E114 - 2023 in Review

Join us for the grand finale of AIAW Podcast's Season 7 with Episode 114, "2023 in Review," where we embark on a reflective journey through the past year's remarkable evolution in AI, covering technology, people, and groundbreaking innovations. We'll revisit the most enlightening moments and discussions from seasons 6 and 7, providing a detailed summary of our esteemed guests and the wisdom they shared. Delve into our personal discoveries and the lessons that 2023 has imprinted on us. As we bid farewell to an eventful year, we'll also cast our gaze forward with predictions for 2024, speculating on the trends, challenges, and advancements that await us in the ever-evolving AI landscape. Don't miss this comprehensive and thought-provoking episode, designed to celebrate our journey through 2023 and to ignite curiosity and excitement for what's to come in the world of AI.

Follow us on youtube: https://www.youtube.com/@aiawpodcast

Transcript

Stock Prediction and Holiday Plans

Anders Arpteg 0:00

Mainly one is the stock prediction project that I'm picking up from the summer the stupidest thing you can use AI for but I need to do it so I don't have to do it manually. Yeah, I'm going to do that. But the second thing is basically my girlfriend has an audit of her daycare business, daycare business, yeah.

Anders Arpteg 0:25

And it's really an extensive audit, so they have to answer a shit lot of questions and every question has to like please describe and give your documented procedures for X, y and Z. And everyone is freaking out because it's so much work to do so, but I'm going to help with that. So it's going to be half of my time during the Christmas holidays to work on the audit. Second half is going to be for my own personal enjoyment the AI project and I'm going to use so much of AI power to do the best ever.

Henrik Göthberg 1:01

You're going to maximize for your own pleasure to use AI.

Anders Arpteg 1:05

Not for my pleasure, but partly yes.

Henrik Göthberg 1:07

Of course.

Anders Arpteg 1:08

It will be a bit more fun to work with it when you can use Chatibiti or more Google barred, I think than Chatibiti to make the best ever audit. They have seen Every process so well documented, so well written. They've never, ever seen such well, what are you planning?

Henrik Göthberg 1:29

I mean, this is going to be I know you you're a barred fan, right, but do you think it's going to be barred in Chatibiti? Mainly, are you going to use some other tools?

Anders Arpteg 1:38

No, but they are the top one. I do have access to it, in Sweden at least, unless you use some VPN stuff. But yeah, so probably barred. But I will try Chatibiti as well. But Chatibiti to me has been failing quite a lot recently and it's going out trying to download stuff on the Internet and then they're trying to do the code interpreter and lighting stuff and then it fails and it takes a long time.

Henrik Göthberg 2:00

But you, you like that right, that part is better.

Anders Arpteg 2:02

You like that, I like that I like to support the underdog.

Henrik Göthberg 2:06

Oh, the underdog, Google is the underdog. I like to. I like when Google is the underdog.

Anders Arpteg 2:11

Yeah, so that's going to be one fun thing.

Henrik Göthberg 2:15

But you make me feel very lazy because you're still going to do some sort of proper value adding stuff. I just want to ski and snowboard.

Anders Arpteg 2:24

Yeah, you're going down to Austria, right.

Henrik Göthberg 2:29

Yeah, we're driving down, that's the big thing. So we are spending some time in the snow. So we're going to be at home over the Christmas and then drive to my mom and dad's on the West Coast in Sweden and then from there we're going to drive down from Sweden to Austria.

Anders Arpteg 2:43

In your Tesla car as well?

Henrik Göthberg 2:44

Yeah, in the electric car. That's correct.

Anders Arpteg 2:48

You don't want to say Tesla, oh yeah.

Henrik Göthberg 2:49

I can say Tesla. I mean, I used to be a Polestar fan. I know I'm a Polestar fan boy and Katrina still prefers the Polestar. She thinks it's nicer and better. Yeah, she likes it better.

Henrik Göthberg 3:00

You want to play. No, I mean like it's from a car experience or from a you know. So she has. She's never been a Tesla fan in terms of how it looks and all that From my point of view, driving and if you're driving far oh, huge difference. I drove far this summer. When you have the whole charging infrastructure you use, plug and play, it's unbeatable.

Anders Arpteg 3:25

How long does it take normally for a supercharging with Tesla?

Henrik Göthberg 3:28

I mean like so, I mean like this, the whole normal thing, right? You never go below 20%, you never go up above 80%, unless you're doing sort of you know, starting off somewhere, right? So typically, you know, typically when you are driving far, it's around 20%, 15%, and then you stay for half an hour.

Henrik Göthberg 3:50

It's half an hour, basically, yeah more or less around half an hour and then and then you end up at 80 to 95% and the whole point is like after 80% the speed, the charging speed, goes down so radically right, so you can really see it when you see how much electricity gets, like you know, after 80% it's not so. It's the only time when you stay beyond 80% is like when you really need the 90% to get to your destination, otherwise it's.

Anders Arpteg 4:22

If you're charging overnight, then it doesn't matter.

Henrik Göthberg 4:24

Yeah, so charging overnight? You, of course, but when you're out and about, I don't see any points that you're staying for longer than 45 minutes ever, and typically it's like 30 minutes, so you're short for 30 minutes and you can drive for three hours or something.

Henrik Göthberg 4:38

Yeah, so typically, so typically, when I did a really long trip, I drove from Stockholm to Napoli this summer, in many stages of course. And then it's quite simple because you simply put in I mean, like the way I do it, I plan it, not the whole way I'm planning, what am I trying to achieve today, right. And then when you say that I want to get to Berlin today, or something like that, and then it sort of plans out the optimum charging. So sometimes you look at where the superchargers are. So you know what I'm going to start. I mean it's going to drive for 10 minutes out of whatever city and then I'm going to stay at that supercharger because it's one of the fast ones and we're going to top up 20 minutes. And then you know it's all about also what becomes the optimum flow of the trip.

Anders Arpteg 5:29

I still remember one of the talks at the Nordic Data Science and Machine Learning Summit like a month ago, where a guy from Tesla spoke about the charger stations and they used first principle thinking in how to really and especially in Germany as well, as you are going to drive through. So they really thought through how to put these kind of charger stations in an optimal way to make the driving experience good.

Henrik Göthberg 5:54

And I must say I actually have an AP test driving from Burroughs to Tuscany, because my dad was driving a Volvo full electric and I was driving a Tesla full electric. So we could really you know, as nerds electric car nerds compare our experiences. What was the result? I mean like this obviously, when you're driving across a continent and you're going through different countries, the problem is like if you don't have a Tesla where you just go plug and play, you do nothing right, you just plug it in. You need to kind of get the payment is all the way to fix it, the payment, everything is done right.

Henrik Göthberg 6:41

But my dad, he has to figure out who's the local operator and of course, the local operator in Austria or Germany is not the same as in Sweden. So then he needs to basically fix that. He needs to. You know he can get Ionity or there's a couple of European Like Ionity is the biggest one in Europe, so he basically that's how he does it. He I stick to Ionity but then he has substantially less charging stations than Tesla, of course, and the bottom line was like my dad was turning 70 and this is in the summertime, so we stayed with our families in a house and I had a Tesla charging station literally five minutes, 10 minutes away from where I lived and I could just go and plug it in wherever I lived, wanted, and my dad oh, I'm going to go and charge my car he was gone for half a day trying to find something that worked, you know.

Henrik Göthberg 7:32

So that sort of sums up If you're charging at home, that's the matter, but it's the long trip advantage that I think Tesla has nailed right now. Anyway, but back to your summer project. So this summer project, can you, Without revealing the secret source, what are you working on now becoming a winter project? What do you want to work on?

Automating and Optimizing Investment Strategies

Anders Arpteg 8:01

improving so, as I said, you know, I think it's the stupidest thing you can use AI for is stock predictions. So I really hate that. I actually have to spend time doing it, but the alternative of not using AI for that and you still have to make some investments because otherwise just leaving money in a bank is not a good idea either. So of course, I have to have a data driven approach to this and I'm forced to use AI for stock prediction and I think the service is out. There is horrible. So, yeah, I have to build something myself.

Henrik Göthberg 8:30

What part of the process are you trying to sort of automate or optimize, that you know? Is it basically nailing which stock to take, or what part of the process do you like to try to do better?

Anders Arpteg 8:42

I really want to have a proper, a more scientific approach to what really works. And what really works means you have to do proper backtesting. It means that you have to really have a buying and selling strategy that you test out and you can have data backwards that makes it possible to do proper backtesting. So if you do that and include the commissions and every other kind of aspects of buying and selling, then you can really show what works and doesn't. And when I first tried it, like two years ago or this summer, it didn't work at all. I couldn't hit index. Index was doing so well, so you couldn't really see a statistically significant improvement at all.

Anders Arpteg 9:30

But I could this summer and that was kind of fun and basically meant leaving the technicals, the actual time series on the stocks completely out of it and just going fundamentals and basically events happening for four different companies, collecting them and trying to really understand what does the meaning of some news article or quarterly report or whatever mean? And then from that trying to get some kind of aggregated view is going up and down.

Henrik Göthberg 9:58

But going for this fundamentals view. Is it many trying that view? Is it still very mechanical?

Anders Arpteg 10:03

I mean like time series approach this is not a field of expertise for me at all. I know some other people that are working in it and, of course, the big hedge fund you know are spending billions and billions of dollars into this In all methods. Yes, in every method possible, but it's a big difference in high frequency trading and there's no way to even go into that area. No, this is something else. Yes, so this is more long term and less frequent kind of.

Henrik Göthberg 10:29

But I was remembering like maybe, if I remember right, the summary of you know from the summer was a little bit like yeah, it's, I can beat it now, but I want to explain it. I want to understand and I really want to be, I want to be able to follow my own rationale. I want to sort of A-B test against my own judgment. How is it going with that?

Anders Arpteg 10:52

But I could see the recommendations. Basically, I got from the models to say buy and sell this, and I couldn't understand it. It didn't make any sense to me.

Henrik Göthberg 10:59

And I was still good.

Anders Arpteg 11:01

Well, I haven't followed it. I've done some investment, but I trusted myself more than a model still, so I haven't really had the proper A-B testing. I know, I can do it backwards. I can see it's working, you know, when you do the proper back testing. But it makes it so irrational kind of recommendations that I get that can't even understand. That scares you right, yeah, so.

Henrik Göthberg 11:27

I still haven't really, so this is what you're working on.

Anders Arpteg 11:30

I get some, you know, I get inspiration from it. So I'm sure this type of companies that I had no idea even what that company was, those recommendations I have trusted and did. But then for when I look at the technicals and I try to see how can it want me to buy this now Doesn't make any sense.

Henrik Göthberg 11:50

So you do, this is what you're working on to get this back testing working, the back testing is working.

Anders Arpteg 11:57

It's just that I don't trust the recommendations really.

Henrik Göthberg 12:01

We'll see.

Anders Arpteg 12:04

So I'm going to, of course, use the latest LLMs and you know there's so many more options now and especially, I don't have the huge GPUs you know that can use. I have like small.

Henrik Göthberg 12:13

But can you give us a hint when you're experimenting now, what is the techniques on a high level, without going into detail that you are, that you are sort of that you are experimenting with.

Anders Arpteg 12:24

Yeah, I have some secrets. I don't want to go into that, but trying to understand the meaning and get some kind of aggregated understanding. That that isn't, in short, is the idea of it, so you can't.

Henrik Göthberg 12:37

I'm not going to push you on this one Good.

Anders Arpteg 12:39

But still, you know you have to make it work on consumer grade stuff and then it can't be like 70 billion parameter models. It has to be something we have so many now awesome open source models that are much, much smaller.

Henrik Göthberg 12:52

But then I have. Then I have another question what type of hardware are you optimizing your model for?

Anders Arpteg 12:59

Yeah, it's a small, like 30, 60 kind of then media stuff it's, it's not, oh, no, 38.

Henrik Göthberg 13:07

But the point is to to, as a home investor, what would you want to have as your sort of setup? Right, it should work on that.

Anders Arpteg 13:15

Yeah, I would like to have an H100.

Henrik Göthberg 13:19

That is not really the home computer typical you could. But yeah, so it's a normal machine, normal advanced machine.

Anders Arpteg 13:25

I love that we see the current trend, I think, going from these huge models to actually having practical models, both when it comes to Gemini and these kind of 1.7 billion and 3.2 billion, but also now the recent one from Microsoft, the five 2.0. That is, I think, around three billion parameters and still kicks ass and that's very much possible to use on even very small grade GPUs. So, but maybe that's the segue.

Henrik Göthberg 13:53

Yes, this is the segue you know to do. You want to set up this episode?

Anders Arpteg 14:00

Yeah, yeah, I guess it's the end of the year, kind of Christmas special.

Henrik Göthberg 14:07

Christmas special. Let's call it like that. Yes, end of the year, christmas special. Summing up the year, making some predictions for next year.

Anders Arpteg 14:15

And we're not having a guest this time, so it's just you and me, henry.

Henrik Göthberg 14:17

Yes, I'm on the guest side. Yes, okay, henry is the guest.

Anders Arpteg 14:21

No, but it's kind of it's different not having a guest, but it's also nice to be able to.

Henrik Göthberg 14:26

I mean it's reflecting on all the guests, so like if today we don't have one guest, we have all the guests from episode 99 to episode 115, roughly Some 86, 80s guests from 86. Goran help us. When did the season start? Episode 86.

Goran Cvetanovski 14:47

Maybe that's a good segue. So I'm feeling like a judge in Yugoslavia.

Goran Cvetanovski 14:58

A judge in Yugoslavia. Yes, it's episode 86. Actually, we started the year with episode 86 with Josef and Viktor from Moodly, so that was actually the first one. It was focusing on MLOPs a very early one actually, because even then we were talking about, like, the future of generative AI and the upcoming AI Act. I was just reading to the transcripts and it seems like the AI Act we have been focusing on the entire year and then finally made happen on the end. But yes, 86. So in total this year we have had Goran is calculating in the set 160 minus 86.

Goran Cvetanovski 15:53

So, 34, all right 34 episodes in total, so that is not actually very bad. I also reflected today when we were speaking with Anders. I thought that this year and these two seasons went very fast. We didn't have so many guests but actually we did have a long summer break actually.

Henrik Göthberg 16:13

Yeah, we had a long summer break, that's true.

Goran Cvetanovski 16:16

But it was a very, very amount, a big amount of people actually.

Goran Cvetanovski 16:20

Very big diversity of people this year In the past five seasons before these two, so we are right now ending the seventh season. I think we focus mostly on the practitioners, so very technical and very business-oriented people trying to demystify the AI. That was the initial idea from the beginning as well, and this year we have been focusing quite a lot on diversification and especially how to implement this in larger organizations and public sector and etc. And what was very interesting and profound for me and also very satisfying is that we managed to be a little bit spontaneous and get panels instead of yeah, I mean there's a number of news.

