E164 - The Fifth Acceleration - Mathias Sundin

In Episode 164 of the AIAW Podcast, we’re joined by Mathias Sundin. He is a former Member of Parliament and Co-Founder of Warp News. He is joining us for a wide-ranging conversation on his upcoming book, The Fifth Acceleration and the societal transformations AI may unleash. We explore why Sweden is falling behind in AI adoption and literacy, the bold idea of free AI access for all citizens, and how public discourse around AI, including political controversies, shapes national momentum. Mathias unpacks how AI could accelerate human potential through what he calls “Practice Levels,” and shares his vision for a future where AGI, if guided ethically, empowers people rather than replacing them. Join us for an honest, forward-looking episode at the intersection of policy, optimism, and technological disruption.

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

Chapters

• 0:00 Who Runs .se and Why It Matters
• 2:30 Sweden’s AI Adoption Lags Neighbors
• 5:20 Waiting for Permission: Cultural Stalling
• 7:40 Meet the Angry Optimist: Matthias Sundin
• 10:10 Media Negativity Versus Fact-Based Optimism
• 15:30 Building Warp News and Its Mission
• 19:30 AI in the Newsroom: WALLY the Writer
• 24:00 Scaling Optimistic Journalism with AI
• 28:00 The Fifth Acceleration: Language as Interface
• 33:20 Four Earlier Accelerations in Human Progress
• 41:30 What Language Models Change at Scale
• 46:00 Sixth Acceleration and Human–Machine Merge
• 50:10 Why Write the Book and for Whom
• 55:40 Writing Process, Interviews, and AI-as-Editor
• 1:05:10 AI News Break: Decoding Animal Language
• 1:08:40 Claude 4.5 and On‑Demand Software
• 1:17:40 Shopping Inside ChatGPT and Agent Economies
• 1:27:30 Sora 2, Cameo, and Safety
• 1:31:20 Inside Sweden’s AI Commission
• 1:38:20 75 Proposals and “AI for All” Reform
• 1:44:30 Missing Pieces: Product and Engineering Focus
• 1:53:40 Value Now vs. Sovereignty Later
• 1:59:20 Education: Individualized Learning with AI
• 2:07:30 Utopia, Dystopia, and Real Constraints
• 2:13:20 Decentralizing Power and Participation
• 2:19:20 Final Takeaways: Use the Tools, Shape the Future

Transcript

Mathias Sundin: 0:00

uh dot se and that's where they get their money from and uh they're sort of the I don't know guardians of the Swedish internet or something like that. I don't know. So they approve and decline uh requests to get domain names under.se I don't know how practically it works but um they're part of the bigger internet I don't know uh structure that you know runs the internet uh in Sweden and but what they've done uh for the international listeners that they started quite early to have a survey and try to measure and understand Swedish internet behavior. I think it's almost 10 years it's more than 10 years and of course uh they released the report the yearly report of on changes in behavior and now of course also within a very much more deeper AI part right more and more AI for the last three years yeah so the Swedish Internet stiftelse or internet foundation or however you say it in English they released this new report this week right what was the content of that well um from the internet side is like same as yesterday last year same procedure as last time yeah exactly so so internet usage has gone from 97 to 98% or something I don't know uh so it's not that part is not that interesting anymore but like we said they've added AI the last three years and and this year many questions on on AI but the main one for me at least is um how many Swedes use AI and and and in this case uh the bar is very low so the question is have you used generative AI or generative AI tools um um uh sometime during the last 12 months so and for example that could be ChatGPT Gemini uh blah blah blah um and and the number is 39% um so which seems to be very low uh internationally uh there's no what I found no really good comparisons um when they ask the ax when they ask this the exact same question the same time period and all that but um if you look at countries that we usually are uh ahead of like Denmark Norway uh Germany France uh we're behind and and way behind and even if even if you take into account that they asked people from uh 16 years old and we asked from 18, you know even if you you you know correct for that we're still behind um so we seem to be behind in in using generative AI yeah um Denmark we said was about I think Denmark was 48 or something like that.

Anders Arpteg: 2:42

Did they motivate it somehow or can we guess on why Sweden is behind?

Mathias Sundin: 2:47

I that's what I've been trying to figure out for the last I don't know two years. And I'm not not sure that I've that I have I have many different answers maybe but one is that we could be um fat and happy so we were we were number one in the world for a while there in the 1990s early 2000s in when it came to using personal computers and and the internet or broadband internet. And digitalization in general exactly and out of that came Skype Spotify and a bunch of these startups that we all know the names of um and so we maybe uh so maybe we thought we are good at these kind of things these kinds of changes we're good at those uh but they don't just happen uh we have to actually do them but it also seems like uh so I I was part of the the government's AI commission and we tried to sort of meet every sort of organization that represented a part of society so I think it was like 150 different organizations and the conclusion only after a couple of months of meeting them was that everyone is waiting for something it's like what are you waiting for? They they weren't sort of against using AI you know or anything but but they they waited for for for a for a go signal from somewhere and and they so when they talked to us they said yeah we're gonna wait for your reports like no no no don't wait for our report just go hopefully we can do things that is that are useful but don't wait for us and and uh so that's why we delivered the report after after almost um half the time um instead of using eight months we used uh nine or ten months like November last year right yeah exactly so we had until this summer uh this last summer uh really but we uh we we did that early just because everyone was waiting for us and and but now they're waiting for the government or I don't know what they're waiting for really but it's rather rather unusual or unheard of that we actually did deliver report ahead of time. Yes that's I think it's the for for Sweden it was the first time in in world history where we sort of called them up and said hey is it okay if we delivered the report early? Yeah that was like you mean later? No earlier what that's never happened before but yeah I guess that should be okay. I love that that's so great. Yeah so I don't know why why and also people seem to be afraid of a lot of things. I I don't think it helped that um Max Tiegmark said well was it two years ago or three years ago um in his summer program for you know two million listeners that you know in in the first 60 seconds he killed the entire humanity uh in 2030 years we're all gonna be extinct because of AI I don't think that helped that's really I'm really angry with that as well yeah um so yeah I mean awesome and um I'd love to welcome you here really glad to have you Matthias Sundin here as a guest in the uh after work podcast and um I I love what you say on the LinkedIn profile the the angry optimist yes it's such an awesome uh description but you're the editor in chief of uh warp news yeah right you're also a famous speaker and uh been around uh uh I guess educating a lot of people being part of the yeah the the AI commission that we had and also politically active right in different ways and being part of the Swedish parliament as well yep are we still there no I quite in 2018 yeah that's uh that many years can you see what happened to the country since yeah yeah exactly all down since then yes but it gives you a perspective of the machinery so to speak exactly yeah and I think that gives you a lot of insight into how things work so I'd love to hear more but perhaps you can start by just giving your version of who is really Matthias. Absolutely yeah I'm really I really am an angry optimist I've I've learned from many different people like Hans Rosling and Johan Uber and Steven Pinker that a lot of very important things in the world are are heading in the right direction and had been for hundreds of years and and for decades. And even now with a lot of problems around the world um things are also in parallel with that a lot of things are improving um education and and lesser poverty and greater wealth and many many many things are getting better all the time but then what makes me angry is that um we know so little about that what Hans Rosling showed that we we we think it's not that we don't know how much better things are we we think they are constantly getting worse most things are getting worse which which is not which is just not true and that makes me angry that we don't know that. But also especially that the news media is constantly pumping out that kind of um false information or maybe the information is not always false it could be correct but they're only giving one side of it. So um it's rather opinionated but why do you think that is why do you think media need has to always you know portray the negative angle on things I think there are uh two main reasons the first is that uh it sells um you can you can that is that is old knowledge in news media that these kind of headlines sell but uh now when they also can measure it um much much more uh much much better and they've seen that uh you know angry headlines sell better or click get more clicks than than others and that drive and if you have a clearer opinion from the newspaper that drives subscribers because you identified with something the maybe maybe the most successful media company in the world now is New York Times and they they have a very clear profile um uh of who they are so so so that kind of sells but also um journalism or journalists are taught that journalism is um uh showing people problems describing problems and scrutinizing um power uh which of course is extremely important in a in a in a democracy um but if you only scrutinize it from from one direction um you get a um you be you become sort of stupid if you if you believe that I listened to just the other week I listened to Anders Boy the former finance minister in in Sweden and I talked first about AI and then he came uh on stage and talked about the world economy. So I said during my talk after me is Anders Boy he's gonna talk about a world economy so that's gonna be much much uh you know uh less optimistic it's gonna be very pessimistic and a lot of you know bad things going on. And then he went on stage and he said no no it's not uh you think there's a lot of problem as in in the world economy that's because you read the newspapers so because I didn't I I don't dive into um economical statistics like he do uh uh so that's why I had a I had my opinion was that you know the world economy is is in trouble and he who who reads all this resource statistics he sees that no no of course there are problems somewhere but you know a much of things are are sort of improving or going in the right direction. So it's very easy to get fooled even if you're aware of this. And of course that makes you a bit stupid and makes you think um the wrong things and and especially take the wrong decisions investment decisions for example if you think the world the economy is going um getting worse you in you'd make decisions based on that.

Henrik Göthberg: 10:33

But it's it it is I think problematic because you can analyze that that the discussion becomes one sided and media is always on the one side that that spins the negative and ultimately it leads to a very skewed view of the world. So and it's the whole truth or not truth it it doesn't even matter in the end. It's the perception that our AI doesn't work or it's the perception that the economy is bad flavors everything. Yeah. And that is problematic I don't know what's our solution to that easy question.

Mathias Sundin: 11:20

In one way but but on the other hand it's easier now than ever before to start your own media whatever the media is um so so it's easier and and there's a lot of good in media not the traditional news media but if you look at YouTube um whatever you're interested in there's a nerdy channel that's super interesting about that. So so there's a lot of information out there that's you know counter to that. But the news media at least what we hope to do over time from Warp News where we write fact-based optimistic news uh we want to show uh news media both that this is a business that you can make money from um and it's also it's also real journalism to report on um optimistic positive things um uh it's not it's not the things that you know someone rescued a cat from a tree or or whatever that that's fine to report on but that's not important news but many really important things happening in AI for example that people need to know about uh not through a negative filter but through you know different versions of of of that and uh and and also there's a lot of organizations out there that uh they started when they started the problem they're trying to solve was a big problem now it's a much much smaller problem but they have no incentives of incentive of saying okay the problem we talked about before it's a tiny problem you should give money to other organizations would handle bigger problems they know incentives to do that so so they keep saying this is this is a really big problem we need to do something about this and the the news media reports that instead they could scrutinize them and say hey yeah it it used to be in the 50s this was a huge problem now is it really you know but they never scrutinize organizations from that point of view and I think that's as important as people claiming something is much better than it is um so and just to circle back to who is uh Matthias like how did you end up uh starting warp news or you know because you were in parliament for some time and stuff like that yeah so how what's the journey that takes us to warp yeah so um in politics I um I tried to cover tech um and tech policy super super broadly um since no one else was interested in that I had to cover everything from space to to AI um and then I met a lot of people entrepreneurs uh scientists and and a bunch of other people that that you know drove this progress in the world um and they talked about their visions and their dreams and what they tried to achieve there. So that made me very very optimistic and the and the knowledge we just talked about um from Steven Pinker and others that made me very optimistic but the other side made me very angry um that shaped you a bit yeah it really did and I felt okay I need to uh so after a while I I I started complaining a lot about people complaining and then I realized hmm am I really making a difference here? So I said okay I need to quit politics um and and try to do something about it. And who should I who should I try to do this with? Well it's the people I met the people who are sort of the optimists the the people who are moving things forward. If I can gather some of them in a community and do things together with them we could probably have an impact um on the world. So so I started Warp Institute that's a foundation that is the community for these people. And then we very soon realized okay we need to share news in this community of things going on. So we started that and and but then we also realized okay this could also hopefully we can make this into a um a business model that works that we can show other media that okay there's actually money to make here so so that's why we started Warp News then uh five five years ago uh almost almost to the day uh five years ago so I'm the editor of chief of of Warp News now and and chairman of this and you send out newsletters once a week or how often yeah exactly we have a free newsletter that comes out uh once a week uh with and and we write on on um uh technology then science and and human progress these and fact-based optimistic news about about that and I love uh actually having proper objective journalism here happening also bring on the positive side so yeah kudos for doing that thank you but it becomes even an angle where I I can even say like if I'm gonna pay money I'm so fed up with all the negative news I'd rather subscribe to to the angry optimist yeah I think it's a good uh angle yeah we we have what we call uh the newsletter is free but we also have what we call premium supporters and they're they are premium but they're mostly supporters because they think is this is important so they want other people to read this um so so we have uh some of them uh Steven Pinker is one of them for example a premium supporter um so yeah it was pretty cool when he signed up I was like Steven Pinker sign up my newsletter which makes me very nervous every time I send something out because I know Stephen Pinker's gonna read it.

Henrik Göthberg: 16:26

What's the commercial model here or how do you fund it or how do you have time to do it?

