E186 - AI Love Stories - Beatrice Bushati & Rebecca Oskarsson

Join us for the Season 12 finale of the AIAW Podcast as we welcome Beatrice Bushati, COO and Co-Founder, and Rebecca Oskarsson, Co-Founder and CTO of Pirr, for a fascinating conversation about one of AI’s most unexpected applications: love stories, emotions, and interactive storytelling. We explore how Pirr is building an AI-native platform where users co-create personalized romantic narratives, why some of the most engaging AI experiences may be driven by imagination rather than productivity, and what the rise of emotional AI means for creativity, entertainment, and human connection. From AI-generated chemistry and storytelling to the future of authorship, relationships, and emotional intelligence, this episode examines how artificial intelligence may reshape not only how we work, but also how we feel, create, and connect with one another. 

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

Transcript

Compute Scarcity And Token Pricing

Anders Arpteg 0:11

And I think that's kind of interesting. For one, Anthropic is heavily limited by the compute that they have. You know, that they are limited in terms of how much they can sell, but they can't even sell enough because they don't have the compute. And um and and in my view at least, I think the data center need is something that they will see and they need to invest in. What do you think? Can that be a reason for it?

Beatrice Bushati 0:38

Well, I'm I'm also thinking that when there's scarcity like I bottleneck, usually prices go up. I mean we we've all talked about the cost of tokens going down over time, uh, and there's the fear that tokens cost will go up. And definitely that could I mean there is one scenario, at least, or a couple of scenarios where that becomes uh a reality. Um so like either they could also like if you don't have the data centers, you know that your prices or your margins can be squeezed. Uh so that could definitely be a reason. And at the same time, we're seeing DeepSeek, uh, they are guaranteeing very low price for their models. Uh and yeah, you have to compete.

Rebecca Oskarsson 1:26

Yeah, and and I guess they they know that they're not as good. So they will need need that advantage, the price advantage. Yeah, yeah.

Anders Arpteg 1:34

Now I'm gonna it's uh such a big day tomorrow, then, when uh SpaceX, I guess, will do the IPO and uh for 1.75 trillion. And it's nothing even close to that before in history for for evaluation in terms of a company. And then uh as you said, as you said, you know, but then we also have Anthropic do an IPO soon, but we don't know exactly when, but they file for it and so have OpenAI. And now like three major AI companies are doing IPOs at the same time. It's it certainly points in some kind of great need for capital here.

Beatrice Bushati 2:10

And also a great need for Elon and his bonus.

Anders Arpteg 2:15

For sure, and it's a good uh way there, I guess, in some way.

Beatrice Bushati 2:18

I think uh it was was it three trillion uh that he will get as a bonus when we have a 10 million population on on Mars, something like that. That's one of the KPIs.

Anders Arpteg 2:30

It's such a crazy scale for for everything.

Beatrice Bushati 2:32

Yeah, we love aggressive KPIs, so it's just yeah, and I guess you you have some here as well.

Anders Arpteg 2:38

Exactly. Anyway, I I guess you are excited about uh the IPO tomorrow as well, and we'll be following that, right?

Beatrice Bushati 2:46

Yeah, I think we should take a day off from the engine watch.

Anders Arpteg 2:49

It's like the same when they do the Starship launch. This is like uh yeah the the IPO launch, uh, and we'll see what happens then. Are you going to invest in it or are you are you into investing as well yourself?

Beatrice Bushati 3:02

I I used to work as an investor, so I did I did that for many years, and now I'm investing in in PIR in our company, so everything from time to capital. So that takes up so much of one's time. Um and also, but on the other hand, people say like don't put too many, many eggs in in the same basket. So uh maybe I will buy some uh some more and some energy stocks or you know something else completely different. Oh, okay. How are you gonna invest?

Anders Arpteg 3:34

Yeah, for sure. But you know, Tesla has been going up and down now for some time, but uh it's it's interesting what happens with you know, we'll see what Terrafab will do with SpaceX and everything. But then it's going to be such a volatile period, I think, you know, after the IPO and then it's gonna be up and down, and then we'll see what happens. But the long term Yeah, but but still the valuation is so high, so it's I don't know, interesting times.

Rebecca Oskarsson 3:58

Yeah, I hope. Is it better to wait? Like is it gonna be two hypes tomorrow, yeah, tomorrow, obviously.

Anders Arpteg 4:04

But uh I think you know, since they're only selling off like five percent as well of SpaceX, it could be an opportunity for some investors to sell off, you know, whilst it goes public. So it could actually cause a yeah, uh a drop like in the price for who knows.

Rebecca Oskarsson 4:20

Interesting.

IPO Mania And Capital Hunger

Rebecca Oskarsson 4:22

Yeah.

Anders Arpteg 4:22

Well, with that, so interesting, and uh very welcome here, uh Rebecca Oskar, co-founder of PIR and CTO of PIR. And we also have uh Beatrice, sorry, uh Bushaki uh Bushati, yes, Albanian, also like Mira from Right, Mira Murat.

Beatrice Bushati 4:43

Yeah, exactly.

Anders Arpteg 4:43

Yeah, amazing. Uh CEO of PIR, and uh we would love to hear more about the story of PIR and uh how amazing we done you have uh with building that company and using the latest AI for love stories, right? Yes. But before that, perhaps we can have a quick um background about you too. If we start with with you, Rebecca, uh who are really Rebecca? How did you come to start Peer?

Rebecca Oskarsson 5:10

Uh oh, I guess I don't know where where to start. Like are we maybe from the beginning because I just shared uh the other week um a picture of my uh my grandmother and my mother. Um my grandmother, she is from a small village in northern Sweden, uh, where it's not more than a hundred people uh living, and that's where I grew up as well, uhmaness. And uh she actually brought internet to this small city, a small or small village.

Anders Arpteg 5:39

What year was that approximately?

Rebecca Oskarsson 5:41

Uh yeah, I mean she started the company with like data education, like going around giving courses about data. And this was you know, late 80s, early 90s. Um, so I think uh she probably had internet, you know, yeah, late 90s, I would guess. Yeah, it was it was really early. I don't need to ask her the exact uh exact year. Um, and I think that um you know it's always been there, the the interest for tech and and SATA DR tech. Um, so it was an easy choice when going into studies and yeah, academia to go to uh yeah, engineering. Uh so that's that's my background. And uh then I have um uh worked uh with SaaS companies and uh uh partly you know a lot of data, like data management, data um analytics. Um but then when I I met with um our third co-founder Anna and and Beatrice and got presented with this uh fantastic uh idea of this was before Chat GPT hit, so uh it was just not well known what generative AI like yes, the experts knew uh and knew where it was going, but but just seeing um this rather new you know new technology and where it could could go, it was just amazing.

Anders Arpteg 7:12

And like I what year was that approximately when you were thinking about this?

Rebecca Oskarsson 7:16

Um yeah, this must have been 2020, I think, right? Yeah, yeah, yeah.

Anders Arpteg 7:22

It's after Chat GPT launched.

Beatrice Bushati 7:23

Actually before.

Anders Arpteg 7:24

Yeah.

Rebecca Oskarsson 7:25

Yeah, because ChatGPT launched in November. Exactly. Yeah. And uh yeah, so I mean it was it it was an easy choice, even though going from something secure to startup is always always been a dream of mine, like my grandmother, you know, running your own business. And uh so that was dream come true.

Anders Arpteg 7:43

But let's hold off of the actual origin of peer.

Rebecca Oskarsson 7:47

Yes.

Anders Arpteg 7:47

But before that, then um you also worked at some companies or yes.

Rebecca Oskarsson 7:52

So so yeah, the the the main player would be Visma, the Big S company. Yeah. Um and um yeah, you uh you you asked who who I am. So I think that and now I guess I am so much peer, so that's why we always tend to go there. Um but um what more do you want to know?

Anders Arpteg 8:11

Like any personal interest besides you know the engineering and uh anything else that's exactly when I'm not by my computer.

Rebecca Oskarsson 8:17

So uh of course, as a romance fan, it's always you know listening to to a good book. Uh and uh romanticy is the favorite genre right now, so fantasy romance.

Anders Arpteg 8:27

Has it always been a passion of yours?

Rebecca Oskarsson 8:30

Um yeah, uh reading, I have had like a break from reading because I've just you know I I also have uh horses, which is also passion, so so that's uh something that's you know competing with with time a bit. But um growing up, it was, you know, I read so many, you know, then it was books about horses and horse girls, and now uh that's you know Twilight and uh Hunger Games and and those popular series. Uh and then now yeah, the romance genre is just growing, so there's so many good series and books to read.

Anders Arpteg 9:04

I also had horses when I grew up.

Rebecca Oskarsson 9:06

Oh, that's lovely. Yeah, yeah.

Beatrice Bushati 9:07

It's actually a peer trait to be a horse person. So a lot of there's been a lot of horse people in interesting connected to peer.

Anders Arpteg 9:14

Perhaps it's an opportunity for me as well.

Rebecca Oskarsson 9:16

Yeah, I mean, there are research. I mean, there is research on like horse, uh horse girls or or people, I guess, uh working with horses. Uh, you know, you you are learning leadership from from the get-go, and you know, needing to be clear in communications and and uh knowing what you want to do, and then give this instruction to a horse. Like there's so many things that you can bring with you from so it's not a coincidence.

Anders Arpteg 9:43

You learn some leadership skills, yes, you do. And Beatrice, please, who are you?

Beatrice Bushati 9:49

Yeah, I was also extremely passionate about technology very early on. I didn't have the grandmother, unfortunately, but uh I had an Eastern European mother that was uh bringing me cheap tech, and she was bringing me video games, um, fake Nintendo video games from Poland, a very crappy 386 computer in the 90s, and like my biggest goal was just to hack the very poor graphic, like the memory in uh in the computer to be able to play games like Quake and Doom. And this was like I was relentless in this, so I was bringing different boys from class to help me to help me change uh or to upgrade the computer, and unfortunately, it actually never worked, so it like feels like a huge failure to me, but it's great memories.

Rebecca Oskarsson 10:47

What if you had Claude? What if I had Claude?

Beatrice Bushati 10:51

I think I also understood that we'll come come back to the story, like the later story of my life, but also that I understood like okay, you need you need a little bit of money to buy to buy good tech.

Anders Arpteg 11:04

Yeah, um did you play I played a lot of Quake, you know, in my days. Uh one of the only games I really played a lot. But did you also compete in it?

Beatrice Bushati 11:13

I actually competed in chess. So also an Eastern European thing. So I was doing chess, I was doing gaming, um, I was playing the piano, I was doing everything that an Eastern European perfectionist child should do. And the last thing is that I started coding. So um I didn't we didn't have internet in my house at that point. So I was saving the code on my that I tried on my computer without any internet, took it to the school and tried it out, checked the bugs, and then took it back to my house. So I was always obsessed with technology and again being crazy about making it work. And I think that's where we are today. Um and but now we're making it work. Finally. Um, but I went uh coming back to the money, money issues that I decided that I want to do something really, really difficult and I want to earn a little bit of money, so I went into finance.

Anders Arpteg 12:08

Right.

Beatrice Bushati 12:09

Um and I became an investor, so I was doing that for more than 12 years, and my specialty was tech companies, of course. So everything from gaming to consumer tech companies, entertainment, of course, marketplaces. I was always obsessed about consumers and technology. Uh, and then peer came along. Um, I was about to actually launch a couple of tech funds at my previous job, but peer took took my my heart. And uh and just one day crossing the street to to my old job, I decided that I cannot cross the same streets once again. Uh I need to do this change, and I didn't regret it at any point in my life since then.

Anders Arpteg 12:54

Amazing. And and we have to go into this. So, how really did Peer get started? Can you

Early Experiments With Story Models

Anders Arpteg 13:00

just try to explain?

Beatrice Bushati 13:02

It was uh so it was our third co-founder, she's the original founder, the OG. Uh, she actually read an article, this was in 2020, that okay, AI can actually start writing a couple of sentences. I think back then it if it was GPT-2 or yeah, I think it was, yeah, right. Uh, and she thought, like, I want to do something with this, and she wanted to do something for for women. Uh and romance was happening because this was during COVID times, and also back in the days, also, we were more focused on erotica. So, this was we wanted to create uh erotica for women, uh, to improve um women's health and also fantasize. Um, and she put those two ideas together and she came to me. And at the first meeting, you knew each other from before somehow? Through an equ through a friend of mine. Uh, and at the first meeting, I said I could quit my job for this. Uh, because at that point, this was 2020, um, the idea came along. Um, you had the Netflixes, uh, HBOs, etc. There was this huge peak in spending on media content creation. Um, and I've never seen anything like that in terms of like interests were were at zero, and you could spend as much money as as possible. So I thought, okay, but with AI, of course, you're gonna be able to do it so much more cheaply.

