The $2B AI Startup Backed By Nvidia & Adobe | Synthesia

15,964 words · about 64 min to read · YouTube’s automatic captions

0:00 The first chapter of AI video is going to be around using new fancy AI technology to do old media formats. As in linear video, you watch from A to B, right? Broadcasting. But of course, once you get with code, you open up so many more new possibilities. But eventually, you create code. You're not entirely delegating video creation to the AI. You're assisting. We very much believe in augmenting humans. Putting the human at the center of all this just net you the best content. For starters, you have code that describes a video. So you have the sort of PowerPoint structure and that transforms into video and then you have other helpers that you know are more similar to our current world coding assistance and I guess in that sense when you're creating things of that nature the user is co-creating your kids kids are not going to be reading and writing they're going to be watching and listening almost exclusively text is an amazing technology the world is built on text but the question is if the friction to storing information ideas and knowledge in a video format versus a text format is as easy as that right will it mean we will still be reading

Replacing cameras with an enterprise AI platform

0:56 and writing Hello everyone. Welcome back to the AI native dev. Uh today we have uh Victor Riparbelli with us on it. Victor uh I've known Victor for maybe about five years now or uh roughly that and he is the CEO and co-founder of Cynthsia which is a super exciting company and the leading company today uh when it comes to AI native video. uh and we'll talk a lot about kind of creation with AI and uh and and decisions made along the journey and also a little bit about the journey itself which is really interesting in its own right uh and apply you know what what type of learnings uh can we take from that to the world of that so Victor thanks a lot for coming onto the show of course happy to be here so Victor maybe just to get us started a little bit tell us uh a little bit about what Cynthia is uh and what it does sure so theia is the the world's largest leading um platform form for AI video for businesses.

1:57 Essentially, what we do is we help our customers land their messages with the highest degree of effectiveness, right? And the highest degree of understanding of whoever is on the other end of that message. And in 2025, that's not with slide decks or text. That is with video. Today, people want to watch and listen to their content. Um, everyone is different. Of course, I'm sure you read a lot. I also still read a lot. But if you look at the average consumer, right, people really want to watch and listen. That's what they do their free time when no one is watching. listen to podcasts, watch Tik Tok videos, YouTube, Netflix and so on and so forth. But when it comes to our work life, most of the content most of us consume is still in text format and in slide format. And we've essentially built a platform um that kind of kind of started with this avatar technology which really was a way to replace a camera that you use in the real world with a sensor. We go around and you film things with AI so we could generate the footage if you will um of people talking specifically. Um but since then we've now built an entire platform around us. So we have a video editor that enable you to actually finalize the video. You can add in your you know screen recordings, text animations, whole bunch of other things

2:59 to actually finalize the video. Uh it's a content management system. It's a publishing platform with our own proprietary video player. It essentially is one platform that helped you go from an idea all the way to delivering that video to your end consumer and understanding uh how that how does this is is impacting uh them. One of the key things around our platform, which I'm sure we'll talk more about today, right, really is this this aspect of enabling people who had never made a video before to make a video, right? And so when you enter the app today and you begin making a video, if you've ever made a PowerPoint in your life before, you should be able to pick it up in in kind of 5 minutes. And that means that what we've essentially done is in some of the world's largest companies, we've created thousands of video creators who used to only be able to create documents and slides, but now they're they're making videos. Um, That's kind of like where we are today that you the next uh 12 24 months. One of the things that's always been a huge part of our thesis is that the first sort of chapter of AI video is going to be around essentially using new fancy AI technology to do old media

4:00 formats as in linear video you watch from A to B, right? Broadcasting. Um but of course once you can generate video with code, you open up so many more new possibilities. videos can be interactive. They can be two-way conversational and they can do a whole bunch of other things we couldn't do with normal video, right? And so I think the next couple of years that's really what um what's going to define the next chapter of the year. Yeah. No, super exciting and we're going to drill a lot more because like each of these things is uh is sort of a a whole topic to unpack. Um maybe so starting so I love how the sort of the platform has this ability for someone who hasn't created a video to create it today. I use it as the kind of uh really kind of a great example of uh how you can kind of enter disrupt uh the business world by really taking something in the world of AI that allows you know powerful creation but requires a lot of change you know a lot of uh it's a very different it's an entirely different way of creating video uh as compared to whatever taking a

5:02 camera and uh and starting to shoot it. Uh so I think that's really uh powerful but it's also probably like a a reasonably high friction type uh environment on it. So maybe first of all like describe a little bit more like what is involved when someone does create a video in Cynthia like what is that experience just for context and then maybe let's talk a little bit about who's like the ideal creator in this

Assisted creation bridges the storytelling gap

5:28 context. Does it even matter if they have previous video creation skills? I think there's so many um you know parallels to to to death uh you know to cook here. So I think when you make a video right we tend to think uh a lot about like the production of the video right which is like technically how are kind of each of the frames kind of created right is it with a camera is it with AI you sort of technically make the video um and that part I think you know we've made super easy right that's essentially you go in you select an avatar you type your script you use this like basic powerpoint editing functionalities to put something together and most people who are office workers I think should be proficient like figuring out how this works you know that's not that's not not that difficult especially if you compare to like using Adobe Premiere like some expert tool where most people would go in and just blank right they wouldn't know what to do. This is a Is it fair to sort of think about this as the story board or the like like it's the it's this it's the script it's the narrative like whatever the chapters of the video that yeah it's it's kind of if you think of a power like making a slide

6:29 deck right like you have different scenes um you have the scripts which what your app will say we made a whole bunch of like design decisions around not giving you a timeline for example which are different layers in the video um because as soon as people see that they begin to freak out they don't know what a video is right so we intentionally kept it almost like PowerPoint with speaker notes and kind of relatively simp simplistic UX kind of I would say from a technical standpoint most people can create a video right but the thing is that a video is not just technical part of it the video is also storytelling right it's how do we actually put together what is a good video right how long is a good video what should you include in the video what should you not include in the video and I think what we found generally speaking right is that actually that's actually the thing that people struggle with the most by now most of us know how to write a document like we know how to write an email we know how to text with people but video is a new format, right?