Mathias Sundin: 16:29

Yeah so it's practically yeah um practically it's been really hard um I think running a news media company is hard uh even if you're negative but if you if you if you're if you're possible extra is extra hard so so yeah it it it is been it has been hard um but um uh so the the business model is is really the premium supporter membership um and and but also I'm uh out there giving talks now um and that of course drives in new subscribers but also um a revenue stream uh a revenue stream um so but I I think um ai will help us scale um uh not not right now uh now it has made things easier for example we have um I've created um simple um ai bot called Wally uh who writes the news um so and and there the news of the so it's not it's a play on the what name Wally exactly exactly and a play on that and on Walt Disneyws sort of yeah um so uh and and she it's a it's a she um she um she rewrites news so it's from press releases or other news articles um which of course ai has been good at for you know for a very long time two and a half years or something like that um and so before it took you know um a simple news uh article like that took maybe two hours for me to write um to write and translate to English and you know find an image and publish it and all that two hours you know in general and when I started using Wally uh it um dropped to maybe 30 minutes so still some pretty many you know almost always I had to edit something you know and it wasn't you know really good but but now it's down to 10-15 minutes and I it's I barely the the editing I do is maybe I feel like okay this segment is you know it's not bad but it's you know it's too long or whatever. I remove a little bit things like that. And but mostly my thing is now copy pasting stuff you know that that's what I do. So in a way finding what to feed. Yeah exactly I find the news um and and she writes the news um I edit it a little bit and copy paste and the editorial editorial editorial in chief that sort of picks the flavor, find the news and and masses down to the theme or the you know and then using AI in order to craft it into exactly uh a framed but it feels like it's pretty close to um setting up a feed that the AI can choose from um and the AI does all the all the all the copy paste stuff also. But also then the next step is that it could translate it to a hundred different languages and also far away. No no exactly it it it could be done today but then to a certain cost. So we we can't afford that you know uh today not not the money or the time but give it I don't know two or three years probably you can translate it to a hundred different languages and publish it but also use that text to create different kind of videos uh reels and longer videos and podcasts and news podcasts and and all that in a hundred different languages so for for a very little cost uh and with with uh the quality can't um be lower then no one will read it or you know look at it.

Anders Arpteg: 20:04

And if you use some open source model it could be more or less for free as well.

Henrik Göthberg: 20:07

Yeah exactly so it's it feels like and since we don't have since there's unfortunately no competition for our news we we can just scaling all the news in you know all the world could be a a pretty big thing um for us in a couple of years there is no competition you think you think this the angle optimist angle is actually not so taken is that what you're saying yeah there's very few of these news or they are out there but they're there um it's also it's almost always in a small local community or it's in a science newspaper somewhere who who a few thousand people read um but we try to make it you know available to more or understandable to more more people so there's almost there's a few newsletters out there but there's no real competition you know but am I getting it right then I mean like you said it really well before uh how to succeed in the new in the media landscape you know one recipe is what New York Times are doing they they they are creating a very strong identity and profile and they are sticking to that even if it's negative or call it vocal call it whatever you want and you can essentially do the same thing but from your core mission or your core values.

Mathias Sundin: 21:20

Exactly yeah yeah exactly and that's really what I'm trying to build now um now a lot around me um and sort of the center of this but I over time I I would like Warp News to have sort of its own voice like the economist has a voice. It doesn't matter who writes the article if it's a human or an AI or whatever the economist has a voice. I would like Warp News to have a voice like that but also and not just write the news but also help people understand the news and also an AI could be really good at that based on the material we've written everyone else has written okay if we write about um electric vehicles and and battery costs um it it's one thing to write um something but if you oh here's a new car that costs less than than the previous car okay but what has happened over the last 15 years well the battery cost has dropped over 90% okay why because of right's law because we're making new more new batteries and and blah blah blah so help people understand things like that and and sometimes we we understand what people want to know and and write that other times of course they could use sort of our AI and WALL-E or whatever to help them is water available for everyone or is it just you using it internally? Not available to write news uh but I I created a version that people can chat with a GPT um that people can chat with um so I haven't updated that in a long time so I don't know if anyone uses it.

Henrik Göthberg: 22:49

Who is your target audience for warp news?

Mathias Sundin: 22:54

It's everyone who's sort of um curious about the future um and want to be part of creating that um future um so yeah cool um thank you for doing this work and actually having some positive or more objective I would say news being sent out there that's awesome and I hope a lot of other news media will copy and be inspired by this me too. But you do more things you also are writing a book yeah right the fifth acceleration is that yeah can you please uh let us know more what why did you choose to start writing this book and what is it about yes so it's a book about uh AI um and my my thesis in it is that uh for humanity has made progress for like three million years or something like that um before Homo sapiens um and and we've always made progress but uh a few times uh over over time um uh we have accelerated um suddenly progress has been faster uh the first time was probably then like three million years ago when we had our first idea um which is fascinating to think about that we there used to be a time there were no ideas there were there were instincts but no one had a no one had a thought an idea a conscious exactly um more instinct based exactly um so based on that first idea we started slowly slowly slowly accelerating but that was the first forward movement really for for for humans um and then uh a few times um uh we've accelerated the last acceleration or the fourth that we are uh um you know at the end of now what I think was uh several things that happened uh combined you had uh you had uh what we call the enlightenment so we had we had science um the Royal Society in England and we started to have real science to to to figure things out uh but uh even more important we we started to having uh democracy that we could think freely um for the first time and and sort of ideas got rights and we got rights to think so we we uh we had many many many more many more ideas and we could critique them and we had science to help us polish those um ideas how many years are this back approximately this is like three four hundred years so in the late 1600s and early 1700s um and out of that came uh the steam engine uh which of course started the industrial uh revolution so these three things combined science democracy and and industry uh uh really accelerated uh humanity way more than than ever before um suddenly you know uh uh wealth and GDP and and uh living standards and everything you know suddenly went through the roof has been almost flat for millions or or hundreds of thousands of years so uh but I think now we are uh at the end of that and at the start of the the fifth acceleration um and that is because of um ai but especially that I see as the the core thing um that we have achieved is to get uh machines to understand our language uh our human language and that's uh that is a before we go into the fifth too much but you mentioned the fourth um you know three four hundred years ago and we started to have more of a scientific kind of uh discovery but if you were to quickly go through the previous three ones which are they so the first one was the first idea um uh and the second one was like 90,000 years ago something like that um and that was the first time we started to having uh we're accumulate ideas add or add remember some knowledge and add new knowledge on top of that because for a while there in Africa we lived um close enough together uh so ideas could actually spread you know more than every few hundred thousand years um so we didn't have to restart from scratch every time a human was born so to speak exactly so for the first time we got some sort of collective intelligence it was really bad you know and and very very very fragile but for the first time uh we started to remember a little bit more than we forgot um and and that was the uh a a big difference so instead of a few hundred thousand years but between every sort of step forward it was just a few thousand years um so still very slow um and that eventually led us to um like um 10,000 years ago that we had enough knowledge to to stop moving around and finding food we could settle down and and grow the food and and put the cattle um in in a barn or you know keep them close to us um and we so we became farmers and of course then the same thing happened that happened uh 90 000 years ago that when you're when you're when you're in one place and you have neighbors um knowledge you preserve knowledge even better um and also uh of course then uh a very important thing that um happened 90,000 years ago or so was that we could sort of first created the human language that we actually could start speaking which of course is crucial to be able to transfer any knowledge to anyone else uh but then um when we settled down as farmers uh not just um uh not just um that we could speak to each other we could actually start writing things down which of course also helps enormously if you should want to remember something yeah um so we could specialize a little bit um so not everyone had to do the exact same thing most people did the same thing as farmers but a few people could do other things and specialize in becoming a little bit better at making shoes or wheels or whatever um and and we we out of that we could create cities and civilizations and and all of that but still progress was very very slow and much much faster than before and that led us to the point uh three four hundred years ago in England uh where the the place where thoughts um were freer than anyone else every anyone else in in the world um and and at in that point uh we created these things science and and industry and and and democracy then people like Newton and Copernicus and whatnot started to actually question some of the fundamental truth that you otherwise believed in right exactly but the really the big thing with um with what happened in England was that you didn't have a lone genius sitting somewhere and thinking things and reading a few books um they were very rare. So the Royal Society um they um They said science should be done in open. You should show other people what you've done so they can learn from it and also try doing it themselves. And that was the that was the big, big breakthrough. And we had geniuses like Newton there. But he was part of a group, and that that made the whole difference.

Henrik Göthberg: 30:22

Do we even talk about this as in some ways the birth of the more modern scientific method where we start critiquing each other and stuff like that? So we've done it before in different ways, but here now you started to get a fundamental open structure. What we do today in academia, etc.

Mathias Sundin: 30:40

Exactly. So that was the first time we really started doing that in an organized way. So of course, every you can throw whatever idea. On another meeting, they had a guy who said, you know, if you if you pour unicorn um powder around a spider, the spider can't run away. People are like, Yeah, okay, show us. So he poured some powder. I don't know where they got the unicorn from, but he poured some powder and he put a spider there and it ran away immediately. It's like, no, that didn't work. So it was really all over the place.

Henrik Göthberg: 31:27

The scientific method, highs and lows.

Anders Arpteg: 31:29

Yes, exactly. But it's also interesting, you know, with this kind of ad hoc reasoning that people try to achieve all the time. I believe it was Copernicus who first built these kind of binoculars and you can, or telescopes and you can watch the moon. And suddenly you saw it. There was a common belief that the moon was perfectly spherical, right? And then he said, no, it has valleys and some kind of you know whatever it's craters and stuff. And they said, No, no way. So then they came up with this argument which is you know not possible to falsify, which is that there is an invisible uh layer of the moon. So it's still spherical, you just can't see it. It's kind of funny how you like crater. Earth is flat, or and or we and we will kill you. Earth is flat is another interesting, you know. Actually, that kind of you know, movement is growing in the world.

Henrik Göthberg: 32:20

The flat earth is back on track, yeah.

Anders Arpteg: 32:22

Yeah. And it's it's kind of sad. If we just speak about that, you know, the science in general and the evolution here, and especially when it comes to AI science, the the traditional kind of conference you had in journals you usually published in, it seems like less and less are using them in some sense. You just put push something on archive these days, and then you put a blog article on it, and that's it, more or less.

Henrik Göthberg: 32:45

And keep running.

Anders Arpteg: 32:47

Is are you are you thinking that science could actually be on a downturn in some way?

Mathias Sundin: 32:52

I don't know. Uh I don't think I know enough about that, but it's it seems like from whatever I hear that it's sort of uh over um over bureaucratized that you know um it's caught in different models how you get um funding uh for for things. So one one person who I've interviewed in the book, uh which is also uh he's also a premium supporter, uh David Deutsch, um uh he's a professor at at Oxford um and sort of the father of quantum computing. Um and his his key paper um was uh physics with a lot of philosophy in it. Because to under to understand the quantum world and quantum computers, you couldn't just use physics, you also have to think from a sort of philosophical uh perspective. And and he liked that. He he's a he's a big fan of Karl Popper, and you know, so he used he used both those um sort of uh physics and philosophy, and he put that in a paper, and that was the um foundation for quantum compute computing. Um and he says there's no way that paper would have got accepted today. It almost didn't get accepted then. Um it was it was really, really close. Um he sent in a similar paper a few years earlier that was rejected, uh, and then he sent in this. There was a sketch, but it was uh just because of one guy on the committee. Um so he said there's no way that's gonna be you know good uh because it's not a physics paper only. Um so um and and get funding for that kind of thing. And you have to, I don't know, follow the trends and all that. But that's sort of an outside.

Anders Arpteg: 34:36

I'm biting my tongue here. I have a lot of thoughts about quantum. Yeah, but let's get back to let's get back to the book.

Henrik Göthberg: 34:40

Yeah. Can we dissect the fifth and now we now we understood the the journey up to the fifth acceleration and you started? What is the what are the components that makes up the fifth acceleration?

Mathias Sundin: 34:52

Yeah, so if you think how important human language has been for humans then, and I I asked Stephen Pinker about that in the book. So what would be what would we be, what would humanity be without human language, the kind of uh uh the kind of language we have. He said, Yeah, we will be basically be shimps. Uh that's what we would be. So, and now we can use that language to talk with the computers, and they can talk with us. Um, and and that makes suddenly uh AI then or these AI tools available for for everyone, at least everyone with an internet connection, it's like six billion uh people, uh, which is a big a big thing in itself. But what I think happens then when you use these tools is that you unleash a lot of uh creativity and and potential that is sort of stuck here, or that you have to use I I I I suck at mu uh music. I can't play an instrument, I I can't sing uh anything, but suddenly I can make music. Um and and now pretty good music uh uh with Sunu 5.0. And and and and that hasn't changed the world yet, but it unleashes a little bit of creativity in in my head, and I can suddenly now have music. When I give talks, I can create songs about the the company I give a talk for, and people laugh and think it's fun, and and you know, so and that's a tiny thing, but when these small things and sometimes bigger things happen in billions of heads, and you you you get you unleash this creativity, uh that's gonna unleash a massive wave of creativity and potential. Uh, you're gonna come up with new things, and you and me, and especially when these new things meet and converge and and create something entirely new, that's when the real um change happens.

Henrik Göthberg: 36:42

So, assuming out here, instead of going down the nitty-gritty, we look at the super macro, the super trends, so to speak, or the super waves, and and then the argument is then that uh we are we are releasing the next wave of creativity because if you think about it, all those waves are waves of creativity in different ways. Yeah. Learning or collective learning and collective learning leading to collective creativity, and now we have taken this like we we'll we are looking at an inflection point where we can now do this with you know machine-empowered exactly.

Mathias Sundin: 37:16

So if you look at the previous accelerations, uh all of them were really that we uh uh uh we spread ideas more and better. Uh for every acceleration, we we we came to a point where suddenly we can spread more ideas, learn more things, have a stronger collective intelligence. Uh and now I think when uh the breakthrough with under when machines are really understanding our language, so we can express our thoughts, not through a programming language or some other type of language.

