they had like 10 days left of compute or 20 days left of compute, and they were like, okay, we've done it more. Should we just go hard or should we change strategy? For the last hours of compute, and they went a little bit different. I don't know if you've heard about all this. It's very, very interesting, right, Because you get a little bit behind the scenes on the training strategy and then ultimately you end up in the big debate, you know, or big discussion and their marketing spiel, of course, why it should be open source and why we need to get more transparency and less secrecy around.
Henrik Göthberg:How do you tune and work models? What do you actually do? So there are many angles on this. The last thing I want to point out is that it totally makes sense to me that Databricks who sort of they're open sourcing the model because they actually want to make money on the compute and they want to make money on the lock-in effects to Databricks, so they can actually find a killer app to drive the revenue model in another way. So of course, they can open sources because what they lose on the carousels they can take back on the swings. So it's interesting, right. So depending on which player goes into the market, it opens up different business models.
Jonatan Raber:What was their strategy with the last 10 days of compute? What was?
Henrik Göthberg:their strategy with the last 10 days of compute. It's called I don't. This is where Anders should have been. It's called curriculum training. So basically what they did. I need to find it. I need to read the article. This is the wrong. I have it here somewhere. So for everyone who's interested, there's a wired training here. So if you go into wirecom inside, the creation of the world's most powerful open source model is sort of the clickbait title. It's a cool title, by the way. So startup DWSG3 is DBRX, the most powerful open source launch model yet, eclipsing Lama 2.
Henrik Göthberg:And then it goes down and sort of goes quite into detail of how they did that. They're talking with the guy. I mean like they hired a guy that came from. I mean, like the starting point is that Databricks acquired a company I forget what the name is that was quite strong on this. Let's see Inner Workings. Let's see if inner workings, let's see if we get down here Sorry, sorry, open up. I missed it now, but I think it's called curriculum training and you know I'm not the kind of guy you should ask on this detail level. It's fully beyond me. But you know, not only brute force of more hours on the same setup, but actually starting to tune it or giving it different data sets. So it was this combination Do we train it on specific data, do we groom it on a certain problem? And I think it was cool. Let's see if I can find it.
Kiryl Zhdanovich:I think it's always quite interesting when the companies do something and you cannot actually say they're doing this because they're nice or just because they want to earn money. No, what is it then? No, but it's like because, like you know, every time someone's open source something, we're like yay, nice, amazing, go open source, we all love open source. But on the same time, probably as you said, they're driving, you know, attention to their like to their business strategy yeah open.
Henrik Göthberg:So if everyone haven't noticed that open source is hardcore at business choice, it has always been, has always been. You know, when they back in the day, when they released the BERT model or whatever they did, when the TensorFlow was released and all that, it was ultimately to get massive performance improvement for free in terms of people working and fine-tuning and fixing the bugs and all that. So open source must be a business strategy. There is no other strategy. Then you can have what is the optimum sustainable business strategy that is also sustainable for the greater good of humanity. I think that plays on that. You can have different values on this, but to naively think that open source is charity, no, no no, of course not.
Jonatan Raber:We actually had one of these discussions today. Parts of what we're developing should be open source.
Henrik Göthberg:Yeah, and that could be a very good idea, but it needs to make business sense.
Kiryl Zhdanovich:No, of course.
Henrik Göthberg:Okay. So different team members threw out ideas in Slack for how to use the remaining week of computer power. One idea was to create a version of the model to generate computer code, a much smaller version for hobbyists to play with. The team also considered stopping work on making the model any larger, instead feeling it carefully curated data that could boost its performance on a specific set of capabilities, an approach called curriculum learning. So that's in the end they made they were not. You know, the team was sort of debating and then the head guy said we're going to go for curriculum learning. So instead of feeding it carefully curated data that could boost its performance on a specific set of capabilities, an approach called curriculum learning. And they made that. And the core point was that they could measure a significant uptake in model behavior by basically then tweaking the training you know. So there's so much alchemy going into this. Now, right, that's quite interesting. That's a cool topic. Do we have any other one? Do you want to add another one?
Goran Cvetanovski:Yeah, I think what is interesting with this topic is how much money nvidia made on this. So the dbrx was trained on three thousand set. There's three thousand seventy two nvidia, h hundreds connected to thirty three point one terab, what's each one? Each uh, h, no, no, how much is it? Approximately like $50,000. Maybe a little bit more. Actually, that is a good question.
Kiryl Zhdanovich:Yeah, how much is it? Because he has like a new.
Goran Cvetanovski:He just announced the new one, which is H200 or 300 or something like that. So H100 NVIDIA.
Kiryl Zhdanovich:I've seen today a photo of Mark Zuckerberg and the NVIDIA. I've seen today a photo of Mark Zuckerberg and the NVIDIA CEO. So probably Facebook bought a lot of NVIDIA's and NVIDIA CEO allowed to wear Mark Zuckerberg to wear his jacket his famous leather jacket.
Henrik Göthberg:If he allows him to wear the jacket, I think that is a news by itself yeah so it's like a very interesting.
Goran Cvetanovski:Give it a jackpot. Jacket right, jack ma right no, no, not jordan no no, no, no, sorry, sorry. Yeah, it looks like it's 43 000, yeah, 44 000.
Henrik Göthberg:So 44 000 times 3 000.