7:17 You can't just like write a PowerPoint doc and then just copy paste that into Synthesis and make a video. That's not going to be a great video. In a video, you have a visual language that you want to use to communicate, right? You have animation, all these other things. Um, so there's like the technical part of actually making the video and then there's kind of more the creative aspect of like actually putting a good video together. And that's actually what people struggle with the more in Tunesia. So, we built a bunch of of tech to to to ease this, right? um we've

1 Highlight · Assisted creation shifts the user role to editor

7:43 built we call assisted creation which is essentially kind of like an agent that we've taught how to use the editor on your behalf. So what that could do is you can give it either you can give it just a script you've just written kind of free form or you can also take a URL it can take a PowerPoint deck it can take a PDF document basically any content existing content you have and it'll take it'll parse it then it'll turn into a video so it'll write a script for you it'll delineate into different scenes it'll also do the design of those different scenes you know some of those scenes may have a big image in it and it'll pull images from from stock um stock writersh'll use some of your own assets and other things it gives you kind an outline. I would say from where it is today, it's not like a final video end to end good to go, but it essentially changes your role as someone who makes a video from like coming up with something from scratch to being an editor, right? The draft is maybe like 80% there, 85%. You still want to go around change part of the design. You want to add something, you want to remove something. Um, and it's it's been pretty interesting to see the progression of uh how many more people

8:42 we can kind of bring on to be creators when you give them a draft to work from with their content as opposed to asking them to come up with something from uh from that's awesome. If I kind of echo back a little bit of of the pieces here, you know, the first part is okay, you kind of you use the sort of the lowest common denominator or sort of know the most also I would sort of say maybe the core piece in which you want the human to decide or sort of the creator to decide which is what is it that you want to say and and I guess the storytelling piece and the video uh formation or like how to tell that story and video format is the part that if you want to democratize it then you need to help people uh and so you assist in that but you have also mentioned like another aspect it's not like you can imagine that type of assistance happening as someone types you know autocomp completion or a little reviewer or something like that and I think you even had you know and have probably still some of those capabilities um but you're describing something that is more than that which is uh just kind of point me at sources of information and I will

9:42 help you form that is that are those just on boarding exercises of like give me a PDF and I will create a thing but now you will engage with something that is um that is back into that format of I've got the slides like you just sort of pre-create the slides but then from here on you work with the slide or are these actual kind of degrees of freedom where you're sort of leaving it to the LLM say here's a PDF like extract info from here extract from there and maybe if you run it again it would extract it a little bit it is like an entry point to making a video which a lot of people use as like the starting point it is just like an onboarding exercise and you can you can kind of tune you basically when you upload your content you can tune um the agent in in like different ways, right? You can tell it how long do you want the video to be, what audience are you making it for, um and it will kind of give you first an outline of the content and then you can actually in bullet form. So it will give you like a bulleted formatted list of like what narrative will be and then from that you can actually edit the bullet yourself.

10:36 So you can add another bullet saying add in something about blah blah. You can take a bullet out then after the outline which is sort of the easiest way for most people to sort of have an overview of like what's going to this video. Then from there it'll expand that into individual scenes which is an actual video and then you can of course edit that out. Right. Right. So it's it's um it's less iterative than say like a completely free from like texting where you just like keep prompting it over and again just spits out new new versions.

11:03 We do ask you to kind of like fine-tune it a little bit in the beginning to like what specifically that you're after. Obviously this is like you know this is this is like work in progress and I

Prioritizing utility over one-shot hype

11:11 think I think in in all modalities right now people trying to figure out like what is the right degree of automation for different types of tasks right for example I think if you're making a Super Bowl ad which is not our target market at all there I wouldn't expect automation to be uh a low degree of automation right because this is a really important piece of content like of those 30 seconds every second uh really really really matters doesn't you want that to be absolutely perfect and that probably means you want to have a human being very much in charge of like every single pixel right on the other hand making like a corporate training video and your main goal is to understand help people understand like I don't some your vacation policy in your company or something like that right they want a high degree of automation because it's a much kind of lower stakes piece of content um and we operate more in that category and we operate in the super category so what we're kind of figuring out is like what's the right UX for people to get to that draft but then also iterate on the draft after Right.