Anders Arpteg 14:33

Yeah.

Beatrice Bushati 14:33

Um, which today feels extremely obvious, but in 2020 that was not the case. Um, so we actually she got uh a third person uh in who is uh to help us, who is called Magnus Solgrian. Um that is an AI celebrity. Um perhaps he's been here at the podcast.

Anders Arpteg 14:54

Not in the podcast, but we know him very well.

Beatrice Bushati 14:56

So yes. And he uh he was helping us. So at the first meeting, he said, Can I say we? Let's be a team. Um and he actually asked his team to help us with train and build the first models. So we were actually secretly training them on some places where spicy content should not always go, but it went and usually actually spicy content moves the internet, right? And tech development. So so um so it was extremely interesting.

Anders Arpteg 15:31

Uh can you if you just take one of the first models you try started to train, can you give some uh tech uh details about it or remember what it was?

Rebecca Oskarsson 15:42

Uh yeah, it was uh a GPT-based model, yeah.

Anders Arpteg 15:47

Uh as far as I think you mean from an existing GPT model?

Rebecca Oskarsson 15:51

Uh we tried both, uh, but we did train from scratch as well. Uh but I don't recall exactly the architecture on that one.

Beatrice Bushati 15:59

We did use uh Luther AI at some point, which was like completely, I think, random group or rather.

Rebecca Oskarsson 16:05

The researchers, yeah. Researchers very early. There they had like a GPT-j model. Yeah.

Beatrice Bushati 16:12

Uh so like the it was not coherent, you could say that. But sometimes we just felt a bit of magic, and we felt like okay, it it can actually write a couple of sentences, which just feel amazing, and then it went back to writing gibberish, so it was like a fever dream.

Rebecca Oskarsson 16:28

So it's started off, but then moved into yeah, yeah, kind of, and you you had to just be no, you know, open-minded, and then it was a lot of fun.

Anders Arpteg 16:37

So you had some starting models to start from DPT, or it could be a Luther, um, and uh and you trained from scratch. But what about the data? Where how did you get the data for this uh to start with?

Beatrice Bushati 16:49

Um well, it was um of course we can go into we were looking in a lot of open source stories that are published online, so um and because there is like the thing also with which we'll get back to to our type of content is that there's a lot of hobby creators that are sharing their stories uh online. Um but back then also like there was a lot of manual labor to uh label the data, to remove parts in the data, cleaning, yeah, cleaning it up and so on. So it was hyper manual back in the days.

Anders Arpteg 17:27

Um yeah, and perhaps we need to just clarify what is really the mission of peer?

Beatrice Bushati 17:33

Um the mission of peer is for everyone to be able to tell tell their story. And so we are a storytelling platform, so you co-create together with our AI. Uh we are very creator focused, so we want everyone to we want to give everyone a voice. Uh we focus today on romance, uh so that's uh stories which have like the majority, it's about the love story. Um but over time we want to do all types of all types of stories.

Anders Arpteg 18:07

So okay, you elaborate a bit more. So they uh the first version of uh of the peer platform or product or what you'd like to call it, the app MVP.

Rebecca Oskarsson 18:18

MVP.

Anders Arpteg 18:18

How did that work?

Rebecca Oskarsson 18:20

Yeah, the first MVP so uh yeah, the first one that we that we released uh to openly was uh you could create by adding what we call snippets. So you got a random intro, you could select from I think it was three or four random intros uh that was uh uh generated,

MVP Design And Latency Workarounds

Rebecca Oskarsson 18:40

and then uh you got the next paragraph suggested, and you could you could you know swipe between selections and go with that. And one of our uh USP features back then was actually like uh broadcasting this live. So you could have other people watching as you created your story, and it's just it's just just a fun idea, but people absolutely ate it up, like they loved it. To see others to be able to swim between them and yeah, being able to read while other were creating and also sharing while you were creating. I think that there was something exciting about that, knowing that I mean that's why you know we're sending live today, like there is something with this live audience and feeling that from you know, people could hit hit like and and um and you can see how many was watching, and yeah.

Anders Arpteg 19:28

But the selections then were they pre-programmed, so to speak?

Rebecca Oskarsson 19:31

No, so they were generated on the spot. So back then it was really slow. So this was uh one of the biggest challenges we had back then was the speed um and the latency of the models. And uh it was, you know, it was fun because like we were working with cloud providers that you know they suggested to use you know their automated and you know managed uh cloud infrastructure, and we were like, it doesn't work. Like your like your VRAM is yeah, it's too slow or it's too small. Like we have bigger models than that. And now they're so tiny compared to the big models today. But back then it was, you know, they were they were large uh in what what the cloud providers knew. Um so that was a big challenge to find the right the right uh infrastructure to run these models. Um but it was I uh I learned a lot and like the whole, yeah, we learned a lot, the whole team. And uh um we actually figured out that we could it was cheaper and faster to like pre-generate. So instead of like if you had like three or four options to choose from, behind that was already generated the three or four options. So we had like a you know a tree of text paragraphs that were presented. But they were all generated step by step. So this was when the latency was you know a minute long. Like that doesn't happen today. Like you, you know, you expect like a second at the most before you know she's something should start happening.

Anders Arpteg 21:03

Yeah, it's like the uh beam search uh technology here about that. That's like how the first trans um uh transformer worked, you know, it actually generated like three different person or versions in parallel, right?

Beatrice Bushati 21:16

Yeah, that sounds exactly like P. Yeah.

Anders Arpteg 21:19

Yeah, yeah, no, nice.

Beatrice Bushati 21:20

But one of the first um a very fun pitch moment, like we released our first version two weeks before ChatGPT, uh, and we went to slush. Oh yeah, the huge this huge conference in in Finland, and we we we wanted to showcase the app to the investors, and investors, I would say 99% did not understand what we were doing. Uh, of course, consumers like users did not understand like AI. What is even that? That was not on anyone's lips. But we were showcasing the app that you could also actually, besides the live function, you could co create with another user. So you and me, we could co create a Story together. And we had an intern sitting in um behind some wall, co-creating with each and every one of our investors. Oh, really? Yes.

Rebecca Oskarsson 22:13

And I remember this was before we launched because we didn't have real users in the app at all at that time. And we were the whole team were creating to keep the live feed. We had like a live feed with a bunch of stories, and we were all creating like while you were. You were like, we're going into an investor meeting now. Please create. And we were like, okay, like let's let's generate stories.

Anders Arpteg 22:32

Cool.

Beatrice Bushati 22:33

So it was five people on the platform. Yeah.

Anders Arpteg 22:38

Amazing. And how okay, so you you got started in somehow. Did you start by your own and putting some money behind it and then build the first product? Or how was the first like uh couple of months and years when you started?

Beatrice Bushati 22:52

That's a good question because since we were working with something at that point, something extremely innovative, uh, we actually got uh grants from the Swedish government. So uh Vinova grants, we're hyper thankful for those. And we got several tranches of that, so that took us through like the first one, one and a half, maybe even two years. Um, again, since this was super early. And during this time, it was very difficult to to recruit as well. So, because there was very few with the experience. So it took us some time to build build a team and find people like find people like Rebecca. Uh, but after that, um we went into getting funding from from investors.

Anders Arpteg 23:36

I mean, I guess it's always always tough in the beginning of a startup as well. It's hard to really withdraw a salary or anything in the beginning, but uh but you managed to get through that period in some way, right?

Beatrice Bushati 23:46

We managed. Uh, but of course, yeah, it's uh it it's been quite many months without salaries. But it's always uh it's always worth it.

Anders Arpteg 23:56

Yeah. I mean, uh great to see. Um okay, so what happened then? And um if you just take it from the first couple of months and when you actually start to see some traction, you know, how did that go about?

Beatrice Bushati 24:09

Well, Chat

Organic Growth After ChatGPT

Beatrice Bushati 24:10

GPT happened.

Anders Arpteg 24:10

Oh yeah, of course.

Beatrice Bushati 24:13

We were um we were spread in the like peer was spread in different communities, like everything from uh like I think was it Discord at that time and 4chan and uh different websites listing peer, different communities listing peer. Yeah, yeah, and it absolutely exploded. So uh we just like at that point we were thinking like how will we even even go from like five to ten users to twenty years to hundreds, and then all of a sudden we were getting many, many thousands of users every month, and it was all organic, so um there was huge interest. Um, and we scaled that to almost half a million downloads globally, yeah. Uh but at that point it was still quite spicy. Spicy, but only spicy in the sense that it was at that point was it was female-led erotica. But what happened was that we actually got a lot of men on the platform. Um, and there's nothing we wanted to create a service that worked for everyone, of course, it doesn't matter what gender or sexuality you had. Um, but men are also always very forward in technology, and where women were more afraid of AI. Uh, at that point, like we're there's still a lot of fear about AI, but at that point it was other types of fear because completely new. So we got a lot of men on the platform uh and uh a lot of downloads, a lot of activity, but we understood that it became a bit too spicy, so a bit too much erotica. So we understood that it will be difficult to scale this further, like not scale from 500,000 to 1 million, but scale to 10 million, scale to 100 million, because then you want to be able to advertise.

Anders Arpteg 26:02

You want to be able to get real you didn't do an advertising at that time?

Beatrice Bushati 26:06

No, and we still haven't. Um we have done a small batches of tests, but still not the real single dollar into into advertising. So everything has been organic. But we wanted to scale, we wanted to be able to distribute it properly, get like the the big the big funding. Um and at some point peer might also go public. So right.

Anders Arpteg 26:32

Yeah, speaking about the big uh SpaceX and peer going one trillion dollars as well.

Beatrice Bushati 26:36

Exactly.

Anders Arpteg 26:39

Okay, and and perhaps you can speak a bit about where are you today, you know, in terms of number of users and and what the functionality has led you to, and how is appear different today compared to when you get started?

Beatrice Bushati 26:52

Yeah, so we we made um what's what startups call

Pivot From Erotica To Romance

Beatrice Bushati 26:57

a pivot. So we pivoted from from erotica to romance, and it sounds like it's a small difference, but it's actually a big difference when it comes to the actual content and what type of audience that you want.

Rebecca Oskarsson 27:09

Yeah, exactly.

Beatrice Bushati 27:10

And uh yeah, on the tech side, I mean, Rebecca, you you were doing uh changes in the AI. Of course, yeah. Um yeah, do you want to add it?

Rebecca Oskarsson 27:18

And yeah, I mean, uh we have some core features that are very similar uh to what we had back then. So we have stayed true to some parts, and I think that that's uh maybe we're gonna get into that so you can yeah, stop me if we should maybe pick it up later. But that would be that uh keeping the human in the loop. So we are we have never been about you know writing one prompt and getting a story. Like that's not what we're about. We want to uh we want to support the creativity in creating a story. And we want to be able to, that's also something that we're probably gonna go into later about uh being able to provide like putting a copyright on a story, like on something that you create, because it's not just the AI, it is you creating piece by piece in the story and controlling the narrative as you create it, and that's something that we kept. Yeah, and that's something that we kept from the get-go. Yeah, uh, which was actually first a kind of a limitation of the early models where they couldn't output much more than a few sentences, but then we have kept that and working with that more and more. Exactly. Um, and people like that. Like that's what our users uh are are you know always you know very um protective about. Um yeah.

Anders Arpteg 28:37

Okay, so less of erotica and more into romance in some way. How did you make that switch? Was it more the data that changed a bit, or was it also some other changes?

Rebecca Oskarsson 28:48

Yeah, exactly. It was it's it's a lot about the the underlying uh data and how we and the branding as well. Like, I mean the models can still write both. And as Beatrice said, like they it's it might sound like a small pivot because uh a lot of the components are the same, but the difference between let's talk about like difference between erotica and like a spicy romance novel is how the the story supports the spicy parts in a romance. Like it's not just about that spicy scene or the explicit scene in a romance story, it's about everything before and everything after, like what uh these how these characters grow, how they fall in love, how they maybe hate each other at first, and then they realize that they're in love. All those things are uh crucial to a Roman story. Meanwhile, erotica is not does not need that uh supportive, necessarily the supportive romance.