Goran Cvetanovski:yes, exactly all right. So it's only that that's a lot of leather jackets. That's a lot of leather jackets. Yes, for sure. On another note, when we are talking about Zuckerberg, of course, and Meta, do you know about their glasses, right? The Meta views. So next.
Jonatan Raber:Oh yeah.
Goran Cvetanovski:I tried these actually.
Jonatan Raber:The Ray-Bans.
Goran Cvetanovski:Yeah, I have them as well. I bought Australia, but they were completely useless in Europe because they're connected with the app and the app was not basically applicable due to regulations.
Goran Cvetanovski:But now, it is, but now it is in eight countries or 15 countries or something, so it's not in all of the countries. It's in Sweden, yes. And now there is a different thing that is adding. So they are adding now multi-modal AI into the glasses. So basically, you can prompt by speaking and then it can, because it has two cameras on the side, it can actually see and then it can give you information. Like, this is a museum, you will find this and in the future, probably you will be able to buy some stuff. I think it's going near to what Apple is trying to do.
Goran Cvetanovski:But what is interesting is, where are we moving with all this technology? So, if you see, three emerging technologies are really, really starting to pop out. So it's AI, ar and robotics and it's all over the place. And if you looked at the NVIDIA's uh release, like uh one week ago, two weeks ago, it was all about that. So they were basically they're preparing for the chips for this. They are also now making like an open source training models that you can use and deploy directly to train robots and etc. And I think, eventually, why uh ellen was super pissed at Sama? It was not because of the open AI, it was about the robot, yeah, so very interesting.
Jonatan Raber:All right, I actually tried these this weekend. I was really impressed.
Goran Cvetanovski:Yeah, the sound is still a little bit shitty because when you run it it's just like a little bit flat, Like he's here.
Jonatan Raber:Okay, I didn't try running with it, but I do think the music part of it, like the sound, is, uh, pretty cool yeah and you can uh talk with them, you can take basically notes and stuff like that.
Goran Cvetanovski:I think it's uh great. So, um, yeah, but uh, it's also dangerous. You don't know where you're going to be filmed.
Jonatan Raber:No, oh exactly, I'm just assuming somebody's always filming oh yeah, at this point, in this point, in time at this room, it's easy.
Goran Cvetanovski:I was in a conference in Singapore we were doing Data Innovation Summit in Singapore and we had actually one of the speakers coming from Meta. He's coming downstairs and we're having like a I think it was the first coffee break in the morning. So we're having a breakfast or something. I'm looking at him and I'm looking at the glasses and I can see coffee break in the morning. So you're having your breakfast or something. I'm looking at him and I'm looking at the glasses and I can see, because you during, if you have had the meta, there is no way you will miss that somebody has those because of the two cameras that are on the side right so it's like you have meta.
Goran Cvetanovski:Sounds like yeah, I uh you know, I film sometimes and make it like this, like it's very, very you know, so he has met us that it's not sunnies. Were they sunny, exactly so I have the ones that are completely sunglasses and now they are the other ones that are basically you can change you, so it's like a yeah yeah, good all right, so let's move back into the topic.
Henrik Göthberg:I want to change gear a little bit and when andres listened to this, he will be very mad at me if I didn't even try to have a little bit techie conversation. So I think there are a couple of topics. First of all, a little bit like what is the data and engineering challenge that you're working on and let's not stay on the macro discussion. But let's please, it's going to be more detailed than I understand. I don't care, there aren't listeners who you know. So what? What is the actual uh engineering problem and what is the uh algorithm problems concretely are working on?
Kiryl Zhdanovich:Um, where to start? Uh, so many things. Um, I think, me and my colleagues, we end up maybe second or third times in our engineering careers where we work with something that is not stable enough. So many of my colleagues they're coming from iZettle and then they're like saying, okay, when we started iZettle, or we're like at the beginning of iZettle, payments were like unreliable, like not possible, like sometimes you can charge user 10 times, unfortunately. And the same thing.
Kiryl Zhdanovich:I was working in a startup where we're developing offline maps for, like, android devices as well, when you know, someone would report us a problem that you know, my very bad chip Android device cannot render your maps. So and then like, okay, what can we do in this case? I think the very basic challenge is, now that we're having that all of these flexibility assets that we're dealing with is quite unstable and there's like sometimes like batteries and then electrical vehicle chargers. They're quite hard to control, but we should do this and like, we should be like stable as possible, because you know we're quite hard to control, but we should do this and we should be stable as possible because we're dealing with electricity and grid and balancing, so there is no way we can do our work bad. So this is a very basic problem that we're having. We're writing software for unreliable devices.
Henrik Göthberg:What are the key topics to think about when you write software for unreliable devices? What do you do?
Kiryl Zhdanovich:The first thing is more like thinking okay, what should we always think about? Okay, if we ask the device to do something, will it do it? We ask, like, do it two times? Or like, okay, we check it didn't do it. Okay, should we send a new command? Should we wait? Should we send a new command? Should we wait? Should we ask a new device to do something? So, like, key thing is more like to learn from our previous experience what should happen.
Kiryl Zhdanovich:Let's let's take example electrical vehicles. So what happens in the grid when the frequency drops until certain level? We need need to stop a certain amount of devices. And let's say we have 500 devices. Now we need to stop 100. And then this is not only 100, we need to pick a certain load of these devices. The first problem that arises is okay, which chargers do we stop? A very interesting algorithmic problem called the sub-sum problem. So, basically, which devices you want to pick, so their sum is actually equal exactly to the number you want. You can go a little bit higher, a little bit lower, but then if you go too low, you don't do your function. If you go too high, you kind of depleting your fleet in this sense. Okay, then we're sending commands to these devices, and then some devices, they say whatever.