12:07 We have a bunch of tools now. So, you know, once you're in your video, you're editing it. You can of course use LLMs to kind of help you extend the scripts and make it know some more happy, more sad, like do a whole bunch of things in there. But we we very much believe in in augmenting humans as opposed to like trying to kind of replace like creating like a fully automatic content generation machine. I'm not saying that will never happen. I think we'll that's definitely the direction the world is moving in. But we very much believe that putting the human at the center of all this just net you the best content, right? Like there's so I I think we've seen so many startups both in our space but also in other spaces the last two years who try to do completely like end to end oneshot generation of XYZ and it's a cool demo but very often right it just doesn't actually work if you want to use it in production right if you get a video that's going to like 70% there but it's not your brand's colors h it doesn't really speak about the product the way you thought it should speak about then then these things are just they they kind of remain like very cool demos um but don't have a lot of utility,

13:08 right? I think that's that's one thing we've always been like obsessed about is really working backwards from like utility, which which sounds very bal and obvious, but I think in when we have these hype cycles, right, it could be easy to to to forget that sometimes. Yeah. Because it might introduce a bit more friction to sort of the initial creation versus just sort of saying a thing and creating it, which I think I'd love to draw some analogies to sort of the world of software development right now, which you know feels fairly clear here. So you have uh you know for starters you have code that describes a video. So you have the sort of PowerPoint structure and that transforms into video and that is in that case the sort of the the text or whatever sort of guidance or configuration that you have about every slide or every one of those portions uh uh gets compiled if you will into into into kind of a video format.

13:57 Um, and then you have other helpers that, you know, are more similar to our, you know, current world coding assistants in which you can text your way or autocomplete your way or such into into creating it. And so that lowers the barrier because you might engage with the system with less video creation uh familiarity or less, you know, even code familiarity. I think in that world it's it's a little bit easier because even after you do all that process eventually you get words so everybody can engage you know with that sort of text that describes what will be said uh but still you might not have video familiarity so you have more of those uh interactive LLM powered like tap the intelligence that is built into the system to be able to create the narrative um and I guess both of those are like avenues for improvement but eventually you you create code so you're not you're not hiding, you know, in uh or like you're not entirely delegating video creation to the AI. You're you're

Maintaining an editable intermediate layout layer

14:55 assisting. Yeah, for sure. And and I think one of the one of one of the things that was always very important to us, right, is you could you could approach video generation as kind of like a completely end to- end task, right? Like you put in text and on the other end you get out pixels basically, which is how a lot of the like you know uh like the very generalized video um

2 Highlight · The necessity of an intermediate layer for editability

15:13 generation mod work. Like if you look at like BO3 for example or runway PK that that approach has certainly has like a lot of of of great things about it. Um, but we always knew that for our customers and for our workflows, right? We needed to have an intermediate layer of editability, right? So, for example, in in in kind of our videos, our videos are composed of the avatar models, right, which produces pixels and footage essentially like you would get it off a camera. And then you have all the other elements of the video which would be, you know, you have like a lower third like I'm like I'm having right here, right? That's the piece of text on screen. There's some animations. there's um a transition to the next scene which winds out everything and fades in or like there's all these things that makes a video a video, right? Um and we were always of the opinion that we need to actually build that as like representations we can we can control, right? So that you can just simply move a shape around the screen, you can change the text, whatever. If you had to prompt that into the video, right, that that clearly just doesn't work. And so what we did was we said, "Okay, we have the footage and we have the editor." And we invested a lot in the editor. That's

16:14 a really uh you know that's a pretty hard piece of software to build well because the craft really matters of like how what it's like to use an editor right if you think about how do you represent how do you sort of convey uh all of these uh uh all of these designers and is like you know building Figma pitch like all these kind of creative tools right that's that's that's actually you know it's very hard being building a great product there but you need that because if you have that everything is fully editable right and that's great because then you can start to help people again like you can prompt something and it's if they can edit everything after the fact then it's

16:47 great, right? We've seen some products like you know there's like uh slide deck generators and video generators that are that try to do everything end to end and when things aren't editable um and you just have to keep reprompting and reprompting and reprompting. It works for some use cases, you know, like if you're if you're if you if you're using some of these general video models to clip clips for a music video or B-roll footage for an ad or something like that, then it's probably okay, right? cuz you're not trying you're just like you just need a 10-second clip or something and what you want there is like a lot of creativity in the model.

17:18 The ability to just like you know throw in 40 different prompts, get 40 different outputs and just pick the one that you like. But for the use case that we're targeting, it's very different, right? Like here people don't want to prompt five times to get a video that actually works, right? Because it's like a volume game. People are creating lots of videos and be able to edit everything out of the fact. And so when we think about automation, the important thing about automation was always that the automation whatever the whatever the the the agent spits out, right, is something you as a user can sit down can correct yourself and edit and also gives us a great signal in terms of like what's the agent good at and what's the agent less good at, right? So that kind of data that we start to collect there is actually pretty interesting. It's almost like RHF data um in in in some some way.

17:56 Yeah. In what people edit. So I love that and I guess that implies a certain level of determinism uh between you know taking taking that those instructions. I don't want to say that text cuz it's more than text, but that representation of the video that you have in every every scene, every slide uh in the deck and creating it. I guess how do you think about um uh kind of I don't know the room for creativity or or like the degrees of freedom that you want to leave uh the uh the LLM. I I'm I'm certain that you know if you take the same description that you've sort of written today uh and you you you generate or something that you've created a year ago and you've generated a video from it that video is not going to be the same one that probably gets generated if you kind of run the same thing today a year later uh just by virtue of the sort of the the system getting better right the platform uh improving I guess how do you how do you envision today and tomorrow this kind of determinism versus um um kind of freedom to uh to evolve uh

Controllability drives business video model design

18:59 on the MLM side. Yeah. So there there's like two components of this. There's one the first one is like the actual you know video models that think of those as like generating the footage right that would come off camera and then the other part of it is kind of like the video editing which again is like all the the lower thirds the shapes the animations the transitions all the stuff that fire these are like today they're the two kind of distinct things with two different models right the first one replaces the camera essentially gives you a footage of someone talking to the camera and the other one take that footage and edits it into something that is uh a coherent video into it right the first one of avatars. And the thing that's that's unique about uh and I

3 Highlight · Controllability versus creativity in AI video

19:36 think the way we look at like the AI video model market, there's there's a bunch of different ways you can approach AI video. And of course, it is again super use case dependent, right? So if you really try and and cook it, really try and like distill it down. I think the the two main and most important dimensions of AI video models are creativity and controllability. So creativity is a bit as we said before, right? How creative can the model get? And creative models are super fun, right? It's like when you use Delhi, Sarab or free, whatever. You can type in literally anything in the model and something will come up, right? You can have dogs skateboarding on the beach.