Anders Arpteg 17:04

I guess for the way we had the podcast this year, one is the panels that we tried out, but also the news section. It's the news in itself.

Henrik Göthberg 17:13

I think it worked out surprisingly well.

Goran Cvetanovski 17:15

Yeah, it did.

Henrik Göthberg 17:16

So two small innovations for the way we had done the podcast has been panels and the news section.

Goran Cvetanovski 17:22

Yes, and we needed to change the supplier for the ham.

Anders Arpteg 17:27

Yeah, that was a terrible one. Yeah, but it is, for. I thought it was snacks we have every weekend.

Goran Cvetanovski 17:31

That's what happens with AI comes in play.

Anders Arpteg 17:34

But we also had a lot of big news, I think just last week, right to go around for the Christmas gift, for the podcast, so to speak.

Goran Cvetanovski 17:41

Yes, so we ended up being listed as number 16 out of the top 100 AI podcasts you listened to in 2023.

Anders Arpteg 17:52

In the world or on the ground.

Goran Cvetanovski 17:54

Yeah, globally. So this was actually a really nice thing to see in here, and it's also acknowledgement that we are yeah, we still make impact while having fun, because I remember some years back, a couple of years back, I think it was the same list. Yeah, it was the same list actually, so it was on the 47, I think it was Okay.

Goran Cvetanovski 18:17

So the list has dramatically changed this year and of course, we need to ask as well how do they construct it and etc. So it would be a little bit more interesting to hear. But there was a lot of podcasts that we used to hear about and they were there, but they were not there this year on that list. But it doesn't matter. For me personally, that list doesn't make any significance, but it's good to feel that we are making a difference while we are having fun, and I think this is the there's some sort of reward, you know, because we're doing it a lot, we love it, it's fun and we're learning.

Henrik Göthberg 18:52

But it's also rewarding to feel, to get recognition. I felt super proud.

Anders Arpteg 18:56

It makes some use as well. Not just us having fun, actually, people are also finding value from it. So that's what I'm saying.

Henrik Göthberg 19:03

We still try to understand how many is listening and how that works. It's still really hard when you're disseminating on all platforms Exactly.

Goran Cvetanovski 19:10

I mean keep in mind that we have been doing three and a half years. We have been stuck in this dungeon every Thursday Stuck.

Henrik Göthberg 19:18

I don't know.

Goran Cvetanovski 19:20

Well, you see, we need to make it a little bit painful.

Henrik Göthberg 19:23

Yeah, it's very painful, yeah, it is extremely painful.

Goran Cvetanovski 19:26

So three and a half years and we have done over 10,000 minutes of content. I think that we are still far away from Joe Rogan and the amount of content that he has put down, but I'm still happy.

Henrik Göthberg 19:39

I think we have done 10,000. Yeah, of course, because we're doing long formats, it sort of stacks up fast in some ways.

Goran Cvetanovski 19:48

And it's been an extremely good journey and a lot of learnings and great contacts, actually and people that we have had in this podcast.

Anders Arpteg 19:56

So I'm super satisfied and I like the diversity that you said we've had especially, I think, this year as well. It was all from AI for education. Of course, we had AI for healthcare, but also these kind of a bit more strange ones like AI for dancing with Kaisa or, yes, ai for fashion and these kind of things, and gaming as well.

Henrik Göthberg 20:16

Yeah, yeah, we had creative directors and we had gaming and we had, yeah, so many angles, so many angles, and we can go through some of them.

Goran Cvetanovski 20:25

I think is very interesting. But what I'm trying to get at is that a lot of things have happened this year, especially with the chat, GPT4 and BART and everything that has happened, and we finally came to a moment where AI is immersed in almost every beats of our lives and even the grandmas are talking about AI right now. So the purpose of what we have been talking and working with the past 10 years, which is demystifying AI and making sure that people understand the value of it, we're already there.

Anders Arpteg 21:03

I think the big change is so it's very important we always believed. I mean, we've been or I've been working with AI for over 20 years and I've been trying to tell people that it will have a huge impact on society, but people are just rolling their eyes and saying yeah, it's just another technique, it won't really make a big difference.

Anders Arpteg 21:19

But now, when everyone actually can use AI, even non-technical people, that can just go to a website and ask it any question and it gets surprisingly good answers and you can see it is some kind of intelligence across the board. You know these words. You know who said some kind of intelligence across the board? That was actually the Kasparov. When Deep Blue beat Gary Kasparov for the first time in 1997.

Henrik Göthberg 21:48

Some across the board has a deeper meaning.

Anders Arpteg 21:51

I can see. I would like to remember the exact quote, but something I can see, something. It's a new type of intelligence across the board and he could see it done in 97. But now I think everyone can see it across the board. Exactly that. You know, it is some kind of intelligence happening.

Goran Cvetanovski 22:08

Yes, but then I have a small rabbit hole about this. It's actually a little bit like a question, do they?

Anders Arpteg 22:16

I mean, do they have intelligence, or what? No, no.

Goran Cvetanovski 22:18

Do. People will ask like, oh wow, this is AI, it's intelligence. So, basically, I think that the point with technology is that it's you know. You know that this technology is going to change everything when people do not even question it. And people usually question technology when they don't understand it, what they say, but eventually they don't understand how they can get some value out of it. What do I get from it? Right? So for the past seven years, nobody got it and everybody was afraid of it. And suddenly, now you have all the girls and kids in the school using it and everybody is using it. Nobody is even questioning like, okay, what is behind it? Oh, it's a large language model. How many times you have spoken with your kids like, oh well, it's a large language model?

Anders Arpteg 23:02

behind it. It can be commoditized, or you can just do it some way, exactly Very quickly.

Goran Cvetanovski 23:05

Very quickly.

Henrik Göthberg 23:06

Exactly, very fast.

Goran Cvetanovski 23:08

And that is my point with this. It's like so. As the this year is passing by and we will talk about the biggest breakthroughs in this, I think that it's even more important to have platforms like this, where we talk about this technology.

Anders Arpteg 23:23

And also perhaps explain the history, because people think, you know, it's a natural part of everyday life these days, but it wasn't this case just a few years ago.

Goran Cvetanovski 23:32

Yeah, One year ago it was not the case. Then there is a lot of I would say there is a lot of confusion. What is out there? There are a lot of Right now. I was looking yesterday at some training programs about utilizing AI in all sorts of things from people that I have never actually seen in my life. Of course, they are certified trainers and etc. But they are not per se AI people, right. So there is a lot of people that are just utilizing the hype in order to make money, and even more people will do it next year. So I think it's important that there is a platform where you basically can hear about like okay, this is how it's done, this is what it is, and etc. So our purpose with this podcast is even becoming more important in the future.

Henrik Göthberg 24:17

Because demystify to some degree has always been about moving away from the hype and, more practically talking about how does it work? What do we work on If you're a business person? What are you working on If you're trying to implement AI? If you're a data scientist, what are you actually working on If you're trying to implement AI? If you're a data engineer, etc. Etc.

Henrik Göthberg 24:39

And I think the challenge with the hype that, when it goes up like it has done this year, is that we get bombarded with messages around what we should do or what education we should take or what technology we should buy, and this is sort of building the FOMO, the fear of missing out, and in reality, it becomes super hard, I think, to step into this arena and start doing something useful with this stuff and then.

Henrik Göthberg 25:08

So then I think it becomes even more important. You know, like everything else, right, you need to start being careful. Who do I listen to Exactly? Who is my personal private? You know, trusted advisor is the wrong word, but who do I listen to, who knows me and who can guide me in this? And then I think you know, when we have 116 guests plus over these seven years, we have literally started taking off the real data and AI community in at least Sweden and, to some parts, nordics and Northern Europe. You know that these people is a little bit like they are here because we know what they have done and the community know what they have done.

Goran Cvetanovski 25:53

They are interesting to listen to. Exactly, and we always said like we are making a time capsule of the people that are actually creating the future as we know it. And these are the people signed there on that piece of painting and etc. Like 115 of them, and they are the ones that are actually making this happen.

Henrik Göthberg 26:13

But it's a personal reflection. It's like, out of those 160 people that we've had here, how many of them has sort of popped up on our radar based on generative AI in the last year? Maybe one or two that we're doing something super, that is doing something exciting. I'm thinking about the fashion AI. I mean, like we did not, we would never have found that angle. But if you look at the broader picture, dude, we know these people are people in the community, right?

Anders Arpteg 26:41

I think you said that well as well, henrik, and I think last week you said I actually have the news that it's not related to large language models and it's kind of absurd. You know what focus it has been on one part of AI. And in reality there are so many other parts and aspects of AI that most people are working with but doesn't really get the attention at all in media or in people's minds.

Henrik Göthberg 27:03

But maybe we should wrap that up and sort of what's the structure for today? Do we have an agenda?

Goran Cvetanovski 27:09

Maybe it's a good segue actually, because maybe we can start first with your personal, basically summary of the year nothing aspect of the conference, but basically on the learnings that you have done, your personal take on the podcast and the people that we met and then maybe we can move to the second part, which we'll be looking at the year from a perspective of what actually technologically has happened and how this has impacted actually public and private sector and the enterprises and how people are actually utilizing this and, to that extent, even society, because we had some guests that were coming that are doing quite well in that aspect. And then the third one we can actually look at some of the favorite moments of the year.

Anders Arpteg 27:58

Yeah right.

Goran Cvetanovski 27:59

So a little bit of a summary so we can do that. If you want, I can start actually going through the entire 32 episodes very shortly and then why don't we get an executive summary what happened this year?

Goran Cvetanovski 28:12

All right, so let's do that actually. So this is going to be around segment over five minutes I will try to squeeze in. So actually, as I mentioned, we started, this is actually right now we ending the seven season of the podcast, so three and a half years into this over 10,000 minutes of content. We started a year actually with the season six on 19th of January. So the first guest was Joseph and Victor Obert from Moduli and we were discussing ML Ops at that point of time. Then we had Stefan Vendina from Neo4j at that point that point of time, basically, but we were talking about social engineering and graphs.

Goran Cvetanovski 28:53

If you remember, it was a very fun episode when he was talking about how he was manipulating social media algorithms for different purposes. I thought that it was a little bit off what we have been doing, but it was a very fun as well thing to listen to. Then we had Luis Vanerel, who was here actually she's from the upskill company and we were talking about tech talent and recruitment and reskilling, which I believe that still is a very underlook at the aspect of AI in general, because AI equals change and change equals people and we are not talking enough about it. So we should take a little bit more in the upcoming year. Then we talked about citizen data scientists, which is a very, very good topic actually, because this happened before AI got democratized. So, if you recall, the citizen data scientist was this effort to democratize data science across all the users in the organization by making a data science simple for everyone and then change it became and then basically simplified for everybody else.

Anders Arpteg 30:00

It's like an element of AI, but for everyone in the company in some sense. Everyone should have a basic understanding of what AI really is.

Goran Cvetanovski 30:07

Yes, and how to utilize it in general. So then we had like a Martin Svenson and Mika Ljungblum from AI Sweden we were discussing of. We were discussing AI in public sector and especially how the Nordics can be better in many aspects when it comes to AI, from talent to enabling organizations in the public sector to be better with that, and Martin Svenson is right now part of the AI Commission as well, so we are extremely happy that he was on the show. Then we had that Reha's best.

Henrik Göthberg 30:38

Actually, I need to jump in. We had at least two guests who are on the AI Commission. Erik Heinz and Martin Svenson Right.

Goran Cvetanovski 30:46

I think we have another one.

Henrik Göthberg 30:47

And one more.

Goran Cvetanovski 30:48

But we'll come back to that. Then we moved in with the predictive analytics type of a topic which I think that is also under looked part when it comes to deriving value out of data, and that is inside. So Andreas was talking about pricing analytics. He's coming from Ignis, a new company that started after Naveti. Yeah, great guy.

Goran Cvetanovski 31:10

Then we had the same date when the Open AI launched the API for Open for a charge, for I believe we had actually Arnold and Sameral Mumbai from Furghatt here, Right, so keep in mind that this was in the same day, so that was cool, and the API was released basically the night, so they were coding the entire night and then when we came here, we had like a first demo of how is to implement chat GPT into a living robot or not living robot, but the living is the physical robot, physical robots, and it was about the text prompting translated into speech and voice and sound, so to speak, and we could then interact with it with a fur hat robot, I mean, it's kind of extreme to be able to do that in hours.

Henrik Göthberg 32:10

In hours.

Anders Arpteg 32:11

Imagine if we have enterprises that could move innovation that fast. That, of course, does not happen, and normally it takes many years for an enterprise to move from an idea to a project. They could, they could, they could in theory.

Henrik Göthberg 32:25

Corona showed some of it. Yeah, exactly.

Goran Cvetanovski 32:28

Right and then to continue. So basically we're talking, then, ai and self driving cars, with Daniela Skilde here. That was a great, actually.

Goran Cvetanovski 32:37

Episode A lot of learnings there, yeah, a lot of learnings and Daniel is a very outspoken speaker and really actually some good gems about self driving vehicles and AI in general and not generally if AI we are talking about traditional applied AI. Then a lot of things started happening with Google, if you remember, and Bard and everything else. So we called the Luca Oliver and Alexandra Kafka to have like an overview of everything that was happening, because at that point of time, generative AI would just basically it just exploded.

Henrik Göthberg 33:09

It exploded was in all media everywhere and I think at that point in time I remember this vividly because I missed the episode and I listened to the episode in my car and I was like I was blown away by our pod. This is by far the best storytelling and communication about what's happening right now.

Goran Cvetanovski 33:28

It was a really dynamic. It was one of my favorites.

Henrik Göthberg 33:30

But because the level of quality, the level of know how, but also the level of communication.

Anders Arpteg 33:37

I was sitting there. I think all these people are so good in communicating.

Henrik Göthberg 33:39

That's the point, right.

Various Topics in Podcast Discussions

Anders Arpteg 33:40

Luca and Oliver and Evelina, and then the whole community Alexandra was there Okay yeah, they have so good skills in doing that, so that's really impressive.

Goran Cvetanovski 33:51

Yeah, and then we continued basically with productivization of data. We saw we have Christopher O'Grain from Atelier here and they did a fantastic job with their crowd.

Henrik Göthberg 34:04

Inside. Inside yes, Like they used, I mean like they really were contributing to how we managed the Corona epidemic in Sweden.

Goran Cvetanovski 34:13

Yes.

Henrik Göthberg 34:14

And they were helping out in many ways. Yes, and they did it in a way that was compliant and very, very robust. Of course, they had a head start, because they've been working on these topics, but they could take something they've been working on and tweak it. And then they made them, they could basically make a big difference.

Goran Cvetanovski 34:35

And make a product out of it. They were in the right position to do it as well with their IoT sensors and mobile towers and everything. Then we had another episode where we were looking at AI and for electrification with Gerovul Zedvits and Sebastian Kau so both coming from Skania and Wattenwald which was a very good intersect between two big giants, First time meeting on this podcast and being able to start breaking some ideas about how they can collaborate when it comes to.