Beatrice Bushati 29:48

Yeah, it's uh 80-20 type of rule. So erotica is 80% the spice and 20% buildup, and in romance for a spicy romance, it's the other way around. So it's 20% spice, yeah.

Tropes And The Happily Ever After

Anders Arpteg 30:01

But okay, so I guess the data is one part, but I guess you can also change the system prompts or yeah, and that that we have.

Rebecca Oskarsson 30:08

We have like specialized it towards the the you know romance, uh the romance genre and what's you know the the tropes as they as it's called, different like um specific because the romance the romance reader is it's a it's a special special breed, uh they know what they want and they want to know these tropes before they start reading a story. They want to know what is tropes, the tropes, for example, it's enemies to lovers, then you know that oh, the characters are gonna be enemies first and then they're gonna fall in love. So or you have friends to lovers, uh, or you have one bed trope. So there's just gonna be one bed when they maybe they go on a trip and oh whoops, it's just one bed. How it was finished here. So that are kind of tropes, and and you can get really into those tropes. I I really love enemies to lovers, so then you want to read that.

Beatrice Bushati 31:02

Um over and over again, yeah. And you can also combine them, so it could be like enemies to lovers, there's one bed, and hockey romance, like super popular right now, exactly. So it could be on a hockey trip, yeah. Uh X, Enemies to Lovers, X, and then it could be also another trope like Grumpy X sunshine. So one person is grumpy, the other one is is happy.

Anders Arpteg 31:25

So it's like a library of tropes that you're using, and yeah, you can see what people like and absolutely, yeah, exactly.

Rebecca Oskarsson 31:33

And and since since that's so common in the Romance uh Romans community, uh people know, you know, they want to create their story based on on certain tropes. So that's how we have we have uh taken in more of the romance community into the app uh to shape it towards what they like.

Beatrice Bushati 31:51

And when you have a user that really knows what they like, it's like perfect for perfect for us, perfect for the AI. I mean, they give us the prompt, we know how to build it up. And there's actually one thing that every because 90% of romance is consumed by women. Uh and there's one thing that every woman wants. Do you know what that would be? Or every romance reader wants?

Anders Arpteg 32:16

I guess I can't do that.

Beatrice Bushati 32:17

That's a three question.

Anders Arpteg 32:19

Um no, sorry, yeah. Uh I don't know. Okay. Okay, please tell me.

Beatrice Bushati 32:27

Yeah, the good excuse, yeah. Yeah. Uh it's the happily ever after.

Anders Arpteg 32:32

Happily ever after, of course.

Rebecca Oskarsson 32:34

Or happy for now is also one. Yeah. But this happy, the happy ending.

Anders Arpteg 32:40

Happy ending in double cells. Yeah, okay.

Beatrice Bushati 32:42

Yeah, exactly. So women, like, they know exactly what they want, and like with romance, you you know that it can be a super tough story. Like they can break up, fall in love, move, never see each other. People are getting killed. Exactly, that as well, it's super dramatic. But then you always can finish off with uh happily ever after aura. He A, as it's called in uh in the romance world, so the acronym. But the good thing to add on uh on the pivot is that we were extremely good at the spicy part because that's what we have been doing for several years, uh, and everything about the prompting and exactly what should happen when. So we could reuse that part. So, like this erotica part or the spicy part, we could put in what was 80% before, we made it into 20%, and then we added added the character buildups and the narratives arcs and so on. And a couple of years ago, this was not possible, like AI couldn't couldn't support the long narrative format. Like you could do five minutes, 10 minute stories, 15 minutes. Uh, but now you can really push push the limits in the story arcs and adding to memory and uh remembering what your character went through.

Rebecca Oskarsson 33:55

I mean, the context windows weren't large enough even a couple of years ago.

Anders Arpteg 33:59

But given these troops, um, before it was swiping, you know, which one you liked, is that still the main functionality, or can you actually give prompts or hints?

Rebecca Oskarsson 34:08

Yeah, you can exactly you can instruct the model uh as you what you want to happen next, and uh if you want something to yeah, take a certain route. Uh, but you are able to just click your way and create a story as well. Like you do not need to we give suggestions on how you could like continue and and all that, but we have also uh included a lot of elements where you can uh write yourself or push the story more and control the story because that's something that was is is important for for our users as well. Um, they can some are very particular in what what they how they want it to go, and then they need that.

Anders Arpteg 34:49

Okay, so but how how do you build up? I mean, I guess a lot of the IP, so to speak, in peer is a lot about understanding what the right probes tropes are and how to do the prompting properly in the data that you have. I guess you're building most of the data yourself now more and more.

Rebecca Oskarsson 35:04

Absolutely, yeah.

Anders Arpteg 35:05

So you you can actually use the data you use in the app as training data today?

Rebecca Oskarsson 35:09

Yeah, yeah, we can.

Beatrice Bushati 35:11

And and also on the data, I mean, from from one story we get 14 times more input from a user than so. If you create a story in peer versus you go to ChatGPT and create a story, our users are so much into so much more control, give us so much more user input. So we get a lot of more insights from our users than if you would prompt a chapter on a story at ChatGPT, and that we can do a lot of a lot of things with. Um, of course, we can personalize and we can remember, uh, and we can create like the we can improve the service all the time. And and also we you could think of peer as the Instagram moment, what Instagram did for photos in your telephone that were just sitting there, nothing was happening with them, um, and then all of a sudden Instagram appeared, and it was super easy to to share your photos with your friends. Like this is what Peer is doing to storytelling. The thing is just storytelling is much more complicated. Uh, and but now we can actually do it for our users. And when Instagram started, um, there was a predecessor. Um, was it hypso-matic or I forgot now what it was called, but it was like ultra complicated. It was super complicated, it had too many functions, it was not not very easy to use, too, way too complicated up. Um, they were making money, but they they were not growing. And then InstaGame came with a couple of couple of filters and super easy sharing. Uh, and that is when this um this tagline was uh established, uh like come for the tools stay for the network.

Anders Arpteg 37:01

Right. The network is tensor. Okay, amazing.

Model Strategy Prompting Over Fine Tuning

Anders Arpteg 37:05

And and uh for the tech-interested people here, I mean you started off with using kind of GPT models and even training from scratch, etc. What do you use today? Anything if you don't want to share it, that's just let me know. But what can you share about the current tech stack, so to speak?

Beatrice Bushati 37:20

Well, we were we were using um the thing is uh when when we started uh GPT2, we could fine-tune and uh fine-tune with spicy, spicy content. We were using open uh open source models as well, like because they were allowing for for the spicy part, uh, but that has been a restriction, so and that is still a restriction with the ChatGPT.

Anders Arpteg 37:44

Um they said that they would allow so the guardrails currently in ChatGPT, etc., is too strict to exactly, exactly.

Beatrice Bushati 37:52

So so we are using different models and we're combining them uh from different parts in the app.

Rebecca Oskarsson 37:57

Exactly, yeah. I think that's that's a good way to put it.

Anders Arpteg 38:00

Um and you are still fine-tuning and uh not anymore.

Beatrice Bushati 38:05

Okay, uh not anymore. Um now it's much more about prompting and extracting, extracting the things that we want from the models. And it could just be like a couple of uh a couple of words or like one word can change can change the results or the output.

Anders Arpteg 38:25

I mean I I guess the expertise in how to properly prompt it and do the context engineering, so to speak, is getting increasingly important.

Rebecca Oskarsson 38:31

Right.

Anders Arpteg 38:33

And and also if we just think about that, you know, why we don't need perhaps to fine-tune as much as we did before. I guess before it they were so stupid, so you needed to to fine-tune it, but now they're getting so intelligent and general general in some sense, right?

Rebecca Oskarsson 38:48

Yeah, I mean it's it's hard to like it it's um it's a re there's a reason why you have the big players dominating the market. I mean, there's a reason for that. And that reason is it's uh you need extreme resources. Yeah. And and it's just I think it's it would be uh naive of us to think that we could, you know, if if we're gonna focus on providing the best uh Roman story creation experience and you know going beyond romance at a certain point, we cannot focus on building foundational models from scratch. No, no.

Beatrice Bushati 39:28

No, you can you cannot compete with that. Uh but what we can compete on is again extracting the right type of output from the AI models. And when we go on, like we we do extremely big amounts of research uh and user interviews and so on. And when we go on Reddit, which is a big there are big communities in writing and also in romance uh contents, um, there's a lot of people that are complaining about uh the output from OpenAI or from Claude, etc.

Rebecca Oskarsson 40:03

Generic, yeah.

Beatrice Bushati 40:04

Yeah, very generic, and uh like it's it's not on par with what you want. So we have we have a huge edge there because we're hyper focused on a vertical, on a huge vertical yeah.

Anders Arpteg 40:17

And you didn't mention which your top favorite model is today. And and if you don't want to say it, I I completely understand. But I think you know, is there something you can mention about that? What what is what is working for you and not?

Rebecca Oskarsson 40:30

Yeah, I mean it's I'm gonna give the boring answer that it depends on on what you want to write, and of course what you want to do with the model. So that's that's what you see in. So you might think that, oh, let's grab the latest model that has you know the best benchmarks. And if you look at the benchmarks, there's you know, there are some that measure creative writing, but it's not it's not always the best indicator either. It's usually like you know, LLM as a judge metrics when it comes to creative writing, um, which is then it's difficult to get it right. So I yeah, I can say like it's it's not it's not always the the biggest and the latest model that's the best. Um and uh of course there is a subjectivity to storytelling and what you prefer, like you have a writing style, a stylistic that you that you might prefer above uh something else. Um yeah, I don't know if if you want to say something more about that, Beatrice.

Beatrice Bushati 41:28

Or I'm thinking for myself, like the biggest unlock for me. I mean, of course, we've used AI, we're AI native from day one, but the biggest unlock for me that looks at product and looks at strategy, uh I'm obsessed with what are users like. I think how can we create the best type of experience for them? And then we need to look at the data. And before, I would say like a couple of years ago, we we did not understand how important very granular data was. And also it's it was expensive and slow to extract it, like in the sense that you need a data analyst and you need to like and now I can slice and dice the data so much more. And personally, this is the biggest unlock for me.

Anders Arpteg 42:16

I

Hiring For Flexible Minds

Anders Arpteg 42:17

mean that's a super interesting topic, and perhaps we can talk a bit more about that. And and you also had to build up a team uh at Peer, etc., and you have to hire people, engineers, and analysts, etc. And also I really love that you are a female-based company and are building up these in a technical area like this. So I really respect all that. But perhaps we can speak about that. You know, what do you think about you know when you are hiring people now? What are you looking for? What type of traits and how do you go about recruiting at peer?

Rebecca Oskarsson 42:47

Yeah. Uh so the most important um, like if you if you talk technical skills, it's more and more important, and I guess maybe everything everyone is saying this, but you must be able to understand the requirements of a technical product, be able to express them and communicate them because that's what you're doing to the AI model today, like the the the coding agents. Uh so you have to have that fundamental understanding of both the the technical aspects, uh, both of the product that you're working with, like the domain knowledge of this product and of the user, of course, because you you need to identify like, is this the best for the user if I create it this way? Um but then also something that you always want to look at, like as at peer. You know, we we always you know love if someone is is a reader or if they have you know some kind of connection to it, doesn't have to be romance. And one of our developers he loves sci-fi and reading a lot of sci-fi. So I mean that that also creates a connection to to literature and understanding of of storytelling. And so that's something that's always like a plus. It's not always a must be a expertise. Yeah, it's always yeah, it's always valuable. Um, and then always, you know, someone who is willing to, you know, kind of stay like test trying new things because that's what the world we live in now, that something that you're using today is is, you know, going to be a hundred times better next week. And that's why you have to have people that are willing to do that, to not be stuck in their old ways, but trying new tools, new, more efficient ways of doing it, being interested in, you know, why is this happening this way? Can I do it faster? Can I do it better? Um, so that that's more like personality traits that I'm looking for, like that drive to always improve, always become better. And I think that's uh necessary to have fun at a startup. Right, right. And Beatrice, I think you probably have maybe some other things that you would look for. Yeah, yeah, of course.

Beatrice Bushati 44:50

It depends on who you're hiring or what role. But I I love uh uh following up with you, Seth Rebecca, I love a flexible brain. So when you can when you can twist and turn things and be flexible in your thinking, uh, and I think uh it's it's not an easy task. Um and uh as a but as a startup, you are thrown so many challenges at the same time, and you need to learn to be flexible and find solutions to them because you're extinguishing so many fires and you need to be creative and be willing, willing to change, willing to experiment. And of course, the classic is to just make mistakes.