Henrik Göthberg: 37:44

This is limiting how many can talk with machines.

Mathias Sundin: 37:47

Exactly. How you talk and how many who can talk with them. Um suddenly I can do that. I can code or program anything, suddenly I can do that. Um, or get the machine to do things that used to um uh need programming. Um, so all of that, so so the same thing that happened uh three, four hundred years ago uh is happening now. We're getting way more ideas, way better ideas, more we can do more things with the ideas, it's much easier. And that's I think with with um uh with Lovable, who went from um zero to a hundred million uh dollars in annual revenue in in eight months. What they did was that uh uh use just uh the interface was language. Uh you put the the interface language, and behind that you put um some tool that can create or build apps. Um you could do that with ChatGPT, or you could before that you could hire a programmer or learn to program and create apps. So it shouldn't be that big of a difference, but it is. Just having language as the interface for the user makes an enormous difference uh because it's so much easier to unleash your creativity. And when this happens for billions of people, that would accelerate all of humanity.

Henrik Göthberg: 39:05

And now you're pushing me to become the angry optimist because sometimes aren't we stuck in nitty-gritty fucking bullshit conversations and we are missing the big picture. We're missing the picture totally that you know regardless of how fast we get to AGI, regardless of what bumps in the road now, it's it's a fundamental pivotal moment potentially, and it has nothing to do with exactly when the techniques, this and that, but maybe about this. Is that is that a core essence?

Mathias Sundin: 39:34

Exactly. So then you have of course, uh, one can debate uh the limits of of LLMs and and uh how much data is gonna be needed, how much data is available, and all these things. Are we have we run into a wall or not? Of course, we can discuss that, but that's not the core thing. The core thing is the breakthrough here. So even if LMMs runs into a wall, we will, based on this discovery or this breakthrough of understanding with language, we will move move past that. It it might take a few years or whatever. That doesn't matter.

Henrik Göthberg: 40:11

Sorry, my language. When people can't see through that, but get stuck in the nitty-gritty and only see problems.

Mathias Sundin: 40:18

Exactly. And especially especially when they do that, it's like, yeah, this is the problem. Yes, it can't be solved. It's it's it's a wall here. It's like there's no fucking wall here. And if it if there is a wall here, we will go right through it, or we will go past it or under it. That's what we do as humans. It might take a while, but you know, constantly thinking that this is unsolvable. No, it's not. It's not. And especially when um when you have a breakthrough in the background that is so big, you know, there's suddenly way more people who can suddenly, or not suddenly, but over time can actually also start developing the models themselves. It's not just the people who have done that in the past. We're adding tens of thousands of new people who can think of the this in different ways than the people who created this in the first place. So they're gonna come up with new ways to do this. So um I I I think it's important to zoom out. I get your point.

Anders Arpteg: 41:14

But if if we let me just add this question, but the fifth acceleration, then I guess in some way it's about accelerating humans, yeah. Still could it be a point where it's not about humans anymore and we still have a development of the society that is beyond humans, or is that something else than the fifth acceleration?

Mathias Sundin: 41:35

I think that is the sixth, maybe, but of course we're uh in a way uh transitioning over to that. Uh but I think uh but I think we're gonna be more and more uh intertwined with technology. Um if you think about uh the phone, it's always with us. Um it's not you know uh it's not stuck to our bodies, but almost. Um so so technology is very, very close to us all the time. Um and I think when we move forward, we if we get to AGI and even better AI, um, we're gonna connect with that in different ways, our brains or whatever way we figure out. So we're gonna be really connected. But also it's sort of that we've um for the first time we've also sort of uh created an intelligence that is outside of our outside of our brains. Um so and that's that kind of intelligence you can um you can all you can spray it. It's sort of like a spray. We can spray it on ourselves to become more intelligent and do more things, but we can also spray it on dead things, and suddenly they turn smart or at least a little bit smart, or alive in in some sense, uh not alive in in the sense we are alive, but you know, from completely dead uh to having some sort of smartness or intelligence and and can talk to each other also, um, of course.

Anders Arpteg: 43:07

And if I may get a bit philosophical here, I think it was uh Mark Andreason that said something like this. But I think you mentioned in the beginning that of course humans before in in first or second acceleration don't like start from the scratch all the time, they can actually learn and continue to build up this kind of community and that way have some collective society or knowledge that they have. Um but I think what he said was more or less that of course we as humans now do not have to start from scratch when we um when we are born, but AI have a very different approach. They can actually literally copy themselves infinitely and never start from scratch. Like humans still have to, yeah, right?

Mathias Sundin: 43:53

When we're born. Yes.

Anders Arpteg: 43:55

So if a new AI is born, you can simply deliterally copy it very, very easily and uh create you know many copies of it and it never has to start from scratch ever again. Would would that be the 60th acceleration then perhaps? Or what do you think?

Mathias Sundin: 44:11

Yes, something like that, or uh when at least when we are uh I I think whenever that happens, we are gonna be connected to that um in in some some sense. It's not gonna be humans here and machines here. We're gonna be uh connected. We we are already not just through smartphones and internet, we are sort of connected even if we're not physically connected. Um but we are. But then uh also maybe physically connected to you know, we'll figure out uh to run the computation that runs in our head, we can probably figure out how to run that on uh on other hardware if if that's how it works. But maybe it works that way, so we could uh run our our software on on other hardware or run other software in our heads, you know, merge in that way with with technology, maybe even without drilling holes in the head or whatever. So so I think we're not gonna um some some people might you know wanna go uh a different uh path, but um I don't think so. It's it it feels dramatic when you think about it. Oh shit, we're gonna really merge with AI or merge with machines, but when we already have it, some yeah, exactly. We already have, and it's it's not dramatic. Um and it's not gonna be you know more, it's dramatic when you think about it, but not when it happens.

Anders Arpteg: 45:37

Isn't Elon Musk you know I have to throw in some quotes from him, you know, every time. But uh anyway, I think he speaks about you know the three layers here. One being that humans have the I think he calls it a limbic system, you know, the instincts that we had from the start as human sapiens, but then we got more into the cerebral cortex that we have, in the at least we as humans, perhaps not all animals have it, but we as humans have rather strong uh cortex which allows us to consciously reason about things. But in in practice today, we already have this tertiary layer which is all the laptops and the mobile phones so that we can quickly get the data from Wikipedia or from current LLMs that we do have. So in reality, we have three layers already and are cyborg more or less.

Henrik Göthberg: 46:20

Yeah. In some form, people forget that we are cyborg as long as we carry our phone with us in the simplest form, maybe. Yeah. And what we are now looking at is you know, how can we increase the input output speed and stuff like this?

Mathias Sundin: 46:35

Exactly. And and Elon often talks about you know the limiting factors of the thumbs that we write with.

Henrik Göthberg: 46:43

Exactly.

Mathias Sundin: 46:44

So and and and that's that's true, but with using our uh and now we can actually speak to the computers, um that has a high bit rate. Exactly, much higher, and and and we can't do that all the time. Um but uh but it but it adds it it at least makes it faster. But of course, if we could connect it, it would be never as fast as computer can communicate, yeah. You can communicate so much better. And that's why I think that's the sixth acceleration.

Henrik Göthberg: 47:11

When you get the full-on consciousness, computer to computer, then you get to the next level of speed on this uh on this acceleration of creativity. Yeah, that's an interesting definition, actually. So when it's a pure computer-to-computer consciousness and creativity unleashed, that maybe is a way of looking at the sixth. But I need to ask, why did you have to write this book? What compelled you because it's it's an undertaking, right? Uh you know, what what was your mental process and what what what what did you want to achieve? What's your goal with the book? All you know, all this.

Mathias Sundin: 47:44

Yeah, for first of all, I'm I'm uh I'm sort of that's writing is my that's my skill. Uh that's what I what I do best, and that's what I enjoy most. So I enjoy writing things. Uh, but writing a book um is what what's so uh there's many things that are really good with writing a book. First, you have to research a lot, so you learn a lot. Then you have then you have to, or you you interview a lot of people, so you learn from that. Um and uh but then you have to write down your thoughts, and uh they develop your thoughts develop a lot when you have to write them down, and especially when you have to write uh them down, not just a few paragraphs, but several pages of something. And I've discovered uh at least a couple of times ideas when I write them down and I try to expand, it was like, no, this is you know, I I you know I it sounded good in my head, but it's I can't get this to work. You know, it's it's you know, it's not a good idea here. It's it's this is a bad explanation. Um so okay, so scratch that. Uh but uh you have to organize your thoughts and uh and also think, really try to think how how does other people see this? The things that I understand that other people don't understand or see from a different perspective. That's very, very, very useful.

Henrik Göthberg: 49:06

Um But is it is are you describing now even uh because I can even relate to that that okay, I I almost want to write a book and I don't give a shit if anyone else reads it because I this is my thought process, or I'm exploding with my thoughts and getting it down on paper and organizing it is a one is a way of controlling or sorting it. Yeah, so this is one angle you're talking about, but but what but but what's the external angle? What's the mission with the book? Or did you have a mission, or is it or is it more fuck I need to write this book to get this on paper?

Mathias Sundin: 49:40

No, I I I I want people to read it. Yeah, so I have to. Yeah, because I think my my my main mission with the book is that I want the people to to read it because since I think we're heading into a new era for humanity, then we're gonna need a lot of pioneers in that in that era. And I want to help people understand that this is a big change. So you should you should prepare that a new era is is coming. But more important, that you could really be part of that era. You you don't have to be just a traveler, you can really be a leader in this. And it's not just, and many people think, oh, it's AI. So I I can do anything with AI. I I don't know how to code, I don't know anything about AI. Um so no, if if if that's the case, you should not go into trying to develop new AI models. But that's not the only thing going on. That's just um James Watt and the Steam engine. Um, that wasn't the industrial revolution, that was the start of it. But uh James Watt wasn't the one who had uh made the most money, for example. It was the people who used the steam engine. And since language is the user interface now, uh every human can use these AL tools to do things. So you could be a pioneer in that way. You can use you can be the lovable of a thousand different things and many, many other things. But also, I think there's a sort of a third category of pioneers and new leaders that are needed. Because if we look at the Enlightenment and the Industrial Revolution, um uh that's that really changed society, how society works. Um the steam engine, when we put that into um factories, um, people moved from um uh from farms to working in in factories. Uh, that was better than the farms, or at least more reliable. But uh after a while they realized, okay, there's some really shitty terms here. You know, it's really dangerous and low pay and everything. Okay, we need to organize. We can organize. And since we have uh freedom of speech, we can we can complain about things and we can go to we can form unions and you know, and we're stronger together. And based on that, you need new ideologies. You have liberalism, um, but you have communism and socialism and conservatism and all that. And that was created in this era uh with people who didn't know anything about steam engines uh or or science or whatever. So we're gonna be needing these kind of people also, uh, not just uh ideologues or or you know, but new kind of leaders in this uh society. So I tried to show people in the book that uh this is a big broad change, and your opportunity to be part of and be pioneered is bigger than ever before. Um, so if you want to, you really should.

Henrik Göthberg: 52:26

I really love this analogy because not everybody needed to be able to build a steam engine, but but what happened in that era created a lot of different debate and different discussions and formed new jobs or new ideologies. And you can do exactly the same parallel here. Yeah, not everybody needs to be AI engineers or data scientists, but you have a you can contribute in terms of building the new society for the uh that is relevant for the fifth acceleration.

Mathias Sundin: 52:54

Yeah, exactly. Um so that's why I've interviewed a lot of different people for the book, but also trying to you know yeah, help the reader um figure out it's their own role in this. I I can think of their role, so they have to do that. And also uh I tell people that you you are truly unique in this because only you, even if we're similar to other people, um you it's only you that have your experiences and your knowledge, and that's that's only you. And if you um if you what I think is the key component here is to mixing your human intelligence with the artificial intelligence, um if you if you mix that um using you as a human uh with the AI, you get something unique out of that, and that's for everyone. Um if you're good at that, uh something unique will come out of that, whatever that is. And and that could only be only be you. You can't do under, you know. Uh you can you can try, but but it only he has his experiences.

Anders Arpteg: 53:59

Could you do you have any thoughts? I know I do, but but do you have any thoughts about you know what the differences are between AI and uh humans in terms of what AI is good at and what potentially humans are good at and AI is not good at, so to speak. Do you have any thoughts about that? Um otherwise I can give you an intro on my thinking. Yeah, go ahead first. The way I like to phrase it is to use the open AI's um pyramid or path to AGI, and they basically divide it into five levels. And the first level is basically conversational, and then you have reasoning, and then autonomous agent, and then you have innovative, and then finally organizational capabilities. And then the claim is, at least for me, that you know AI is uh today better than humans on the first layer, meaning just working with a large amount of data, taking that together, summarizing it, even writing it, right? So from a like a language almost point of view, they are actually far superior to us. Then when it comes to reasoning, uh of course we are trying to add reasoning to AI, but I would say humans still are significantly better and doing more in-depth reasoning. And I think even more so when it comes to actually taking actions. We can help an AI to take actions by writing MCP servers and doing the integration for it, but you as humans don't need that, and but we still need to help uh the AI a lot if it should take actions. So it it is still significantly inferior to humans that, and even more so for level four and five. But still, you know, AI is improving all the time and slowly you know becoming higher, higher up in this pyramid, and humans then should move higher up in this as well. But I think few people actually speak about you know what AI is bad at. And um, I think you can't simply say AI is better than humans or vice versa, you have to differentiate it a bit more nuanced. Right?