Henrik Göthberg 35:08

AI, and this is, of course, one of my favorite episodes. We can talk about it more but, of course, having worked both with Skania and Wattenwald and then now more deeply understand the transport ecosystem and then understanding how this starts to interrelate and the data is sort of an underpinning capability, all these algorithms, that was super cool, but it was fun to see two different, on the surface different industries.

Goran Cvetanovski 35:35

They found each other and they, you know, but they were not, because, if you look at it, if they combine their efforts, they can actually there was.

Goran Cvetanovski 35:44

I mean, if you listen to the opportunities there If you listen to the episode, if somebody is smart, they can just basically take that idea and make a multimillion business out of it. So that is episode 96, by the way, if you listen, if you're curious, go there. Then actually we had UC here, so we were talking about limitations of current generative AI and LLM models. Keep in mind we are still early. This is April where we are talking about generative AI, and UC just joined Silo AI at that point of time, right, and I believe, anders, you were taking the lead on that one.

Henrik Göthberg 36:18

Yeah, he's a very knowledgeable person.

Anders Arpteg 36:21

Of course, you see Carl Green right.

Henrik Göthberg 36:23

Yeah, but this is also another favorite, because here we were not. Sometimes we have too much agreement with our guests.

Anders Arpteg 36:29

Yeah, that's the problem.

Henrik Göthberg 36:30

But here actually we had some good debating going on. I love that.

Goran Cvetanovski 36:34

Yeah, Then we of course Henrik is breaking some things. Then we moved actually with Dutchel, Henrandes and Innovation and Inventions and Diffusion. So this is where Henrik was actually having a very large debate with Dutchel about innovation versus invention versus diffusion.

Henrik Göthberg 36:57

Diffusion yeah.

Goran Cvetanovski 36:58

This was before the Data Innovation Summit, so you can see that you were preparing for it.

Henrik Göthberg 37:03

I was preparing already. Then I was testing some material yeah.

Goran Cvetanovski 37:11

Then we moved on with basically the state of digitalization in AI in Sweden again, but now we had for the first time a politician active politician on the podcast, and she was amazing.

Henrik Göthberg 37:28

So, Maria Stockhaus was a true pleasure. Out of this and we all had, let's be honest, right. We thought are we going to have the politician speech and we're going to have the politician answers? And she was so authentic, A real person and really switched on curious. I was like, yep, that's the person I want to follow and help.

Goran Cvetanovski 37:51

Exactly, I mean absolutely one of the gems, and I was mind blown by the transparency as well and how everything works and whether the hinders and challenges and opportunities with this area.

Henrik Göthberg 38:02

This was amazing learning, because this is how is politics done, man.

Anders Arpteg 38:06

It's like it's a craftsmanship in itself to get stuff done and it's so hard to be a politician, they have to know everything about everything, and they can't, of course, but it's glad to see that a lot of people do have at least a really good understanding of even advanced techniques.

Henrik Göthberg 38:22

But this is so important then now. So this is a craft to be a politician and to craft laws and regulations and budgets and investment for society, and we need to know how that works in order to be part of it and influence it, and that was a fantastic episode for me.

Goran Cvetanovski 38:40

I agree. Then we had basically the legal aspects of AI, and this is actually also one of my favorites. We had Claire Ignon Bogus and Yanika Tunkvis here and this was mind blowing basically in terms of the upcoming act and regulations, and especially intellectual property when it comes to generative AI.

Henrik Göthberg 39:03

Yeah, that was the main thing. That was intellectual property. Yes, remember that.

Goran Cvetanovski 39:07

Yeah, I can remind that so many things has happened. But before many of these LLM providers, they changed their terms and conditions quite a lot. I was actually now last week on Bumi. There is this application where you can generate AM music and now when you enter, I haven't been there for a long time, so I needed something. So like OK, let me generate something, because we are recording another pod in the same studio now.

Goran Cvetanovski 39:33

So I needed to make music and so like, ok, let me generate AI. So I went there, logged in and the first thing that pops out is basically the terms and conditions and it's like OK, I will read it this time, especially who owns it. And they have made it very vividly that basically they own the intellectual property. At any point of time they can claim it Right. So for a producer like me, when I produce something like this, if I put it on YouTube and everybody else, if the music and everything else belongs to somebody else, of course, then the credits and the monetization and money goes to that person or that identity. So, only with that is a huge problem. So I had to basically went back. So I had laid a bit like around three hours, so I created my own music back home.

Goran Cvetanovski 40:25

I just sit down on a piano, the whole fashion way, let's do something. But this also shows, actually, that we are so early with the generative AI and everybody is actually thinking that they can get the benefit out of it very quickly. But you know, devil is in the details, so you need to understand what you're doing.

Henrik Göthberg 40:42

And I think the whole intellectual property game is evolving at the rapid pace now and you have you have updated terms and agreements, but also, I think that I think the tech giants, especially, is recognizing this problem.

Anders Arpteg 40:56

They are trying to, you know, for their generative services, have this kind of indemnification service where, if someone gets sued like this, yeah, I heard that actually will pay for the damages.

Henrik Göthberg 41:08

Yeah, that was Microsoft that came up with. Who?

Anders Arpteg 41:10

Now everyone, everybody came out to basically say they're putting away a buffer right. They will cover the legal expenses and the damages, potentially if you get sued. So I think you know that's the simple everybody recognizing that we this is unknown territory.

Goran Cvetanovski 41:28

We need to help each other Will come there is no running away.

Henrik Göthberg 41:31

We can talk about the big tech giants stepping up.

Goran Cvetanovski 41:33

Then, to sort of to move forward, but that episode is actually something that many people should listen to because I think there are beautiful gems about it and if you need any assistance with this, you should contact Claire and Janneka, their experts in this area. Then we finalized the season six on the same day when we had the 100 episode pod fest here in the yeah, that's right, that was the studio. And that was with stuff, stuff and through a real legend.

Anders Arpteg 42:04

Yes, a real legend from talking officer at the future.

Goran Cvetanovski 42:07

Chiefs of talking.

Anders Arpteg 42:08

I love him. He's such a good knowledge of a person as well.

Henrik Göthberg 42:13

But OK, so can we say that record a future? Can we claim it as a Swedish company?

Goran Cvetanovski 42:17

Yes, definitely yeah, but it's, but because they have so many times.

Henrik Göthberg 42:21

I know, but I think they're there. They're there. Oh, we almost lost them to Silicon Valley. We almost lost them to yeah for me, that episode is actually.

Goran Cvetanovski 42:29

If you're looking at the cyber threat and you want to understand how exposed people are and how organizations are looking at this and how much money is drawn to protect identities and organizations from cyber terror right now, that episode is something that you need to know and, if anyone, if anyone wants to get a glimpse on how AI can be used in in security cyber security and what is actually the cut, cutting edge in my opinion, around thinking and first principles of how to use this.

Henrik Göthberg 43:02

I think record a future has been on this game for a long time compared to many others.

Anders Arpteg 43:07

They're very nice visions. Well, to be the world's largest private intelligence service.

Goran Cvetanovski 43:12

Yeah nice vision and very super positive person. I haven't never seen a more positive person in my life. I thought that this was just like I was.

Henrik Göthberg 43:21

I was fell bound, yeah and you know we were saying it before the pod like he's been a long time in the industry and he you know Giri has a schoolboy. Yeah, so cool.

Goran Cvetanovski 43:31

Amazing so that was it for a season six, season seven. I will speed up this because it's not five minutes it's usual, but we are still still good. I think we started actually with the update, so we got Luca here and Evelina and Oliver again.

Goran Cvetanovski 43:50

So, that was one of my favorites, for sure, and you know this is immediately after the summer, so somewhere in September actually, we started and there's some beautiful gems there as well and updates of what works and what not works and at this point of time we started to see actually a differentiation between all of these LLMs and how they work, and Luca was actually testing most of them and he was saying like which one they were talking about, which one actually is good, which one is not. Oliver was also discussing a little bit more about investments in the area and what is going on with the startups and how organizations should focus on data. There was a nice diversity in this palace and this panel.

Goran Cvetanovski 44:28

And then Evelina, of course, ai and regulation, which is and VC as well, vc as well, which is beautiful. Then we moved with AI governance and ESG data management with Vanessa Ericsson. This is something that will be more important even in 2024 as the taxonomy is coming in and most of the companies are focusing on the environment safe.

Henrik Göthberg 44:51

Social and governance.

Goran Cvetanovski 44:54

And, as we know, now the things are getting even more stricter, with all the countries agreeing to even be more strict on the policies regarding CO2.

Henrik Göthberg 45:05

But it's interesting because this topic has really had two dimensions, Because it's the dimension of adhering to ESG compliance regulation and data.

Goran Cvetanovski 45:16

I think we should call Aurora again back.

Henrik Göthberg 45:20

Maybe yeah, and then, because that's the whole thing like that, this is actually quite complex data problem to deal with this type of reporting and analytics, and then you have the fundamentals of how we can have more energy sustainable data and data management overall. So this is sort of the underlying what we are trying to achieve with being more sustainable, and then it's actually the data problem of dealing with the reporting around that. And these are two major trends I would say that will kick into 2024 with the regulations around this in the US, especially around this reporting.

Goran Cvetanovski 45:58

I completely agree. We should look at this actually from I would love to have a CEO or somebody very high just looking at how they are thinking they're going to actually make this possible without having data and AI.

Henrik Göthberg 46:11

The interesting point was that, in my opinion, the wrong type of consultants has been advising on this, because ESG reporting is a data problem.

Goran Cvetanovski 46:20

But of course, okay, consultant is there to sell your problem, not to sell your symptom, not to sell your solution. Right, not all of them. Not all of them. Just to repeat Then we moved actually with AI in education and transformative role in AI in education. This was Anders Ensjøma and this was awesome Again. Yes, A person that is really involved in education, not an AI practitioner or super expert in the field, but actually AI practitioner user, not AI engineer.

Anders Arpteg 46:54

Yes, but also a great communicator.

Henrik Göthberg 46:57

But a great communicator, a great evangelist and a great lobbyist, I would add, in terms of how do we use AI in education. If you see how much it puts up in social media. He's part of the debate. We had some really brutal conversations about some of our top politicians that really have no clue. In my opinion, if I take Anders' insights with me, what do you think is the future of education? So I think he was spor on.

Goran Cvetanovski 47:29

One of the things that he said basically which stuck me and I was speaking with my girls at home he mentioned we are preparing kids, we are educating for jobs that they will not exist, so we are preparing for something that actually is not going to be there in. Cambodia.

Goran Cvetanovski 47:47

And I think it's super profound and very honest and truthful, because if we are denying utilization of these tools and we are not educating people how to use them, they're going to find a way. They will use them, but they will use them in a different, maybe even wrong manner. So we need to start doing that. We need to prepare them for the future, because this escalates.

Henrik Göthberg 48:11

And what becomes super clear? When you have someone who is a professional educator himself, so he can't sort of sit talk, you know he needs to implement this as well as a gymnasial error, right? Then it becomes super obvious how important it is for a whole profession to be on top of it, and not only the profession to be on top of it, to the regulators or the people putting up the frameworks, because otherwise our education will be relevant.

Goran Cvetanovski 48:42

Yes, there are many discussions that education has been irrelevant for many years. I think that the change has been no, but everybody here is.

Henrik Göthberg 48:50

but the whole thing about adaptability and innovation in education has to keep pace highly recommend it, so it was good.

Goran Cvetanovski 48:59

Then we had actually investing in the Nordic AI, so we had Rebecca read her read off and she's just awesome. She's a person that he has been she's been monitoring this area for a long time and she's lead the investors Could you mention where she works. She is and she used to be an investor, but right now it's called put you on the spot.

Henrik Göthberg 49:23

Inventure right, Inventure right.

Goran Cvetanovski 49:25

I give a second. I will find it right now.

Henrik Göthberg 49:27

So because she's been in the. She's in the VC space so of course I was just gonna do a shout out. Sorry for putting on the spot.

Goran Cvetanovski 49:34

No, no, it's fine. I mean I cannot remember everything, but it's good. So now we are having a yeah, it was in venture. Good, you see. No taking it. No taking Sorry about being so late.

Goran Cvetanovski 49:47

Great episode if you want to understand how, basically, to start a company that is working. Or if you have idea that will use AI generative AI or you have like an idea for a company that will help the development in this area. That is a great episode actually, to listen to what investors are looking at and how they are investing these days, especially when it's everything is about AI. Then we had two amazing episodes, which were again not AI experts or data experts, but actually users of the technology, and that was actually Magnus Ostegren. So he was talking about like AI and democratization of creativity, so he's like a game producer. That was amazing. And then we had like Carl Oxel of Wallstrom, who is basically the person that created the first AI fashion magazine. Have you seen how famous he is right now? Yeah, it's all over. I saw him every crazy how he's like, but a bit of a scope there.

Henrik Göthberg 50:54

We had him early.

Goran Cvetanovski 50:56

Yeah, but it was amazing and if somebody wants to really understand how prompt yeah, he's prompt.

Anders Arpteg 51:03

engineering tips that he gave really learned me a lot, so I learned a lot from a more interesting thing was like to really create images yes and have the proper techniques, which is very different from what you think, yes, so I would really encourage people to listen to this if you want to learn it.

Henrik Göthberg 51:18

Was it a little bit like on learning assumptions and then learning no, no, no. This is the way, this is the experimentation way that seems to work.

Goran Cvetanovski 51:25

What my biggest takeaway there for the prompting part was that I thought that you need to write, for example, write me, give me a butterfly in the sky, blue colors, this one, that hard work like hard works, like sky butterfly and things like that. So but also it broke down this stigma that you know you just utilize AI and everything is perfect. No, it's not. Behind that 300 page magazine, every single picture took at least like 20 hours or something like that to be produced. So first you had the prompting of the picture, then you had like a lot of retouching and making sure that is like correct and all these other things.

Goran Cvetanovski 52:11

So it's not just some documentation right and it's like an enlargement and all of these other things, then putting it and curating all the magazines, so it's not that easy.

Henrik Göthberg 52:19

No, and in the end, it's an AI process with several AI tools yes, or algorithm supporting tools in different ways, even down to Photoshop, which is also, of course, algorithms.

Goran Cvetanovski 52:32

the back Then we had Josephine Rosen. She's like a diva almost when it comes to regulation and AI, so she's basically she was talking about the future of trustworthy AI here, and Josephine is also a winner of the Nordic Deir Awards for responsible AI, so we had like a great discussion about the upcoming act.

Henrik Göthberg 52:57

It was a good discussion. Actually, I still wonder if I remember and we talked about the executive order and other things.

Anders Arpteg 53:03

Yes, Also the Bleschli Park. Accuration just happened, and so did the executive order, so so many things were happening very timely to her podcast.