Henrik Göthberg:Try again. I don't listen to you.
Kiryl Zhdanovich:I continue doing what I'm doing. Or some devices yeah, I'm stopped now and then in one second I'm started now again and you're like yeah, but you know we had this conversation before, please stop. Some devices not replying at all, some devices saying thank you for your try. I'm actually leaving now because my user wants to drive somewhere, and so you know this is very basic engineering problems.
Henrik Göthberg:So this is software engineering when you have a lot of different scenarios, uncertainties and you need to think about to make it flexible. I guess.
Kiryl Zhdanovich:Absolutely, and then you know the thing is is okay, we can learn from data. So what is our strategy in this case? Okay, this time we tried to stop 100 chargers. Let the x of them, x percent of them, stopped. How? What is our strategy next time?
Kiryl Zhdanovich:So, okay, this charger, or like this home battery, was misbehaving like. We can maybe like, when it comes to home battery, we can, okay, okay, talk, like send message to users. Okay, you know your battery is malfunctioning. Maybe you should do something with it. What can we do with the electrical vehicle charger? Should we like exclude it to like, what should we do? What should we do in this case? So, and it's kind of like trial and error, so we run like a lot of experiments in this case, but it's also experiments are quite costly in this case, because it's like actual users and sure we can stop devices.
Kiryl Zhdanovich:What is quite important also, we shouldn't forget to start devices, because some users I guess most of them, they're known for this they will be pissed off if their car is not charged in the morning. And then they say, yeah, amazing, amazing flexibility of things, guys, but I need is not charged in the morning. And then they say, yeah, amazing, amazing flexibility of things, guys, but I need my car charged in the morning. This is kind of important for us. So, and this is also this is also a problem like we're dealing with on. On one hand, we're dealing with uh tso, that you know we want to fulfill our obligation. On another hand, we're dealing with customers many of them, exactly and so, like I would completely understand if something happens with, like I cannot drive my car, and then like, if I cannot drive my car, if emergency happens, it's you know it's even more complicated. So, before even we're going to some like data land or like something.
Kiryl Zhdanovich:So this, we're now dealing with devices which are haven't got, you know, a lot of attention during recent years. It's changing, it's absolutely changing, but it's also this industry's, like the energy industry, getting a lot of attention. So, like many companies, they're becoming much, much more better, they're hiring better people or like more people. Yeah, everything becoming better, but it's still like we're we're still not there yet. So like we have a lot of, in first place, engineering challenges. How do we like, even if we did the best predictions in the world, how do we operate our electrical vehicle fleet? So first, there's like algorithmical problems how do we don't deplete fleet and there is like software engineer problems. How do we store our state fleet and there is like software engineer problems. Okay, how do we store our state? What if our service goes down? What if data centers that we're hosting our services goes down? So this is all very complex challenges that we need to overcome and quite important for us that we try to pay attention to it from day one.
Henrik Göthberg:But I think this is quite important profound lessons learned here that I find when I look at more analog companies that think they're going to do analytics and AI and they don't have a real software engineering background the naivety in relation to what is the product? Oh, it's the algorithm. You know, going back to Google's, the technical depth of machine learning, it's so obvious that you need to, in the end, even in very basic, much, much simpler examples of BI or analytics, you need to think software engineering, you need to think end-to-end system and then put your attention where it really counts. So what you are saying is that any machine learning or AI or any cool gadget that you're playing, first of all we are building software and then in that software this comes fancy algorithms if it's Rags or generative AI or whatever, but don't forget that what you're really dealing with is software engineering.
Henrik Göthberg:Would that be a fair summary?
Kiryl Zhdanovich:Absolutely. And then you know, like some very basic things like on-call.
Kiryl Zhdanovich:So, you know, if battery go bananas, we need to know about this, we need to be waken up. Or I sometimes feel like sorry, not sometimes. I always feel very sorry for my colleagues when I kind of mispredict something, when I say, when I am like, too positive though, we're trying to be as conservative as possible, but we're trying to predict a few hours before operation if we're able to fulfill our promises to TSO. So basically, like sometime before that, we have time to call and say, yeah, sorry, this is emergency, we did it like this, we didn't expect it, we tried our best, but this is how it happens. It usually never happens, but if it is, we need to do this and this should be in place. No questions asked.
Henrik Göthberg:So step number one when we talk about the engineering challenge, what we're doing it's a software engineering challenge, and it's a software engineering challenge and it's a software engineering when we still have unstable or not so mature devices to deal with. Got that Humor me. Now, if I still want to talk about what is the algorithmic dimension of this? What are the types of machine learning techniques? What type of mathematical algorithmic problem is this optimization problem?
Kiryl Zhdanovich:Yes, so for now we are not solving that much optimization problem as I described about future. So let's say I will name two main problems that we have now. First one we need to predict how much load we will have from electrical vehicles and then how we calculate it. So basically, you have a graph or like you can like think about it like every day, we need to predict in the future 24 hours how the day will look like for certain customer in certain area. So this is like the load for the grid, like saying this hour we will have one megawatt of ev like power during this hour. And how do we do this? How we do this like predictions, yeah so what is?