Mathias Sundin: 55:52

I I haven't thought about it that way, but I've I've thought about um uh since since uh there's a lot of claims that AI will replace humans in different ways from jobs to everything eventually and and then kill us off. Um but so one thing that uh that AI um is bad at is being human. Um I'm interested in you because you are human. Even if you were uh a Westworld AI uh that looked exactly like you and talked like you and everything, uh as soon as I know that you're not human, uh that's a difference for me because I'm interested in you because you're a human. That doesn't mean I couldn't couldn't enjoy AI music or whatever, or a lot of AI stuff, or have a some sort of relationship with a robot, or you know. Absolutely. But I think I think we're always going to be interested in in humans because they're humans.

Anders Arpteg: 56:53

It's like playing playing chess. We know you know computers are much better in playing chess, but to still watch human humans play chess.

Mathias Sundin: 56:59

Exactly.

Anders Arpteg: 56:59

Right.

Mathias Sundin: 57:00

Exactly. That that's one of the examples in in the book uh that uh yeah, we know that the you know the the chess uh program I have in my phone beats the the best uh human players that uh you know out there. Uh but still I want to I want to play chess and I I don't want to see or might want to see uh two AIs play chess, but I I'm much more interested in humans playing chess. Um so um so even though hum uh even though computers have completely outcompeted us in chess in every way, um chess is more popular than ever before. And the chess players are better than ever before because of course they can compete against much better opponents that can teach them how to play. So so that that's why I don't think uh you know AI or machines are never going to replace us. We're always gonna be interested in that.

Henrik Göthberg: 57:51

It's a very interesting topic. Chess is probably more popular than ever, and it should die out because what's the purpose of playing chess? Exactly.

Mathias Sundin: 57:59

And what is the purpose of what's causes sucks at chess compared to all the computers?

Henrik Göthberg: 58:03

But and just look at that on YouTube or look at how much we even see chess on television now that you didn't see in normal television 10-15 years ago. Yeah. But but I think that puts a finger on what you said. Yeah. It's not good at being human, and we are interested in humans. So and we that won't go away.

Mathias Sundin: 58:23

That won't go away. We've had a number of robots on Mars. That's great. It's it's nothing even compared to the first human setting foot on Mars. Yeah. So uh and that's I think gonna remain over time. That doesn't mean that AI will replace, of course, both jobs and tasks and and many things, and and sometimes that will hurt people, of course. Um, but overall, I think uh no matter how good the AI is, first we're gonna be connected to that, so we are also gonna level up. But uh even even just the the dumb meat uh here, we're gonna be interested in that because that's us, I think.

Henrik Göthberg: 59:01

Before we move away from the book, I have a couple of nerdy questions, and it's uh it's almost uh for my own interest, uh, but I think it's interesting anyway. I mean, like, what was your thought process on structuring the book? Or how you know so it's a combination of like um selling the book on what can I read in the book, but I'm more interested in your thought process or headaches to string it together. How were you thinking? What was your process from the potential of me or someone listening? Uh, how to write the book or how to structure it and used learning from your experience?

Mathias Sundin: 59:38

Yeah, it's uh I sort of had the the basic uh structure or the basic idea uh of the fifth acceleration. Um, or or I didn't call it the title came after a while when I wrote about this uh these accelerations and uh but it it It sort of had the basic theory there. I had a basic theory and I had that for many years that it's really ideas and sort of freedom of ideas that drive drive progress. And now I see that in AI. So I had a basic structure, and I also I always love when what what I love about technology is that it decentralizes power. So that's sort of my my driving um But that's a good that's a that's a discussion on its own.

Henrik Göthberg: 1:00:29

That's a deeper AI-divide discussion. Yeah, but let's continue. Please continue. II style.

Mathias Sundin: 1:00:35

So and and I saw okay, this is this is the technology that gives people more power than than any other uh technology uh out there. So I wanted to show people okay, a big change is coming. How can you be part of of that? But that was sort of the idea. Then to get all of this together, and especially when you interview people, you run into the problem that they say very interesting things, but after a while you realize it doesn't fit into my book. It's very, very interesting. But if I write it here, people like, why the hell is this here?

Henrik Göthberg: 1:01:07

And was the was the format of having an interview based book or weaving interviews that was from the get-go, how you wanted to write it? Yes, because that's because that's the narrative strategy.

Mathias Sundin: 1:01:17

Yeah, because I I and I also know a lot of very interesting people that have interesting thoughts on from different perspectives. So I really wanted to, but then I didn't want to write an interview book that here's an interview with Steven Pinker, here's an interview with Buddy Econ, here's a because that's very boring. Uh so I wanted to weave that into a narrative of, you know, so so that was the tricky part. And at least my process, my writing, some some writers are like they write a synopsis and they follow that, boom, boom, boom, boom, boom, boom, boom, and they just write it out. Um, I I can't do that. I had an idea of the book, and then really when I had written the first time, that's where I saw, okay, this doesn't fit here, this shouldn't be moved here, this doesn't go into the book at all. And I had, you know, there's this for writers, there's a thing called kill your darlings, uh, really. It's like, and there was uh scene in the book that I had when I actually sent it to the publisher the first time, there was a key scene in the book that I had to cut. And I love that scene. And it's like I want to write a book just about that, but you know, it doesn't fit into the book, so I had to remove it. Um, so it's like and it and it and it hurts. It's just there. I I love it, and no one has read it.

Henrik Göthberg: 1:02:37

So this was an iterative process. And it was uh how many versions until you were sort of happy?

Mathias Sundin: 1:02:43

Um many. Um of the first chapters, I have like I think they're up to 15, the the 15th version, but but a few of them are like, I don't know, three or four versions, and then there are smaller changes uh to them. But I also had uh, of course, great help from from AI. I really tried to do what I preach uh here. So not have AI write the book for me, that would be not very interesting for me, at least. Uh, but really help me write a better book. Um, really try to mix my intelligence with it. So uh I created um I I have a human editor who's really good. Uh but uh he's a human, so he's not with me all the time. And I I can't tell him to, can you give me 50 ideas for a title? It's like, fuck you, there's four. So but an AI doesn't complain like that. So but the problem with uh the problem with large language models when you want feedback is that they're so polite. So everything I you know fed it, it was like, yeah, this is really good. And first I was like, yeah, that's true. But then I realized, okay, that can't be true for everything here. So I had to tell it, okay, be more critical. You should be more critical. And it became a little bit more critical. But as soon as I said something back, it was immediately like, no, no, no, yeah, yeah, yeah, that's good, that's good, that's perfect. Yeah, you're right, you're absolutely right. Um, and I was like, this is not helping. So eventually the the sort of prompt that made it really useful was that I said, you should be the Steve Jobs of book editors. Super rude. Don't think about my feelings. Your your task here is to sort of squeeze the best book out of me. Uh and then it became really, really fucking rude.

Henrik Göthberg: 1:04:35

If you give it a personality, it works better.

Mathias Sundin: 1:04:39

So we've constantly, since then we've constantly, I'm constantly arguing with it. And and it's so bizarre because it's a machine and it it is a machine that I've told to be rude, and I constantly get pissed at by that. It's really it's really weird. Um so but it has been very helpful.

Anders Arpteg: 1:04:59

Um yes, cool.

AINewsJingle: 1:05:07

It's time for AI News, brought to you by AI8W Podcast.

Anders Arpteg: 1:05:14

Yes, so we usually do this kind of break in the middle of the podcast to just reflect a bit on the latest AI news. And we try to keep it short, but we usually fail. I speak too long about the news. And when having uh an expert here in the news field, I'm sure it will prolong for a long time. But um let's start. Do you have uh Matthias some news that you'd like to discuss?

Mathias Sundin: 1:05:38

Uh yes. Um let's see if I can find it. Uh it's it uh oh, I turned off internet smartphone. Otherwise, we could start and you can yeah, I I think I sort of remember it. Um it's on um I wrote about it, or Wally wrote about it recently on Warp News. Um it's it's how we I've written several news about this in the past, but it's sort of how we're using AI, but AI and algorithm computers to understand animal language or animals languages. Um and with there are many different uh science projects out there. Uh I think they come the furthest with with whales that they've seen. They have they have they seem to have some structures in their language that reminds reminds of of human language or part of human language. So but what I think is so fascinating is that we're sort of approaching a point where we actually gonna understand uh animals, what they what they say, what they mean, uh and and be able to communicate with them in I think in a in a in a way that mentions a specific animal or um the whales I've heard. Whales, um, but also different kind of birds and um uh squids that you know how they use their arms to communicate.

Henrik Göthberg: 1:07:03

If they I don't remember now exactly, but if they raise two arms that's read your article on this, yeah, because I was gonna quote something and I'm quoting you.

Mathias Sundin: 1:07:13

So but I I you know how so of course it's sort of basing basic machine learning in this that we have suddenly we can collect a lot of data and and help uh you know use AI to analyze that. But but over time it's it feels like okay, we might be there in a few years where we can really understand what they mean. Wouldn't it amazing if you can speak to your dog or something? Right? I would love that. It's yeah. Exactly. That that's what's fascinates me so much, is like when we get there, that we could actually have a much, much deeper understanding of each other then, because they will we will find ways of helping them understand us. And what would that do to uh eating meat, for example, if you can actually talk to the cow? You know, it's even harder to kill it. Uh so uh and you know, yeah, uh a lot of repercussions down the line there, but but it's I think that's fascinating.

Anders Arpteg: 1:08:13

So we just put in a future into ChatGPT or Gemna or something and say translate to dog language, and then you and then it barks or whatever. Yeah. Wouldn't that be amazing? Yeah, yeah, exactly. Yeah, yeah, yeah. Uh cool stuff. Cool. And um this week I think has been rather heavy in terms of different AI news compared to last week, that was a bit poorer. Um, but let me choose one that I think you will like as well, because it has a bit to be with the creativity to do as well. And um, then Athropic uh launched the new Claude 4.5, which you know normally is better than anything else in different things, and of course, Claude is really good in programming, and yeah, it's better than ever on Sweebench verified and whatnot. So that that's cool, and it's great, and all the models are continuing to improve faster and faster. Great. Um but they also I mean, just to give some stats, I mean, they have you know, computer use is one of the things which which actually enables a computer to take actions more properly instead of having you know proper APIs that is pre-built in a many way, they can actually start to use a keyboard on my mouse in in different ways, and then have a computer use or browser use that is actually letting it take action. It's still really hard, and humans are still of course better, but it's getting better and better. And at some point it will you can just unleash your AI onto your computer, and it can you know navigate to the different windows and do whatever a human can. Uh so it's getting better there uh as well. And of course, math, etc. And uh they also coding is is supposed to be you know clause uh favorite thing, and it's amazing in coding. I'm not sure if you tried using I guess we use lovable at least, right?

Mathias Sundin: 1:10:00

Yeah, exactly.

Anders Arpteg: 1:10:01

Yeah, and also you know, one of the things is adding is the um the ability to um to create like PowerPoints or Excel files. It's surprising that hasn't happened earlier. I know Microsoft also you know spoke about vibe working now, and they can inside Excel or PowerPoint also just ask it to create slides and whatnot. So I think you know all the PowerPoint engineers out there they they will soon get very relieved. Um so that's really cool to say. But I think in one thing that you will like is they had a new uh feature called uh Claude Imagine. Now, this is still a preview and only available. I haven't been able to try it myself, but what it enabled was basically you it does coding in real time. So meaning you start off saying um build a calculator and then it creates like an interface for it, but not without the functionality underneath. And then you press some of the buttons, so you press two and eight, and then plus or something, and then while you press the plus button, it then generates a code. Oh and then as you continue to iterate, it's just building the code base more and more in real time. It's a completely new like idea of how software will work in the future, right? Where it just continuously adapts itself, so it's on demand.

Mathias Sundin: 1:11:29

So if someone never have tried to, you know, you you have some uh you have a user that's okay, I want to do this, and they try to do that, and no one has thought of that or programmed that, or you know, but then it could do that for just for that user. Wow.

Anders Arpteg: 1:11:47

And you don't even need to prompt it in some way, right? You just you know, use a user interface and it continuously just add the functionality to highlight you're using it. I mean I I remember Sam Altman spoke about on-demand software at some point, but then he simply said, I think this is more extreme than that. So on-demand software is more like um if you need to have some software to do something for you here and now, then you just ask Lovable or Cursor or Replit or something and tell please write it software, and then you use it once, and then you throw it away. Like previous software engineering is you build something for months or years, and then you let someone use it for years and years, and it stays the same, and you perhaps you know improve it slightly, but still is static in some sense. But this is you know on-demand software is something where you just build something for here and now and then throw it away. It doesn't exist, but this is even more extreme, I would say. This is not that you ask it to build something, right? Exactly.

Mathias Sundin: 1:12:43

It's real time, it understands what the user wants and in real time.

Anders Arpteg: 1:12:48

And it's just continuously changing all the time. So I think this this is even hard to grasp. Yeah, yeah, yeah. But imagine that point where software is a constantly dynamic movement where it just adapts all the time and you out without you having to prompt it in any way, right?