Henrik Göthberg 53:11

Yes her podcast and with her as a guest. I mean, it was so cool when she was, she could sort of name drop. Oh yeah, yeah, my boss was standing next to Biden when he did the executive order. Yes, you know, that's cool.

Goran Cvetanovski 53:23

That was really nice, that's cool. And then for the final ones, we had basically Magnus Schellberg here, which we were discussing AI in health, back in healthcare, so that was a great presentation or basically session to listen to. Then we had Uldes Akhrison.

Henrik Göthberg 53:42

That was a funny one actually, but a lot of practical examples of how to utilize actually AI in journalism, journalism, serious audio, and what really stood out as an interesting point with Ulle is and I can see many similarities. I come from the business background and really when you get to really hardcore, trying to adopt AI in any operations, you need the people who has started to learn about and piloting and testing AI. But they but his depth in understanding the media landscape that needs profession is profound right. So that combination is super. It's sort of a mind opener that you need to have both the business view and the AI very closely together.

Goran Cvetanovski 54:31

Yes, and super interesting what is happening actually and how they're thinking about it. So they're not shy to utilize this technology which was very surprising and very encouraging, so to say, how to build LLMs with Ariel Akhren. Now, that was a very interesting discussion because also we had for the second time when people do not agree, so it was very interesting. So it was a topic where we're discussing about the Swedish language model that was a Sv3. Which is basically now available for all organizations to utilize and etc.

Goran Cvetanovski 55:16

Yes, exactly, and then the last two is actually NLP and engineering with Kajsan Nuri, but also it was AI and dancing episode right to their dancing show as well before the episode.

Anders Arpteg 55:29

So she had a premiere in Dunstensus. I think it's still playing if someone wants to go to it. So they used AI to actually produce the choreography and the instruction for the human dancers that later actually performed it and it was, I might actually go.

Goran Cvetanovski 55:47

I'm here for the Christmas. That could be interesting. And then, finally, last week we had here Patrick Ekkemo and Daniel from sorry I cut you because you were eating chips, and Daniel from Bullock's Verket and we were discussing AI in public sector. We were discussing AI act because Daniel was a very it's been through the whole process in different roles. And in a timely manner, because the AI act was just basically pushed forward as well, the same week.

Goran Cvetanovski 56:19

The same week. So we had like a lot of good scoops there and I actually got a very good feedback about that episode that people really liked.

Anders Arpteg 56:28

Patrick has been in Dig and so many other places and has been a long, very prominent person in AI. He spoke about Bullock's Verket and they want to be a role model in how to properly digitalize or digitize an agency and I think that truly are doing that. So it's great to see.

Henrik Göthberg 56:46

Yeah, but it was also very interesting to hear about the European consortium, the European collaborations around the identity, the wallet. I forget the European EC. What is it? Ecw?

Anders Arpteg 57:01

EWC? I think yeah. Ewc sorry, the European Wallet Consortium.

Henrik Göthberg 57:06

Yeah, which is basically then looking at many different facets of why do we need to have a very seamless way to deal with identity and data?

Anders Arpteg 57:17

in many ways in Europe so fast. I think they're planning already in 2026 to have like a European wallet and potentially some means of identification, which is surprising if they actually managed to do that in just a year or two, so I'm hoping they are correct.

Henrik Göthberg 57:34

But what stood out to me is I really like the way I mean like there's a couple of things here. I mean like if we're a little bit rude and cynical in Sweden, we kind of tag along in the EU and we have we sort of Other people are sort of lobbying more or getting more money or using more money and we kind of used. We haven't sort of been. I don't If I'm going to put it brutally play the game well enough, but in this case here we have Bullock's work leading one of the consortiums and the way they do it with a combination of private sector, you know, expert actors and others. I think it's something really interesting to look into and learn from and do more of.

Anders Arpteg 58:25

I think they're on the right track. Yeah, that's truly our role model in that aspect as well, and also, once again, we had the luck of having a very timely episode with.

Henrik Göthberg 58:33

David.

Anders Arpteg 58:34

Magod, that actually was part of the Swedish team for the AI Act and now we have the big agreement coming forward in. Europe for the AI Act as well, just in time for this episode.

Henrik Göthberg 58:44

So if you want to learn more about that, yeah, and you know, and to be able to dissect it at the after, after work. Nothing is said, nothing forgotten.

Anders Arpteg 58:54

Yeah, we could take part of the most interesting part of the podcast after we turn off the cameras.

Goran Cvetanovski 59:00

Yes, but in summary, last year, when we sat down and started basically pencil what is going to happen this year, we said, like let's move from the justification of AI to actually how organizations can actually put this in practice and show.

Goran Cvetanovski 59:16

Yeah exactly and showcase the solutions and the use cases of these things. And we didn't expect that everything would go so fast but when I look back actually on all the speakers and the episodes that we had and the guests and etc. It was really, really we were really on the dot because this is a user AI user year, I would say it's the year of the user.

Henrik Göthberg 59:42

right I was. I had that in mind.

Goran Cvetanovski 59:44

Yes, so Interesting. I'm super happy actually with the entire two seasons and I'm very thankful to everybody who actually made it to the podcast, and we hope so that we'll continue like that in the next year as well.

Henrik Göthberg 59:58

All right, so that was not five minutes, yeah, but I think that I managed. Five minutes was a deal.

Goran Cvetanovski 1:00:04

Five minutes per person. Exactly, not five minutes per person. But I think we also managed to do some A little bit of a summary of the year as well because, yes, you combined the two. Basically, yes, exactly.

Henrik Göthberg 1:00:14

So yeah, but okay, but from here should we now maybe. How would you summarize the year, Andres, in terms of what happened? Yeah, what are the top five, six topics that we want to highlight?

Anders Arpteg 1:00:26

For me. I just want to repeat once again to sit on the side of the podcast table and just broaden the view in ways you would never do otherwise, this is a very great learning experience. Yeah, just doing these podcasts, even though we prepare very little for them, we learn so much. Yeah, and I think that's the first reaction to it, and it happened this year again, which is great, yeah.

Henrik Göthberg 1:00:50

Did you have a favorite learning sort of thing or something that stood out as a main learning point of view for you?

Anders Arpteg 1:00:56

Always learn a lot when Luca is here. Yeah, I agree, but also, yeah, I find it's so many things and I think in the news section as well that we started to do this autumn was surprisingly useful. So, thanks, goran, for making that happen and yeah, it's really a lot of news that happened. That was very good.

Henrik Göthberg 1:01:18

Yeah, one of my favorites as a learning experience was actually a little bit talking to Maria Stockhaus to getting a little bit more deeper understanding of the political process, so to speak, and I think what the main takeaway for me was how important it is that we, as a society, understand the political process in order to be democracy, in order to influence it, and I think there is some gap here. I mean, like we are all experts in our fields, right, and if we want to help ourselves, we need to help the politicians and we need to then understand the process.

Anders Arpteg 1:01:54

That was a big for me. Yeah, I agree. And one concept that Luca spoke a lot about and I also was thinking more about was the idea of theory of mind. So we were saying that Shatty Pity potentially had the possibility and the capability to put oneself into the mind of another person, and I think this is actually what we are doing a bit you and me, henrik, and I think everyone should. You know, if you have a politician, for example, if you don't understand their view of things, to put yourself in their mind, to have a theory of mind of what the people that are potentially using it or being politicians, or having some being a policymaker or a user of it, if you can't put yourself in the mind of that person, you will not be able to make a change, and I think that's one of the bigger learnings. That's Maria and other guests.

AI Trends and Developments in 2023

Henrik Göthberg 1:02:50

Now you sum it really well, because now I need to feel almost obliged to put the year in perspective of myself and their ducks. So, because this is a, I've been working closely with TetraPak this year and we actually been working with the, with the central data science team. Alberto Barroso is leading that and he has several sort of team leads and stuff like that.

Henrik Göthberg 1:03:13

Yeah, he should of course, alberto should be on the podcast. What haven't he been here? Rasmus in his team has been in. Rasmus Thunberg has been here, but not Alberto.

Henrik Göthberg 1:03:23

The main topic we've been working on is to have a more adoption-centric approach in how we drive investments and projects. So the typical blind spot is exactly what you're talking about how do we put ourselves in the other person's shoes, in the user's shoes? How will the business leader experience this? So it's so easy that you so. In reality, alberto doesn't have a data science problem. He has so good people. You know. The main challenge is how do we navigate in a very complex landscape as TetraPak now, because now it's not only one market right, you need to not only put yourself in the shoes of the guys in Lund, you need to go. You know it's all about putting yourself in the other person's shoes and navigating from that perspective, which is now that's why I wanted, you know. So the adoption or this sort of what you said how did Luca phrase it? I love that way of the perception, the theory of mind idea.

Henrik Göthberg 1:04:25

The theory of mind idea right.

Anders Arpteg 1:04:27

You can actually put yourself in the head of the other person that you're speaking to and potentially AI can do that, and I certainly think it can already today.

Henrik Göthberg 1:04:35

So we maybe can get help. So in this change in adoption, I should maybe ask the AI how should I frame my communication for?

Anders Arpteg 1:04:42

a salesperson. Mind over salesperson over a politician or a Lovely or.

Goran Cvetanovski 1:04:47

Keep in mind that the chat GPT right now accepts tips, so he's getting lazier because, now he has not incentivized enough.

Henrik Göthberg 1:04:56

Yeah, you should. If you want him to be good, ask him please, If you want to tell him not to be racist. He's getting really lazy chat, GPT.

Goran Cvetanovski 1:05:05

And corrupted. Obviously, because if you're like, he will do it for the highest bid. That's funny.

Henrik Göthberg 1:05:11

Okay, any other tea learning story?

Anders Arpteg 1:05:14

I think the open AI drama was still such a weird thing to happen and the quick turnaround on that. It's like the. It's like a story that if you saw the manuscript for that for the movie, you wouldn't believe it. It's too absurd.

Henrik Göthberg 1:05:28

It's too absurd, right.

Anders Arpteg 1:05:29

Yeah, basically the dictant over the reality of the word.

Henrik Göthberg 1:05:34

How do you say that in English the fiction beats reality? Yeah, in some way.

Anders Arpteg 1:05:39

Yeah, it's so weird.

Henrik Göthberg 1:05:40

But I was trying to sort of summarize, you know from 2023, some four or five main things, sort of. We were talking about this before, so we had one. I mean like so one big moment, of course, if we go outside the pod now and talk about what happened in 2023. So you kind of need to put one sentence like AI in mainstream media. Ai is discussed by my grandmother, so in media, but basically in all people, in all people. And we had anecdotes. I had anecdotes. I went to an event has nothing to do with AI. I speak with an elderly couple and she doesn't know what I'm working with. She's just talking to AI doomer speak. You know, and you know it's interesting, right, and that does one. And that led me into the second one. The whole AI doomer movement. That sort of grew right With Max Teagmech's summer speech yeah, that's.

Henrik Göthberg 1:06:40

I'm reading Max Teagmech's Leave 3.0 right now. It's actually really good. Yeah, I mean, he's a brilliant person, but I'm really sad that he came across as a doomer. Yeah, because when I read his book, he doesn't come across that bad of a person, that bad of a boomer, to be honest, and what else. And then I put another third one. We talked about this, the model, and going to production frenzy in 2023. We saw so much.

Anders Arpteg 1:07:08

Yeah, both saw a lot of models coming out, but I think what was even more impressive this year was actually big companies putting models in production into their products, into all the operating system of Windows, into the web browsers, into the whole Office Suite. Google did the exact same thing into the workspaces and you could keep up, right.

Henrik Göthberg 1:07:30

I mean, it was so much going on. I mean you were trying to oh, we could summarize the year in this way. And then you showed me a slide with, like model ABC. It was used to release after release, after release during 2023. You know, your head spins right. And then we had all the models and then we had the plugins, right. So all the products that came out very fast in terms of different types of API, plugins, approaches, so there was a lot of stuff going on in that field for a while there. And then, yeah, so pop-ra. I already mentioned you mentioned the so pop-ra, so media doomer models, so pop-ra. Yeah, anything else that sort of caught our attention. If you want to summarize this, yes, you're too much.

Henrik Göthberg 1:08:17

You're five or yeah you have, they lose my five. We kind of made it.

Anders Arpteg 1:08:21

We've seen on this in the beginning. But I think just to put some of the big stories out there and I think you know Gemini was kind of weird when it came out. You know I was speaking about. You know, oh, it's so good in multimodality. And then it turned out a week later they basically faked a lot of the videos and it's really put me off, I must say. I mean, google did it again. They did the same with Duplex and other things. What are their marketing department really doing?

Henrik Göthberg 1:08:47

Because you want to laugh at them?

Goran Cvetanovski 1:08:49

No, no, they're doing exactly what they're supposed to do. They only shit here that because they got caught.

Anders Arpteg 1:08:54

But they got caught the last time as well and it's ah why and they don't have to. I was a bit surprised with that, but still it's glad that openly, I get some competition at least. So that's good, I guess. But also one of the things that we haven't spoken about is, I think in the election in Argentina in October I think it was when they there were basically two finalists into who will become the president in Argentina, and this one guy was this millennia person who was having this kind of change. So when it's going to cut up the government and whatnot, not a strange person. Anyway, they use made you massive use of generative AI on both sides. So they used generative AI to put them on the positive note, Hugging a teddy bear or whatnot, or they used gen AI to put the other competitor in like a zombie picture or someone else, and both sides did it. So it was perhaps one of the first really big like Genetic AI elections.

Anders Arpteg 1:10:00

Interesting and thinking about coming year. There were so many elections in EU, in US and so many other things that will be interesting to see. What happens, Of course, if we sum up the year.

Henrik Göthberg 1:10:12

There is one more theme. Maybe that could be summed up as AI safety. I mean, like we had the AI Act whole thing coming around, we had the executive order coming out from Biden and we had the I can't say Bleachly, lechly, lechly the AI safety summit. So two quite big topics that really, if we say genetic AI was on the agenda, I would say AI safety is now from 2023.

Goran Cvetanovski 1:10:42

We forgot the memorandum that they were writing when Elon was saying please stop AI, Please stop AI.

Henrik Göthberg 1:10:49

That was group. But it was AI doomer. Yeah, the AI doomer is the heading and the memorandum is right out on the net. So typical cry me a river so I can win time so I can release group.

Goran Cvetanovski 1:11:01

Yeah, but exactly that was so double standard. Elon Musk is signing thing and then he's going to be on the agenda.

Anders Arpteg 1:11:07

And then he's building rock AI. Yeah, but robots and shit like that.

Henrik Göthberg 1:11:12

But anyone who followed those articles when they came out. I mean, like there was equally amount of articles explaining how Silicon Valley works and how you, you know how the whole Silicon Valley needed to find the next golden nugget, because they had some challenges and any technology that is so scary needs to be super, super fancy. So I need to invest in that. So to this, all this discussion and cynical debate also, you know how much is this doomer stuff is real and how much is used Very, very smart marketing.