Kiryl Zhdanovich:the techniques. Yeah, yeah, so this is like time series forecasting, so everything. Everything works here like in in a sense.
Henrik Göthberg:So uh, like you could try different machine learning techniques for time series type problems yes, exactly.
Kiryl Zhdanovich:So you know, there's like some very simple ones starting like that when we're starting, like with arima, prop head libraries. So when, then, we tried like a boosted, boosted trees, which are, I think, working pretty well right now, uh, the next step is more like um, we try more like deep learning approaches or like more modern approaches, I would say in this case, and now it's the Occam's razor type topic.
Henrik Göthberg:Use the right algorithm for the job or the right eraser so you can go bananas on deep learning. Why would you go deep learning or why wouldn't you?
Kiryl Zhdanovich:So I would say, like this it's also like the amount of data which we're talking about, like it's not huge. We're talking about like 24 values per day and then we have, like I don't know, a year year maybe of data, or like year and a half max of data. So can we utilize it, so how this will be utilized. I think what is interesting here is more like okay, we have few customers with similar fleets. Like, okay, we have few customers with the similar fleets, how can we train maybe the one model using all of their data, if they're fine with this, like because it's anonymized and it's like federated machine learning like something like this, and then like, okay, can we?
Kiryl Zhdanovich:because we always have like when we have a new customer, we have a cold start problem. So when someone comes to us and they say like okay, how much time do I need to wait for you to start participating in the market?
Kiryl Zhdanovich:because we cannot say okay, we will just 100 kilowatts you need to have the data, you need to learn their profile in order to understand how they can participate yes, but then, like, we have like an assumption, let's say, like that the users of this, like because you know it's, let's say it's not industrial, it's like EV chargers from people's home. Do we think these customers are similar in a way, or are they different actually? Or can we train some core model in this case, where it maybe will be a deep learning model, when we have much more data, much more inputs in this case, and then maybe we'll fine-tune it in this case? Or maybe, okay, okay, we have a new customer. Can we utilize this master model with as little possible data that we have right now?
Kiryl Zhdanovich:But our goal is, like, is to reduce time to market, to market in this case. So, and I'm like, maybe I'm boring, old schoolish, I'm trying, you know, baselines first, models next, and then try to make things complicated. So what we spend maybe the first year at Fever is more like, okay, we want to have data in shape and ready at any point of time, because if new algorithm pops up because Google recently announced that they will have, like in GCP, they will have a service for time series forecast If this works better than whatever we did at Fever, like, why should we wait? We should have data prepared for this.
Henrik Göthberg:So, in a sense, now you need to think about optimizing the data, but actually, the more you're collecting data and having it in a good quality setup, you can actually feed that into new technology when it comes absolutely.
Kiryl Zhdanovich:and then this so this smart, yes. So the thing is following because you know what, like when I was working at spotify, there was like um at the beginning. When I joined, like cto, spotify said there will be no r&d department. We want all departments being r&d departments. After some time there was r&D department. We want all departments being R&D departments. After some time there was R&D department and these people were doing like a lot of things and then suddenly OpenAI publishing their tool where it can Spotify suddenly convert podcasts into different languages. And then you're asking okay, we had this R&D department, they probably were doing some useful thing. And then this small startup did something like incredible and suddenly years of our work kind of of useless, but we have data ready. We have data ready and then we can convert it to maybe something better.
Henrik Göthberg:So the bottom line, the research and collecting the data and working on that problem made them ready to now jump on the train much faster.
Kiryl Zhdanovich:Absolutely. And then we're trying you know, because our big data is actually small data, yeah, so we're trying the same. So we want to have like data in a good shape accessible to anyone in the company. That because it's kind of the same as the early days of Spotify. We believe like anyone can do something with the data if they want to do something so and want to do something so. And I mean I always get like a little bit shy when people say, okay, what algorithm you're doing?
Kiryl Zhdanovich:And then they probably say yeah, we're using this version four neural network from this crazy paper. I'm happy, we're not. I'm happy, we're not. I'm happy, we're like delivering results at this point and we're learning market.
Henrik Göthberg:Yeah, but I heard back to Pontus Falin, who is working with data analytics in TSO. I mean, like when we're talking about national security class data and stuff like that, he pointed out the explainability is so important in what we are doing. So it's actually it's not really good practice to use too sophisticated stuff. We don't really know what it's doing. So he was almost arguing like deep learning when it becomes more black box is kind of difficult. Do you agree with that?
Kiryl Zhdanovich:Yeah, but I think at this point of time, the law of big data kicks in. If something works, do you actually care why it works? Yeah, so like I mean, I don't want to say this about energy, but it's a tricky point.
Henrik Göthberg:When we have music, right, I would say the law is real, yes. When it's energy, when it's this, you know national infrastructure, you know sensitive stuff, right, I guess you're right.
Kiryl Zhdanovich:But then you, you, your threshold, you know, you're just you, you know it's not the same game yes for us, for for us for now is actually we are like what I love about here, we're constantly in the learning mode. Yeah, so and then? And actually we are trying to understand, okay, if these prices are like this, why like? Why this Is it random? Is it because of this amount of I don't know wind? Is it because of this how a hydro plant is not working this day? So why did this happen?
Kiryl Zhdanovich:The same thing is about actually electrical vehicles.
Kiryl Zhdanovich:I have a colleague, daniel, who has a car and he's like he educated me a lot how he uses car and actually I think he might qualify as an average EV user, and he told me a lot about how he thinks about car.