Anders Arpteg 45:33

Right.

Beatrice Bushati 45:34

Uh, this is something that I learned the hard way because I'm a perfectionist, again, Eastern European, you don't you don't fail with anything. So um learning that through the startup way, it's extremely empowering that you know that nine arrow things, ten things are gonna fail.

Anders Arpteg 45:53

It wasn't that Elon Musk that said basically that um he basically had this 820 rule, meaning that he makes a decision even though you only have like 80% of the information, meaning like 20% of the decisions is maybe wrong. Even to the point that if you're not making 20% of decisions that are wrong, you're moving too slow. Meaning there is a value in itself in making actually like 20% of bad decisions. Would you agree with that?

Beatrice Bushati 46:20

Or yeah, I think so. I think you just want to move forward. You want to move forward, you you need to learn quickly as well. Um and be very perceptive.

Rebecca Oskarsson 46:29

Yeah, and something we work with a lot is of course, like if you there's also I don't know where it comes from, but it's also saying that if you um if you release your product when it's perfect or that you're happy with it, then you release too late. Right. Yeah. So I mean that's also kind of the same thinking here. Um yeah, so yeah, don't be afraid to fail. How how do you do it?

Anders Arpteg 46:52

Yeah, but I I like what you were speaking there about you know not only the hard skill, of course, but the soft skills. And I think you know the the ability to communicate is actually one of the things that I look a lot for. And so many people are bad in communicating. It's it's scary when you think of it sometimes. And if you just ask someone to you know explain something complicated or or being able to express it in an efficient way, most people will fail. So I think that's almost like a way to see also for one, do they understand complicated things and know something? But if they don't have the ability to express it, it doesn't really matter if they know it or not. I mean you still have to be able to communicate about it because that's you know what's necessary to make a team work. So I really like what you see saying that as well. And being able to communicate and express you know these kind of things.

Rebecca Oskarsson 47:44

I think the best people that I've hired are the ones that early on in like the first weeks they are asking so many questions. Yeah, you know, this almost that I get tired, like like oh my gosh, she or he is writing again. But but I love that because that that shows that they they they expose what they don't know and they are willing to find out.

Beatrice Bushati 48:06

Yeah, and you need to create a culture for that. Uh it's not like I think it's a difficult task to create a culture where like please go fail, like do it, uh, because that's not the nature of it.

Anders Arpteg 48:20

I mean, it was you know, like Doni Lick, you know, I worked in Spotify for a long while back, and he he said something uh very interesting, I think, uh, which is Spotify is going to be the company that fails the fastest. And when you first hear it, it it sounds really stupid, but when you think about it, it's actually really profound. It is the ability to quickly make failures and uh and accept it and fail fast. That is super important, especially for startups, I think.

Rebecca Oskarsson 48:48

That's interesting. A lot of uh writers and authors are saying the same thing about their first draft. Like write your fast first draft very quickly, don't go back and you know correct mistakes or anything, just just write it right and then you can go back and and you know find those plot holes and fix it. But but just getting getting it out.

Anders Arpteg 49:08

I mean super interesting.

Guardrails Moderation And Dark Romance

Anders Arpteg 49:10

And also if we move into this kind of topic, it and and you know you have to build models that are going to do things that are on the edge sometimes on what's ethically like sound, and and then it could go over the edge, and perhaps you were a bit too erotic in the beginning and then a more romantic. How do you uh build that in? I mean, if we just compare it to the latest type of AI models of today, anthropic missiles and cybersecurity problems, and if you don't put the right guard rails in, then it did it's it's doing a lot of stupid things. So they have to like program the model to to to not do the wrong things and not being able to abuse it. What are your thinking here? How do you like measure if it's going too erotic or too too off the guardrail, so to speak? Do you have some framework for that? Or how do you work with this kind of balance of how to put the proper guardrails in place?

Beatrice Bushati 50:09

I mean, we were ultra strict with the guardrails in the beginning. I mean, and back then it was very clunky, like you just have to forbid certain words, and back then it was like if you went into a cloud of certain words, then we knew, like, okay, now you're crossing into a territory that we don't want. And that's frankly the the problem also with working with erotica, uh, is that you get into like the ethical discussion about it because it's always like fantasizing that's on the edge of things very often, and you want to create a a product which allows for that but still has guardrails, right? Uh which is a very rare case, like that's not how people operate in general, like uh the rare, like it's you want to create productivity or whatever, and then someone goes a bit, someone goes cuckoo and tries to build something that you don't want to. But for us, we were like very we had to live like closer to the edge more often and to understand like what is what is the uh importance to create like these fantasies that people want. But it also it's difficult to work with like moral, like or ethical um subjects, right? Um but Rebecca is like the big master in uh in the moderator uh on what we allow allow and not. But the good thing I will I will finish is that like we do have good uh already moderator filters built in into some of these models as well today.

Anders Arpteg 51:38

Can you elaborate a bit more how it works? I mean, one of course would be simply that you put it in a system prompt and says, you know, don't use these kind of topics or something. In other way, it could be that you have some kind of uh harnesses or cardrails looking at you know whatever the users are prompting it to do or what the output is, etc. Can you yeah?

Rebecca Oskarsson 51:57

So since since we have uh people can write themselves as well, so it's not just about what's what they are prompting to the AI or what they are or the AI outputting, but it's also about what the user adds to the story in clean text. So that or pure text. And that's why it's important for us to have like ex like external, we cannot rely on the uh the LLM, the generative AI model to not generate those type kind of topics. We do need that external uh which which consists of machine learning models today and a separate model in some yeah, exactly. So it's a separate model, it's cheap and fast and uh somewhat accurate. I think that we can, you know, that's some one of our things in our product roadmap is to uh improve those uh those uh moderating filters um to be more uh precise and yeah and accurate.

Beatrice Bushati 52:49

Um but it can be like even in in romance, like for example, like coming back to tropes, a very popular one is called dark romance, and that is when you have stalkers, and like that's more like a oh, I don't know why is this guy stalking me, but I kind of fall in love with him. But in romance, like the nice ethical twist is that it actually ends well, so the guy becomes goes from red flag bad guy to a good guy, right? Uh and then we're all good. But but sometimes when I've been like another uh dark romance is mafia uh topics. So you have this mafia boss, uh, usually like uh it's very popular in New York, for example, because New York is more of a dark city, and they see the kind of differences around the world, indeed, yes. So it's actually hyper local what different cities like in New York dark romance is huge. So they love, I guess, sopranos, they love the mafia, they love some dark topics. And I was creating myself um a dark romance, mafia romance in in Pyrrh. Uh, this was a a little bit a while back, but I was I had some people dying actually because they were shooting, and it was also a little bit spicy, and then like the AI did not understand like is this like where's the romance, where's the violence, what is okay, and not. Uh, and also like Romeo and Juliet, you have like people die, actually, and sometimes that's difficult to handle. Right.

Anders Arpteg 54:23

Okay, but are there some what what are the guards rates or what are the limits?

Rebecca Oskarsson 54:29

I mean, to try that. I mean, I mean, I can think a bottom line is that like so so one thing it's it's it's not illegal to write about illegal things. Like you can always write about it, but we have drawn the line where if it's I think kind of yeah, I think there are exceptions, but if it's illegal to do, it's it's not okay to write about it in PID.

Beatrice Bushati 54:52

Like we we prefer to have hard regardings than not. Yeah, yeah.

Rebecca Oskarsson 54:56

So so than a loser.

Beatrice Bushati 54:59

Like if I would if if I'm like in an uh like in an ethical dilemma, like should I should I allow this or not?

Anders Arpteg 55:05

Okay, it's like a safer rather than sorry. Exactly.

Beatrice Bushati 55:08

And we we believe that for women that's that's also very important because it's one thing that maybe you cannot write a certain thing that maybe you want to push towards, but rather you don't want as a woman content that would not be comfortable.

Rebecca Oskarsson 55:23

It can be so off-putting too. Exactly, yes, yeah.

Anders Arpteg 55:27

I mean, some concepts or are super you know difficult, like suicide or something, you know. If you get into this kind of really if you have a person or a user that is sad because they lost their husband or whatnot, uh, and and then you know it could push people in certain directions. And we know OpenAI got sued for for actually having a person speaking to to Chat TPT and then they committed suicide. I mean, horribly bad, of course. So it is difficult topics, right?

Rebecca Oskarsson 55:55

Um and it's so many nuances as well, in in especially in you know, you're you're writing fiction, right? So and it is so much about how you write about it. I mean, you can write about a murder uh as a murder mystery, or or you know, it's it's something that that happened and it's sad, and you know, you try to solve this murder, or you write it and it's gory and it's you know, it's you, you know, you kind of you know, yeah. It's it's not it's it's nuances and and you need to determine like okay, is this how the user writing about this, is that okay? Or is it not okay? It could be the same topic, the same words almost. Uh exactly. So that's the difficulty, and that's why we we decided to to uh train specific uh models for that.

Anders Arpteg 56:41

Um so it's um LM as a judge in some way, but yes, yeah.

Beatrice Bushati 56:46

And I also do we read a lot uh ourselves as well. So every story that is published, we read Yeah, we didn't tell yeah, exactly.

Rebecca Oskarsson 56:53

We have human filters as well.

Anders Arpteg 56:55

Okay, I mean that's okay.

Beatrice Bushati 56:57

Human Yeah, and our users judge as well. Yeah, we our users can also report stories that they don't they can like stories, but can also report. Uh so we have several layers of filters. I mean, we will always be developing this over time, uh, but again, it's been super important from us for us from day one.

Anders Arpteg 57:16

Some

Can AI Understand Feelings

Anders Arpteg 57:17

people are saying sometimes that AI could never understand morals or AI could never understand you know the ethical dilemmas of what's right or wrong in some way. But what do you think about that? Is can AI do that? Or what's your thing?

Rebecca Oskarsson 57:33

I guess we need to define understanding here. Because I mean, can can an AI model understand really? I mean, for for me at least, that would be like a fundamental definition for like what what are we talking about when we say understanding? Because it's still you know repeating patterns that it's has seen before.

Anders Arpteg 57:51

What do you think? Can AI understand?

Rebecca Oskarsson 57:53

Uh not today. No. Okay. It can I guess interpret and you know, um and in that sense it can seem like it's understanding. Yeah. But but no, I wouldn't say that it can um yeah.

Beatrice Bushati 58:12

Sometimes I think like, does it matter? Uh because probably not. But like we we were talking about like what will today, what will the future of AI be with like AGI, like what's the next big thing? And Rebecca mentioned like it's gonna be physical AI, like the robots, etc. And we were talking, do you want the physical AI that can understand your feelings or not? And I said, like, no, and then and I said yes. And uh one of my favorite examples or uh yeah, that I think about quite often come back to quite often is that uh quite a while back, I don't know if it was 30 or 40 years ago, there was this AI experiment. I'm sure you know about it, where you had like a kind of AI that was acting like a shrink. Yeah, maybe it was where it was just repeating. Yeah, it's a lysa. Yeah, I I just love that, and people just loved it because you just repeat whatever that's like. And so for me, that's that's good enough, actually. If if the consumer or the the counterparty feels understood, yeah, if it looks like a duck, if it quacks like a duck, yeah, you know.

Anders Arpteg 59:24

Yeah, it can be good enough.

Rebecca Oskarsson 59:26

But what about you? Would you like like if you talk about robotics, like do you want to have like if you have uh yeah, help in in like a housekeeper or something helping you out at home? Would you like uh that robot to understand and your feelings?

Anders Arpteg 59:41

I would, I think. Yeah, um, yes. I I have a hard time to seeing why what the bad side would be. What would it why do wouldn't you like that?

Rebecca Oskarsson 59:52

I first I like I did I didn't want to like hum humanize it because you know it's uh it yeah, like it could feel wrong to ask it for to, you know, I don't know, do my my services that I could do myself. Like that that's my immediate feeling. But then then I quite quickly turned around when we were talking about this earlier. That but I would love it if it was like, oh, I see that yourself. Would you like a cup of tea? And like, you know, take care of it. Then I I would absolutely love that. It's your love language, exactly right.

Anders Arpteg 1:00:23

The robots we are so as humans, so easy to humanize you know technology, even like a twister or your car, you know. Sometimes, oh, the car is feeling sad, it's such a cold winter night, right? You humanize it even though you know it's not a human, but you still connect feelings to it in some way. And I guess isn't that in some way you can just view feelings as a way to just you know understand what some complicated system works like, even though it doesn't have feelings in the same way.