Anders Arpteg 1:11:49

And also some people that are trying to do the right thing I mean, so many people are, I think and the question is should you have closed models or closed weights, or open weights or open source? And you know, jan Le Koon is very clear on this and he believes, you know, in open source and that's a more safe approach. Potentially, other people are thinking differently, but then we have, you know, mistral also doing it very publicly, openly, and I think so much happened from a regulation side. But also, you know, we were speaking. We know there are a number of court cases happening now against the big generative AI companies like Midgernie and also Dali and Open AI, and towards, you know, using co-pilot and GitHub for Microsoft, and all of them are getting sued during this year and we are starting to see some decisions coming in. Some court cases actually got some initial findings and basically it says that the generative AI output is too dissimilar to the original work for it to be copyright violation. So it's on the side of AI, if you call it that.

Henrik Göthberg 1:13:00

Yeah.

Anders Arpteg 1:13:01

It's that will be. If it were the opposite, it would be really interesting. Now, basically, it's saying, okay, we are probably a bit more safe than we think, or that we thought before. So, yeah, we can wonder what that will mean. And we had the big strike, you know, for Hollywood artists and the screenplay writers and the actors.

Henrik Göthberg 1:13:24

They had the Hollywood. Was it called memorandum? No, it was called the Hollywood. They came out, you know, like the union came out with some sort of agreement agreement or whatever you want to call it.

Anders Arpteg 1:13:39

So many things happening around. You know what are the impact of AI being so powerful. I still think you know it's. I think we will adapt quickly to it. I am a more of a positive kind of person. I think we will find a really good way to do this. I think in general, I guess, if we I'm eager to go to 2024, but maybe, maybe that's the segue because you it's, not because I haven't had my head because I think that you I want to fill out the gaps where you basically didn't.

Goran Cvetanovski 1:14:17

For me, this year has been a tremendous amount of drama. I think this was very dramatic here from many aspects. It's been a shitty year for many people. First of all, we had like a lot, a lot of layoffs and all of these things, which doesn't have right. We didn't say that.

Goran Cvetanovski 1:14:32

We started actually with the layoffs. I was actually worried how this year is going to end and then the money became very expensive. So you know, companies didn't get the money and funds and actually so they went to see shit and etc. But if I need to summarize the top five for me, and very shortly, these are the top five. First, this has been a year of user experience. Yeah, good one.

Henrik Göthberg 1:14:57

Good one.

Goran Cvetanovski 1:14:58

Because what chat GPT brought they didn't talk about this is how you can use convolutional networks in sales. Sorry, you used to do that in Voltarium. They actually just provide you one simple user experience and then they let you give you know, take use your mind in any way possible and be creative as possible to utilize the technology Right. And it was, it was beautiful, and I believe that was the main actually win for them. It was not the AI behind it, because, as we know and we have been talking about this for a long time on this podcast this is not a new technology. Their principle, their approaches, may be a little bit different, but this has been happening since 2018 and before Right and under your witness of that.

Goran Cvetanovski 1:15:46

So you, this has been a year of user experience and I think that even now, when you look at companies that are trying to implement all the application, ai applications, they're, they're, they're entering a hinder where, basically, even the users they say like, yeah, great that you're doing this, but I want the user experience to be like chat GPT. I want to talk with it, I want to converse with it. Why should I have like a dashboard of I don't know click tableau, whatever it is when I need to play with all of this data on a site, like apply this parameter, and then you need to remote this filter and this thing. I just want to speak with it, like give me the last quarter, how do?

Universal Text and Change in a Year

Anders Arpteg 1:16:27

you look like. I think Luca said a good thing there, which is that text is surprisingly universal. It's universal in the sense that most, or basically any person can use it. The user experience is very simple with text.

Goran Cvetanovski 1:16:41

Yes.

Anders Arpteg 1:16:41

And image and audio, but it's much easier than a weird, complicated form.

Goran Cvetanovski 1:16:45

I think text is better because you can actually formulate yourself. You take time to write your set name.

Anders Arpteg 1:16:50

And the second part of that is not only the user experience. That is easy with text, it's also easier to build applications on it. So basically any company have data in the formal text. So it's surprisingly easy to build a product and that's why we've seen so many product deployments in this year, because it is simple to build. It is simple to take in a large language model and apply that to the data you have and integrate that into your product. So it's surprisingly universal also in terms of building products. So both the user experience and product development is easy with a universal format like text.

Henrik Göthberg 1:17:26

But I think you can go even bigger on. This was the year of the user experience, because working in traditional enterprises trying to adopt data and AI, I think the trend for 2022. When we did this a year ago, we talked about how we had the trend around data mesh and data as a product Okay, good, so now we're starting to productize things. But in this year, we came all the way to something that I think Anders and a lot of people have known for years you need to build proper solutions and products. So when you know so I can imagine when you start now in your new company, you are thinking the end to end stack, from the beginning, from technology, to what is the user experience of my users inside my organization and this I saw a couple of times now where you know we are talking with clients and all of a sudden, okay, we need to have some UX guys, we need to have this and this and that, and I have not really seen that.

Henrik Göthberg 1:18:26

That that big understanding in the traditional enterprise has been are the engineers over here, the little scientists over there, and then we do something in power BI. That is not a product thinking Now. Now it's an end to end product thinking that I am trying to push in my in all my work and also where the clients are more mature to understand we need to go product. So I think use, and when I say product now, I mean the user experience as the end game, right? Yes, because you understood that years ago and you've been doing that in Spotify years ago.

Anders Arpteg 1:19:00

Right, but we haven't in traditional enterprise, not so strongly and I'm trying to bite my tongue off to go 2024 here, but just speaking about the gap between the companies that are making proper use of AI in those that are not is something that I think we, during this year, have seen increase as well.

Henrik Göthberg 1:19:21

Yeah we can speak more about that. Yeah, but you stop now so he can do his best.

Goran Cvetanovski 1:19:25

So the second one this has been a year of year of change. We are not talking about change very often about this, because AI equals change. Right, it's a change problem is not actually a technology problem. What we are having right now and I often speak with my team is that we always need to be able to not dwell on the ways how we work and do things rather than be adoptable and agile at any point of time, because people that actually have built a, let's say, some kind of understanding of how AI will look like in a five, six years from now this was in 2022. This is completely off. It's gone. They needed to recalibrate every single thing about it. We also.

Goran Cvetanovski 1:20:11

It was a year of change because now, when you have the users using this technology as much as they can, they actually understood the power of it and many people took it very negatively, like AI is going to replace us, instead of thinking like a, this is not going to replace us. We just need to change. We need to adopt this technology so we can be better. Right, if I need to write like one. If it takes me one day to write an article now, I can write it maybe in half an hour. That means I can write three articles today, maybe, maybe even more, I can be more productive or I can dedicate my time to doing something else, but we need to be able to change, and this year changed everything the perception, the mindset, and I think that collectively, as a society, as a people, we have entered a new level of consciousness and human, maybe evolution.

Henrik Göthberg 1:21:09

I agree and I don't agree. Super good, because I think what you're talking about is profound and true. And we even wrote an article, me and Mika Lecklingval. We learned how to hyperscale and we are trying to formulate what comes after economies of scale as an economic dogma, and we did this two years ago. So adaptability beats scale. How do you scale efficiently? And I even talked about that and used that as an example on the innovation summit two years ago and I gave example about economies of learning and I did a really nerdy one in Dubai on this topic when it was fresh. They didn't get it, they didn't get it right.

Henrik Göthberg 1:21:53

But what you are saying now is that the only thing that matters that now becomes a super true is economies of learning and adaptability. So the way you go faster, the way you're improving your business, is not about sharpening 5% on the old way. It's about constantly adapting and learning new ways. It's a completely different mindset than a ball game. And you are now claiming that this change happened In some ways. Yes, we felt it. Everybody could feel maybe what Kurzweil is talking about, with accelerating returns and all that. It's going fast now. But if I look at how the real industries, the traditional companies are working. They haven't really learned this yet. The old steering models and business models are still fucked because they're based on the old school way of efficiencies.

Goran Cvetanovski 1:22:49

I will come back on that. On the point five OK, right, so we saw it. Now I think that many of the organizations that we have, traditionally they are afraid of change. Why? Because they don't understand it and they are afraid of it. And then when they invest, they invest for different reasons. No, they're not geared.

Henrik Göthberg 1:23:08

They're not organized for change.

Goran Cvetanovski 1:23:11

Which leaves us to the number three, which has actually been a year of value. Why? Because I think that, no matter which technology comes at us, we need to understand that it's just a technology. It's just a tool, a tool that is there to provide bigger value for what you're doing. So, if you're a restaurant and your job is to serve a very nice food, using AI and robots to increase the service, but you have a shitty food, nobody will come and eat it, right, and many people actually find themselves stranded, that they need to revolutionize their companies and start reimagining themselves instead of focusing on what their value is on the market.

Goran Cvetanovski 1:23:57

The perception Right, if you're a tracking company, you're a tracking company. You shouldn't be something else now. Right, but they actually took that part, many of them, and I think that it was very negative to some extent. But now you can see companies are coming back because they are finding a way how to derive value, because the only end of the on the end is actually how do you actually get value from the technology that you're investing? I am investing in the generative AI because it's, let's say, increasing my marketing capabilities. I can do different things, but we need to find value. So, no matter which technology it is, the same aspect comes to it?

Anders Arpteg 1:24:41

Is it a change with this technology? If you call it that or not, I mean, in some sense you can say you have to do what you always have to do find value.

Goran Cvetanovski 1:24:53

Start from the value. What are you trying to do? I'm trying to make cheap what is called best apparel for the cheapest price out there Is that H&M. So why don't you use AI to make even more fashion? Better fashion for cheaper money? But because that is your slogan, why should you revolutionize now fashion in a way that you don't understand?

Henrik Göthberg 1:25:17

But, to be honest, maybe it comes further on. But I can flip this whole thing and say we're no better this year than any other year, because we're talking about the technology generative AI and we're giving all the stage. It's a beautiful new technology. But in the end, back to basics, back to Anders' comment here. Right, if you don't know what you're going to use it for how to work.

Goran Cvetanovski 1:25:41

But that is the point. This is the year when we realize that, no matter what technology comes along, you always need to start with the value. But do you think?

Henrik Göthberg 1:25:49

that's happened Because, in one way I think, we had four more in generative AI everywhere and very little value. No, people are starting to understand.

Goran Cvetanovski 1:25:56

People are starting to understand. They jump on the generative AI. I mean it's ridiculous, right.

Henrik Göthberg 1:26:02

We are all the different reports. We're going to put 30% on generative AI in our budgets for marketing, blah, blah, blah. And it's tech-driven first right, Instead of really truly understanding what is the problem use case and then reverse engineer into the technology and usages and then therefore the techniques relevant. I don't think we're that much better.

Goran Cvetanovski 1:26:24

We are not, but that is the point that we are actually. Now you can understand that, no matter what you do, you always need to start with the value.

Henrik Göthberg 1:26:34

So if generative AI, okay, it's all hyped, it all goes out and everything looks like new. But in the end, for more and more people, back to value. Please, please, back to value. Yes, okay.

Goran Cvetanovski 1:26:46

Yes. Then number four, disruption. Year of disruption. I think that basically, whatever spotifies and Amazon's and Google's and et cetera if you remember beginning of the year, google's stocks went down at the beginning because they were afraid how chat GPT is going to influence their search engine. You remember that?

Goran Cvetanovski 1:27:09

Yes, right, yes, and there have been a lot of discussion about this.

Goran Cvetanovski 1:27:14

So my point is actually that everything that we know today with this generative AI and if we are talking about co-pilots, every single thing that was a disruptive force in the 2000s everything Uber's and et cetera I think they're going to be disrupted again in the upcoming five to 10 years, just because we are in a completely different. We're living in a completely different world right now with generative AI or the way, actually how we will find new, innovative way, how to do things and summarize them and make profit out of it. If we are entering this hyper personalization type of era where I don't need to go to a web page to search for shoes I mentioned this couple of times rather than I speak with my co-pilot to do that, imagine what is the infrastructure behind it. The payment gate to is how my co-pilot is speaking with the co-pilot of the company and the co-pilot of the company is speaking with the co-pilot of the supplier and the co-pilot of the supplier is speaking with the supplier of the textile, and et cetera, et cetera. Then everything again will be disrupted.

Anders Arpteg 1:28:28

I'd like to challenge that with little bits, because I think for one of course you're right Disruption will happen and a lot of products and business models will be changed and they should. But I think also, generative AI has another pillar, so to speak, which is simply to augment the human. That means that the current product you're doing, the current process you're having, the current business model you have, can stick, but you simply can do it more efficient. If you're an ad copywriter, if you're a programmer, imagine all the developers we have today that can use AI to help them code in a much more efficient way and in a more high quality way. They will simply be augmented by generative AI. One part, I think, of the value from generative AI at least, will be simply to augment the human, but another part will be also these kind of new business opportunities that will arise. But I don't think we should dismiss the simple value that augmenting the current processes and the humans that work with them will have. I think that will be really big.

Henrik Göthberg 1:29:35

I agree with Anders here that there are two parts to this. We had this conversation before the pod and I was trying to shape it in a formulation that you did it better now. I was trying to say actually a lot of the AI investments we've seen in the past has been about improving process or reinventing process. Now we can have all this opportunity where we have AI tools, where I don't try to reinvent the whole business model or process, but I give tools to the humans to figure out how he can be more productive in the old process. That's my word saying exactly the same as you said now.

Henrik Göthberg 1:30:12

I think, then, there are two different. This is also a trend for 2023 that not only are we talking about AI to improving our business processes, but actually augmenting me as an individual.

Anders Arpteg 1:30:26

Individual and also products. So if you take what the Microsoft and Google and did? They didn't change the product. No they augmented them. They added a co-pollet on the site for the office for the word for Excel, for the web browser for Windows. They didn't change the product, they augmented it.

Henrik Göthberg 1:30:42

It's augmenting our products and individuals that we saw right front and center in 2023. And the narrative change from reinventing processes to augmenting processes and individuals.

Anders Arpteg 1:30:56

And then there will be, of course, new opportunities as well, of course as well.

Henrik Göthberg 1:30:59

But the low hanging fruits.

Anders Arpteg 1:31:00

It's also you shouldn't dismiss the value of simply augmenting the existing humans and products.

Henrik Göthberg 1:31:06

Yeah, and what happens now is that people, you can. Now we are moving into the trends of 2024, because we are now talking about innovators and early adopters, and you cannot be an early adopter as a company, but you can be an early adopter as an individual.

Anders Arpteg 1:31:23

Also the product improvements we saw during the year that Google and Microsoft and Meta did. That was simply augmenting existing products.

Henrik Göthberg 1:31:30

Yeah, really no new products.