20:10 You can have an elephant walking around Bloom. You can have all these amazing things, right? And that's super important. If you're creating creative content, storytelling content, meme content, viral content, whatever, then you really that creative dimension is incredibly important, right? On the other hand, you have controllability. Controllability is um in our case for example, does the avatar look exactly the same every time you make a video, right? But does it change slightly depending on the brand? Like maybe sometimes, you know, you'll have a beard, maybe sometimes you won't have a beard. Um does it say exactly what you put in the script? Can you even actually decide the person says how fast is it to run? Um you know, there's all these things that that that are essentially around like controlling out. How deterministic is it? Does it work every single time? Are we in the kind of slot machine world of of using big video models where you just put something, you hope something gets out?

21:00 And here there's a very clear distinction of like what people are focusing on. We're very clearly focused on controllability, reliability, costs, ease of use. Uh right, our avatars need to look exactly the same every single time. They need to say exactly what we put in the script box and all this stuff. And then there's other models that started this creativity. Of course, we are moving towards more creativity in our models. the new flagship models we're releasing soon, they'll have, you know, you'll be able to like prompt yourself into new environments, prompt your avatars. We're getting like we're taking some of that magic, but it's all

21:29 kind of built around this idea of like controllable models because that's what our users need, right? On the other hand, you have other models who built around the creativity and sort of idea, right? This is where you get more the your creative directors like um agencies, etc., right? who who don't mind like prompting something 50 times to get to the right scene because if you have to prompt something 50 times, it's still I mean 100,000 times more affordable than actually going on shooting 50 scenes or even shooting one scene. Yeah. Shooting an elephant on the moon in a studio somewhere, right? Yeah.

22:01 But of course, everyone wants to have controllable models that also very creative. So I think what we'll see is that eventually probably most of all AI video will converge on like a model at some point and the the different products will will be will be you know kind of rooted in the same models but the the the use cases will will kind of determine um what the products look like and the UX will be the thing that differentiates different products right on the LLM side of like actually taking

Moving from templates to video language

22:28 that footage and then putting it into videos um that is essentially a task of like generating um uh creating a model right that understands like our editor that can move shapes around write text etc. The main vector of improvement here is just like can we teach a model to go from more today it's more kind of templatized right so the way it works is sort of like you you write the scripts of course like LLMs to do that and then based on the scripts we try to figure out like okay what kind of design which is basically like a whole bunch of what's the sentiment do you sound angry do you sound sort of laughing this is more like around like should there be should there be like a process diagram or should it be like a big picture or something like that right and then the model will figure out like okay from this scene I think it makes the most sense to have like three bullet points or I think it makes the most sense to have like a big image here or process diagram something like that um and that works pretty well actually you know like cuz ultimately when you're making like simple videos that can be it's a bit

23:28 like think of how like you have like 150 PowerPoint templates and the model sort of picks which one works the best for your go to market slide which one works the best for your product vision slide and so on right and it fills those in with your brand colors your fonts a whole bunch of other things that makes it look real. The thing we're working on now, right, is though, how do you go the way you you kind of go above that is you actually teach a model the language of video, right? How could you do that? Well, you could have a model that watches a lot of video and tries to figure out, you know, not just like what kind of template should I pick in, but like what should this video actually look like? What does great editing look like, right? Well, that's something with like, you know, you'll zoom a little, you're talking, the camera kind of slightly zooms in and animation pops up on the screen. This is much harder of course but obviously this this is going to be possible right like I'm sure if you sat down a model down and watched like a lot a lot a lot of videos and we have a whole whole bank of videos ourselves that make for us right can we learn to instead of just using kind of a templated approach to like taking you from that you know idea to a video to actually just learning what does your

24:28 brand look like what does your font look like what is the language of video like for example if you look at modern video people cut every 8 seconds right it's like this very like ADHD way of like making videos, but people like so the model will probably learn to automatically cut that, you know, do a side angle, do a straight on angle, do like for cool animations. This is like, you know, definitely more in the in the kind of like research phase at the moment. But if you imagine that you at some point, right, you will have models that can spit out content and footage that looks as good as Hollywood films and is as controllable, you can maintain dead end, you can do really whatever you want it to do and you have a model that understands the language of video, right? that you can combine the footage you can generate with the kind of video ed like with the world's best video editor that can take that and turn it into something. And I think that's that's the path for um that that's the two main vectors of of of kind of how you you get to kind of Steven Spielberg

AI video will diverge from traditional formats

25:19 as one AI product of creating. I I I love a lot of these uh a lot of this sort of analogy and I think actually it it it I might be sort of uh you know a little bit when all you have is a hammer you know you sort of everything looks like a nail but I think of it a lot as being specentric like you are defining to a level of specification uh what is it that is in the video and I think what you've built in synthesia is you've built a mechanism that allows a fairly substantial level of control probably as compared to code still low level of control right you still don't say hey uh increase or sort of raise the corner of the mouth by sort of uh whatever you know 2 cmters you know at this point you're still you're you're actually still allowing a decent amount of interpretability uh to the uh to to the model because it's just kind of unfeasible to describe the video at that level of granularity.