Beatrice Bushati 1:00:55

I mean, what uh uh another example that I like is like how upset a lot of uh OpenAI's users were when they took away the the very friendly chatty version. Was it when GT3 was replaced, right? Or maybe No, I think it was later. Exactly. Yeah, no, sorry, it was later, but like a lot of people just became upset because they didn't get this personal touch. Yeah, uh and it it matters for people a lot, and actually, but it doesn't matter that it's not uh human that understands you, yeah. Uh it's enough, and and also physical, physical AI or or robots, they can help. Like we always talk about like it can make you lonely, and I agree. Like, I think there's a divide, and there's gonna be many different good ways and bad ways. Uh, but another nice example is when you use robots in elderly health care or elderly homes.

Anders Arpteg 1:01:51

Exactly. I was getting to that. That's a great point. I mean, in Japan, for example, they're very much into robotics for elderly care, etc. And if you just take an example, you have a robot dog, for example. You could easily see that a dog, a proper dog, so to speak, could be uh a great companion for someone that's lonely and they can have some kind of you know relationship with that. Um would it be okay to have a like a robotic robotic dog, right? If it would create some kind of you know less loneliness, would that be okay? What do you think about that?

Rebecca Oskarsson 1:02:30

Yeah, I I think that would be okay. Uh I think that there's a lot of benefits to that, just thinking about like animal well, you know, welfare. But but I I have I have a personal hard time to see that I would you know would go like I I I would feel that that connection to a robotic dog. But if someone would do that, I wouldn't, you know, I wouldn't stop them. I wouldn't say that that's wrong. Of course, of course you can you can do that if if it comforts you and you know makes you feel better. Because uh I would totally understand that that a real dog would do that. I mean, I have horses, you know, of course they make me feel good, and I'm I'm happy with it when I'm with them and everything. So so uh if you can get that from a robot, why not?

Anders Arpteg 1:03:14

Because in some way, I mean Pierre is giving like an uh robotic kind of companion that you can speak with, right? You could have the same kind of collaboration or communication with a human where you interact with different ways to to have the story, but now you can do that through machine in some sense, right?

Beatrice Bushati 1:03:35

Well, there's a big there's actually one big difference, is uh there's a lot of uh AI chatbots which we are not uh like AI chatbots as in AI girlfriends or AI boyfriends and AI companions. Um that's a big market, and there's a lot of players in there, but that is less I would say healthy than creating a story. Um yeah, and consuming romance is actually there's a lot of research done on not only reading because reading is in general good for your health, but also romance is very good, is a good way to to relax, to escape, and it has positive health benefits. And I think it's amazing to create a service that is actually good for you, yeah. Um, and for chatbots or girlfriends, boyfriends like that can become more lonely. Um, because I I guess you could easily do a pivot and Twitter in in Peter where it goes to the girlfriend, boyfriend kind of case, but you wouldn't do that, or no, I mean what is important is characters, so you want to have uh uh you want to give the users the opportunity to create rich characters and remember what those characters go through. Um, and in one way it can be a bit of a relationship with a character, of course. Like there's you can fall in love with a character, you can fall in love with a character on in a movie or in a book. Yes, um we do a lot, right? Yeah, exactly, over and over again, and but uh but it's still there's a difference when there's a story, and that's more about escapism and also creating like you can create for yourself, but you can also create for others, so that's a social experiment, experience, and storytelling that is a social that's one of the most basic human needs, like uh consuming stories, listening to other people creating stories and creating yourself. So that's what we're doing.

Stockholm’s AI Boom And Ecosystem

Anders Arpteg 1:05:40

It's time for AI News brought to you by AI8W podcast. So we usually have this kind of small break in the middle of the podcast to just um share some ideas about recent AI news, and then we get back to the podcast with Rebecca and Beatrice. Um but to start with that, do you have any AI news that you heard about recently that you'd like to share?

Beatrice Bushati 1:06:05

Yeah, peer. But uh on AI news, I think what's what's extremely interesting is the boom that is happening in Stockholm. Right. So many investors are coming here from the US. It's really fascinating. And of course, it's amazing for us and the ecosystem.

Anders Arpteg 1:06:23

And that is interesting, right? I mean, it is one of the top uh please disagree if you don't like, but it's one of the top European at least AI hubs that we have in Europe, right? Would you agree?

Beatrice Bushati 1:06:34

Definitely, definitely.

Anders Arpteg 1:06:35

London, of course, is big and Paris is big, but still, you know.

Beatrice Bushati 1:06:38

No, it's it's it's happening for sure. And um we we see like why combinator was here for a month ago to recruit Swedish startups uh to their program. Like I'm a person I love, I I want to I want to keep a lot of stuff European if if if that allows. Um but they were looking for Swedish startups and people were flying in from I was meeting people at the event from uh Finland, from Spain, like they were all over the place coming to Stockholm, and so it's really exploding right now.

Anders Arpteg 1:07:10

What do you think that is? What is what is making the Silicon Valhalla explode?

Rebecca Oskarsson 1:07:15

Is it a support system that we you know many people know each other and get uh can get that yeah, the support from uh from a strong community in some sense? Yeah, community, yeah.

Beatrice Bushati 1:07:27

Yeah, there's a lot of events happening, and like when I think Swedes are amazing, like we have an amazing way of growing up. Uh that we are allowed to be creative, we're allowed to be entrepreneurs, uh, and Swedes are as you know, being at Spotify, we're extremely good at smooth, sleek user design experiences. And when you have um when you have that combined with with AI, I think it's like it's like the perfect spot for things to happen.

Rebecca Oskarsson 1:07:58

And maybe the the like we have, you know, we have safety nets in Sweden. You can you can take that risk of leaving your job, your day job, and and and going after your dream. Like it's it's actually possible, and you're not you're not necessarily gonna end up you know on on the streets because of that, but you you have that opportunity, yeah.

Anders Arpteg 1:08:18

We have a strong safety net for sure, yeah. And strong universities and great education and people coming out there, and and and rather good capital, given what you just said, right?

Beatrice Bushati 1:08:28

Yeah, it's happening, uh it's happening right now. So um 90% of our all AI deals have been done in in the US, uh, but definitely capital is moving here, and we also see that of course it's going to be trickling down in the system because you have Swedish AI founders, they're making exits and they will be able to invest in, etc. So uh I definitely believe in the strength of the Swedish system.

Anders Arpteg 1:08:54

We have one of the biggest, I think, uh, number of unicorns per capita in in Sweden. So it's uh something we're doing right now. It's impressive. Yeah, cool. Um, another big event, uh, besides the big we already spoke about the IPO tomorrow, you know, SpaceX doing that, and then we have Anthropic do another IPO soon, and then OpenEI is going to do that. That's going to be super you know exciting to watch and see what happens there. Uh and you you with a background in finance as well. It's it's great to hear your thoughts there as well. Um, so that's going to be a big thing. Perhaps one other thing is um yeah, they released uh Fable 5 or the Mythos 5, and and uh it would be fun to hear what you're thinking here if if that's something you would like to use or not going forward. But but just to give some background here, uh we had a big myth about mythos, or you know, Anthropic released or created this model called Mythos uh that wasn't released before, so they had a preview version of that that was very, very powerful, but too powerful potentially to yeah, to release. So because of cybersecurity reason, they didn't release it, and then they uh had a glass wing project where they let some of the biggest uh software companies um use it internally first to patch the security vulner vulnerabilities before actually releasing it. But now they are releasing it. And um they are having like two versions, one is the more original mythos, which is the you know, not so many guardrails in, but that's one is not being publicly released, and then they have the other one now released called Fable 5, which is uh heavily guard railed, uh and uh that one is now getting released, so we can start to use it, and I guess it will be also for Peer to start using it as well, but also super expensive. So it's going to be doubly expensive than what Alpus is, the previous most expensive and biggest model. So this one is twice as expensive, probably many, many trillions of parameters, if not like you know, five to ten ten trillion parameters. Or um, I actually have uh my own theory here saying it may not be uh simply the number of parameters uh that is big in mythos. I think it could potentially be uh that they have uh these kind of internal uh loops, meaning uh they actually do not do a single forward pass, but they actually stop in the middle of the layer and do like 16 uh rows in the middle in the semantic space. So there was this open mythos repo that someone tried to reverse engineer what the mythos did, and they came to the conclusion it must be doing this kind of latent space reasoning thing. And that of course would also be more expensive because it spends a lot more time for each token. Meaning that could be the reason it's double expensive. Anyway, I'm just guessing here, but I I think it could be that. But uh in impressive anyway, that uh you know Anthropic has been doing so well, they have been leading for a long time when it comes to most performant coding AI tool, and most performant also in terms of our genetic workflows uh which they're releasing. So they're doing something right in Anthropic. Yeah, and uh they also surpassed OpenAI in uh in having a bigger revenue while it's still just being a few years old, it's it's uh super fast growing here. So impressive in many ways, uh, but also kind of scary that AI Moles get so powerful they have to have so much card rates that they have to wait before they release it.

Mythos Fable And The Future Of Safety

Anders Arpteg 1:12:30

So, what do you think about this? Is this the future we will continue to see that AI Moles gets too powerful? So we need to have so much security to prevent cybersecurity or people creating biological warfare, you know, weapons with it, or whatnot.

Beatrice Bushati 1:12:47

I mean, if if I start, I'm I mean, we're the guardrail, guardrail girls.

Anders Arpteg 1:12:53

That would be a t-shirt. That would be great.

Beatrice Bushati 1:12:57

And we're also European, so like I believe in a certain a certain path of regulation, a certain part of guardrails, and like it's uh comparing uh open AI together with with Anthropic, I find it great that Anthropic is pushing the guardrails as well, and like that's that's the European thing, yeah, the European culture. Um so I think it's great that they're doing that. And uh regarding like you were talking about the revenues and the ARR of Anthropic versus uh OpenAI, and uh as a startup, like investors always want to see some type of exponential growth. Like if you're linear, you have linear growth, then you are boring, and they just want you to do um exponential. But Anthropic actually very recently had linear growth uh up until they released Claude Code, and that's when things really took off. So, like considering the size, it took them quite a while to like to get to that crazy inflection point. Um, and we were checking the bench benchmarks earlier today of um of the new model, uh, and I think it was around like 10 to 20 percentage points better in terms of coding. I mean, we'll have to test it on complex complex builds, but it's exciting for sure.

Rebecca Oskarsson 1:14:20

Yeah. Yeah, so I I recently read a report from Anthropic as well on how how uh they have uh uh optimized their development with with AI, and and you could see just uh yeah, the difference when they uh went from I think it was Opus 4. Was it 4.6 uh and then to the Mythos preview that they had internally at Anthropic and just the uh incredible like four open-ended problems and like big complex problems that that was like just huge difference with the model with with Mythos, and that's gonna be interesting to see now how because I think that there are problems that don't need that power, of course. Like it's it's probably gonna be people who are you know using it for things that it's just not necessary. Um and of course, it's also interesting with yeah, making it available for for everyone. Like it's not with those prices, of course, it's not gonna be available to to everyone. Um, I think Anthropic did exactly the right thing in in not releasing it a right way. I think it was uh very responsible because just imagine their the growth they could have got uh gained if they released it. Yes. Um so I think that that was, you know, um I was very glad reading that. Uh now it was a couple of months ago. Uh so it's it's interesting to see how, yeah, are these guard rates going to be a problem or not? And and um I would love to hear more and maybe some expert in what those guard rates are and what it could mean for uh yeah, democratizing uh or the democracies in in like is is it is it hindering anything for us? Like how is it censoring or or what's what's the risks to society?

Beatrice Bushati 1:16:23

What are we removing in the guard rates, you mean?

Rebecca Oskarsson 1:16:25

Exactly. Or or is it like is it only good? I'm guessing there and and I think it's gonna be also a while just as any very powerful tool or weapon. Um there's gonna be a while before we know what it can do and feel secure in that and have kind of built the infrastructure around it. Yeah. And I just I'm just thinking, you know, with the internet. I mean, of course, there were you know major security vulnerabilities everywhere because it was just those, you know, a sudden change, and you didn't really know and what was you know the uh what was dangerous and what was not. And and I think we're there now as well. Like we need to we need to just uh figure this out as we go. What do you think?