Mathias Sundin: 1:13:09

Because we're I'm I'm part of a uh a company that we're uh developing uh AI tools for for schools, first for teachers and um yeah, first for teachers and other school stuff, then eventually for students. But uh one of the things is is of course, we you know, like like every project like that, you you try to understand what does the user want and and how how are we gonna, you know, when they start using it, how are we gonna figure out what they want and fix that, you know, of course, for every every project like this. But uh in in this case, then we you know uh they could say I want to use it for this, and and it's a completely different type of software, right?

Anders Arpteg: 1:13:50

Wow. But I think that could be the future. Software will not be the static thing, it's something that constant constantly understands what you want and just do it.

Henrik Göthberg: 1:13:59

Right? It's also the background to that uh story. I it's supposed to claude uh 4.5 sonnets the sonnet story here, and now it went 30 hours and beyond.

Anders Arpteg: 1:14:07

Yeah, that's another thing. Of course, it's really good in doing agentic tasks as well and can um write 30 hours of but you took the you took the deeper dimension of what they at least another thing called the Claude Imagine. Oh, this is Claude Imagine, okay. And that is what we spoke about, and I think it's uh it's it's really amazing, it really tells what the future will be in terms of software. Software will be such a different thing in the future.

Henrik Göthberg: 1:14:31

But this is why we are back to the fifth acceleration, even that we we are unleashing creativity and and we are still stuck in very old world models in understanding right now what it what it can do and what it should do. So we are thinking from an old mental model and and framing it that way. Yeah, exactly. And as soon as the creativity is unleashed, we get to Sonnet Imagine, and we are getting to oh my god, this is a completely new coding paradigm again.

Mathias Sundin: 1:15:00

So I uh Yeah, and that that's what I one thing I I love so much about being alive right now and being part of this. I'm I'm so grateful that I'm that I'm part of this, or I'm because I've never in my life been so often mind-blown. You know, there's a few times in in life when it comes to technology that you sort of I I remember the first time I used Google, you know, you came from the Alta Vista world and you and you Google it, it's like, oh shit, it found it. And it's like, okay, let's try this. Oh, it found it awesome! And you're like, yeah, this is and then Spotify, you know, where you had downloaded all this music and it's like it's always crap, and you just like and then and then type this in, and yeah, and then it's a famous song, and it found it to me this day. Wow, this is amazing. Do this, do this, and it found it. It's like, and now this happens almost every week.

Henrik Göthberg: 1:15:55

It's getting mind. I mean, like, and it's almost like we're getting fatigue to mind blowing. Exactly, exactly.

Mathias Sundin: 1:16:01

Exactly. It's like, yeah, yeah, yeah.

Henrik Göthberg: 1:16:03

Um we're not used to be impressed anymore, having noticed.

Mathias Sundin: 1:16:06

Yeah, yeah, yeah, yeah. Exactly. Yeah, the bar has been raised. It's like Chapity 5. Ah exactly. It's like, ah, yeah, yeah. It's you know.

Anders Arpteg: 1:16:14

But did you talk about 4.5 already? Yeah. But in very brief terms. But you know, for one sense, in a sense, it's just you know, one better model, better in coding, better in agency, better in math and whatnot. But we see that all in time.

Henrik Göthberg: 1:16:27

But I I still think if if jumping from seven hours to thirty hours in a long-term task, is that is that is that even relevant?

Anders Arpteg: 1:16:35

Or I think perhaps the that type of benchmark is interesting. So if we speak about benchmarks in general, we can see, of course, AI is saturating one benchmark after the other, and it's really hard to come up with something that humans do that AI can't do better, and uh you know, the human last exam and whatnot, and and they are you know starting to saturate one more after the other. But this kind of metric mark is doing something else, it doesn't measure you know what the accuracy level is. It's trying to say if it takes this long for a human to do something, how long would it take an AI to do it in some way? And then in the beginning of like GPT 3.5, it could do things that um took uh humans like seconds to do, and then it starts to do more and more advanced things. So um now it's up to like 30 hours of doing thing uh things that you know will take humans even longer to do. So you you measure progress uh instead in uh uh you know how long it can do things or how advanced things how long it would take humans to do something else.

Henrik Göthberg: 1:17:40

I think it's a way to understand abstraction level that you're that it's can is able to work on. Because if you give an an instruction and then it can go effectively for 30 hours, that's that's a lot higher abstraction level of giving delegation than than seven minutes.

Anders Arpteg: 1:17:59

So for me, this is I think this is quite. I think some of the media reporters on this did it wrong, as usual. They said, oh, it can reason for 30 hours. It's not hard to have something to reason for 30 minutes, it can do it really, really poorly and do it for days. That's not the point. I don't think that it that it reasons for 30 hours, it's that it can do things that take humans a very long time to do.

Henrik Göthberg: 1:18:20

Yeah, but but and I'm looking at this, I work with this in you know enterprise setting and you know, thinking about how do we adopt or how does it change organization. I mean, like the technology is there now, but think about it when your mandate structures and your delegation structures is not in minutes or hours but in months. I mean, like uh that's it that's a completely different work environment, and it's a completely different leadership.

Anders Arpteg: 1:18:47

But once again, you know, be clear. I mean, for a human, if they were to write your book, for example, or do the research you have done, you would you spend months, if not years, to collect, right? It's not really the time it takes for the AI, it's the time it would take humans to do a similar thing. That's the big thing. Yes. Right?

Henrik Göthberg: 1:19:05

Yeah. So it's it's it you need to wrap your head around it. I'm I'm I'm I don't want to go into the weeds of the details. I'm thinking about it a little bit like on the macro philosophical level. Now you need to start wrapping your head around what leadership means. I think I think that's you uh we can go into very nitty-gritty details on how we define things, but if if I use zoom out, it tells us something about the trajectory of how to lead, how to organize work. This which I think is super interesting, and I think is part of becoming data and AI ready or like being able to reap the benefits of this, then you need to put your delegation in in accordance.

Anders Arpteg: 1:19:44

But it's also that it can reason better. I mean, it can without having to ask a human continue to do a number of tasks and continue to operate. So that in some sense, I mean it's really improving the reasoning skills, of course.

Henrik Göthberg: 1:19:55

Is it the best now?

Anders Arpteg: 1:19:56

I mean, like forget about cursor, forget about everything. Now it's clouds onet. I would love to see. I haven't seen a metric for the ARC AGI uh metric. I don't think it beats uh Groc4 yet. No, but I'm not sure. I don't think so. So that's another tough metric.

Henrik Göthberg: 1:20:10

Do we have any any any other ones?

Anders Arpteg: 1:20:13

Yeah, I have a number of ones. Uh do you have something you want to do?

Henrik Göthberg: 1:20:16

I mean, like so I think we need to talk a little bit about the uh I mean like there's a couple of news. One is more of the uh collaboration between Databricks and Open AI. I think that is that we can take that later. If I'm just gonna uh exhaust the technical ones, uh we have a chat GPT in shopping. Yeah, that's an interesting one. Because that's an interesting one, right? Uh you know, we're talking about you know what's the ads, you know, what what's the what's the revenue model, so where we're going with this. And now together with Stripe, which is one of those, I guess, payment management stuff like this, like Larna, but American if I'm I mean it's uh you can easily interact or integrate. So bottom line is you you start instead of doing search, and from a search going to a web page, you you're you're writing with you know with chat GPT, like your perplexity. What's your favorite, what's your favorite? Oh, I really like cowboy boots. What cowboy boots should I buy? And then you do search on that, and you get a couple of brands, and then ah, can I oh those look nice, I want to buy them. And in the background, then agentically, it sort of together with Stripe, goes out and purchase a pair of boots for you. So you don't, you never, you never touch the uh the Amazon web shop, you never touch. So this is quite interesting in terms of what you know, many, many dimensions of how to open up value streams or revenue streams, how can uh OpenAI make more money? Uh and but it's also sort of showing, okay, this is a completely different ad space all of a sudden. All this stuff, right? So it's there, and we we and Goran's been the sort of six months before the trajectory of uh anticipating this. It's happening now. What do you want to say? I mean, like Goran, you're the ones who commented because this has been your pet peeve. Follow the money.

AINewsJingle: 1:22:06

Yes, exactly, follow the money. Now uh you have Meta now utilizing your prompt chats with Meta to sell you ads.

Henrik Göthberg: 1:22:15

Yeah. So it's an as a variant of this. But how do we feel about shopping straight in Chat GPT? What do we think about that?

Mathias Sundin: 1:22:23

I think it feels very natural, but but how does sort of the um what is going to happen to all sort of uh everything sort of behind the the the interface, the chat interface? You know, how how how how are how how is the AI going to find your boots? Uh of course you have all these online um shops, so that's one thing. But how do you um how do you or you know if you if you run uh a cowboy boots online store, how do you make sure your boots end up in in chat GPTs? I don't know, I know nothing about that.

Henrik Göthberg: 1:23:02

You know, how the the first the the the starting point is that they make a partnership with uh uh with a brand called Etsy. So Etsy is a big web page uh or brand or sort of shopping site. And secondly, they make a central deal with Shopify. So Shopify is you know the the application that helps you build an e-commerce store in minutes. So one all of a sudden you have one million Shopify stores. So all of a sudden it sort of drives it's an interesting thing because it drives everybody. Shit, I should have put built my e-commerce site in Shopify. Otherwise, I'm I'm I'm not part of the OpenAI Chat GPT checkout. So I don't know. I mean, like so you can imagine all these partnerships. I mean, like it's a little bit like uh Spotify creating all the partnerships with all the different major labels, right? Something similar happening here. But they go they're going for Shopify, they're not going for the single store. I think that's interesting, right? So that's one million stores in one go.

AINewsJingle: 1:23:58

Yep, this was also my pet beef as well. For now, for a year, I have been screaming about agent to agent and B2B and B2C. You will have B2C, B2B, A to B, A to C, and then you have A to A. And if you look at the the screen there, you basically you will see how it's gonna go. So, for example, HM, they will have their own uh uh logistic agent that understands what is actually in a supply uh in uh supply and demand, what is in the warehouse, uh, which sizes and everything else, right? So they will have their agent that is speaking with your agent. Now, uh your agent will also need to have access to your bank to check how much money you have, right? So that uh disrupts as well how we are right now purchasing things. Now the agent of HM will have to speak with the agent of the logistics, let's say that this is postnule to order it, uh-huh, and when is going to come to you and all of these other things? So it's basically we are talking about every single company within five to six years, they will have agents that will talk with the other agents and your agents exactly. Right. But now the question is which was the pet beef, is like if that is the case, uh why do we need to have web shops and websites in five to ten years from now? So everything that we are seeing right now, in five to ten years, it will change.

Henrik Göthberg: 1:25:17

Right, we don't know, we don't know where this is going.

AINewsJingle: 1:25:19

Yes, but the beautiful thing about it is that this world is shaping right now, and you can actually be the first one to disrupt it. So the disruptors of 2015, the Ubers, the Netflix, and etc., are about to be disrupted, and Arnas and Spotify and everybody else, because it's a new way of working, it's A to A, A to B, A to C, and all these other things. So I think it's a very interesting future to do that.

Henrik Göthberg: 1:25:42

But and this and why we want to talk about it is that this is something you need to keep your eye on. I mean, like a large enterprise, or if you are you know e-commerce, but even as our shopping behaviors will change, right? Right. So as an individual, as a consumer, I think it's also about what do we want. Yeah, you know, so if we have an obligation to just not follow the herd if we think it's a bad idea, but make me know I I use the word vote with your wallet. You know, so if some some behaviors come up here where they want to steer us in a certain direction, I also I think that's why we as consumers need to be aware and understand to think about uh you know, where do we put our money? Do we do we make it easier for them to go in this direction? Or uh if we like it, yes. If we think it's fundamentally wrong in some ways, or ethically or risky, or you know, I think I think it impacts the consumer, it impacts the value chain, and it impacts sort of big enterprises and startups to understand these mechanisms.

Mathias Sundin: 1:26:45

Fascinating.

Anders Arpteg: 1:26:46

Yeah, shopping coming to AI. That's it. Perhaps one we have more.

Henrik Göthberg: 1:26:54

Uh no, I mean like the Microsoft announcement as well. No, no, boring, boring. But uh you know, say something, yeah, say something about it. So we we covered it.

Anders Arpteg: 1:27:10

Of course, we know AI working really well for coding. So vibe coding, like lovable, but also uh when you have more of an engineering standpoint, like cursor or reply, that uh that enables you know you to use AI to be extremely more efficient when it comes to coding. That's cool. Now, this what they called vibe working, what they mean, I mean something else, but they meant basically in Excel or in PowerPoint, you have the co-pilot, of course, that now is getting a bit more agentic, and you can ask it to create content for you and do different visualizations and summarizations in Word and whatnot. So I I think they are stealing a bit the the real meaning of the word vibe working because I mean something else, and they're stealing it.

Henrik Göthberg: 1:27:51

But I I think it's funny because I think they're also diluting all the definitions themselves. They have copilot, and oh, I have co-pilot in in in uh I have co-pilot in in Excel mode. Isn't that a kind of agent? No, no, no. You have copilot, I have agent mode, vibe working, you know. Come on. It's bullshit bingo uh uh on a high level.

Mathias Sundin: 1:28:10

It feels like if Microsoft says it's vibe, it's not broadly. It doesn't have much vibe, I think.

Henrik Göthberg: 1:28:16

But do we understand the technical difference between the copilot route and the vibe working route and their definition of agents here? What's different?

Anders Arpteg: 1:28:25

I think you know, I think in theory, if we speak about the vibe working kind of concept, I think that's very interesting to consider. Now, in this case, I mean it's simply that it can create things you know in the format that is Excel and PowerPoint, which is good for all the PowerPoint engineers and didn't you do that with Copilot?