26:09 So as compared to code is probably still already has some built-in degrees of freedom but still you you control it uh uh substantially. But then the other side of that is adaptability because I think you talk about creativity and that's great but but and and I agree with that but that in turn enables adaptability. So that you know if you were to generate a video 10 years ago or for whatever like 40 year olds like me you know and and you might uh go easy on the ADHD and sort of you know allow a moment for set of scenes to change while if you're sort of doing it for a teenager uh you might be much more sort of snappy in modification. So the same video can be generated in a way that is adaptable to to that context right and I think a lot about that in in the context of code because you know code has its own stacks like sometimes it's financial needs sometimes it comes back again to sort of user needs and sort of preference and do people want to you know get a a clean page or reach page do they want it corporate looking or do they want it modern looking uh and and

27:09 even in the functionality right do you want to optimize for latency do you want to optimize for costs do you want to uh you know use a specific specific stacks. All of those are different contexts, right? And you will adapt uh the code to that. But you need that spec that allows you to control that sort of slider between predictability and adaptability. Uh and and I think you're using the better words for the world of video, which is uh controllability and uh and uh creativity. Um but is that am I on track here in terms of describing it's essentially all the kind of knobs you can tune, right? and and different knobs will will work different like different people, different users will have will have different needs. Um, as I said, I do think all these things eventually will convert, but it's kind of a matter of like what's your wedge into the market and and and how do you how do you position yourself? Um, the exciting thing as to your point here, right, is of course that, and I kind of also said that a bit in the intro, I think, is um I think we're right at the brink of I think right now like video and AI video look very similar, right? It's it's

28:10 basically like how it was produced. That's that's different. But that's going to change pretty dramatically, right? Um as to your point, right? What we're going to be able to do when every video is generated with code and there's no sensors involved. Everything is literally just like zeros and ones. Then we can start to make an entire new type of video experience where like maybe um you don't like the ADHD experience of like Tik Tok cutting like eight sec every 8 seconds. That that should really annoy you if you had to watch a video that's built like that. Mhm. You could just not watch it like that. You'll get a version of the video that is maybe a bit, you know, more kind of slow paced.

28:44 It's not as like Tik Ty, but then maybe like, you know, your daughter or someone else who's like younger than you prefers to watch that style of information. They and they can do that, right? There's a lot of parallels. What you usually use is like, you know, when the first websites kind of were conceived uh back in the day, they kind of looked like newspapers on screens to some extent because that was kind of like what we imagined back then, right? was like the most obvious thing to do because everyone knows what a newspaper is. It's a page with some information and let's just replicate that on our screens. And of course that was a great product and I think that's kind of like where we are right now in AI video. But then eventually like we figured out well actually kind of a website and newspaper are two very different things right a newspaper like you print it every single morning you send it out to 10 million people all read exactly the same thing.

29:27 Um on a website you can update the information in real time right you can have a feed that's personalized to whoever is actually watching it. you can drive interactivity between users but one click you can access a gazillion pages on the internet um and all the things that makes a website very different than a newspaper right so I think what will happen is we'll like these mediums will diverge and I think what we think about as AI video right now right will eventually be that right it'll be more about a system that has been instructed by someone like me for example a user to communicate a message and then the system can kind of figure out how itself like well maybe guy wants to deliver this way Victor wants to

Co-creation through multi-layered instruction

30:02 deliver this Okay, that's built across linkers, body languages, what avatar do you want to use, etc. But that's very clearly the way all all of the stuff is going to go, right? And and that's always actually been the thing that excites me the most. I think in our space there's going to be two main directions going to be one direction which is going to be the Hollywood film uh entertainment thing. There's like so much to do there still of just like enabling some great Yeah. just to sort of replicate to the current status quo or like you know produce the thing that can be done uh today with sort of physical uh means or digital but uh but production means but in a much much cheaper and sort of uh iterative way but eventually producing the same type of like output. Exactly. It's kind of like when people you know use a computer to write a book, right? You write the book and you publish it. And of course that's like I mean way easier than sitting with like a typewriter back in the day or at at some point carving it in you know stone or something like it's it's much much easier and much much faster and has a huge kind of value added to it. But if you're someone with a with with a message, right, a lot of like young

31:03 people say that they'll start a blog or there'll be a blog and and Twitter, right? That's like constantly evolving and it's not really like the format of a book. It's something very different. And I think it's like it is all the all these trends that we see with every time we invent new media technologies, this exact same thing happens. And that's about to happen for for AI video as well. And that's definitely direction we're more excited about. I think there'll be other companies who'll be focused on like, you know, entertainment, fictional content, like video as we kind of know video today.