Anders Arpteg 1:17:14

It's so many interesting topics here. One is you know, should we continue to do open source or not? Open weight models, for example. What's the consequence of that? When it comes to mythos and fable, I mean, it's nice to see this kind of split now in one with little guard rails and with a one with a lot of guard rates. Will this be a trend now? So everyone is starting to do this kind of split where they have a limited release of something that has less guard rates. I think that will happen. And uh, I think also you know, uh even Trump have said, you know, they want to have a 30-day uh review period of any kind of big model, so to see if they it's safe or not, for example, or it could be for other purposes, but I'm not going there. Um and given how powerful AI becomes, the risk of abuse will of course just increase more and more and more. So it seems to be a trend going towards, you know, we cannot open source too much without having proper guard rates. If you do open source, it's so easy to remove the guard rates. So what what will that mean, right? So then it means that the super powerful models probably will have a super hard time to be open sourced. Even if it's not open sourced, it still will be separate releases here, with one that is you know more safe, but then very restricted, even saying as soon as you're prompt it or ask anything about biological weapons, it will just not reply. But there could be very good use cases for I want to find a vaccine for the coronavirus, but then it's just going to say sorry, I can't do that. It's like they hell 9,000 here is sorry, Dave. I can't do that. But but that may be the price that we have to pay. Because anything that is going to be publicly released rather be safe than sorry. Similar to what you're saying yourself, right? You'd rather be safe than sorry for the way that you're building the service.

Rebecca Oskarsson 1:19:16

And give that um first prong in Swedish, the give the head start, yeah. Give a head start to the good guys, to to setting up uh the security that's necessary for when this like like Anthropic did, like, okay, let's fix this fix all these bugs first, yeah, and then we can release it.

Beatrice Bushati 1:19:38

But also HAL went rogue, right? So it did.

Anders Arpteg 1:19:41

It didn't go well there. No, and I think that's so important then that we we can go into more philosophical topics in the end, but this is I think a great topic to to talk more about, and we are going to get more much more philosophical in the end here. So that's let's go that soon. Anything you would like to add, Goran? Or should you know? Yeah, yeah, yeah. No, no, Mike. Good. AinRite. So AynRite, I'm just repeating what you're saying here so people can hear it. Okay, Ainrod went uh IPO and they went uh unicorn and it was a spec, was it not? Or was it was it um amazing, but let's get back uh to here uh and speak more about that. And

New Entertainment Formats And Vertical Drama

Anders Arpteg 1:20:31

um if we just speak about, you know, it is a female focused, you know, audience or consumers that you do have. And uh for one, we can speak about the the entertainment business as such and and book reading or storytelling in general, and then how does that relate to what we're seeing in Hollywood or with the movies, etc.? Because yeah, can you just elaborate? What are you seeing in terms of what the trends are moving towards?

Beatrice Bushati 1:20:59

I mean, in in Hollywood content creation, I mean they are embracing AI. Uh, so you see some AI companies, some big AI companies, they are making uh taking advisors, like the biggest directors are becoming advisors, and like the most influential people are actually leaning into AI. But I was speaking to an American VC yesterday that is within the entertainment that was telling me, like, yeah, they are, but still very still old school, to speak. But they're trying to fight to keep the old yeah, other like otherwise, I don't know what's gonna happen with um like they want to fight it, but they're also trying to lean in, but it's also becomes like asymmetric in a way, like how much are you leaning in in what way? Like, um, I was uh having lunch today with a big gaming entrepreneur here in in Sweden, and they were looking at like what is happening in the gaming sector. And like he sees that a lot of a lot of investors are looking for like a one-size-fits-all or like a tool that can do everything for them. And he has a big gaming studio with a lot of developers, so they do very intricate video games. And but he he wanted to have like, oh, we want this tool to do this thing in particular in our game, and etc. So I think the market is still extremely young in entertainment, as in like getting eyeballs or like competing with Netflixes. You have the vertical dramas which are becoming extremely popular.

Anders Arpteg 1:22:30

So, vertical dramas, but what is that?

Beatrice Bushati 1:22:32

That is like TikTok style. Uh either it's human gener, like with human actors, but now it's becoming more and more AI generated, it's very popular in China. Uh, the latest trend was like having cats fighting each other. It was dramas, but they were in clothes, and uh it's a specific, like very niche in a way, but it's a huge niche. And um, what they've been able to prove is that they can definitely monetize that type of content. Um, and you pay like per episode, and very every episode is a couple of minutes, so it's very TikTok style type of entertainment, right? And Netflix is following this, they are re-remodeling their mobile app, so it's going to become much more vertical entertainment, like much more leaning towards into this. Uh, but it still has not really, it's it's taken some market in in the US or in the Western world, but not as much as in in the East. Um, so there's a lot of different things going on, but it's still fairly early.

Anders Arpteg 1:23:36

But couldn't you say that peer is is taking the book industry in some way and transforming that a bit to be more of an interactive book, right?

Beatrice Bushati 1:23:44

100%, yeah. Yeah.

Anders Arpteg 1:23:46

And wouldn't you think could could peer be moving to interactive movies at some point?

Rebecca Oskarsson 1:23:51

I I like, yeah, this is uh a great question. So um yes and no. So how we see it is how Peer can this is your pitch usually, Beatrice, but you can you can jump in, is that we can become an an IP engine for the future, future movies and future series. So that building the scripts for yeah, because what we see now, which is uh huge. I don't know if you maybe you've heard about Heated Riverly or or off campus now. It's these uh big like romance books and series that have become series on uh prime video. I think Amazon has been very quickly picking this up. You have in Netflix, you have uh Bridgeton, and you also have uh one of the absolutely biggest fantasy series, romantic series right now that's uh about to you know do casting so they are in you know production already or soon. Um and the Empyrean series by Rebecca Yarros. But that is what we see, like the books moving into Hollywood. So how and and that's where we see that it is uh a lot of uh possible potential for for um yeah, monetizing and monetizing for our creators. Like we can we can take you there, like we can we can be that passage into um yeah, into selling royalties and selling the the corporate.

Anders Arpteg 1:25:17

But given how you know the latest license Gemini, Omni, etc., it can do video generation surprisingly well, and uh it can very quickly, or not super quickly, but in in minutes at least, generate a really nice video, and then you could potentially tweak the story going forward. Wouldn't that be an interesting path to take? Or what do you think?

Beatrice Bushati 1:25:41

I mean, we we're always like as a startup, we want to be hyper-focused on what we're doing today. Yeah uh and uh and that is storytelling in in words. I mean, we've been experimenting with sounds, you have we have an audio audio shelf that we know a lot of our users love. Uh over time there could be a possibility to move into video as well. Um, but first, like our first most important task is just to nail storytelling. We want to give you the best story, yeah.

Peer As An IP Engine

Beatrice Bushati 1:26:12

And also, like I think about uh as an ex-investor, and uh knowing what what the investors want, what we want is is retention, and there we can do a lot with personalization. So the the things that we see that you do, or like the data that you give us, we can personalize your experience and you we can remember your story, like the characters, the worlds that you create. Uh, I think that there's a much, much deeper defensibility in that, either if it's video or not, but like just knowing all of these things versus vertical dramas that you just scroll, and you the input that you give is like okay, you can you continue watching this series, but that's it. Like the data is much shallower, and we want we're on the deep end, uh learning what's really what you want and serve super serving that. That makes sense, and yeah, on just on the book market, uh it's the publishing market, I would say they've done a lot of they've done innovations, but the big innovation was the printing press, like 500 years ago. And since then it's still very much passive consumption. So you know if 51% of all Americans wants to write a book, but actually basically no one does, and if you write it, like the the percentage of creators that get published, it's almost zero.

Anders Arpteg 1:27:45

But but can you be an author in some other way than in the future? Or what is if you okay, let me just put another crazy idea in your head here. It's actually partly from Spotify as well, and and uh you can see Spotify. For one, a platform for consumers, but it can also see it for creators. And uh in your case, you are having users that are creators in some sense, right? And then if you would like to support them to not only do it for fun, but potentially even create a business around that, could that be a future that you could go into to say that you become a platform for them to not be like kind of traditional static book, but they actually are the the best like interactive platform that people want to subscribe to in some way. Does it make sense?

Beatrice Bushati 1:28:36

Or at least being today we're hyper focused on the creators, uh, but that is the path to creating the marketplace for them. So that's the next step.

Anders Arpteg 1:28:50

So you're thinking about the marketplace, right?

Beatrice Bushati 1:28:52

Definitely we are. Uh, and we have investors that are very good in marketplace as well. So uh I'm very excited about like that opportunity for us to build and create. And the thing is, if you are an author today and you publish you actually manage to write a story. I mean, that's a huge hurdle in itself, but okay, you managed. Then you want to publish it, you go to to a book publisher first, and if you manage, they they plus the retail store will take up to 70%. Like you you can have maximum 30% of that. Uh if you publish it by yourself on Amazon, like you go on Kindle Kindle Direct Publishing, at best you can keep 70%, but that's the maximum. So what we want to build is to uh have a platform where the creators can keep much more of their content, and we believe that that also gives us the possibility to scale scale the platform quickly.

Anders Arpteg 1:29:55

I think so.

Creator Marketplace And Publishing Disruption

Anders Arpteg 1:29:56

I mean, and and also it's a bit sad about the the the normal authors in some sense because you know, why would you buy a static book that is you know two years old potentially and that is not interactive?

Rebecca Oskarsson 1:30:08

Exactly. And also also the all the all the people out there that that has the potential to write but haven't gotten there, maybe you know, this just life got in the way.

Anders Arpteg 1:30:20

Yeah.

Rebecca Oskarsson 1:30:20

And and those those people we are able to serve thanks to Genitive AI, uh, helping to get over the hump. And um it that's also something that we see that we're not we're not uh and and never have like aimed for the professional authors because they are already supported by the the industry. Like we're talking about all those people who are not in the industry and and can can get there so so easily uh with a little help.

Anders Arpteg 1:30:53

I mean, amazing. I mean I I think you know, as we are moving towards you know AI being able to do so many things, we we still want to enable humans to find new ways to change perhaps job roles into the future type of job roles. Um in in doing this and more directing what the agents are doing. I guess in a future, peer could be like an agent marketplace where you uh design what the agent should do. And then people can subscribe to it. It it seems like a an attractive future in some sense, right?

Beatrice Bushati 1:31:24

Yeah, I mean we're uh we want to put the creator in in the focus. So whatever whatever can help people to create and also to connect, like as in as in Spotify, you you want to connect the the right creators with the right consumers. Yes, um, and that is also a very important future step for us, uh because there's a lot of content being created. Uh, I would say a lot of it is AI slop, um, but there's also qualitative content being created. So a lot of people think that uh everything is AI slop, but if you look at the publishing industry, also established authors are actually producing much more content today than they did pre-Chat GPT.

Anders Arpteg 1:32:12

Yes.

Beatrice Bushati 1:32:13

So an 80% of writers they use AI, and um but there will be a huge amount of content out there. Uh, we believe that creators will become even more important. Um, that's super important to differentiate, I mean, between different types of content, but also like in our world where it's so important to get exactly what you want, you want to be able to be matched to the correct story, like the right story that you would love or that we know that you're gonna love, and also the creators that you would love to follow. Like 50% of all Reddit searches uh in the romance channels are about I want to find this specific book that it's gonna be about a a redhead and a hockey player, uh, but they become vampires and then they break up, and it's it's very, very, very specific, and again, we're here to super serve that.

Anders Arpteg 1:33:08

Interesting. But but thinking a bit longer ahead about Peter here, and Peter, of course, in the name is connected to romance in some sense, but still what can you see the future could hold? I mean, marketplace, of course, is one could be it could be expanding into other areas than romance potentially. I mean, it can be a lot of people.

Rebecca Oskarsson 1:33:29

Yeah, that that has been part of the plan as well for quite some time. That that romance is is one part of fiction and it is the uh the biggest uh fiction genre in in the Western world, and and being able to expand into other uh fiction genres is absolutely an opportunity. Uh not yet, but uh but for sure.

Anders Arpteg 1:33:50

Yeah, just fiction. I mean, just doing let's say an AI podcast. I mean, if you were to listen into an AI podcast and uh you wanted to steer it in some direction and having some kind of interactive generation, couldn't that also absolutely, yeah, yeah.

Beatrice Bushati 1:34:09

From our predecessors uh that are storytelling platforms, but that are not, they have nothing to do with AI. Um, 60% of their content is romance because that is the biggest genre, and the rest is sci-fi, can be thrillers, it can be fan fiction. Uh, fan fiction is also very close to romance as well. There's a lot of romance living in in fan fiction, and vice versa, uh, and fantasy, of course. So there are really adjacent topics that go together, and we are learning like step step by step, bit by bit. So definitely excited to to expand that over time.