Kiryl Zhdanovich:And I mean, obviously, if something works, we don't understand why, at some point of time, we can be fine with it. At some point, I guess in the future, we should say, yeah, actually it was bad that we didn't understand, because we need to learn this now. And we are now, I believe, like starting simple, but already good enough. You know, like to attract customers and say, okay, like this is our performance, and then you can compare it to like with other companies. So we're like, from my point of view, I'm happy where we are now, and then I think we're have like a good foundation for the future. Good foundation for the future which is like, which is important, I believe, as you said, like with the all the advancements of advancements of um other lamps and everything. So I think we just should be ready for, like, what comes next.
Jonatan Raber:I'll get, if I can just add. I think the um you're. You're completely right in that. Uh, it can't be too black box right now, because we're also on a kind of journey together with the TSOs to help them understand and build confidence in what we're doing, because they're not used to this and so if we can't explain how we're doing things, they will never trust us to handle parts of the critical infrastructure.
Henrik Göthberg:Yeah, so this is also part of the maturity game in terms of the whole society and our decision makers. This needs to be done stepwise. Yes, I mean, there is another angle to the same question, and that is a little bit too. Instead of looking at the engineering problem, what's the data problem? How is data collection? What is the challenges of collecting data in this ecosystem?
Kiryl Zhdanovich:you know what is that all about since so far we're not participating in like real-time trading. Yeah, so for us it's more, more or less, like they're things like okay, what were the different type of prices, what was like the energy expert in different parts of Sweden, like in Scandinavia, like all of it, and other things is like, okay, how our batteries are behaving, so like our batteries actually provide to us like a lot of data. The same thing comes to electrical vehicles.
Henrik Göthberg:So let's sort me out here. Like so we put the battery, let's see if this is one of them more like I would put at home. So then you basically connect the battery and you set it up so it sits sort of close to your main box, fuse box, fuse box, right and then it needs to be connected. What is the data you are then measuring and looking at? You need to get the feeling for the profile of that house right how is it used? Or what is the data there? Is it smart meter data, kind of stuff that you're tapping into?
Jonatan Raber:So we, yeah, at times we look at the household data. Yeah, Kind of to observe.
Henrik Göthberg:Not on the inside, on sort of.
Jonatan Raber:Yeah, like for the full household. Yeah, but for us we use that to set constraints on how different assets can be used. So for a battery, if you know there's a constraint being the the fuse box, yeah, the fuse size of of that household. We need to know at what times that fuse is also being kind of loaded by other other consumption, or if there's production in the house, meaning solar panels.
Henrik Göthberg:Yeah, so this is the. This is the household part of the data collection. And then you talked about the battery itself. What, what type of? Is that how the battery is behaving?
Kiryl Zhdanovich:absolutely so like, since we have, like, different types of batteries, the most critical data is more like how battery reacts to our commands. Okay, so, if we ask battery to discharge a certain amount of power or something, how fast we were to receive this signal, because it's crucial for our qualifications and performing and like how like, yeah, so like reaction to our, to our commands is like the most crucial data. But it's also like in the real time saying like, okay, are you functioning? So like, is there like some heartbeat to the battery? So because disturbance no one tells you in advance, the disturbance will come. So, basically, like we assume that everyone should be ready and then, yeah, you know, there can be a lot of problems because, like, home internet is not stable, maybe it's like in garage, like away from Wi-Fi, it's like I don't know, maybe Wi-Fi doesn't work today, so things like this.
Jonatan Raber:So the data you know it could be everything from the state of charge of a battery and if they're different battery cells, have they drifted? If they're different battery cells, have they drifted? New battery technology tends to drift a lot in terms of state of charge, depending on if it's warmer or colder where you have this battery. Sometimes the bottom sections of a battery are colder because it's in a garage and that screws with the entire performance of a battery. So we do a lot of testing to understand the dynamics of the longevity of a charge.
Henrik Göthberg:So this is monitoring data. Is this real time? Is this sent in batches, in minutes, or how much do you ping your battery on these topics?
Kiryl Zhdanovich:I guess we have batteries that send us at least 10 messages per second. So this is some type of battery, another one, maybe we don't want to talk with them that often, so then, and then, like charger manufacturers, they also kind of, because most of the so there are different technologies, but there are technologies when messages from EV chargers go to a charger manufacturer cloud, so they also don't want to, maybe, to send them often so, and this data we use more like for like, not for machine learning, but more like for data analytics analysis. Okay, how chargers behave, how batteries behave, what should we? How we can improve our future algorithms when we control the fleet of batteries.
Henrik Göthberg:And what is the data collection or data sending in the other end, towards the TSO, because if you have a balancing service, they are sending signals. What is the data?
Jonatan Raber:transfer on this side. Actually, do you have your phone on you? Yeah, okay, so I'll show you what happens. So turn on the slow--mo video slow-mo video yes and let's see if we have these type of lights. I don't know these likes work, but hold it up to towards the light, so I should, I should. Basically, does it flicker?
Henrik Göthberg:I you're talking about the flicker and the frequency.
Jonatan Raber:Yes, yeah, so anybody can do this, I guess, yeah, I can't see it.
Henrik Göthberg:Oh, I see it. Over there it's flickering.