Goran Cvetanovski 1:31:33

No new products, yeah. And the last one then it's people. We are usually forgetting that basically all of these things user experience change, value disruption, regulations all of that is impacting people. We started the year with a lot of layoffs and et cetera, and we are talking often about AI divide, and recently we talked that there is a new trend where basically more people that are working with these technologies are now coming back to Europe than actually going away from Europe to US, which is super trendy but which is super good.

Goran Cvetanovski 1:32:09

But in my perception of this, basically I think that this has been a year of people that first, if we are talking about, if you want to be good with this technology, if you want to be advanced with this technology, you need to have the right people and skillset in the place. We are talking about AI divide. We are talking about AI in Sweden, Norway, nordics, germany, europe, wherever we want to, and we talked about quite largely with Maria Stokhous from the politician about it. She asked me actually, what do you think what we should do? So we need to get as many people in this country that are working with this technology, because once you have that innovation, this mindset, then you can actually start creating things.

Goran Cvetanovski 1:32:48

The second thing was the very important about people is actually that they are the ones that determine which product is going to fly or not fly, and this is very important as well. So, in the end, and because this technology is going to impact them eventually, how they use products, how they use services and everything else and then we have organizations. For example, we are usually forgetting the end users or the people in the sales and marketing that will be using the technology by itself. So data scientists are right now sometimes creating a product that they have never spoken with the person, if they like it or not, if they basically need that solution or not, and in the end, it's all about people, right, because the value is about people.

Henrik Göthberg 1:33:31

That's the whole adoption story. User experience is about people, user adoption and user experience.

Anders Arpteg 1:33:35

And perhaps you should get you know in 2015, google said they're going to be an AI-first company. So they are reorganizing and changing focus and prioritizing according to be AI-first company and not of other follow-down. But perhaps you know the coming thing, because I do agree with what you said Goran is going to be we're going to be a people-first company. Yes, I think that actually is a good slogan.

Henrik Göthberg 1:33:58

But actually I'm going to test. I'm changing my Derek slogan the AI adoption company Maybe.

Goran Cvetanovski 1:34:09

I have a better one, we will talk about that later.

Goran Cvetanovski 1:34:11

I have to think.

Anders Arpteg 1:34:12

What will the acronym be for that?

Henrik Göthberg 1:34:14

No, no, the data AI readiness the. Ai, the data AI readiness deal holds. Yeah, it still holds, but so finalize the year?

Goran Cvetanovski 1:34:23

Yeah, completely Okay. So this technology is too young and it's too, is developing too fast. And it's almost like you know, if you take an atom out of water, it becomes very, very, very you know eager to get the atom back. So I think this technology is also something like that it's unpredictable and it's developing very fast. But all of the things that you were talking about technology, regulations, me, people and et cetera we need to understand. This technology is very young, it's too early and we need to give ourselves freedom to actually make mistakes. And right now, when we are starting regulating all of these things, something that is not actually fully shaped, is almost like regulating a fetus that is in three months and is not being born. We don't know how it's going to look like. Yes, we're going to put guardrails and all this other stuff, but in the same time, we in Europe probably will be shooting ourselves in a foot. If you do it in the wrong way.

Goran Cvetanovski 1:35:22

Yes, because the technology is too young. We don't know what we're doing. We were talking about intellectual property. We are talking about people actually putting things in chat, gpt, without understanding where this data goes. We had like a league going on when they were showing how they were training data. All of this technology is very young and we need to accept it. That is very young and it's going to be stupid and it's going to make mistakes and we're going to do a mistake with it, but we cannot fear that. We need to move ahead and utilize it as much as we can so we can find a way how properly to use it.

Henrik Göthberg 1:35:55

Yeah, well, put Should we. That's it.

Anders Arpteg 1:35:59

Yeah, too early, I would object to a bit. But that it's moving too fast I agree with. But certainly I think we shouldn't tell companies to wait to start to add up to AI.

Goran Cvetanovski 1:36:10

No, no, no, it's opposite, they should be utilizing it even now, it's not too early really.

Anders Arpteg 1:36:15

No, no we are too early to regulate Too early to regulate into. I think we need to regulate even earlier as well. Okay.

Henrik Göthberg 1:36:24

I like this but let's sort this out now because I think we're talking about the same things. But it's a good segue. The devil is in the detail. So regulating like something is a mature product. When it's not, it's very, very dangerous because it's very hard. But to have no guardrails or no regulation at all, even if it's immature- but I don't think anyone is really promoting like no regulation. No, no, no.

Anders Arpteg 1:36:49

I think we really need to have that and as any technology, if you have nuclear technology, but regulate.

Goran Cvetanovski 1:36:57

Why don't we standardize and then move to regulations? I mean, I went to Australia. It was beautiful 60 companies got together to make a standard how to implement AI in an enterprise so they can move faster. So standardize first, then regulate. Why do we need to regulate and then standardize?

Anders Arpteg 1:37:17

Should we go into this topic?

Goran Cvetanovski 1:37:18

Yes, we can. If you want to, no problem.

Anders Arpteg 1:37:22

Let me use perhaps the old Elon Musk comparison to the seat belt. I think that's a very pedagogical way to describe it. And before, when they built like automotive cars, the industry in the 1950s were very opposed to having seat belts in it. Because if they put a regulation in place where automotive manufacturers have to put seat belt, no one would ever buy a car. And so awkward and horrible to have a seat belt. That will be horrific to the whole industry, even though all the statistics showed that you get killed much more if you don't have a seat belt. So that was really destructive to the whole society that it took so long for regulation for seat belts to come in place.

Anders Arpteg 1:38:12

Now, with AI, it's moving so fast that if you wait for a lot of horrible accidents to happen, it can be really really nasty. It's moving so much faster. The AI technology is moving so much faster than the car industry did in the 1950s. So I think we need to prepare to have regulation that will ensure that we are using AI for good and not for bad, especially in these times where it's so easy to have people being augmented with AI. And that means all people. It means people that use it for good purposes and people that use it for bad purposes For cyber security, for warfare or whatever. That is something that we don't want to wait for to happen. We need to stop that before it happens, and I think we need to make sure that we have regulation for these kind of bad use cases that protect our society, and I think we need to do it quickly.

Henrik Göthberg 1:39:12

I fully, 100% agree to your statement. But how do we regulate? So it's not a matter of if, it's a matter of how. How do you regulate the fetus, if I use the word? How do you regulate something that is experimentative still?

Anders Arpteg 1:39:31

I think you can, but I think we can also say how you should not regulate it. That's easier to say. If you just regulate because of the technology, that's really bad To just say that because it's a chatbot, we should allow it or not allow it. It's like one of the original versions of the AI Act was basically saying if it's a chatbot, it's not that dangerous, it's just a customer service thing. What not? That's low risk or minimum risk. But if you use a chatbot for grooming purposes online for children or for using it as a therapist for people that are suicidal, it is not low risk. That's a typical example of if you regulate technology and not the use case, it will turn bad.

Goran Cvetanovski 1:40:18

Yes, but that is what we are talking about. It's basically like this is air, abortation, abortation.

Henrik Göthberg 1:40:28

But let me draw out something that becomes more and more strong in my head right now as we speak no-transcript. When you're regulating something that is super mature, like when you're selling a mature product, that is easy to understand. If I sell a phone, I don't need to consult around the phone or how the phone is used. I can just put it in the shelf and someone can buy it. If I'm buying, if I'm selling a product that is highly complex, like an ERP system, I probably need to do a lot of consulting and guidance in order to implement something that has more risk and complexity.

Henrik Göthberg 1:41:09

This is what I now think is happening with regulation. If you're regulating something that is very simple and clear and known to men, then you can do a very simple regulation and then lower don't hit yourself, don't hit anyone with a hammer. It's simple and you don't need to have any consultative approach around that. But when something now is very gray or complex and difficult, the package what is around the actual hammer or whatever you want to call it, the AI becomes the more important In sales. We talk about traditional product sales versus more consultative, complex sales. I think this is what's happening now when we are regulating. Wrong is when we are trying to use a very simplistic approach to use put the regulation out and there's no package, there's no consulting, there's no guidance. I think this is the major problem with regulating wrong.

Anders Arpteg 1:42:02

I don't think you need to understand how to take something like.

Henrik Göthberg 1:42:06

Now how to act, how to do it, how to Take autonomous weapons or something.

Anders Arpteg 1:42:10

I mean, we know that autonomous weapons that are stupid, are super, super dangerous. I heard Jan-Likund speak about these ones, which really made me think a lot. But should I go there? I'll go there. Anyway, you can actually say that if you have people or humans out of the loop, for some use cases that means basically the only thing that needs to be there. If humans are not, there is AI For some use cases that are completely autonomous.

Anders Arpteg 1:42:43

I think we need to have regulation in place that speaks about autonomous systems that is not in control of humans. That is really dangerous. I think for some of these use cases, we need to have it AI specific. I do think that are examples that, even though we don't understand how the AI works, even if it's immature, even if it's we don't know if it's a large language model or if it's an Alpha-zero reinforcement learning reasoning and super advanced thing that is working underneath we know that if we have a system that is going to be able to target a thousand people at once using some swarm autonomous drone technique, it should probably be regulated just as chemical weapons need to be.

Henrik Göthberg 1:43:28

And my point is, of course, you're right here I think we need to spend a lot more investment, money and effort on the how around this regulation. The regulation is there, as it should be. As you highlight, the what is then defined in the regulation, but I think, because we are all experimenting, we need very, very clear approaches. How should I go about experimenting? This? So it's not the what is not enough when it's so immature. It needs to be a lot more how in this regulation, which is maybe not a regulation anymore, but I would call it the package.

Goran Cvetanovski 1:44:08

I would pull the support system, but this is a good segue to 2024.

Henrik Göthberg 1:44:14

Yeah, let's stop there, let's stop there.

Goran Cvetanovski 1:44:15

No, no, we can continue, but we can Because one of the biggest things that will happen for Europe is probably going to come in January.

Henrik Göthberg 1:44:26

Yeah, so what is the prediction of 2024? Let's start with the HEA Act.

Anders Arpteg 1:44:30

Yes, let's start with discussion here. By the way, it's awesome.

Goran Cvetanovski 1:44:34

So let's start with this. So let's talk about regulation first. So I think that regulation is really good. The only problem that I don't agree that we are regulating basically the hammer and not the usage of the hammer Right, and we are all in agreement that that is actually.

Anders Arpteg 1:44:47

I think, we are all in agreement in that.

Goran Cvetanovski 1:44:48

Right. So, and it's going to be very interesting. So, first of all, this is not like a law already, rather than it's something that is put in motion. So in January, if the European Parliament will vote on it, keep in mind that there are very big forces that might actually boycott this, because some countries that are super involved in all this and invested in this, like Germany and France and Spain and Italy, where you also have one of the major supercomputers in Europe and they're developing LLM models, of course, maybe this year We'll see how it's going to turn out, but once it's voted, I think that next year is going to be a year of fear for me Personally, Because if they vote yeah, act.

Goran Cvetanovski 1:45:34

It's going to happen what happened in 2018 when the GDPR came out.

Henrik Göthberg 1:45:38

I guess we need to start our AI regulation, yes, so.

Goran Cvetanovski 1:45:44

I already bought some domains about.

Henrik Göthberg 1:45:45

That will be very regrettable business. Yeah, exactly.

Goran Cvetanovski 1:45:47

Corona I already talked about. Yeah, I already bought some domains as well, so my prediction is basically if this comes in motion, then companies will have two years to be compliant and two years for the companies, two years for the countries to have a support system.

Goran Cvetanovski 1:46:09

Keep in mind that what happened in 2018, some of the companies just basically not to have this headache with GDPR. They completely killed their entire database. Just pulverize it with one bottom delete. Okay, start building it from scratch. What do you think is going to happen for organizations that have like hundreds of models that they don't have any lifecycle management on? They don't know who created them, they don't know which basically model from a GitHub or whatever it has been used. What are they going to do about it? Delete.

Henrik Göthberg 1:46:43

How will you deal with this? Yes, this is what I'm saying. So you were talking about this fucking simple point. Where is the support mechanism, the support package? Where are the guidelines? Where are the examples?

Anders Arpteg 1:46:54

Where are the?

Henrik Göthberg 1:46:54

case studies. Where is the regulatory library?

Goran Cvetanovski 1:46:57

Yes, let me tell you how this is going to look like so next year, and if you look at LinkedIn, this is very obvious, right? And many of you might hate me at this point of time, but I know that you're gonna make a lot of money, so this money is going to come for you next year. Next year, the consultants are going to just basically make so much money out of this form and they're going to play the field the people that basically have never had anything to do with AI.

Henrik Göthberg 1:47:23

They're gonna have so many actors Exactly, and then we'll read to companies.

Goran Cvetanovski 1:47:27

This is exactly what you need to do. Your model should be doing this. Ask him what the model actually is and what the model is. He will not know how to put it in place and, yes, there are some that will know, and we had some very knowledgeable people here, but many people are not. And if you look who is actually happy about the AI Act, look at every single consultancy. They are pushing this so hard because they know this is how we are going to be relevant in the AI sphere.

Henrik Göthberg 1:47:57

There are so many management consultants who have been dropping behind, one by one, in the real engineering game and I've seen it clearly and now they pop back up on the surface again and they are lightweight, exactly Because what you know, these consultancies.

Goran Cvetanovski 1:48:16

They try on this Dig God, and it's a very easy sell. You play a golf with some cell. Okay, how is your model going? What model? Well, you know it's 25% of your global revenue. That will go out right. What models, like? You have a data science team. Yes, I have All right, they're good. 25%. Oh, come on Monday, you need to talk, let's talk. So it's going to be so much Then I believe that you will. We will find quite a lot of new companies, business models that will be called audit companies.

Henrik Göthberg 1:48:46

I want to have your whole, your business.

Goran Cvetanovski 1:48:51

Because I bought all the domains.

Henrik Göthberg 1:48:56

You're being so lazy, goran, so you're not wanting to build the companies anymore, you just built the board. I just want to sell the domains.

Goran Cvetanovski 1:49:03

Yeah, audit red of visiting all of these other things. I bought them all, but no, but my fact at my point on this, joke on a side, this will bring actually companies quite a lot in a fear to understand, and it for a good to understand. What do they have under the hood? And this is where data governance and AI governance comes in place, because if you really want to find you know how your model works, you need to have data governance. You need to work with data. If you want to understand what is the life cycle of your what is called ML models, ai models you will need to have AI governance. And then, finally, you will need to have Phenops, because all of this costs like a lot, right to produce all of this, llms, etc. So even now we are getting quite a lot of requests for Phenops topics. What?

Henrik Göthberg 1:49:54

is.

Goran Cvetanovski 1:49:54

Phenops. Phenops is basically cloud cost optimization. So data management cost optimization. How do you actually optimize your processes of cloud usage so you don't over burn your?

Henrik Göthberg 1:50:10

financials. So how do you optimize the setup of your platforms from a costing? You know, compute versus storage cost, etc.