31:29 And what we're the most passionate about is like how do we evolve video into something new, right? I think that's that's that's just both on a personal level, I think it's very very exciting and fun to do and but I also think from a from kind of a business opportunity perspective um there's just so much so much knowledge to be driven here if you take training and education which is something that's you know very close to our use cases today. um when you can watch a video and you can ask the video a question or the video can quiz you or examinate you on the content you've just consumed. Um change the content uh depending on what you're interested in, right? You may have a kid learning mathematics and this particular kid loves football, right? So let's make all the examples that makes everything that you're learning about math is about football. That makes when if I'm learning math, it would be about music because that's the thing that that I'm the most passion about. If it's you, it's something else, right? Just the the the the societal value we can create with these kind of like personalized learning. It's essentially like everyone

32:29 will have kind of a personalized tutor in some way, right? That's so exciting. And there's I guess other examples, you know, in in in the corporate world, in the business world where where amount of value this will create is just going to be it's going to be immense. Uh I I love uh I love that both kind of you know the sort of the mission aspect of it. uh but also uh the technology and the kind of lens on creation and I guess in that sense when you're creating things of that nature you the user is co-creating uh with uh with a platform and you end up having sort of these three-way uh creation uh whatever combo here right in which there's the original creator that scaffolded the system and defined you know these things have to be there and they are um they're set in stone there's no there's no degrees of freedom over here like it would always be these brand colors it will always be you know these are the topics that need to be examined on etc. Um then you have the the AI platform itself uh or the sort of the platform in this context uh filling in the gaps and sort of uh uh uh

33:29 implementing simplifying creation and bringing it and then you have the user kind of with whatever their input is maybe it is whatever the favorite uh step that interacts with the platform. So, it's a it's a new type of creation, not just delivery, but you need to enable the delivery of that, otherwise you're you're hindered. Sure. Yeah. Super excited. I've got like a million more sort of uh topics to sort of ask

Surviving the AI winter of 2017

33:53 about uh about the creation side. Maybe we'll come back to that, but I want to uh switch a little bit to talk about the the journey of Cynthia because it's just super interesting. So, you started the company back in 2017 uh uh kind of the AI winter type period. at the time it wasn't uh it wasn't like the best fundraising strategy uh to come uh and come up with a you know today you're this sort of you know sort of high-f flyier you know I think you announced that you're over 100 million uh in ARR and you grew there like massively fast and uh and that's incredible uh but it didn't look like that in the early years of it can you tell us a little bit about just a bit of an abbreviated story of the uh of the early years uh yeah for Sure, for sure.

34:38 Um, yeah, I think in many ways I think we're a kind of a a classical story of, you know, looking looking like fools for three or four years. Um, but then, you know, eventually turned out that that the kind of bet we made back in 2017, we found the company was uh was was right. But the very story is that I'm from co making Denmark like grew up there. Um, did a bunch of things by myself, worked in the daily startup ecosystem. uh you knew that I wanted to build a company, knew I didn't want to build like accounting tools um and and kind of business process things. I love science fiction. You say you haven't. You've sort of built something a bit more creative than that. Yeah.

35:16 Uh yeah. And and back then I was like building like HR tools, stuff and it was like I learned a lot from it, right? I was just like on a personal level I love sci-fi. I love Frontier Tech and I wanted to do something that so I moved to London and I spent a year trying to figure out like what I wanted to build. This was just when AR and VR were kind of like in in in kind of a hype cycle. The Oculus had just came out and everybody thought that by now we would definitely all be in VR headsets all the time. Um I I I love VR still to this day, but I I kind of back then felt it's it's not really there yet. Like there's this just like there's like too many obstacles. It's out of my control to build a big company around it. And but I met Professor Matis Ner co-founder today who um had also looked at VR but he kind of looked at more of the content creation angle. And he done this this paper called face tof face which really was the first time the world saw a neural network producing photorealistic video frames automatically. Um and this paper like made quite a lot of quite a big big stir and and and I saw it as well and I just got obsessed with it.

36:12 you know, I felt like this is super early and of course it, you know, it barely worked and we look back today, it it doesn't look very impressive, but I just felt like something very magical here, right? I this is going to change first how we create content, then it's going to change content itself as we've just discussed. And I just felt there was so many parallels, you know, like uh I love music. Music is my big passion when I'm when I'm not working and especially electronic music. Uh and I saw so many parallels to how we, you know, like by now like a couple of years ago, we started to digitize how you make music, right? You can sample things you're having to use real estate. You can build synthesizers. You can create sounds yourself out of nothing, right?

36:48 And everything got brought to us in terms of like new music genres, democratization of access to music production and so on. And it felt like, you know, what's what's happening now is just that we're digitizing um video, right? And there's so many analogies to all the creative music tools that we're using. So I got very obsessed with that idea. Right. Yeah. By the way, that uh I didn't clue into that analogy. That makes a lot of sense, which you've gone from whatever a physical guitar to being able to reproduce that sound on uh on the computer as an analogy to you could produce a thing with a camera. Uh and you know, now you can generate that on on screen. So, good job. Just like I love that analogy. Yeah. I think there's actually so much of like um my experience like tinkering around these tools uh in my in my in my childhood that really you know it's one of the things it's like in your back of your head you're not like consciously think about all the time but understanding how you make content digitally I think it was definitely helpful anyway um we felt like you know let's start trying to start a company around this uh it took me a very long time to convince Matias because at this point in time I was 25 years old I had no I had absolutely zero

37:50 qualification to start this company at all but I was just obsessed with the idea Yep. And we managed to get a get kind of a team together and we're trying to raise some money. Totally impossible. Got turned down by like 80 100 investors something like that. Everybody just thought uh you know the idea was kind of crazy which it was like it it was you have to extrapate a lot to to see this actually panning out. But also at that point in time especially in London here like everybody wanted to fund like PhDs right who would go out and they would hire like the best free PhD friends and they would like go on some journey to