Anders Arpteg 1:34:47

Awesome.

Control Versus Autoplay In Stories

Anders Arpteg 1:34:48

But if we just go back to the question about the you know, AI, does it really understand or not? For example, as I spoke about. Um, and then thinking, you know, in your area with romance, it's really a lot about feelings all the time. Still, AI is able to tell a story at least that is connected to feelings and surprisingly well. What's your thinking here as agents or AI gets increasingly good? Does that is that is that a bad thing or not? Um what do you think about AI getting potentially the one that actually creates the stories, the feelings, the the the experience that humans then start to consume? Sorry for the fluffy question, but I hope you get to get what I mean. Could it be a problem that AI is the one driving the entertainment business here?

Rebecca Oskarsson 1:35:47

I I haven't thought about it.

Beatrice Bushati 1:35:48

I don't have a clear like it depends on like how far maybe you go because we see it still as um as a tool, um, a very, very advanced tool. Um but we still see it as a way to to express something else or or like something new or to extract something that you have inside of you and to empower empower a person. Uh so that's all like positive positive feelings. And we were discussing today, like what um again, what do we think that AI is going to be able to create super like very good stories? I mean, on its own. I think yes. Um, I mean Rebecca, you you agree as well, right?

Rebecca Oskarsson 1:36:35

Of course, of course. I I think it's it's going that way. And and I think for me the the the difficulties in the difficulty in the question was how can can it hurt? Yeah, yeah. Can that be hurtful for the and yes for the industry, I guess, because I mean it's the classic, you know, people lose their jobs. Uh but but of course there's gonna be other other types of jobs, and we're still gonna be, as long as we are the consumers, we are still controlling the what what's being written, even if it's AI that's writing it.

Anders Arpteg 1:37:08

But but if AI gets much better than humans ever could in writing a good romantic story, is that a problem or not? Or is that just positive in the future?

Beatrice Bushati 1:37:21

Um I mean we have a two-side, like you can create for yourself, or you can experience experience a story that is interactive. Uh that is one way of the consumption today for our users. And the other side is like when you create and you have this ambition to to share your story. Uh we want both sides to be able to exist. Um personally, I truly believe that like the fandom, the creator behind it, and like I think that's it's very, very important that that's part of the IP itself. Um, so I think we'll be able to have both sides, like we'll be able to have both parallel lanes. Um but I mean a lot of fantastic things will be created with AI, and also like there will be a bit of shakedown as well.

Anders Arpteg 1:38:11

So but if I do an analogy to coding, for example, I think this is interesting here. A lot of people, or I I heard a smart person saying something like this, and and that is that we know AI is working really well for programming, yes? But that mainly means that AI is really good in writing the code. It doesn't really mean that AI knows what to do, it knows how to do it. So given that you have a human that tells the AI what to do, you can do amazing things. But if you don't have the human telling it what to do, it will do potentially crazy things. So in some sense, it seems like humans are having skills at some level that is different from what AI can do. So AI is more executing on what the instructions to it is doing rather than being the true innovator or true creator in some sense. Will that apply for romantic storytelling as well? That AI the human still is necessary to be there to tell you should go in this direction, but doesn't necessarily have to write the exact like wording for it.

Rebecca Oskarsson 1:39:14

Absolutely. I mean that's and that's kind of what I meant as well with it's the still comes consumers that's gonna, you know, control you. So so if you look at it a broader perspective, then I mean that's happening already. Like uh authors love usually their community, their readers, they want to talk to their readers, even if they're maybe they've already released their book series or whatever, but but it's still they're still taking influence from what's like what do people want to read. So so yes, that would be the same, even if the AI is the writer, the full writer in this. And and we we believe that as Beatrice said, like it's important with uh with the human in the loop here to control the store, because yeah, they're still saying like where to take it, right? And and something that I want to pick up on that that Beatrice was was talking about is the uh the feeling of um if it's accomplishment or um you know the being able to produce something as a human, like we're we're knitting our sweater, like we can we can knit something, a sweater or or a scarf. You know, you can still go out and buy a scarf, we can buy a sweater and and and I'm still maybe knitting it, you know, because I I like that process, or I'm doing a crossword puzzle. Like the crossword puzzle doesn't have you know a result or or the an output. It's it I don't, you know, an or an outcome. And still I do it.

Anders Arpteg 1:40:40

It's like riding a horse. You still do it. You don't need it for transport, really, right?

Rebecca Oskarsson 1:40:45

Kind of. And uh I would say that there is you know things to you know, you can compete and you get but uh but still like we have so many, like many users in the app today choose not to publish their story. Like they are not after, they're not after the likes, they're not after the the audience, they just love the process. So I still think that you know, even I understand that wasn't really your question, but like even if we would have we would go into a world where AI is creating um and writing, you know, the full story is still gonna be a human need with that storytelling process. Like, I mean, we have done that from you know the campfires, uh, where we are, you know, telling uh you know, telling people about our day or telling something that happened, like this something that we we have that need to tell the story or write it down or whatever it might be.

Anders Arpteg 1:41:38

So so you mean yeah, sorry.

Beatrice Bushati 1:41:39

No, no, I was thinking about like just AI and what does it make us? Like what's how does it enhance us? And uh Chat GPT has 900 plus million monthly active users. Uh and it's nine, so let's say it's one billion. It's like it's very soon it's gonna be one billion people. And let's say that one percent are extremely good at prompting, like it would be Rebecca would be extremely good at prompting, and a lot of people in Silicon Valley, and and here in Silicon Valhalla. Uh, but 99% like do average prompting. Um but that means that all of those 99% that use AI, they're still gonna have like the average output because the AI is gonna enhance you equally well.

Rebecca Oskarsson 1:42:30

Yeah.

Beatrice Bushati 1:42:31

So yes, we become better in like, oh yeah, me personally, like can do 20 extra things through AI, but you always need to be better than the average.

Anders Arpteg 1:42:42

Right.

Beatrice Bushati 1:42:42

Uh and that's gonna remain.

Anders Arpteg 1:42:45

So you as a human still always have an edge if you just know how to use AI properly in some sense.

Beatrice Bushati 1:42:50

I think any chill, right? Like if uh if we were thinking thinking about when coming back to the horses, uh, like okay, some uh some or or some type of uh innovation when it came, and then people started started using that, and some are better than then everyone is using cars, but some are just better at using it, or driving better, or um competing in something better. So, like we always want to be better than than the average, and that's the way it's always been.

Anders Arpteg 1:43:20

Yeah. So you get good at different things in some way, and that will continue to happen, even with AI becoming increasingly intelligent in some ways. What if it's it it does increase even more? What if you know not only AI can write the the lyrics or of the story, what if it actually is better uh than humans in in saying what should happen next? Will there be a will the humans still want to drive a story, or will it simply be to push the auto-play button, so to speak?

Rebecca Oskarsson 1:43:54

I think we're already there. Like, yeah, and and like because because it's because it's so subjective. Like if if I'm creating a story in peer because I'm not I'm not uh a professional writer, I can be happy with what the AI is outfitting. Yeah. But if but if I am a professional writer and you know I have explored this uh profession for years, you know, I have an expertise, then I'm gonna notice the nuances, I'm gonna notice that oh, this doesn't really fit well, and I'm gonna be very picky. So I think that in a way that many people are already there. Like you can just click and it's probably it's gonna be way better than if you wrote it yourself.

Anders Arpteg 1:44:34

So but okay, I did mean like if take it compared to music, for example. You know, a vision that Spotify had at one point was that the whole like app interface would simply be a play button. There is nothing else you can do. You just press play or not, and then it will always know exactly what type of music you will like to listen to. And it's no need to select artist or genre or nothing. That's of obviously not the case today, but but you know, it could be a vision in the future. And I if you think that for storytelling and and uh I still think there is a need for a lot of people to have control. Uh I think there is still a need that people would like to say, no, I I do want it to go this direction. I don't care really if AI believes this is the best way, but I think the the need for control in some way will still want or create the necessity of having the ability to choose direction. If you see what I mean. Do you would you agree with that? Or or do you think the vision should be an auto-play button and just you know let the story unfold?

Rebecca Oskarsson 1:45:42

We have played around with that idea. It was some time ago now, before I think we we realized how important control is. Like even if even if you, you know, it's enough that you think that you're choosing. Yeah. I mean, that's a good point. It's like, you know, with with Netflix, you know, we all have different, you know, feeds in Netflix, but uh but at least I get to choose, kind of. So so I think that uh yeah, maybe there's just uh illusional control. Could be, yeah, could be that that's enough. Um I think that's interesting. We talked about that, like there's uh you can say there's kind of two different readers um when it comes to yeah, and any book basically where some people are just in it for the ride. You know, you don't really like you maybe you read quite quickly and and you know you it don't matter if you remember something, but then you have the other person who is, you know, maybe reading the same page five times because they know that there might be a clue in here about what's going to happen later. And they understand know that this this particular author, she's really, really good at putting in these clues and and I really want to know it all. And maybe they, you know, if this is a fantasy book, you know, they read up on on you know the the world behind it and and all the characters and all the creatures in this book, and they want to know everything. So I think that that's also different in in Peter as well. Some people are very, very um detail oriented, and it has to be really you know perfect, and then you have the people who are just you know in it for the ride and just you know click away.

Beatrice Bushati 1:47:11

I I mean I'm just I just love entertainment in so many ways. So this is my favorite, I think my favorite topic. Um coming back to being obsessed about the creators, like how important that is. One example is Stranger Things, right? It became huge success on the movie, right?

Anders Arpteg 1:47:29

I'm not into it. It's yeah, series to the series, yeah.

Beatrice Bushati 1:47:32

It's sci-fi. Sci-fi, exactly. It's uh Spielberg meeting. Oh, sorry, yeah, it's uh immense success on Netflix. It's the Duffer brothers that created it. And uh in the last season, which was aired, the last episode was aired, I think it was a New Year's Eve, um, a couple of months ago. And after that, Netflix released an episode which was about the making of. Uh, and people were very upset about the ending. I mean, there's been huge fandom around this series, it's been going on for 10 plus years. It's a little bit open-ended ending. Yeah, it was kind of ambiguous, yeah. Yeah, and uh and then like the making off came out, and uh, it was about like the last season and how much more budget they had, like how much more amazing things they could do in this season. Like the amount of recording days was so many times like they had so many more recording days than they had in the first season. So this was like huge budget. And still people got extremely upset, like the fans became extremely upset. And why? It's because the creators had not figured out the ending when they started the last season. So, like what the creators want and what their vision is, and all of the details that Rebecca is is mentioning, it's extremely important. Like, and especially when you create fandoms, when you create these worlds, and it's like it's very important for us for us humans. We want to find meaning. Like, there is no meaning like that, you didn't have an ending, yeah, and it's not enough, it's not acceptable.

Rebecca Oskarsson 1:49:14

Because, yeah, because you have those fans that want to go back to the first season and understand why it ended the way it did, and when they get it, you know, in like okay, I can't even do that because the creators didn't do that, then you know it's it's like the whole world falls apart.

Beatrice Bushati 1:49:28

There's no yeah, no point. No, but we're still big fans, though.

Rebecca Oskarsson 1:49:33

And it was actually something that I read yesterday today about you know, uh from an author how important it is to know the ending of your book.

Beatrice Bushati 1:49:42

Uh yeah, like some some some consumers, like some readers, they just read the the end of the last pages. Yeah, they want to know that's how romance readers sometimes consume. Uh and we have uh extremely long session times. So when you look at creators, they spend huge amounts. Amount of time in that.

Anders Arpteg 1:50:01

What was the like average?

Beatrice Bushati 1:50:03

The average on an active day is one and a half hour. Really? For for women, yeah. They they create even longer. Like men are around one hour, women are around one and a half. And our subscribers, they spend three and a half hours on an active day. So it's a lot of time. Amazing.

Rebecca Oskarsson 1:50:20

And what about you? What type of uh reader are you or consumer? Are you in it for the ride or are you very, you know, want to know everything?

Anders Arpteg 1:50:29

I get very frustrated if I don't understand anything. So yes, I am very much into uh nagging in and trying to understand why stuff happens. So yeah. That would be annoying to me as well if if it ends a certain way. It's like a sense of anarchy when that's stopped in a very bad way. I was very upset, I can tell you. Yeah. So I wish we could have fear that simply were able to direct it into a different ending.