Goran Cvetanovski 1:50:19

Who has access to it? When do you compute, do you need it necessary and all these things.

Henrik Göthberg 1:50:24

Phenops.

Goran Cvetanovski 1:50:25

Yes, phenops and this is actually we were discussing the other day with a client who is actually responsible for this, and some companies have it and some companies is very distributed. There is this Phenopmanifesto right now. That is very hot, but if you look at the persona behind the Phenops is like 10 different personas, which is basically everybody except the cleaner in the company, right? So everybody who has.

Henrik Göthberg 1:50:46

And then you have a couple of companies where basically, they use their company credit card to get another AWS license or GCP license.

Goran Cvetanovski 1:50:53

So you will have like this ghost thing as well, but in general, you have the CFO or you have the person that is responsible for cloud computing and cloud infrastructure, usually people that are working on this. So this is actually where I think that the most things will go next year, but also, we will push people once they start doing that, and as a consequence of what happened this year is to look a little bit more on the value, because once this, once now you have regulation and you have cost optimization, you as a manager working with AI, you will have hard times getting money, which means that you will need to start working on value. They will tell you it's enough with experimentation. Tell me how much we are increasing value for it, because if not, I'm killing your cloud computing power. You will not be able to do this and I will not allow you to get another.

Henrik Göthberg 1:51:44

So summarize this in what is the short-term prediction or trend.

Goran Cvetanovski 1:51:49

My prediction is that it's going to be a year or four. That will force, basically, organizations to re-evaluate how they approach AI, if they should approach AI, how they govern models in production, how many models they need to have in production. What is the value of that and what is the cost?

Henrik Göthberg 1:52:06

And this is around this, and then bolting on to that is that and we're going to have consultants making a shitload of money on this.

Anders Arpteg 1:52:13

So is it really a fear of missing out, then, or is it the fear of getting sued? Oh, yeah exactly, so it's not.

Goran Cvetanovski 1:52:23

Fear of being sued. Fox.

Henrik Göthberg 1:52:28

Fear of Fox.

Goran Cvetanovski 1:52:33

This goes on linked into yes, fear of being sued.

Anders Arpteg 1:52:38

Yes, good, I think that's good Fox, you heard it here first.

Henrik Göthberg 1:52:45

No, being sued Fear of being sued Fox.

Goran Cvetanovski 1:52:50

I'm making a t-shirt tomorrow.

Henrik Göthberg 1:52:52

We are the FOMOS of the Fox.

Goran Cvetanovski 1:52:55

Okay, so I'm done with the predictions.

Henrik Göthberg 1:52:58

Do you want to lead into the predictions, or should I do one or you do one?

Anders Arpteg 1:53:03

Yeah, I think, still one of my.

AI Adoption Gap Grows in Companies

Anders Arpteg 1:53:06

If you just think for 2024, a company is thinking okay, we understand AI is a big thing, but how do you really make the best use of AI?

Anders Arpteg 1:53:14

I think one of the sayings I've been having for a long time still holds very true, even in the world of generative AI, and that is that AI is good at different things than humans.

Anders Arpteg 1:53:26

Ai is really good at going through large amounts of data, but that's so in a rather superficial way. It can't really reason in an efficient way, especially not chat-tbt and these kind of large language models, but it can go through a huge amount of data in a very, very efficient way, much more efficient than any human going through thousands of pages of text or hours of videos and whatnot, or even chat-tbt that basically have the knowledge of all the web. That's the huge amount of data that it actually can process and manage in better way than any single human ever could. So it's really good at working with large amounts of data. Humans are not, so this is awesome. We can see here that if you want to use AI in the best possible way, let's make use of AI for what's good for humans, what they are good at, and any company in 2024 that is doing that will have a good edge. So let's make sure that we use AI for the things it's good for Humans they are.

Henrik Göthberg 1:54:25

Let's chop up the prediction here. So you have a prediction here that basically says what happens to companies in 2024, that is, using AI or not using AI.

Anders Arpteg 1:54:36

Some companies will learn this, if I'm correct, and they will have huge benefits of these new possibilities and they will increase in value. So I think the gap, the prediction is basically, I guess, this the gap between companies that can make positive use of AI will increase compared to companies that will not learn the lesson and do not make use of AI in a good way. So that gap will increase. Previously we were speaking about the AI divide in different ways and then focusing on the superscalers, the big tech giants, but I think we will see other companies now that actually do understand how to make use of AI and they will also move towards closer this kind of superscalers, but at normal scale. But you will also see companies that are still in the conservative, traditional type of technology, that are unable to make use of AI to augment existing processes in humans. And that gap, the AI gap in this way or AI divide, I think, in 2024 will increase rather.

Henrik Göthberg 1:55:39

So let me try to put my words on it, and we spoke about it a little bit preparing for the pod. So we've always had the real AI innovators and we talked about the AI divide for years and actually one part of why we started this pod to make this more to spot like this. So we have more urgency in Europe, especially so we have the tech giants right. We have been comparing with them. This is the AI divide of the rest of the world and what you are saying now, anders, there is a nuanced view on this emerging in 2024, that we now have the AI innovators to tech giants and then we have what we could consider maybe the early AI adopters. So we have the companies who are, or individuals who are experimenting and basically have increased their own productivity, that basically now are on the ball.

Henrik Göthberg 1:56:37

And I made this example where economists of learning, economists of adaptability when I took the case, that of Kobe, the data innovation summit, the Kuglenbibbuteke have actually been brilliant very early because they adapted the BERT models and stuff like that. So here we have one example here that we can now see in 2024, some people, some organizations, adapting a little bit and starting to be part of the new curve, so to speak. And then the hypothesis for the organizations that are doing nothing. They actually. That divide is now increasing and it's starting to go with some speed now. So there is a difference here that the people who are the laggards or the non-actors at all. That's what we're worrying us.

Anders Arpteg 1:57:31

And I think also what you said comparing companies to humans. I think the same applies both for single individuals humans as for companies. So humans that are able to use AI for their own tasks will get surprisingly better value compared to humans that do not, and the same for companies. Companies that do find a way to make use of AI properly will have a significant advantage to companies that do not. So I think this is a learning for both humans and for companies.

Henrik Göthberg 1:58:05

Yeah, and there's been this quote circulating and I've used it sometimes AI will not replace you. Another human competent with AI will replace you.

Anders Arpteg 1:58:19

So don't think AI will replace humans. It's rather that humans that use AI will replace humans that do not use AI, and the same for companies. Very simple.

Henrik Göthberg 1:58:29

Yeah, but this is a fairly major prediction because it sort of highlights how the whole market is shifting and who do we need to be right. Let me go another way now. I think another big debate that potentially can lead into prediction that I think is crystallizing more and more during 2023 is the way of AI open source versus closed, so to speak, and you will have people like Jan Le Koon being at the different shows and talking quite a forefront and saying like it has to be open source as a way to evolve the world and other proponents of open source. Have a look at where would we be without pytorch or TensorFlow or anything of what we have reached today has been based on open source. Don't forget that. And all of a sudden we have some movements that are sort of the opposite.

Anders Arpteg 1:59:36

And it's really weird because the concentration of power that we're now seeing, with the super skaters running away and even single people like Elon Musk that has this insane amount of power right now, and the concentration to a few individuals or to a few companies we need to break that.

Henrik Göthberg 1:59:55

We need to break that, and it pisses me off that we don't talk about this, and sometimes it's almost like Jan Le Koon stands there alone. I mean, I would argue very, very strongly that proprietary is bad in this sense because it's such a concentration of power, especially now. So how do we make that into a prediction for 2024? Can we even make it into a lobbying? What do we urge people to do and think about for 2024?.

Anders Arpteg 2:00:31

We should have regulation that made that incentive even stronger.

Henrik Göthberg 2:00:35

Yeah, regulation that incentivizes open source.

Anders Arpteg 2:00:38

We're not seeing that, I must say.

Henrik Göthberg 2:00:40

Small, you get a little bit of exempt in the AI Act, yeah. But it's.

Anders Arpteg 2:00:44

like you know, instead of killing open source which it would otherwise it is, at least leave it hanging there. We should instead incentivize that, and that we're not seeing.

Henrik Göthberg 2:00:57

But is there a prediction in this? Okay, where will this I mean, it's a little bit like VHA's beta max, which will increase in momentum in 2024?.

Anders Arpteg 2:01:09

I don't know. In some sense, I think the proprietary approach is increasing still, because, you know, open AI is anything but open. Of course, google is not releasing anything also, and Microsoft of course. Not metais though, but it's the exception to the rule, more or less.

Henrik Göthberg 2:01:29

But can I be blunt If I'm a European who believes in open source, I would put my money and investment in idealistic ways to try to make the maximum out of Mistral. Is that a bad consultant advice or is that a good one?

Anders Arpteg 2:01:52

I have to think about this, but I think it's a good question. How can we really incentivize companies like Meta and Mistral, because both of them are pushing for this, but it really goes against commercialization.

Henrik Göthberg 2:02:06

Nobody's voting with your wallet. I go to the supermarket and I decide I need to buy ecological milk. It's bullshit, but it's like what can I do in order to? How can I vote? When I vote it, I eat more sustainable food. Whatever I do, can I vote with my wallet? What kind of world I want and how do I do that?

Anders Arpteg 2:02:27

Because at the same time you can reason about the AI Act and the requirement for transparency that they are enforcing, saying you need to show what data you have used, how the model works. Good luck explaining that. In some sense, that is actually giving some incentives to open source right.

Anders Arpteg 2:02:48

But it needs to be doing that in a way, without killing the business model of all the companies and without having to give away the IP. For me, I certainly don't have a good answer for this, but I think it's scary, the trend we're seeing right now.

Henrik Göthberg 2:03:03

Do you think the trend is scary? Because it seems like the proprietary world or the superpowers are getting more and more secretive by the minute.

Anders Arpteg 2:03:13

They're saying they're doing it for safety reasons. We really work from a politically correct point of view, but in reality they're doing it from a commercialization point of view. They're using safety as an excuse to keep things closed.

Henrik Göthberg 2:03:27

That was the predictions and discussions about open versus proprietary. Do we have a couple of more topics on my mind? Do you have one in between?

Anders Arpteg 2:03:38

Sure, I mean, if we go for the three components of an AGI system.

Henrik Göthberg 2:03:43

Yes, I was going to go there, you go and I bolt onto that.

Anders Arpteg 2:03:49

One way to. Jan Le Koon also had his version. I have a bit more simplified one saying perception, planning, reasoning and control. If you take a full self-driving car, just to explain it, perception is basically taking the input of all the cameras or whatever sensor you have and putting it into better space or some kind of world model that you have. That is basically what the chat bots are doing today. They are taking the perception game.

Henrik Göthberg 2:04:15

We're getting better and better at.

Anders Arpteg 2:04:17

Yes, and really good at perception these days. That is not a problem. Today. We can do that extremely well. We can have a world model now by understanding this, just as a self-driving car can understand the world around it in a vector space. But it's not sufficient with that. Even the DeepMind paper about AGI that came out said this also very clearly. We are missing a lot. We're only on level one out of five in terms of AGI.

Henrik Göthberg 2:04:42

We didn't mention the paper. We can mention that because that's one of the best definitions we've seen so far.

Anders Arpteg 2:04:50

They tried to have some paper speaking about how to measure the level of artificial general intelligence.

Anders Arpteg 2:04:56

In any case, we are certainly lacking reasoning in the big model. I think that is something that I hope we will see in 2024. It is potentially what the big Q-star thing that OpenAI had is about. Some people say, not, we don't know, we're just guessing.

Anders Arpteg 2:05:15

If we were to have the reasoning power of AlphaGo, alphazero, that can play chess but have no world model at all, but then they have a super high reasoning capability that can beat anyone in chess and go and whatever, no way the human can beat the reasoning power with that very narrow task. Let's take that reasoning power and put that on top of the world models that the chatbots and LLMs have and you get something amazing. But it's still not sufficient. Now you have two parts potentially. We haven't seen this yet happening, but if someone are able to bring in the reasoning power together with the perception of foundational models, that would be an amazing thing. I think we will see movements in this direction in 2024, but it still doesn't solve the control problem of having agency, of taking actions, of controlling a robot or whatnot. One person, of course, is doing it is Tesla and the Optimus Optimus generally.

Anders Arpteg 2:06:13

They came out with the Gen 2 now recently, amazingly being able to work and plan and do reasoning and have perception. But if we get the first one to have all three in a very general sense, both perception about anything, both high level reasoning and the control ability to be agency Jesus Christ, then we get really close to AGI.

Henrik Göthberg 2:06:38

If I now summarize this to a prediction for 2024, perception if you look at these three levels, we are way above on the perception part. Which one do we think we're going to have some announcements on or see something out in the wild?

Anders Arpteg 2:06:56

Which robotics or motorics.

Henrik Göthberg 2:06:58

Not control, so something around reasoning. So have a stab at betting or imagining what type of feasible breakthrough could we see? They talked about Q-Star. Right, and it's been all this Q-Star. The Q-Star is a little bit going in this direction with saying, oh my God, why? Maybe the whole saga in open AI was that they had so much more going on with the GPT-45 or whatever they want to call it. So someone was arguing oh, maybe they have some come.

Anders Arpteg 2:07:34

And they claim it could solve math problems.

Henrik Göthberg 2:07:37

So you've got to solve math problems. Is this the reasoning part? So what are the sort of reasoning techniques? That is sort of because we've been working on this for over a year, more than a year in the different deep mind is working on this. We know this.

Anders Arpteg 2:07:54

Mieta is working on this a lot. The. Jepa paper highlights what they are working on, complaining about the autoregressive nature of foundation models and he says it's never going to be enough period.

Henrik Göthberg 2:08:05

He said that years ago.

Anders Arpteg 2:08:06

It's not the scalable approach, and I, of course, agree with him.

Henrik Göthberg 2:08:09

So where's the research right now on the reasoning part? What could we maybe, if we're lucky, see next year?

Anders Arpteg 2:08:16

I think the JEPA approach, of course, is an amazing one, and just having this kind of autoregressive approach, it's super simplistic. I mean, that's one of the very surprising learnings in the last couple of years. You can get so much value and appearance of intelligence from just predicting the next word, but of course it's very inefficient way from a reasoning point of view. And the one that can go away from autoregressive and have foundation models that can reason in a more high level, that will be the killer app, that will be amazing and that will be able to move into much smaller models.

Henrik Göthberg 2:09:00

But what could be the MVP minimum next step in this? I guess, Mistral is Mistral, right. So that okay. So now we're good. So let's say MOEs is the first step Sure of experts, yeah, mixer of experts type systems. So what we are seeing with Mistral now is maybe the micro level that we are not moving in one mega. Could you explain?

Anders Arpteg 2:09:26

Yeah, it's still not reasoning in a good sense, but at least we have split up the super simplistic transformer architecture, which is very simple, it's just in a high, high scale. Now suddenly have at least split up in different experts In this mixed role that is a mixture of experts. They have seven different, eight types. Yeah, they're seven B but eight of them.