4 Highlight · Landing early backing via Mark Cuban

38:20 create some big AI company. And some of those companies, I think our strength actually end up becoming the fact that I'm not a PhD and that I'm a person, someone who thinks commercially, but people definitely didn't like that. At some point, um, my my co-founder Stephan, he sends an email to Mark Cuban cold before we found this email in a in a leaked Sony hack. Um, and he sent a very short email just saying, "Hey, we're building this thing. We think you'd fit very well. You know, he understands technology, understand media, would be." He replies within 3 minutes. Very interested. Sends a bunch of questions. We email with him for 10 hours non-stop. never get on call with him. Um, and he ends up investing a million dollars with which which was the founder of the company. And the lesson there was that Mark had himself implemented the face toface paper at home with a tutor. He we did not have to convince him at all that in 10 years time you can make a Hollywood film with your laptop. He completely bought into that. He was essentially just evaluating the team, right? Um and um and I think that was like if you're doing something crazy, you want to try and find someone who shares like your crazy vision of the

39:20 future because it's with VCs, right? It's so hard to convince them something they don't they don't already believe if it's not like uh if it's if it's not like part of the current dogma of like the tech community. Yeah. And the the VC world is very much you only need to find one, but sometimes you need to talk to

39:35 so many. I mean, exactly. the the Mark Cuban email is like it's like a part of startup lore now you know sort of like sort of I've heard that story now being told not by you you know of people just sort of in part is like they'll hold on sort of faith on it uh and part sort of the craze of it but I guess what what I don't know is like how many such emails like the ones it's not that you sort of picked hey there one person we're going to email is Mark Cuban I mean definitely hundreds maybe even a thousand right we're just desperately trying to find someone to um kind of help you get going on the journey.

40:12 And and I think you know looking back like it's I understood why it was a difficult pitch like I had no I was 25 years old had no qualifications start the company at all at least on paper. Um it was a crazy idea. It wasn't clear what the real use case if the technology was even going to work. Um but I'm very grateful that you know that Mark took the bet on us. Um, and as it always is,

Finding product-market fit through corporate use cases

40:32 right, like raising money is like a is is is a milestone, but the real hard part began like after we got the money and we're like, "Okay, great. So, no one believes in us. We have a million dollars. What can we realistically like build and sell?" And this is what year is this? Like 2017, you started the company. When are we now? This is 17. So, we we kind of we kind of in spirit the idea originated in 16. We raised the first money uh in in October 2017.

41:00 Um, and I think kind of like what we what we essentially did when we found the contrast, we did what you're supposed to not do, which we bet on a technology, not on a specific use case, uh, or product, right? Um, and so we had to kind of figure out, okay, what can we build backwards from? So we ended up in this idea of like AI dubbing, which is like you give us a real video, not a generated video, we'll translate it to a different language for you. Um, but we'll not just replace the voice over, we'll also reanimate the face so it looks like you're actually speaking in like Spanish, Italian, whatever. And I think the issue we had there was like, you know, makes a lot of sense. It's like you're an advertiser and if you live in Europe where you've definitely seen like really bad dubbed ads, right, where like the mouth's completely out of sync in the German versions. If you can kind of know for 10K or whatever can fix that for you and all your ads look like they were shot natively in whatever language that you're put in, that makes sense. If you're Netflix and you invest a lot of money creating content, then you could take all those films, you can make them look like they're actually shot natively in German or French. that should have some kind of like economical um there's been economic abuse there, right? I I love how this is a a European

42:02 pain point or a painoint felt more in Europe on it. We had Mati on the from 11 Labs on the uh on the podcast as well and they also a bunch of years later started uh in that pain. But I think it was just a you feel it more closely when you live in a in a non-s single language uh ecosystem compared to the very interesting because like the first the first two funny rounds were were very very very difficult. um not just because of this but it was definitely also a thing that like the American VC ecosystem didn't understand the pain right they kind of like you could explain to the yeah I kind of guess that makes sense but you haven't actually felt it firsthand or like you know if you're a company and you're kind of trying to penetrate many different markets uh it's actually really hard right and it's big value proposition of the platform today like it is for for 11 labs but but essentially that was the thesis we we built this thing this is back when you know this is like GANs were were like the thing back then the technology basically It did work if you look straight at the camera without looking away. And we had two PhDs sitting for like two weeks to create a 30-cond clip. So, it was more kind of like a visual effects. We were all kind

43:04 like a visual effects um you know studio with a proprietary technology. Um and after having done it for a few years, we did we did a million dollars in revenue or something like that. Um which is which is not terrible at our stage, right? It kept us alive. But it was very clear to us that first of all, what we're building was a painkiller. Sorry, it was a vitamin, not a painkiller. Um, people thought it was like cool, but no one would like scream and yell if we disappeared the day after, right? We were not in control of the go to market because we were selling this to video professionals like agencies, production companies, Hollywood studios, and we were such a small part of like we did this very famous David Beckham ad, right? And of obviously you know 98% of the bulk of the work of of making that ad was like someone comes up with idea that David Beckham should do this ad someone pitch it David Beckham someone arranges it and films it and then we were like the very small part at the end where we were translated to different languages right so we were not in control of our go to market at all like we're basically at the whim of like all these agencies deciding to work with us um but more importantly also just didn't scale like the economics just didn't make any sense and so we kind of went

44:04 back to the drawing board we just went out and and spoke to hundreds maybe even thousands of people to try and understand like what is video, you know, why do people make video? Um who's not making video but wants to make video.