Henrik Göthberg: 1:28:42

What's different? Is it used a rebranding or is it technically something else going on?

Anders Arpteg: 1:28:46

But it it wasn't really good in creating things. I haven't tried this as well.

Henrik Göthberg: 1:28:50

No, I haven't either. So maybe it's more embedded, integrated.

Anders Arpteg: 1:28:53

I mean, it before it was good to understand what you had in it, but to really create things, it wasn't really working well. No, I don't think it still works well, but I haven't tried it. So let's leave it. But I think you know the point with this is really that we are seeing vibe coding going more and more general. So for vibe coding, we can see it works well in terms of knowledge management, but it also works surprisingly well in terms of reasoning for coding purposes, and even a genetic parts, meaning creating code and building things. But that is going to generalize more and more. So I think the idea of vibe working when it really starts to move to what a human coworker can do, as in in working sense, that's going to be really cool. This is not really that though. Is that what you mean by vibe working? Yes. Okay. So then it means level one to three in this kind of pyramid of open AIs.

Henrik Göthberg: 1:29:42

Yeah, so because it's literally the pyramid shows how we go up to abstraction levels. So you're not working on the task level, you're not even working on the reasoning level, you're working on the innovation level level or organization level more.

Anders Arpteg: 1:29:55

Autonomous level of the autonomous level.

Henrik Göthberg: 1:29:56

I mean, like so we are going in that trajectory and and Working would indicate that you were working one abstraction level higher or two higher than what we understand as work today.

Anders Arpteg: 1:30:06

Right, okay. So I'm sure we will see a lot of this kind of vibe working and going better and better. But the I don't I haven't tried this, but I don't I wouldn't say it's it's super X.

Henrik Göthberg: 1:30:16

Uh should we say something about the partnership? OpenAI Databricks.

AINewsJingle: 1:30:21

Well, I think that uh we are running out of time. I will do it just a setup. We will talk about it next week. I have actually I have been uh I have been in a down rabbit hole about this imaginary money that are coming up. So um now we have circular economy, yeah, has a new meaning. But we will talk about it a little bit more. But it was a very interesting uh article by um um uh I leave it now, it will come later. The name uh Wall Street Journal actually about the AI bubble right now, because right now the bubble is getting bigger and bigger, and now you know we have over uh blown AI initiatives that are valuable for a lot of money and all these other things. But it seems like the most of this money is actually coming because they need to support the Stargate, right? There is new data centers that need to be installed, they started building it, etc. And uh OpenAI needs more money to build uh the future revenue, right? So SoftBank gave them a lot of money, uh Nvidia was there, so um Databricks uh and many other so it's interesting. However, my my uh we will talk about it next week. However, this week I was super excited actually to try Sora.

Anders Arpteg: 1:31:29

Ah, okay, that was my next one as well. Yes, but go there.

AINewsJingle: 1:31:32

It's uh I I cannot use it because it's not available in Europe.

Anders Arpteg: 1:31:37

Ah, of course, as usual, right? Yeah, but I think what's cool with uh Sora then, um of course it's better than generating videos and everything, and you can also have audio similar to what VO3 from Google has been doing for some time. So, in some sense, it's catching up to that. But what's really cool with Sora 2, which I don't think anyone else has, I think they call it the the cameo feature. It's basically a way to insert yourself into the video. But it they need to have security on this because if I can take you, Matthias, and put you in a video, I could easily have a lot of fake news here and a lot of manipulation. So they're trying to have a secure way to add yourself to videos that so no one else can bring it into. So I think in short, you you have to you know validate that you are the one you're saying, and then they you have to look in the camera and you have to turn the face and do certain things, and then it's okay, it is really you now you're allowed to actually create a video with yourself.

Mathias Sundin: 1:32:35

Well, there's also a strong incentive to start using open AI then because if if you haven't done that, someone else can, yeah, okay. And it's interesting, you know.

AINewsJingle: 1:32:44

Uh they created the app for it. Yeah, and now you will think like, well, that is cool. Why would you create an app?

Anders Arpteg: 1:32:53

But I think if I I can imagine for security reasons, actually, but still, what what do you think?

AINewsJingle: 1:32:57

I think it is basically they want to be a little bit closer to the new generation because they want to create a new generation of users. So because right now they have reached their capacity, right now, right? And it's like the Tableau moment. You remember how Tableau, beat Click, and everybody else, they went to the students and till uh give them uh tutorials about how to use Tableau. So if you want to generate the next generation of users, you need to supply the demand, you need to create the demand. And just with this, you you it's it's so instantaneous.

Anders Arpteg: 1:33:28

It becomes I think that's the key thing also. Yeah, that's the point with the app then. I mean, why would that another thing?

AINewsJingle: 1:33:35

So my kids, yeah, right, they use apps all the time. They have facewaps and all of these other things.

Anders Arpteg: 1:33:41

Why couldn't they use the web service for it?

Mathias Sundin: 1:33:43

My daughter who doesn't want to do a web service. Are you getting saying that? Exactly. I think and I don't think it's only I don't think it's only younger people they want to reach. It's it's the people who don't use ChatGPT now because it's whatever it is, you know. Um but if if if there's a cool new app that you can make fun videos with, uh that's great. And that could be sort of the entry point to ChatGPT. It's like, oh, there's a lot behind this, you know. And and so I think that's the thing.

Anders Arpteg: 1:34:14

The normal web interface doesn't really have the functionality and all the security. That is my guess why they need to have a separate app and not being able to have it in the normal interface. So the practical reason as well. Yeah, yeah. Cool stuff. Anyway, yeah, yeah. Back to the story. And perhaps we should, you know, I'd love to hear a bit more, Matthias, about your work with the Swedish AI Commission. Perhaps you can just give a short introduction to how you got into that work and what you were doing there.

Mathias Sundin: 1:34:47

Yeah. No, so it's um yeah, the model here in in Sweden is of course to uh appoint some people to come up with proposals for things. Um and for and actually that's that's part of the the law here that you have to prepare proposals that that way. Not always in the in a report like this. But, anyways, that's that's sort of the the background. Uh another background is that in the 90s we had very successful IT commissions. The first one in 1994. Uh, we had uh the former prime minister then called Bilt as chairman. And uh uh and they they just worked for a few months and but they delivered a very uh an extremely visionary uh report back then. Um talked about um how IT then could sort of give wings of freedom to people and all that. You can just replace um uh IT with AI and you can use that. So that that could have been an even shorter uh work for us. But um uh but also that uh the second IT commission in 97 or 98 or something like that come up came up with um the home PC reform, so subsidized home personal computers for for people that had a uh had a pretty big um uh effect. Um so that's sort of the the background of these these kind of commissions. So now the government won an AI uh commission. Um and uh I guess I became part of it both because I have a political background, there were no other people on that commission that had that, but also that of course now I I uh uh that I uh write and do a lot of things uh with AI and sort of represent the media perspective a little bit. So there were uh we were like 10 people on the commission and then had an expert group um tied to that, or expert group, but but a group of other people. Um so our task was super broad to come up with proposals to use AI to strengthen uh Swedish um uh competitiveness. Um that was the thing. And we could do anything except we we could go into any area except except one school. School education. That's so crazy. Or actually, education was fine if it was university. But smaller children, no, no, no, no, no AI here. That's gonna be very, very interesting.

Henrik Göthberg: 1:37:14

That's an very interesting oh that's a rabbit hole in itself. Please go back.

unknown: 1:37:19

Yeah.

Mathias Sundin: 1:37:20

So um uh we started uh meeting uh a lot of um different uh people representing different parts of uh, like I said before, part of part of society. And we we very soon realized, okay, uh we had 18 months. 18 months in AI time is I don't know, 18 years or something. So we're that's okay, that's one problem. But the other problem was that's that what we talked about before, that people are waiting for the commission, waiting for that. So we said, okay, we we we want to deliver this after halftime. Um and and we did. We we wrote 75 different proposals. The ideas was to mostly do proposals that didn't need um uh much legwork after that. Is you can put money or uh you know decide to do this um and and get rolling pretty quickly with many of the things. And also there was a lot of speculation that we, you know, they're gonna want uh you know gazillion uh dollars or kruna uh for this. But we said also we we want to pick the low-hanging fruits here. There's a lot of things we could do, a lot of things, a lot of areas we could dive deeper into, but we wanna serve the politicians with uh a strategy and especially a lot of proposals they could start doing quickly since we are lagging behind. Um so so that's what we did.

Anders Arpteg: 1:38:41

Um cool. So you got these um directions from the government that exclude the same part, but otherwise it could uh do whatever, and you came up with a very nicely written report, I must say. And uh I guess you interviewed like hundreds of people or something, or yeah, yeah.

Mathias Sundin: 1:38:59

I was yeah, I think it was 150 different organizations, so trying to cover all of society through organizations.

Henrik Göthberg: 1:39:07

So you from the inside, what were you what are you most proud of in terms of how you went about and what came out of the report? And what do you think uh this we could have done a little bit better, or this one we didn't really do us to the ambition level we had?

Mathias Sundin: 1:39:21

Um I I think uh the the part I uh I like the most was um that we both wrote about um that uh that AI now is for for everyone. Uh and I think that's a key message from the government because most people don't know that AI is for everyone. You know, that's why they they need Sora as an app. Um because oh okay, oh you're using AI, am I? Like, yeah, cool, and you can do other things as well. So so that was that was important, but also uh that what we call the AI for all uh reform, which was uh heavily inspired by the home PC reform. So um getting people to use these AI tools for free, the paid versions, um, the pro versions, um, use them for free for a while, for six months or 12 months or something like that. So both send a message then this is for everyone, and everyone can get access to this now. Uh you don't have to pay at first, but you should really use the paid versions because they are much better, and you can use and you should use them much more so than you need the paid versions to get people on board there. Um and and I think that would have sent a strong message to to Swedes, but also a strong message internationally that okay, this is a country that does this differently. Um they say, you know, everyone should have access to these tools, and you know.

Henrik Göthberg: 1:40:52

So um we see other countries have gone down that route. This is happening in the Middle East.

Mathias Sundin: 1:40:58

Yes, exactly. And and I saw that uh UK is at least thinking about it or discussing this with OpenAI. I I did I I didn't want just one, not just ChatGPT or not just Gemini or Copilot, uh, but you know, um many of these tools. So you can, you know, um and and it would be a very nice platform for Swedish companies to launch AI tools, you know, because suddenly there was some sort of platform that you know hundreds of thousands of Swedes could could use. Um so I I think it would be helpful in in many uh different ways. So that that was the part that I um I work with the most and I I liked uh the most about it.

Henrik Göthberg: 1:41:38

And any things you think, oh, we missed on this, or this was a blind spot in the report, or this I think was Steph Muddly uh Stu Modlit behandless.

Mathias Sundin: 1:41:49

Um I think in general uh it's there's a lot of things that we could have gone more sort of in depth to um thought about more and and thing, but um time-wise, you know, yeah. Um that's good. I I think also maybe the uh the the the people in in the commission, uh it was kind of amazing how there's very a lot of people with strong opinions and you know who usually, you know, yeah, the the CEO of Ericsson Envolvo, they lead companies with 100,000 people each and many people like that in the commission. And and we really worked well as a as a group uh in some way. Um anyway, so so that was good. But I think uh we really would have benefited to having people from a few other types of people, maybe someone from the startup world, maybe someone from something completely different, some an artist or whatever, you know. So you know, so there was a lot of big um like um big big authorities, big companies, and they were great, but they have their perspective. So I I think we could have used um others uh perspective. It's not something I can say, yeah, we could have it would have been better because we would have had that perspective. Because I I don't know, but if we would have a couple of you know a couple of people from completely different, for much smaller settings, I think that would have been overall useful.

Anders Arpteg: 1:43:23

But if people just want to have an understanding of what the report really contained, I think you should be so proud of that and all the concrete like 75 proposals that you did include, right? But you also had, which I think is super hard, some kind of monetary kind of estimate for what the budget should be to prove to actually do that, right? H how did you come up with and how do you come up with that kind of number for the budgets for it as well?

Mathias Sundin: 1:43:46

Yeah, yeah, I don't know really. No, we we had a we had a staff of um uh of of people who you know that that's what they do. Um they they try to uh sometimes, you know, some of the some of the numbers are more like okay, we don't really know. Um there's no way of figuring out the perfect number. Uh it's probably 50 million around that. You know, some numbers are like that. Some are numbers are really more calculated and really try to think things through. Um so so that was really the staff trying to find. And and we, of course, we had to sort of try to prioritize. Um, but we uh everyone said that okay, we shouldn't do the we're not gonna propose you know Sweden spend uh a hundred billion Swedish kruner on uh data centers or whatever. It's like uh we we could have said that, but you know, it's it's not yeah.