Anders Arpteg 2:09:50

Yeah, thank you, henrik, the first time, but okay, so that makes it super much more efficient, much better than even the 70 billion parameter of Lama 2, but much smaller and significantly faster. So that's moving into like a small smarter architecture. You're seeing now not simply scaling the architecture but actually having a smarter architecture that is speeding up on the improvement and that type of development, I think, is a small but a first step towards something that's changed the architecture to be much more efficient.

Henrik Göthberg 2:10:28

Yeah, and I was reading some stats how they were trying to explain efficiency of doing that. So the difference by lighting up every single neuron in every single time you're doing a path, versus you're doing some sort of you're doing a reasoning, which expert model is more relevant, and then use lighting up a part of this and it's amazingly less neuron being fired.

Anders Arpteg 2:10:55

I think it's actually, if you wait, where to make a super succinct kind of prediction on 2024, I think it will be the year of efficient foundational models.

Henrik Göthberg 2:11:03

The year of efficient foundational models. You heard it here first, guys. I think you're right. I think you're right because we were talking about this oh, we're going to not have so many massive scale. Of course it's going to be massive scale with more to model, you argue, but because of the massive scale in other dimensions, we need to work smarter, and the smartness is actually how we're going to back ourselves into the reasoning capability. I mean, like we're going to end up in that because of other the reasoning of just briefing the next word is so stupid.

Anders Arpteg 2:11:37

So the one that actually are coming up with smarter ways to have an objective function will be able to create so much more efficient models.

AI Market and Chip Predictions

Henrik Göthberg 2:11:44

So you get more efficient models and by having that objective as your first objective, we actually getting closer to the reasoning capability. So the year of the efficient foundation model. And who is best equipped to win that race Do we know, mr?

Anders Arpteg 2:12:03

But of course, Manta has been speaking about this for a long time.

Henrik Göthberg 2:12:07

But if you talk about the value proposition and being first on the market with the value proposition, I think also deep mind.

Anders Arpteg 2:12:13

You know they have been working on this kind of reasoning techniques for so many years.

Henrik Göthberg 2:12:17

We don't even know exactly what it is.

Anders Arpteg 2:12:19

I don't know I thought Germany would be it, but it wasn't. Germany was just in the same old. The same old, the GPT approach, but multimodal yeah, multimodal, but still GPT. And the one that can go beyond that, that will be the winner, and I think Google has the ideas for that. So I think Google or Manta will be one of the potential winners.

Henrik Göthberg 2:12:42

All right, any other predictions I can, I can leave with another one. Do we have any predictions for the VC market, ai VC investment market, and we were joking about, you know, when should we invest it in Nvidia? When should we have invested in AMD? Why should we have invested in Microsoft? So ha ha, you know, is the? Is the money drying up? Is it harder to getting VC? We have, just like Alina Antelana, et cetera, we talked about this. So anything we can say about the money market to run AI, I think that one is extremely hard to mention.

Goran Cvetanovski 2:13:23

I think that if you look at the stock market right now and to all readers, we are extremely novice in this and we're just speculating, so you're having fun. Just having fun. From my aspect, I think that everything is going AI, so I think that, in any case, ai chips and graphic cards are always a safe bet, because you need to have more of those in the future. If we don't find a different way, how do we actually utilize GPU?

Henrik Göthberg 2:13:51

You went, you went the investing part. I was more thinking also about startups and startups. So, having a startup in 2024.

Goran Cvetanovski 2:13:58

So now the past couple of weeks I have gotten an approach because of the LinkedIn Angel investment title I got. There is a lot of companies that are doing a startup on the accuracy and fine tuning, but if you listen to Rebecca and Evelynna and Oliver, for example, they are betting on the application part of AI. I think that is much more a safer bet because if you look at the companies that are right now working with the accuracy of LLM models and etc. Yes, there is some kind of a. It makes sense, but is it in a long run? Is it safe bet?

Henrik Göthberg 2:14:45

Because we don't know where the technology is going.

Goran Cvetanovski 2:14:47

So I think it's basically another band-aid for a bandage, basically for a wound that will be healing in one year, so maybe not. We will not maybe exist in one year.

Henrik Göthberg 2:14:58

So if we summarized the advice from the VC community that we've been on the pod.

Goran Cvetanovski 2:15:03

No, this is my personal. I would not even put them.

Henrik Göthberg 2:15:06

Don't put it worse, but you could sort of sense for the fundamentals. Do you solve a big hairy problem for a user? If you do that, it's typically quite good.

Goran Cvetanovski 2:15:17

I would more bat in a company that are trying to think how everything will work when, in five years, you will have co-pilots and different way of how do you interact with things. And I think disruption is upon us and companies will think differently and some of these companies will not even see them.

Anders Arpteg 2:15:44

Since I also is in the investment business for some.

Henrik Göthberg 2:15:50

Yeah, you have some good friends that you're meeting with as well.

Anders Arpteg 2:15:52

Yeah, Anyway, I think applications party is one and, of course, just making use of the GPTs and the new type of use cases that you can build on top of the foundation laws.

Anders Arpteg 2:16:04

They will have a lot of potential use cases. They have to be really lean, I think, to make it work, but still we will, I think, of course, see a lot of those in 2024. But I think also I think we mentioned as well Goran, the AI ships and the hardware part, because we can see the hardware space is now finally getting some more competition. It's not only NVIDIA, at least AMD is there. Yeah, amd is there, but every major tech company is now shy and cold.

Goran Cvetanovski 2:16:34

Same old one had like AI chip company.

Henrik Göthberg 2:16:36

Yeah, the rain is going hard on this, the Americans have announced a new ship. America has gone home For a long time.

Anders Arpteg 2:16:42

Tesla had their own ship. Everyone is building their own chips.

Goran Cvetanovski 2:16:45

Yeah, because they don't want to be dependent on somebody else. The problem is that basically the resources for that is going to be.

Anders Arpteg 2:16:53

Taiwan.

Goran Cvetanovski 2:16:54

Yes, exactly so this is where everything else comes to play. So if next year or the year after things get a little bit more serious in the so what's the prediction for the chip market? The chip market. If something happens with Thailand, you just basically sell all of your stocks and go to a fucking island somewhere and shut up until everything is all right.

Anders Arpteg 2:17:18

Yeah, that's a weird market, but at least we're getting. I think we will start to see in 2024 a significant increase in competition compared to the extreme monopoly that we have seen in previous years.

Henrik Göthberg 2:17:31

As anyone follow Jens and anyone else. What are they working on as the next level?

Anders Arpteg 2:17:36

Yeah, it's awesome stuff.

Henrik Göthberg 2:17:37

Could, you.

Goran Cvetanovski 2:17:39

We will have them on data innovations actually, so we will have one of their machine learning.

Anders Arpteg 2:17:45

But there's stuff happening and the NVIDIA is doing amazing things and of course they have so much money that they simply cannot manufacture enough stuff. So they and they're charging insane amounts of money for their hardware. Of course they're making so much money, of course they're spending it on research, of course they're spending it on building the next generation of these things.

Henrik Göthberg 2:18:05

But I heard Jens and I think he was on Eric's work in the New York Times guy who kind of do the panel discussions or interviews, and he asked the CEO of NVIDIA so what's when? You're planning now your architecture on your level? How far ahead are you right now? And this is now everything is happening and it's moving so fast and they need to plan with, he said, 10 years. Otherwise, understanding to what he should develop in order to get the tooling, in order to get the manufacturing capability he needs to envision 10 years from now. Can you imagine envisioning 10 years from now?

Goran Cvetanovski 2:18:45

This is what their bets are on right now, but I think that is what every solid CEO of a company should actually do.

Anders Arpteg 2:18:54

For a super skater. Of course, of course.

Goran Cvetanovski 2:18:57

No, no, even for IKEA, Scania, Tetra Pak, all of these companies.

Anders Arpteg 2:19:01

They shouldn't build their own shit. No, they're thinking 10 years ahead. They should be thinking 10 years ahead.

Henrik Göthberg 2:19:06

I'm not sure 10 years is a very long time. It's a super hard time right now. When I heard him say that how can you imagine 10 years Neuromorphic computing right?

Anders Arpteg 2:19:17

Yes, that's my mission. Okay, so hardware. It will be an interesting year, at least in 2020, for hardware, for sure in AI.

Henrik Göthberg 2:19:27

But you don't want to be specific on any. I think simply what can we see?

Anders Arpteg 2:19:32

It's still a couple of years ahead, but I think the ones moving in that direction and getting some first economically viable solution for that, that, I think, is the future. I can go much more into depth.

Henrik Göthberg 2:19:44

Because in your morphic is when you do everything in one place.

Anders Arpteg 2:19:47

Yes, so when you combine computation with memory into one place instead of having them separate, and you can have them in the same place. Just as a human brain counts, it brings so much.

Goran Cvetanovski 2:19:59

Just investing consultants in this year.

Henrik Göthberg 2:20:02

I've been triangulating what you've been talking about. That and we are talking about, actually, an enormous architect that has been around for how many years? Hundreds, hundreds. Right, but is it because it's unbeatable or is it because it couldn't be made at that point in time and now we can make it? So it probably.

Anders Arpteg 2:20:18

Still really can't make it. That's, Intel has been trying for some time to make it work, but it's hard to find the economy to make it work. But the benefits for the one that do make it work. And of course, Sam Oldman is investing in this company called Rainai.

Henrik Göthberg 2:20:32

Rainai is right. So here we're talking about Sam Oldman's bet in the chip industry, right? What is the angle of rain?

Anders Arpteg 2:20:40

That's the same, the same. The neuromorphic right yes, but I don't think they would release anything in 2020 or 2020.

Goran Cvetanovski 2:20:47

No, but in general, do you expect more drama like this year? Exponential growth.

Henrik Göthberg 2:20:54

Yeah.

Anders Arpteg 2:20:55

Who knows, but maybe Probably. But just speaking about investments, I think application of course one thing We'll see a lot of new stuff there but I think also simply augmenting existing products and people. You shouldn't forget about that. Then we have the hardware, but I think also one other aspect is actually the developer experience and the AI. That is true actually.

Henrik Göthberg 2:21:20

That's a good one.

Anders Arpteg 2:21:21

That will improve and change the way developers work in really strong ways. I mean, I use generative AI all the time when I program these days, and everyone will in 2024. Or the ones that don't will be far, far behind.

Henrik Göthberg 2:21:36

So the co-piloting era is here to stay.

Anders Arpteg 2:21:38

Yes, and the one that actually can get a working developer experience well in 2024 will be the best investment I think you can make.

Henrik Göthberg 2:21:52

So who are now leading the race on co-piloting for developers, would you argue.

Anders Arpteg 2:21:58

OpenAI and Google.

Henrik Göthberg 2:22:00

What about Git? The whole Git have been working on the dev that's OpenAI, yeah of course yes, it's the same thing.

Goran Cvetanovski 2:22:07

It's.

Anders Arpteg 2:22:07

Microsoft. It's Microsoft, yeah, so once again it's the superscalers, and Meta, of course, also have some stuff there, and Mistral. Actually I have also some code stuff. But you can see this kind of I mean I hope you all see the superscalers in the AI divide. They will provide so many values to different types of users.

Henrik Göthberg 2:22:24

This is a good prediction, right. Follow closely what is happening in the co-piloting space in 2024. Who is sort of charging ahead? Because we are talking about software development 2.0, 3.0. 3.0, yeah, 3.0. We are talking about the paradigm shift for the whole thing, how we do. We can't put that first. Yeah, yeah.

Anders Arpteg 2:22:51

You mentioned the term software 2.0, speaking about you have machine learning. Suddenly that can help you to avoid to not have to write all the rules and instead you simply feed data and use machine learning. But 3.0 is basically you use AI to write the code as well. So for the stuff that you can train a model on, you can use AI to do the coding for you as well. It's one step further.

Henrik Göthberg 2:23:15

This is crazy stuff, because we were sitting here and doing predictions and discussing, as a cool thing 2.0. And now all of a sudden we've just flipped it.

Anders Arpteg 2:23:26

This is, by the way, another thing where we should have regulation for AI the one that actually first delivers or builds which someone has tried builds a system that can actually develop itself and program itself in an autonomous way. This is dangerous stuff. This is where you can get over the hash.

Henrik Göthberg 2:23:42

This should be regulated. But is that a fair summary? It's an interesting prediction or advice. Have a close look at how the co-piloting for coders is evolving and what is happening here. Follow this very, very closely, because this is a major big deal. It really revolutionizes the whole software industry.

Anders Arpteg 2:24:06

All right.

Henrik Göthberg 2:24:06

Should we stop there with these predictions? Do we have anyone else? How do we wrap this up, Gron? What do we have left?

Goran Cvetanovski 2:24:13

Well, all standard way.

Anders Arpteg 2:24:19

Let's talk about philosophy.

Goran Cvetanovski 2:24:22

Now, I think we will wrap up with the after-after there are some nice predictions.

Henrik Göthberg 2:24:26

I think that's a cool prediction.

Goran Cvetanovski 2:24:28

There are some, a few more that we can add, but I think it's going to be an exciting year. In any case, I don't expect to be as big as it was this year.

Henrik Göthberg 2:24:37

I'd say, I'd say it's crazy if it is.

Goran Cvetanovski 2:24:40

This was a fundamental change in terms of what is called dissemination of AI across society, but that, of course, will continue, but I don't think that we will have like.

Henrik Göthberg 2:24:53

But you get fatigue.

Goran Cvetanovski 2:24:55

Yeah, I think it's going to develop on already what happened this year, but of course we will see how it's going to go, but it's going to be exciting. In any case, what?

Goran Cvetanovski 2:25:04

we can do and what we can predict, that is going to be a super exciting season eight and season nine of this podcast. So then the yeah, so we are coming back in January approximately. So we are continuing with our new segment, which has been very successful and with a lot of great feedback regarding that, and we will try actually to get even more Get into the top 10, perhaps of the world's best AI podcast.

Goran Cvetanovski 2:25:38

Well, my KPI is actually not that. My KPI will be actually more bringing even more exciting people to this podcast that we can actually speak and have after work with. That will be my yeah, and what happened?

Anders Arpteg 2:25:50

Having more fun? Yes, exactly.

Goran Cvetanovski 2:25:54

And what happens out of that. That is a different thing. In any case, what we promise that we will do is that we will stand here every Thursday and we will basically demystify.

Anders Arpteg 2:26:06

AI yeah.

Goran Cvetanovski 2:26:07

And have fun with and have fun while we are doing it. Sounds great to me so happy Christmas, happy holidays.

Henrik Göthberg 2:26:16

Happy Christmas, happy holidays.

Goran Cvetanovski 2:26:19

And see you on season eight.

Henrik Göthberg 2:26:21

See you in season eight. Yes, thanks. This has been Matlab of KPI.