Jonatan Raber:Yeah, yes, so the flickering is actually the frequency of the grid. Yeah, the 50 hertz, the 50 hertz, so the 50 hertz is what we have in Europe. Yeah, and that is if it's exactly at 50 Hertz, the grid is imbalanced, so you have as much production and consumption at the same time. So if you're very, very good at measuring the frequency of the grid we have pretty expensive hardware distributed all over Sweden we can listen to the exact balance of the grid If it dips below 50, more energy is needed.
Henrik Göthberg:So Fever is doing this is doing this, yes, so fever, in your, in your negotiation and partnership with uh, with tso, you are one of the key data collection point is to listening to the frequency of the grid. Yes, in strategic places.
Jonatan Raber:Yes, as a way to signal actions or but and here's the fun part- yeah, if you, we have the exact same frequency if you plug it in into a socket here. Yeah, as if somebody plugs it into a socket in oslo. Yeah, uh, so uh, we just need expensive equipment to measure this frequency and we read it many, many, many times a second. Yeah, because then we have a perfect state of what the grid needs from us. Yeah, so the tso in this product.
Henrik Göthberg:They don't actually have to tell us anything because it's kind of built into the grid, because it's built into the grid and but what then is happening is that you're listening, converting that into a, an analysis and, ultimately, an action. So basically, this is happening then in real time. So in real time, like in microbursts, get more energy from this guy, get more energy from that guy, stop using that one, and so this is. And what are these algorithms? That sort of this is then balancing what is happening here. Like, what is that? Because then there's a lot of signals coming in, you're listening at a very, very high frequency and and then it should execute according to that yes, when it comes to algorithm, in this case it's kind of like home-cooked algorithm.
Kiryl Zhdanovich:especially when it comes to evs with batteries, it's like a little bit simple, like if you have like industrial size battery, it's like some of them have meters inside them so they can do everything on their own. If they not, then we need to send them a command, basically yeah, so you can think about, it is okay, like, if the frequency is very, very low, you need to operate 100. If it's not that low, maybe 50, and then if, like, if it's not reached, like then zero. So basically you convert frequency and your like amount of energy you promised to TSO to how much battery should charge or discharge right now. The same goes to our electrical vehicles.
Kiryl Zhdanovich:So how it works we're listening to the signal, okay, we detect disturbance, it's called what it's called. And then we know, okay, how severe is disturbance? Okay, it's maybe that severe Like this, like 20% of what we promised. Then we pick chargers, send commands, wait. While we're waiting, frequency may change to better or to worse. And then we need kind of like say, okay, this was our previous state, we're now in a new state. How do we tackle this new state? In a way. So we're now in a new state. How do we, how do we tackle this new state in a way, yeah, so like, do we stop more chargers or do we cancel maybe some commands, if it's possible? Still, so, that's, that's that and that's as you said, like it happens, like I don't know, 10, 20 times per second, but so would you say I would.
Henrik Göthberg:I would reflect over this like this is the real mathematical engine of everything, which is part of the core of what the fever optimization is doing. Is this proprietary? Is it you have cooked that up? Or is it something that you can buy? Or this is at? This is working progress, I guess, or how is?
Kiryl Zhdanovich:it. I would say word mathematical might be true, like just part of it, yeah, and I'm not sure if you can buy it. So it's actually what we're, what we're selling. In a sense, we say, okay, we, we got some like experiments, we got some data, we got some experience with it but you need to code it in such a way it actually works yes, but it's also like in in terms of software, so it's reliable and it's fast but it's also like if we, as I said, like if we stop too many chargers, what do we do when it's actually because worse this is when it's.
Henrik Göthberg:What is the AI? What is the software? It's a system, and you need to optimize for unreliability, you need to optimize for latency and speed, you need to optimize executing commands. And then in here, of all this, what is software? What is data processing? What is algorithm? It's a system.
Jonatan Raber:This is how I would summarize it. I think that's very perceptive of you.
Kiryl Zhdanovich:I think, if it just may, so like the whole the full cycle. I think, when it comes like to software like AI, engineering and Fiverr, when you get flexibility asset, you need to prove to SVCore, to TSO, that this asset is actually can be qualified to participate in this market. And then you need to write certain type of tests, which also involves, like you know, engineering skills. It involves mathematical skills because, like, there's something called Nyquist diagram that we need to plot, which involves, like complex numbers, and then you know anyone can say, okay, this should be provided, but we coded it ourselves because at the beginning it wasn't really working for us. So SVCore provided us, like Windows tool to build it and it didn't work for us, unfortunately.
Henrik Göthberg:You need to build a tool as well. Exactly.
Kiryl Zhdanovich:We implemented, like my colleagues implemented it on their own. And this is like this whole cycle you run tests and then you know you fill the document right, like jonathan has phd in filling in filling documents, yes, and then you send it.
Kiryl Zhdanovich:You wait, and then you are like, okay, I'm approved for this. What happens next? We take this asset with it's like power, we go to a bidding market, so then we predict, okay, what was like should be our price, and then like if it's battery, we know the capacity. If it's electrical vehicle chargers fleet, we need to predict capacity. So there is like already starting, you know some machine learning which is maybe not related to this specific device.
Henrik Göthberg:When it comes, to prices, but this is an ecosystem. I think what are the main points that I want to make with a conversation like this for anyone listening. You cannot start putting machine learning engineers over here and data guys over there and then you know this is a cross-functional team working with a very strong system or product mindset. Would that be a fair summary that? Don't even try this at home, girls and boys, if you're not willing to work together as a team on this, Because I see this over and over again. Here's the AI guys and here's the machine, and this is typical of the analog company. They are not the software companies. They haven't figured this out yet.