Rebecca Oskarsson 1:50:50

And that's that's one of the fun ideas we had at the start. Uh the you choose your own ending, or yeah. Uh like you could have you could have classics and and just you know write your own ending to them.

Anders Arpteg 1:51:03

Where

Where Peer Goes Next

Anders Arpteg 1:51:04

will uh fear be in 10 years?

Beatrice Bushati 1:51:08

Global. All languages. Uh we talked about maybe other media. And let's see if we go like it. Let's see if we go to video, uh, but audio for sure. And we want to be one of the we want to replace one of the big publishing houses.

Anders Arpteg 1:51:27

Replace publishing house. Okay. So okay.

Beatrice Bushati 1:51:30

So we go from the traditional book to the interactive book, and we empower the creators and we allow them to monetize much better than was previously possible.

Anders Arpteg 1:51:42

Could you see an exit in being acquired by publishing house? Or yeah.

Beatrice Bushati 1:51:49

That could be possible. The other way around can also happen. Um that you acquire them. Yeah, like storytelling, they bought snorge debts.

Anders Arpteg 1:51:56

Yeah.

Beatrice Bushati 1:51:56

Uh so that's a classic example. Um, I mean, I'm from I'm from the public equity markets. That's how I was investing before. So um I always had some sort of dream of a public company, and there are public companies like similar to public as in doing an IPO.

Anders Arpteg 1:52:15

Exactly.

Beatrice Bushati 1:52:16

But on the other hand, like we're also like 90% of uh of all startups become acquired. Um, so like it could be publishing houses, it could be gaming companies, it could be entertainment studios. Uh, there's a lot of different exit opportunities that we see.

Anders Arpteg 1:52:34

Yeah, I can imagine. I think it's so many more things than I mean, of course, peer is connected to to romance, but I could see the the core idea being applied to so many different things here as well. Amazing.

Beatrice Bushati 1:52:48

Yeah, we want to see your stories, so share your username.

Anders Arpteg 1:52:52

I I will I could do that not publicly here, but perhaps later. What do you think about AGI then? And do you think do you have any thoughts if AGI will happen? If we use um a definition of AGI like um Sam Altman did, which is when AI becomes as good as an average coworker, meaning you could literally replace a person um and and it would do itly well. And today I think it is very far from, to be frank. Uh I mean there are so many things that a human can do that AI is still really bad at, but it could be a point. And do you think that will happen or what's your thinking here about AGI?

Rebecca Oskarsson 1:53:32

Yeah, with that definition, I think it's uh very close. It depends on which role we're talking about and what you expect from that colleague. Like, do you want to grab a coffee with it?

Anders Arpteg 1:53:45

Then yeah, maybe if you start with an author, for example, like a book author.

Rebecca Oskarsson 1:53:49

Yeah.

Anders Arpteg 1:53:50

And um I think today, you know, with Peter, I think it has the nice balance of both having human directing AI and then using AI together with humans in a really good way, and having human in the loop. But there could be a point then when humans are not necessarily put necessary, right? Um so that could be one type of AGI in terms of just the role of an author. But then it could be for for anything else, uh a software engineer or for uh you know a finance person or whatnot. And and just assuming that that do you think that will happen? Do you think AI will continue to just increase? So basically at some point it will be able to do any kind of job as well as a human. Or do you think there will be some limit?

Beatrice Bushati 1:54:38

Um I I probably think that it could do a lot of things. Uh I mean a lot of things as as in as good as a human if you look at an isolated task. But I also think that a lot of new industries are gonna appear. So I'm optimistic in that sense. And if you're an AI evangelist, uh then you believe that it's gonna create growth the same way that previous industrial revolutions have done. And um, I mean, we maybe we'll all be in space quite soon and exploring completely different things that we're doing today. Um, but of course, it will be a lot of difficult, I mean, we'll go through a lot of difficult challenges for sure, but uh staying staying optimistic that it will unlock a lot of opportunities.

Anders Arpteg 1:55:27

I'm gonna I have to question a lot about you know what happens with software engineers as AI is is really excelling at software engineering, uh not perhaps as much in other fields, but still that it's going really faster. And if we just look at the statistics, there's never been you know, people are still hiring people more and more. Even Anthropic, the leading companies is hiring a lot of engineers, and we just have some statistics from US uh Board of Job Statistics or whatever, whatever it's called, and and it's going up. And and people thought it probably would start to go down, but it's not. And and if I just think myself, you know, if you know comparing I've been programming since I was seven, and and uh if it just takes three years or five years ago what I could do as a programmer compared to today, of course I can do so much more today. Does that mean that I work less with programming? No, it simply means that I can do so much more, and it also means that people that couldn't do programming suddenly can. And perhaps in authoring you know stories, it meant that people that couldn't perhaps do be an author suddenly can with peer, right? Yeah, so so it simply enables people to do more things that they could before.

Rebecca Oskarsson 1:56:37

Yeah.

Anders Arpteg 1:56:39

So yeah.

Rebecca Oskarsson 1:56:40

Yeah, I just want to add something here that I th that I you know feel very passionate about, and that is you know the democratizing of of uh AI, because uh that's something that that I've been talking about since the start of peer, how we want it to be so simple. Uh and and you know, when we started, it wasn't, you know, prompting wasn't even you know used in the Swedish language at least. Um and how important it is that that um everyone, including women who are not uh generally speaking, not as uh early adopters as men, how we also learn the technology. And that means that it has to be available for us to learn. And that doesn't just uh mean you know, men and women, but it also means parts of the world having access. And you know, when the API prices go up for models, then we're making them unaccessible for inaccessible for parts of the world, parts, parts of of uh, you know, huge amounts of people that that still is you know then doesn't have access to this extreme you know optimization uh that can happen with AI and then falling behind and growing the gaps, the social gaps. It could, I mean, and social economics.

Anders Arpteg 1:58:01

That's an interesting like uh dilemma here. For one, you know, AI can really be democratizing intelligence and knowledge in a huge way. On the other side, if you think people are token maxing more and more, yeah, meaning that it becomes too expensive to use it all the time, that only the richest one can really use it to the full extent and the other ones cannot, yes, creating potential an even greater gap than it already is.

Rebecca Oskarsson 1:58:26

Exactly. Unfortunately, yeah. And that that is kind of what you know, if you are a domain expert, and it's I mean, it's not a coincidence that that the AI models are good at coding. I mean, there are, you know, the developers and the engineering team has the one that has, you know, created these models. Of course, they're gonna be great at that. But the same thing is like if we take the the authorship, for example, if we have, if we um, you know, the the professional authors will quickly understand how to use AI in a way that is beneficial for the authorship. Uh, we need to give that to more people, just as we need to, you know, in other industries give that to more people because otherwise, yeah, the the experts and the or or the people that has the money and they they will get richer.

Anders Arpteg 1:59:11

Uh it could be polarizing the classes even more.

Rebecca Oskarsson 1:59:16

Yeah. So that's why for us uh we have we have put that emphasis all the time in the product that it's easy to use. You don't need to write a single word, but you can also add as much as you want. And you don't need to think about the AI behind it if you don't want to.

AGI ASI Regulation And Education

Anders Arpteg 1:59:33

Yeah, so many interesting topics. But then going to a final question here and um thinking even more philosophically, and assuming that AGI will happen, and perhaps even ASI, meaning uh artificial superintelligence, meaning a single AI is more intelligent than all humans combined. What would that lead to? And and then we can take like two extremes here, and and one of extreme, of course, is that we will have the world of the Terminators of the Matrix and the machines trying to kill us all as humans. Would be rather bad. Or it could be the other extreme that uh AI is actually helping us solve all the challenges we have in the world, meaning uh finding a cure for cancer and fighting climate crisis, and perhaps moving towards some kind of world of abundance where the cost of goods and services goes towards zero and and we basically can we don't have to work unless we want to, etc. Where do you think we will end up? What do you think, you know, as we move closer, at least to AGI and ASI, where do you think we could end up?

Rebecca Oskarsson 2:00:36

If we start with you, yeah, just just like a a quick quick one on this would be that you know it depends on uh which hands it lands in. Just just like the you know, the atomic bomb, like it depends on what we use it for. And um that would be crazy. It's not it's not the tech, but it's the yeah the people. And and that's even if it's you know, it's it's of course extremely difficult to to answer because it's about things that we you know we don't know what we don't know. We have no idea what that superintelligence could could actually mean because it could also save the world, like it can save humanity that we we could because we we do need that help. Like we do need some help. Yeah, yeah, exactly. So yeah, very maybe boring, uh expecting question, but the answer that yeah, it depends on who who's making the call.

Beatrice Bushati 2:01:38

I'm thinking what type of um like we're always applying human uh human emotions and human um goals, like as in good or bad, or creation or destruction. I mean, that's of course we are prompting AI or like developing AI to to have human traits, but AI in itself maybe will not care about staying alive. Um I I'm not sure, I don't think we know that yet. Um, like it's humans that are all about not being getting extinct. And uh as a like the human race has been like in the last 50 years, we've been on the brink of extinction. Uh like talking about the atomic bomb, like every generation is is scared, like, oh, should we have more kids? Like maybe we should have more kids.

Rebecca Oskarsson 2:02:33

Every generation thinks they're the last. Yeah.

Anders Arpteg 2:02:35

Yeah, and coronaviruses and whatnot. So yeah, exactly.

Beatrice Bushati 2:02:39

So uh everyone is thinking, should we bring new people to to the world? And I I really I really hope so. And hopefully we want to create AI that can bring people together more.

Anders Arpteg 2:02:50

Yeah, I think we all do, uh, and I certainly do as well. But what can we what can we do to to to try to steer it in that direction then? Very simple question. So but uh it's hard to to uh of course fix. But if we just think about you know what Anthropic just did and they at least try to in some way uh limit uh the access to power or limit the access to intelligence. Is that a way or should should the government be more involved? Should uh should we what should we do?

Beatrice Bushati 2:03:29

It's such a complex question, of course, but uh talking about like the the competition between nations, it's so like we're talking about we're talking about Europe and Europe being a startup hub, or even Sweden versus the US and how and and China of course, and how much they are driving, how much China is competing, and like we know that everyone has been scraping every single piece of data that exists, like uh and all of them have been doing that, except perhaps the Europeans. Like, we've been good people. Look at us now.

Anders Arpteg 2:04:05

Yeah, so far behind.

Beatrice Bushati 2:04:07

So I really hope uh I really hope that we can find like uh a way where a bit of regulation can can help uh instead of obstruct. Uh so far it's been way too complicated. We know that from all the EU projects, and we've been applying to to EU accelerators and so on. Um it's so complicated uh to get those type of funds and get through. So I really hope we can find a good balance. Uh a lot of people are vocal here in the Swedish community uh to drive that type of innovation.

Anders Arpteg 2:04:43

Um Swedish Prime Minister, you know, saying we need to you know reduce the regulation somehow.

Beatrice Bushati 2:04:49

Yeah, um, so but let's hope we keep like a certain amount still there.

Anders Arpteg 2:04:54

Um I mean the idea of regulation I don't think anyone is against, it's just you know, the intention is is clear and we need it, but it needs to be done in the right way, right?

Rebecca Oskarsson 2:05:03

Uh I I would say also learn. Like we need everyone to like if I go in the same you know uh direction that I talked about earlier, like about democratizing, and and that is also about knowledge and learning uh how to how to work with AI, because if more people can learn that, if more people are aware of the technology and and knows it, then we're also better off, I would say. Because then just like if we want to cure cancer, like the more people who are doing research, the more likely we are to make a breakthrough. And I think that goes both ways. Like if we want to, if we want to, of course, it's putting it in the hands of everyone could also mean that it's potentially bad, but it also means so much good for the world. And like regulating could, yes, we it needs some some certain regulation, but I think also education, education, yeah. That's uh um and I think that's also something that uh governments, at least in our countries in in the Western world, need to take a uh responsibility for and make sure that we educate our young young people and uh yeah.

Anders Arpteg 2:06:18

Yeah, finding a responsible

Final Thanks And Farewell

Anders Arpteg 2:06:20

way to use AI, I think you know, is something we should be educating everyone about. And I think actually peer is a perfect example of actually using human in the loop here to make sure that we use humans together with AI to create an awesome solution in some way. So thanks for doing that, and I hope you I wish you the best success going forward with PEER. And I'm very glad to have had you here. So thank you so much for coming to the AI After Work podcast, Rebecca Oskashon and uh Beatrice Buschati. Thank you so much.

Rebecca Oskarsson 2:06:48

Thank you. Thank you.