Henrik Göthberg: 1:44:37

But I there's there are two things I think you did really well. You just the fact that you went to went to a point of putting a number there, for me is is more like you know how you budget with t-shirt sizes. It gives an indication on the ambition level that you need in in order to start something, and which is I think it's fairly realistic, and I think it's it's it's in the end, what is the strategy without money behind it? You know, so so therefore to actually push that into the proposal, I think was a very, very good move. Even if, you know, and I don't give a shit if it's 30% off or something like that, is it's the is the fundamental point of view. And where I'm I'm I'm a little bit like I'm more happy to see um money in a budget than I am to see a written strategy document. Yeah, yeah. So so that I I need to that we applaud very much. So what what what didn't you like or or what what should have been in the report that wasn't? I mean, like we can do it together, but but we we've had several guests now. I mean, like we've had 160 guests and uh and we have our own uh I think uh understanding of this. And and uh if I'm looking at if the profile, you are onto this already. The profile of who is in the commission is either the big CEO or is academia, researchers of different clients or or public sector. What I think is the missing, the the biggest missing point if I if I'm used to looking at Silicon Valley, is the hardcore product and engineering perspective. So if I think about it, I mean like we had Sergei Johnson here, head of Rice's uh AI Center of Excellence, and he said it so well on the pod that you know the and we see it now with Lovable, the second generation startups in AI. It's it's this whole engine that spews out great data and AI engineers that know what great looks like, and what does it take to build that? So it means like uh I was missing sort of maybe these some key uh senior, super senior engineers at Spotify or you know, Clara now or something like that, which is sort of you it I mean there may be a represented in the commission, but it's not the engineering folks. So where's so if you look at the funding and the storytelling on the proposition, proposal, it kind of follows. Oh, we need more PhD, we need more AI research. Do we really, or do we need more, vastly a lot more AI engineers? You know, can we vibe codos out of this? I don't think so. So so that that is sort of my main blind spot of a of a uh there where we're not really picking apart one of the main success factors of Silicon Valley and and looking at you know how are we doing here? Yeah. That's the one that's the main one that stands out to me.

Mathias Sundin: 1:47:35

I I think you're absolutely right. And and like I said, I I think it would have benefited from people from completely different, you know, there's a limit, of course. There's a limit. Exactly. And and that and also it runs into a political problem um that you have to you have a list of sort of different categories. You have male-female, of course, uh uh, but also you have different parts of the country, you have to represent different parts of the public sector, and you know, it's like so you have to find so it's almost and now we're at AGI levels to solve this puzzle. So so so yeah.

Henrik Göthberg: 1:48:10

And it's what were you you we liked it. I mean, like first of all, we really, really liked it. So this is it's actually not a critique, it's more about you know, trying to what's the blind spot and what what would you say, Anders?

Anders Arpteg: 1:48:22

Yeah, but we can really and given the people that were in the commission and the people that spoke to you, can really I I can see who who wrote what, and so to speak. So it's kind of easy to to identify. And of course, the focus on more research is something that I you know don't think is perhaps what we need the most. We already have so much investments in terms from Wallenberg and many other places in terms of AI research. That's really good, and and uh we should have it. You know, I'm having I have a research background, but I what I'm lacking is the engineering part. You know, if you look at what the big tech companies are doing, they perhaps spend like X number of dollars on RD work, but then they at least spend 10 or even a hundred times more on the engineering part, and that I'm strongly lacking. And uh if we don't have that and really have this kind of product mindset or really the mindset of how can we find true value, um, then it will be a lot loss here. So just to give an example, there is a number of, you know, if you just invest in a data center and say, Oh, we can train this amazing model here, yeah, great. But how can we deliver and deploy the the application on it? It doesn't speak about the application, it's a lot of focus on training some model, but not on really delivering the product. And even today, and and this is not on the commission report, but really in general from Europe and Sweden, it's so much work on having a data center that can train a model but not read on serving it, or even more so how to actually deploy an application. Imagine if you had a data center just for training a model and then you have to serve it somewhere else, and then you have an application and you have to serve that somewhere else as well on Kubernetes on and all these kind of other services. That would never work. That's why you know you have HBC clusters that you can't even give away compute on because it's not working from uh engineering point of view. So that's a bit a shame. The other part, and I'm going into rant mode, no, it's it's not really you know a shame on the commission. This is more in general, so to speak, because I think still the commission did amazing work on the report, it's really good, so don't take it personally. But I think in general, we need to have a strategy of how Sweden can quickly find value from AI. Now, there is a lot of given the Trump kind of era here with the tariffs coming in that oh, we need sovereignty in in Sweden and Europe. Okay, yeah, of course we do. We would love to have that, but in reality we don't. We don't have a viable alternative to the big cloud providers today. So, although I wish and I hope that we will start to build the kind of long-term objective of having an alternative to them, but that will take many, many years to achieve. So then we need to have an um like compromise, meanwhile.

Henrik Göthberg: 1:51:14

A two-prone strategy. We need a short-term strategy to suck out what we can do with still relying on them.

Anders Arpteg: 1:51:22

Yeah, but there are a number of alternatives. Even the the big hyperscalers have this kind of hybrid mode where you can run the tech stack you know on your own machines, that's not a problem. But we need to have this kind of reference architecture. I can go into much deeper details here, but I shouldn't. The point is really we need to think what really delivers value for society in our uh corporations in Sweden.

Henrik Göthberg: 1:51:42

In the context we are in.

Anders Arpteg: 1:51:43

We can't we can't wait five years for that. We need to have it today. And then you you have to have to have a realistic look in what is necessary to make that happen. And that I'm missing.

Henrik Göthberg: 1:51:53

Yeah, um, and here you kind of see how those two dimensions tie together because it that's not a research question, it's a hardcore engineering question. What what tools to use, what tool stacks to use, how to build something technology agnostic uh as far as possible, etc. etc. Which which is sort of uh not this is more than applied research, it's it's product.

Anders Arpteg: 1:52:16

And and and I mean like so I even spoke to Rice recently. I mean, there you know, we have the big Invest AI uh program from European Commission, and and that was you know, it's 200 billion euros and 20 billion just on data centers. And if they don't think about this in this way and then continue in the normal way, it will be useless as previous investments have been. So I'm a bit but still I have hope, and I I think with your help and others, we can you know think print value first.

Henrik Göthberg: 1:52:48

It used to be not been bottom line, I think the the Commission report ticks the right boxes in terms of what it has, but we are missing boxes. And when you're adding some boxes, you need to rebalance across the boxes. Right. So we need research, we need talent, all this, all these studies mentioned, all the seventy-five proposals are important. I think there are like 10 different proposals that should have been in there and that are higher up on the agenda than some others.

Mathias Sundin: 1:53:16

Yeah. No, and and just you saying this now, it it would have been super interesting to hear this conversation in the commission because there were there were no you know real pushback against you know uh education or or you know And we're not pushing back, it's just simply balanced. Yeah, yeah, exactly. But uh sort of having is this really the right priority, or you know, should it be more like this, or you know, could we have both? Or you know, so so yeah. Um that's back to the diversity of the room. Yeah. And then you also want something, someone like I said, who has a completely different like engineering. Well, what do we need that for? We need poetry or whatever. It's like philosophy also flip now.

Anders Arpteg: 1:53:56

No, but you know, the biggest problem that most companies have and why they're failing to find value from AI is they get stuck in prototypes, right? It's to build a model that does some kind of prototype, but it's such a big difference to do a proper product, and that is what's missing still, I would say. Anyway, okay. That's our ramped. Yeah, so still Matthias, such a great work with the commission, so applaud you for that. Thank you. Let's get a bit more philosophical, perhaps. And we're starting to run away on time here, so I'm I'm going to skip a number of potential topics that we were uh considering. But let's if we do think more from a societal point of view, and we know AI is going to make a lot of changes to the workforce, to our educational system, etc. Do you have kids by the way? Or one what would you recommend him to learn and educate in uh today?

Mathias Sundin: 1:54:58

I think um whatever he's uh interested in, because it's gonna be uh even easier to I don't know, make a living or or do something, um uh whatever that is, uh, because the world is constantly opening up um with the internet, of course, but uh that's just the start of it. So you're more and more you're adding you're adding more uh different types of skills. Um you can learn skills as a human, but you can add more skills to it, and you have a bigger audience, whatever that is. And and maybe it's not something you have an audience, but you know it's it's a community or whatever, uh it's something that where you can do something that is um um sort of yeah uh something that you really like. Um and I think my and this is of course nothing unique for this time, but um my task as his father is trying to show him all these different worlds, and and for me it already helps with AI because, like I said before, I can I can't sing or or make any music or whatever, so I can't show him that world, that world of music. I I can show him, you know, we can play music for him and we can take him to concerts or whatever, these kind of things, but we can't I can't make mu I can't make music with him, but suddenly I can, at least to a certain point. So uh we've made songs together now. So one song about um his uh uh teachers at preschool and all the kids in his preschool, and it's it's a really good, it's a really good song, actually, and and all with all the names. And he added the names, and we sort of oh, who's more the other day? It's Friedrich and it's Hedda and all them. And now he hears music. Um, and I can just go to myself. Um, now when I've started creating music like this, uh it's very simple music, but it's made me a little bit more interested in how do you actually make good music? I don't know nothing about that, but suddenly I'm a little bit interested in to actually read a book, maybe, because that's my way of learning. Reading a book about how do you make music, uh, you know, so I can use AI to make a little bit better music. So it helps me as a father to show him more parts, and maybe that will make him interested in learning to play the guitar, to play the song that he you know created. Uh, and and that could be done for uh you know a number of different um things. So he could choose you know, or have have as many options as possible.

Henrik Göthberg: 1:57:38

But one thing you're saying here is that in the new world, try to figure out your passions as part of your focus in order to get good at it, and because there is no limits, if you have a passion, you can use tools and be really, really good at it if you if you're creative and putting in the hours. So then it's back to the 10,000, you know, you need to focus on stuff that you're willing to really uh put your hours in and be passionate.

Mathias Sundin: 1:58:04

Yeah, and that's easier than before. Um, it's it's less of you have to go to a factory to make money and you hate it, uh, but you have to because you have to pay rent and buy food and things. And that's of course gonna be part of life in the next decades as well. But it could be more and more your passion, whatever that is, it's gonna be a bigger opportunity to do more with your passion.

Henrik Göthberg: 1:58:27

So with your passion, becoming really good at it, and then you can most likely carve out an a career somewhere. Yeah, that's an interesting one.

Mathias Sundin: 1:58:34

It's at least easier or not as hard as before. It's not easy, probably, but uh you you still have to put an effort into. But you can discover more of the of the world, what it is, and and fight hopefully find uh be better at finding whatever your thing is, and then uh the chance of actually doing that for a living, or you know, what does we think about the educational system then that we do have?

Anders Arpteg: 1:58:58

I mean, uh he has to make some choices in whatever he chooses to educate himself in. Yeah. And uh, some people are arguing that you know we can um we don't need perhaps these kind of long-term educational systems and getting a degree in whatever perhaps isn't as useful. And others saying the opposite. What do you think? You know, how is the best way for kids these days to prepare for the future of tomorrow?

Mathias Sundin: 1:59:26

I think um yeah, there might be ideas out there of uh there are many ideas of a better education system, but but you know, at least uh I think the the idea of a school and an education go to is is pretty good. Then then that school needs to change a lot in how we teach kids, and that was actually my my original plan was to become a teacher. Um so and and how you use these tools in in education that could also change a lot, but I still think it's a It feels like a good idea to learn a lot of different things and then specialize a bit. Because that will help you interact with all the machines and and everything. And you can use the machines to learn and you know, so it's it's merging of things.

Henrik Göthberg: 2:00:13

But I still like the idea of the fundamental idea of an education system is still sound, or we can't really fault. But it needs but then it needs reform. That's another example. Exactly. So it's not it's not throwing out the baby with the bathwater here.

Mathias Sundin: 2:00:27

No, exactly. Don't you know, yeah. School is still a good idea, and education is still a good idea, I think. But also there's a path, especially in higher education, that if you don't want to do that, there's many other paths. Uh you don't have to go to that university if you don't have if you don't want to. You can go, you can choose many other different paths. Um but also school and education for me is not just learning uh topics, it's it's being part of a social system and learning to interact and having fun and you know, all these things, you know, learning to be a grown-up uh that has nothing to do with whatever you're studying.

Henrik Göthberg: 2:01:03

Um but how how would you frame the f societal challenges with the education as it is today? You know, not not solving it but trying to put a finger on what we need to fix or work on.

Mathias Sundin: 2:01:16

It feels it still feels like it's very much um one size fits all. Um and and and and we we we already have tools to make it more individual. Um but uh with AI we where we we can at least approach the the holy grail of education that you can actually um individualize it very much just for you because how you learn and and how do you learn not just in general but this week, um it adapts. And I I think one key thing, one key thing, maybe the key thing in education and as a teacher is to get the students to understand that they actually learn things. So they say, you know more things this week than you did last week. Because I when when I worked as a teacher, I I took over uh some eighth graders, and there were two um uh two two kids in that that were and and they got grades for the first time uh at Christmas. And the first time was in eighth grade, and uh for Christmas they got their grades. And they hadn't learned much, so so I had to fail them, I had to give them failed grades. And they were really shocked, like, what? What the hell? It's like what? We thought we knew these things. Like, no, sorry, you don't. Um and then they started to make a real effort to learn things, but they were far behind, you know. Um so when they uh uh uh when they uh in ninth grade, when they left uh Hug Stadit, um they had learned a lot in these 18 months, a lot, but they hadn't learned enough. So they were failed again. In their minds, they were failures, they hadn't learned anything. They sucked, we can't learn. I saw the progress and I tried to tell them, but they didn't understand it. And and if you interact constantly uh with uh a computer that way, it could help you under help you understand that you actually learn things. You're not stupid, you learn things, and it can adapt the way to your ways of learning so you can learn more, of course.