Kiryl Zhdanovich:Absolutely. So there is certain things that you cannot say I am a kind of engineer and then I cannot do anything else, or like, okay, I can build this model, but you use this model. No, we're kind of like we're all working together, so like when this is when we participated in the market, when we did our predictions. Then it comes to execution part. This is pure software engineering part, but it's also we want to have best strategies. Maybe we can learn these strategies from our previous experiments and from the data and all of it.
Henrik Göthberg:So it's a full pipeline where you know like… and you're a multidisciplinary team, you're working as one entity but you're T-shaped. You understand the problem of the grid, but you have your own super strengths in this team, and then you try to solve whatever comes in front of you.
Kiryl Zhdanovich:Yes, yes, I think like one thing. I'm feeling sad a little bit because we're becoming, you know, bigger. We have like more people and then, before I could write code in any part of the company, Now, I'm already you know I cannot focus on everything right now, so sometimes when people are solving problems, I really want to solve it, but I need it. Probably I will have better input in another place. Wow.
Henrik Göthberg:All right, there's so much to talk about. We're already going for two hours, but I think we should wrap up with a couple of a little bit more like bigger questions. I think one macro question or geopolitical question. Yes, I think it's a geopolitical question that I think you guys should be maybe closer to than others. It's the sole topic of cybersecurity and where we're thinking about the world. And. And. Where we're thinking about the world and where we're thinking about this, because I think, typically, infrastructure, like the grid, is one of the key stuff you want to protect on a national level, and cybersecurity strategies, or cyber war, of course, as infrastructure attacks as one of the main vectors. So what is your thinking and how do you discuss and think about cybersecurity at Fever and how do you try to be resilient?
Jonatan Raber:What is the cybersecurity conversation going on at Fever? I mean, I would say it started from day one.
Jonatan Raber:We know, this is critical infrastructure, even kind of. Our investors have to go through pretty rigorous security processes. So, going back to what you said around building a system, this is a core part of how we built it not only stability, but redundancy, very high security, because if we do our jobs well, we're going to be a critical part of this ecosystem and we need to do it in a way where we're not, uh, the first attack vector, at least yeah, so I must.
Henrik Göthberg:I can imagine like being part of the infrastructure, like a grid. You know that you need to have a strategy for not being hacked or being into combat hacking, and that is one of the fastest moving technology trajectories like there's no new patches, you know. So how do you deal with sort of having your cybersecurity software division being one step ahead? How does that work?
Kiryl Zhdanovich:So, like in a simple case, we have, like our, we have a colleague of ours whose responsibility is, like this part, so he educates us about a lot of things. He has experience, like he has hardcore security experience from his previous works. So we're like, not only employ, like you know, the best practices of, like you know, just basic things like changing passwords, like encrypting everything, so how we work with clouds, how we work with our devices and all of it. So this is even out of the question. This is solved.
Henrik Göthberg:This is hygiene. This is solved.
Kiryl Zhdanovich:Yes.
Henrik Göthberg:But I mean, I'm more about how to stay, because this is a continuous, the operating model is a continuous improvement.
Jonatan Raber:So I'm not in the kind of tech coding, obviously, but I think one thing that we speak a lot about is keeping things simple and nimble and not making kind of our product and tech too big unnecessarilyarily, because that just increases the chances of slipping up. So it's just one thing I kind of observed from kind of across the table.
Henrik Göthberg:Yeah, because I think the security then becomes an architectural strategy as well. Like you have data core framing requirement in everything you do, so you take choices. We don't want to do it like that, because then the blast radius is too big.
Kiryl Zhdanovich:we'd rather do it like this containment and all this yes, and then we're trying like to also like to separate our like services. So let's say, like if, for some reason, our like electrical vehicle for this customer is down doesn't mean that, like, all customers like should receive the same experience in the way. Or like if the battery service, like that controlling this certain batteries down shouldn't influence all our assets at fever. So things like this yeah, they're just like you know, they're very, very simple, very simple approaches that actually saying that, okay, there shouldn't be like single point of failure in this case.
Henrik Göthberg:It's not a coincidence that the core backend team comes from FinTech, autonomous vehicles, Volvo M Because all of them has, in different ways, high safety standards. So I have to have high safety standards in different ways. But where do you think the whole cybersecurity or cyber threat? What's the trajectory on that? Are you anticipating changes even now, with LLMs or different attack vectors coming in different ways of doing it? How are you seeing cybersecurity topic evolving?
Kiryl Zhdanovich:So our tips now to fail Swedish grid. Five tips to fail Swedish grid.
Henrik Göthberg:Five tips.
Kiryl Zhdanovich:No. So what I wanted to say that it's like I think the main point about the future and everything is that there will be a lot of companies and a lot of distributed assets. So it will not be possible to say, okay, if we're killing this point of a grid, or if we're killing this hydro plant or whatever plant, that doesn't mean the Swedish people in different parts of Sweden will stop receiving electricity. It might be a case now, but it shouldn't be a case in the future. So the future of the grid is distributed, so if one part is not functioning, another one is for sure should be functioning.
Henrik Göthberg:So it really means that if we build this right in the distributed world with a high security topic, we will minimize blast radiuses of any attacks or any problems. This is one benefit of a distributed architecture, I guess.
Kiryl Zhdanovich:Absolutely. And then having small distributed assets will have some neighborhoods for some time when things are getting fixed. So there will be no problems like with hospitals or something.
Henrik Göthberg:Yeah, and the counter argument to that is, of course, that there will be a lot of back doors. Now, can you, can you? I mean, like with the distribution, you have more attack vectors and with more attack vectors, potentially, this is where you get the worm in to send.
Jonatan Raber:So there's two different ways of looking at distributed risk. I think the uh, the kind of the new world brings electricity and uh kind of of information much more closer. So if you have a data center that goes down, that could also mean that parts of the electrical grid goes down. So you have a different type of coupling.
Henrik Göthberg:Yeah, this becomes a new type of coupling and a new type of cybersecurity context. So go there a little bit now, because now we're getting back into what is the AI divide and what is the electricity divide and what is the convergence between information and electricity. So I think your starting point was quite interesting, but we are circling around information, AI, electricity and how this is converging in different ways.
Kiryl Zhdanovich:I can absolutely say that there will be probability of something bad happening. So there were, like we already saw, some incidents in Finland where some software was giving some very low numbers on the market. So for some time in Finnish, a grid, you will get 5,000 euros, if I'm not mistaken, per hour if you consume an energy 5,000 euros because of software issue. So this, like, we will absolutely have problems with it. So the question is okay, like I think SVCore, like its role as a TSO, will be in the future. Okay, how we invite more people and make sure that, even if one of the actors is not functioning, how to make sure we still have grid in a good shape.
Jonatan Raber:I think my answer is I'm not educated enough to kind of talk about the new state of cybersecurity.
Henrik Göthberg:But we can back out of the cybersecurity topic. I'm going to test and hypothesis on you where I think that there is. There needs to be a wakeup call in companies like SVCore, because if you have the understanding of convergence between software algorithms and hardcore energy and energy grids, you kind of need to have a team and organization and operating model that reflects that. So if you have an organization where we sort of have the hardcore craft power engineers over here with their view, you're told about the two cabs and then you have new guys who are more distributed.
Henrik Göthberg:I'm just going to add now to you you know what SVCore needs to be 50% software company as much as their own electricity company. So I saw this already in Vattenfall and I'm like you don't get it right. You think you're a big asset operator, but in a distributed world where information and software is going to be so big part of the game and we have just now proven that you can't do shit if you're not putting a cross-functional team together I'm a little bit worried about you know what? Have you woken up that you're a 50% software company and then 50% AI company and then 50% electricity companies? You're in 150% company.
Kiryl Zhdanovich:Right now it's a joke, right, but I think I'm a little bit worried.
Jonatan Raber:Do they understand themselves as a software company? They kind of become the sum of all the parts in the system. And if parts in the system are how?
Henrik Göthberg:will you manage the national grid if you are not understanding what it is you're managing and if part of your management is copper, another part is software and a third part is algorithms. What the fuck are you managing if you're leaving two-thirds of the problem at the table? I don't get it. Yeah, this is a provocative question, but I think we're talking about the Swedish infrastructure here and where they're only doing one-third of their job.
Kiryl Zhdanovich:I have two answers maybe. First one when we're selling, like I said about this qualification process, when we're sending this qualification, sometimes we get amazing questions back about our software.
Henrik Göthberg:Which kind of?
Kiryl Zhdanovich:you know, when you talk to someone and someone asks right questions, you're like okay, so you probably know what you're doing. And when they ask these questions, I'm like wow, I actually feel a little bit more positive Because you know you rarely can say this when it comes to software and government company. This is the first thing. And the second thing I think Anders is an amazing example of a person working in the government organization who worked before in the companies where, like which you know, have high software standards. So I hope actually, when you know, when companies in Stockholm were laying off people, I was thinking, oh, maybe finally companies like SVK or like Migrahundsverket or Skattaverket, they will finally get like talent that they deserve, meaning, deserve like that. Like you know, in my perfect world, smart people should not only solve, like, the problems of how they sell ads, but also, like you know, doing some uh, you know things that help countries where they live.
Henrik Göthberg:yeah, but because I don't want to bash in any way like this, but I think this is it's, it's. It's. It's the analog companies of the world that all of a sudden is growing into a new type of role. So at proud analog history, they've done an amazing job and they now need to pivot to some, to something that you know. No one knows exactly how it's going to look like, but it, but it's an interesting challenge and I and I'm very humble to it. So I'm just I it, I'm just saying oh, I just hope the right people are waking up to it in time. And it's good to hear and reassuring when you say, when you get good questions back, that we know there is, I know there's so much good talent in software and AI over there. I just don't think it's enough.
Jonatan Raber:I think you need more. I think you're right. I think you need more. I think you're right. I'm just happy that there's something pushing back to the Klondike moment here.
Henrik Göthberg:Also that we also need to think about how is, of course, to partner with startups and partner with companies like Fever to solve the game together. And I think that's why I'm so happy to have you on the pod today and because I think that's actually the fastest way they can solve that major talent issue. They would have otherwise. So I think you's actually the fastest way they can solve that major talent issue. They would have otherwise. So I think you're doing a great job, guys thank you.
Kiryl Zhdanovich:Thank you and thank you for inviting and I'm a little bit sorry that we didn't discuss hiring yeah, maybe part two, maybe part two.
Henrik Göthberg:I was getting to it but never got to it. Okay, thank you very much, guys.
Jonatan Raber:Thank you thank.