Henrik Göthberg: 2:03:27

Because it can be that oh, we need to shift the curricula, we need to have freedom so you can study whatever you want. Okay, that's far out there. We can back the tape up to basically it's the same curricula for all, but we have individual individualized learning because we learn differently. And and I have a like a good anecdote with my my 17-year-old who basically uh it's a simple thing when someone changes a teacher, and that teacher all of a sudden, oh I'm struggling with math, I don't get it, I don't get it. And it's a simple path of uh another communication style, another person who's better to meet him where he's at and lift him from where he's at, which to me is a very simple observation highlighting that oh my god, if we could figure out the learning pace optimized and the learning style optimized for each child. Wow.

Mathias Sundin: 2:04:24

Yeah, and we do that in computer games, for example. We're really good at giving you feedback all the time, how good or bad you are. With existing techniques, either. Yeah, so it's not that it's not that complicated. AI now makes it easier or even better and more adaptable, and you know, um, but but the basic uh I don't mean to gamify everything, but you know, the basic idea there of giving you constant feedback or giving you feedback that you can understand. And so I think the key thing is to get students to understand that they actually can learn things. Um I I remember back when I was in school, there was a kid, you know, you know, the kid, uh he was a real troublemaker, you know, you know, skipped school all the time, hit people, you know, you know, the type. And and he thought he was stupid, and everyone else thought he was stupid because he didn't learn anything. He failed every class. But he knew every car brand, he knew every car, you know, every everything, you know, a car drove back, he knew what year, what kind of engine, whatever. He learned everything about it. So he wasn't stupid. It was that these things that he learned him in school, he didn't want to learn for different reasons. Maybe not it wasn't the school's problem, it was probably his parents or something, but still, he wasn't stupid. If you could get him to understand, because he probably thought he was stupid, but he wasn't. If you could get him to understand, you're not stupid, you can really learn things that would probably help him. Um and AI won't solve that like that, of course not.

Henrik Göthberg: 2:05:52

But you know But we can see how technologies here can make a huge difference applied correctly.

Anders Arpteg: 2:05:57

Yeah. So you are more positive or than negative when when AI comes into the educational systems?

Mathias Sundin: 2:06:06

In Sweden, I don't know. It feels like we're going in the opposite direction. So I I don't know. Uh but um over time I I hope uh Jan Stienbeck is right that tech beats politics, and and uh there's a lot of teachers out there that understands this. And as long as they don't forbid that, but I I wouldn't, you know, there's an election coming up.

Anders Arpteg: 2:06:31

Certainly are universities that forbid use of AI. Exactly.

Mathias Sundin: 2:06:34

So and the politicians could, you know, there could be a uh in the next election, some some party, maybe the one I was part of, will say, you know, we're gonna forbid whatever they're making.

Henrik Göthberg: 2:06:44

They already forbidding out of it, and it's terrible.

Mathias Sundin: 2:06:49

Yeah, weird thing. So I'm not not that optimistic about that.

Henrik Göthberg: 2:06:53

So you're you want to see it, but you're not optimistic of how it will play out in Sweden. No. Uh it's um It's a weird. I'll be like, I've been following a little bit. We had Anders Engström here. Do you know Anders? Yeah. Amazing guy, and and you know, and he is I'm not even sure if he's an angry optimist. He is an angry optimist. He's angry. Fucking angry with uh with the bullshit going on. And he's a guy he has a pretty good eye on spotting it and talking about it.

Anders Arpteg: 2:07:22

Matthias. Um let's go for an even more uh philosophical question. And uh of course, uh you mentioned AGI a number of times. Do you believe AGI will come? And if so, when potentially?

Mathias Sundin: 2:07:36

Yes, absolutely. I I don't think there's uh as long as everything can be solved as long as it doesn't break the laws of physics, and I don't think AGI is breaking will break any laws of physics, but um I I don't think it's around the corner. And for me, AGI is I I'm sort of uh I like the David Deutsch version of AGI that it's that's um uh an AI with its own mind, its own opinions. So if you it's not the AI that does what tell what you tell it to do. It could do what you ask it to do if it wants to, like I. You can ask me, you know, so so there's some sort of the same sort of um mind that a that a human has that with you know uh that it can decide to do things on its own. Um that that for me is uh true artificial general um intelligence. And I uh I don't think uh Is it two, five, ten, fifty years? I I don't know, but it it's not not the next 10 at least. I I don't think that it's it's way more sharpening the definition quite a bit when you say it like that. Exactly. So it's not just something that's uh it feels like the the definition is constantly changing and and the sort of the bar is lowered because everyone wants to reach AGI first, so they lower the bar. Of course, they they they make better and better AIs, but they also lower the bar so they can meet here and say, hey, we have AGI. It's like, no, that's not what I think is AGI.

Henrik Göthberg: 2:09:10

So I'm quite intrigued with your definition. I kind of it's an interesting one.

Anders Arpteg: 2:09:16

Okay, but at some point it will come, yeah, and uh then we can imagine uh the world is playing out in one direction or the other, and and if we were to imagine the two extremes where one is the uh dystopian one where AI will turn into the Terminator and the Matrix movies, and the machines are trying to kill us all, and the other extreme, of course, is the utopian version where instead AI have solved cancer and fixed the energy crisis and had educational systems that makes humans superior to any kind of human beings that we have today. And we have perhaps we have a world of abundance, as Elon and others call it. Where do you think we'll end up in this kind of spectrum?

Mathias Sundin: 2:09:57

Um I think definitely more to the utopia side, but I don't think it will be in uh utopia because there's so many things that um and that's one of the things I I warn about in in the book that even if I even though I think uh this is such a breakthrough, it will lead to an the the largest acceleration of human progress ever, or so far, uh I think we should not believe it's gonna solve all our problems. Um, because that's many of these problems are not um problems that technology can solve. It can help us solve uh many problems, or you know, cancer, things like that. That kind of research that's perfect for an AI to solve these kinds of uh problems. But for example, um poverty or uh uh starvation, that's not a problem that technology can solve. We have all the food we need already, uh, we have all the money we need. Um, uh we have all the technology to get the food to people, but for some reason we don't do that, and that's mostly because uh those people don't live in in democracy, so their leaders don't give a shit if they eat enough. Um, and so the problem there is is is is not uh logistics or anything like that, it's it's democracy of lack of um democracy. So and AI is not gonna solve that. AI could help um resistance movements and democracy movements to overthrow dictators, but of course, dictators are also pretty tech savvy, unfortunately. Um so I think there's a lot of a lot of problems in the world that that technology in itself can't solve, or humans with technology can't solve. So that's really up to us, and of course, using technology in that, but it's up to us to solve this. But it will help us solve many problems and especially help us create abundance of many things, uh, I think, which is of course very helpful.

Anders Arpteg: 2:11:52

Um I'd like to just iterate on what you said a bit before, which is uh you quoted someone as saying the tech will decentralize power in some way. Yeah. And just to to argue on that point a bit, I mean, I think today we're seeing a a bit of a or a rather large concentration of power, especially into the Elon companies of the world and some of the tech companies that are, of course, you know, creating immense wealth and having more and more power. Are we really seeing a decentralization of power now with AI? I think, or or will it do that perhaps in the future?

Mathias Sundin: 2:12:29

I think so. Um, of course, you're you're absolutely right. In uh they are also the Elons and those companies of the world are are getting very, very, very powerful. But at the same time, it seems like, for example, language models is not something unique that just a few companies will be able to create. You and I and everyone else will have their own, if we want to, their own language models that are very powerful, maybe not as powerful as the best ones, but they're gonna be really good. That gives you a lot of power. Um so so even if in some ways power is concentrating to companies and and people, um, I think the overall effect is that everyone is getting more powerful and more wealthy. And and if you look at these numbers for the world, the world is getting uh more and more equal uh economically. But we we get a skewed view of that because we see some people that are richer than ever before, and they are, uh, but everyone else is also richer.

Anders Arpteg: 2:13:30

It's not the serious sum game, right? Exactly. Just because some people are getting richer, richer doesn't mean others get poorer.

Mathias Sundin: 2:13:36

No, exactly. You don't lose money just because Elon gets gains another 100 billion.

Henrik Göthberg: 2:13:40

But but these statistics are so hard to penetrate and understand this because if you look at if if you know you can the lies, 10 lies and statistics. I mean, like it's uh you you can look at the um uh GDP of US and you can see that uh the inequalities in terms of how how much wealth is with a few compared to the rest. Uh, if you if you if you follow that line of analysis, you get to how the the divide is growing, right? But then I I think what Rusling was one of the pioneers doing is looking at the real macro data in in in uh zooming out even further, and then you could see okay, inequalities have risen, but at the same time, the whole world is at a different pace, and and we and he she could show how we all have a very skewed way of Africa as an example, right? So I I I'm getting you what you're saying here. But it's when you say it's getting uh we are dem democratizing power. No, we're not. It's the gene-jerk reaction because we are looking at it from one lens. But if you look at the you know the the fifth acceleration lens, I get it. Yeah, we're actually raising the bar across.

Mathias Sundin: 2:14:48

And of course, some countries uh inequality, economic inequality is increasing. Uh, but mostly in in poor countries, they're already poor. Those are the ones who are really unequal. They have one or you know, they have 10 people as they have 90% of everything in the country, and the rest have to share the rest. Um, but in in most countries like the US, um, you can get different small shifts in in that. But overall in the world, if you look at nations, uh, we're getting more equal. Um, the poor countries are getting richer much faster than the rich countries. But also, if you look in those countries, um people are also getting um uh more equal economically. So the poorer people in those countries are getting richer, and the richer are also getting richer, but in less so so it's evening out. Uh by no means a perfect world, but it's it's it's getting a little bit more.

Henrik Göthberg: 2:15:38

But but here we could have a huge debate on this, and depending on which stats we're looking at, I could prove the opposite of what you said now. So I don't think so. I don't I don't have the data but to back it up. But but but I mean like there there is the inequality conversation going on in US, and when you look at how how big percentage of the GDP is now with 10%. And if you look at that curve, if you follow exactly that statistic, it goes in the wrong direction. That particular statistic.

Anders Arpteg: 2:16:08

Uh so uh it sounds like a perfect discussion topic to continue having here. And uh and I also love to discuss that a bit more. And uh I'm a you know, I'm positive in the terms that I do agree that for most people it will be a very positive change where you will have more availability to products and services and wealth in some way. But I am concerned with the concentration of power that is happening to the very few, and that is that's a bit scary as well.

Henrik Göthberg: 2:16:36

And then and then we can all agree upon it. All accelerations like this will have a bumpy ride for society.

Mathias Sundin: 2:16:41

Yes, absolutely. That's something also we can see and also it's it's not it's not automatic. It it doesn't just happen because you know, um, we think it will, we will have to do that, and that's why I want so that's why I want so many people to be part of this because I'm also concerned with not from a sort of a societal level, um exactly based on that, but I'm concerned that a few people uh have too much power. Um much influence on the story here. Yeah, and and yeah, exactly, and and different kinds of but economical power and and different kind of power and the con you know the the empire that is under uh Mark Zuckerberg, he seems okay, but you know, he's no no matter how okay you are as a person, you should have that much power, I think. So and that is a problem. But I think then at the same time, the tools these many of these people are making money from, at the same time, that's making us more powerful. Um, even though I don't have uh you know as much money as Mark Zagbert.

Henrik Göthberg: 2:17:44

I think it's a good uh summary. It it makes some it it makes some people very, very powerful here in a way that could be troublesome, but it's also making all of us more powerful if we choose to tap into it.

Mathias Sundin: 2:17:57

Exactly. That's why it's so important to be part of this. Otherwise they will gain more and more power. And suddenly and and eventually that will turn bad. Because if you concentrate power, that's the worst thing in the world wherever it is concentrated. Trevor Burrus, Jr.

Henrik Göthberg: 2:18:11

The the fundamental idea of democracy is to engage and participate. Yeah. So that's what we're talking about.

Mathias Sundin: 2:18:19

That's the same thing here, yeah, exactly. And but in this case, not just go vote or be engaged with policy. You should absolutely be that. But now we have these tools that we could do so much more with. We could become more powerful if we use these tools, and that's what I want people to do.

Henrik Göthberg: 2:18:33

But then we understand them better, and therefore we can voice our concerns or what AI for good. If you don't like it, if you don't have a clue what it is, how can we define AI for good even? Exactly.

Mathias Sundin: 2:18:44

If you don't like the uh social media empire of Mark Zuckerberg, you actually have a better chance than ever before to create something other. You know, it's not easy, of course, you know, but the chance is better than ever before.

Anders Arpteg: 2:18:57

But I think you know, to end on a positive note here, I think what Sam Altman is saying that there has never been a better time to create a startup or to actually you know take actions today. And if you actually do make use of AI, you have better opportunities than ever possible before, right now. And the worst thing I I think you can do is is to try to avoid using AI. Would you agree with that?

Mathias Sundin: 2:19:18

Yes, absolutely. Just sitting here around waiting, or you know, um that's the really bad outcome.

Henrik Göthberg: 2:19:25

If just a few people then they then they get the power that we didn't want them to get. Yeah. Because we let them. Yeah, exactly. So so it's really up to us to up to us.

Anders Arpteg: 2:19:34

Yep, cool. Thank you so much, Matthias Sundin, for coming here. So many interesting discussions, and I have a lot more uh topics that I love to to hear from you. So I hope you can stay on for some after after work uh discussions as well. But thank you so much for coming here.

Mathias Sundin: 2:19:49

Thank you very much for having me. Thank you so much. So much fun.