5 Highlight · Pivoting to corporate video avatars

44:16 And I think that was one the advantage we have being like very young and being very naive is that you can really approach a space with zero baggage, right? Um and what we discovered was there all these people in the corporate world and like smaller companies and they're contacting us because they they think our technology does something it doesn't actually do, but they really are very intrigued by it, right? And we can tell these people like they're all telling us they want to make video but they can't today. They don't have a camera. They don't have a user camera. They can't get budget from their boss.

44:42 And they're all saying roughly the same thing. Saying like I'm making all this text content and this slide deck content and I know that no one reads it. And if they do they forget everything immediately after. I want to make videos. I've tried to make videos and it works much better but the the scalability of video production is just not there. It's it's it's so much of a pain to make videos. You can't edit them after. And so we we we we spoke to these people and we figured out pretty quickly that these these people are not trying to make Stephen Spielberg films, right?

45:09 They're generally making very simple video like corporate video of like someone talking to the camera interspersed with like a screen recording of someone showing how to do something picture of a product. It's very simple concept, right? And of course for us it was like well that's actually pretty good, right? Because there's two things that are uh that are interesting here. First is the the domain of video they want us to generate is much much smaller. It's basically a person talking to the camera, right? And secondly, the quality threshold is much much lower for these people. If you go to a Hollywood director or some video agency, right, they will not use your AI tooling unless it looks exactly as good as the real thing because, right, well, that's the bar, right? For these people, the bar is not real video. The bar is a PowerPoint. So, they're valuing the outputs like very differently. And and that was the thing that we kind of latched on to. We thought those were interesting. And that got us to this idea of avatars, right? Like you create this sort of digital persona of yourself or you take one of our off-the-shelf avatars, you just type in the script and you magically get a video out. And then you know we we launched that that started working really well. There clearly we hit something there. We built

46:09 the editor around it so that you could finalize the videos yourself inside this sort of PowerPoint type of UX build collaboration platform or content management system and publishing and how we are today. So yeah, it took us three or four years of like pain to really get to think that worked. But um it was also one of those I think it's a classical

Low-stakes versus high-stakes AI generation

46:27 story of like you'd rather be too early than too late, right? Like we we managed to survive as long as it's alive. When when the wave sort of hit us, we were just so well positioned, right? And we're years ahead of of any of our competitors. Amazing story from a journey perspective. also I think super instructional for for the world of sort of software development today because I think when you look at you know agentic creation today with software development and you look at the revolutions I think that distinction about who is it for is is so critical you you get such extremely uh sort of conflicting opinions from people about sort of coding assistance and about the vibe coding and those creations uh and I think what people often times overlook is there's a a key distinction between someone says hey I could have written this code you know in fact I'm very good at it uh and this is just about sort of optimization so does it produce code that really thinks about all the different small details that I think about uh and I'm amazing and that's a very very high bar clearly something

47:28 you'd want very attended um but also you kind of start doubting you know is it really producing value here or am I just you know babysitting more time than I if I had written it and then you got people on the other side who are using you know whatever it is bolt or is 44 or uh or or lovable uh and you know they before they couldn't produce software they couldn't create software just like you know the user base you're talking about which is they just couldn't create a video so now you make it available so like anything that is functional is is already magic uh is already something that was not available it still needs to be good enough to be something that they will embrace um and I guess those worlds eventually converge uh or they don't really converge. They they became they become a gradients of the same uh domain because I guess we all sort of anticipate that eventually AI part creation will will permeate the stack and you know all the way up to the quality but there will still be different tools that you would need at different kind of a quality caliber and

48:31 skill set orientation. Sure. I think I think there's like the in about like is this like high stakes content or low stakes content right? I think it's exactly the same in in in the code dev world where what I think is going to happen right is that we're going to see a lot of this like low stakes like very personalized things like I make the to-do list app. I feel like everyone always have like they think they have like the perfect recipe for a to-do list app, right? And there's nothing out there that really suits exactly what you want. That's probably one of the things for like vip coding it for someone like me who's not like a I can sort of duct tape things together that that might be really cool for me, right? and and maybe I'll also even pay like I don't know $5 $10 a month to just like run this somewhere without any kind of pains. But that's very different than me saying, "Hey, let's like vibe code um a CRM system for synthesis, right?" Like that is such a high stakes environment where the quality of the soft really matters, the security really matters. So it also feels like it'll just be as to your point and and what we talked about before, right? Mine is like Hollywood films or like uh you know corporate

49:32 person just like creating product marketing content and in your world it's like are you vi coding a to-do list app for yourself or something like small fund for your family or maybe like an individual tool for your team that does them like data transformation or are you like building like a CRM system and it feels to me at least a lot of the conversation the last two years people kind of conflate the two or they think of it as one right and I think it's just it's not one like it's it's it's two like very different use case of course there are many others And the only thing is like where we where we've seen a lot of the the new AI tools work well is actually delivering value is in these kind of more like low stakes environments where you enable people who haven't been able to do it before to do something new. Um but in in the kind of high stakes no matter if you're building like enterprise CRM software or if

Bridging familiarity and innovation with the MAYA principle

50:15 you're making like Hollywood films it's the the tools are just like not there yet right but but they obviously will be at some point. I think maybe indeed we sort of use this moment to to talk a little bit about the people uh involved in in creation here. Um and so I asked you before you know whether you know video skill sets uh or video editing skill sets of today are relevant and as I understood it they're sort of like they're helpful in the kind of crafting of the visual storytelling. So look, they're they're a little bit available, but you're mostly sort of making video accessible to people that are that are not uh not requiring that skill. I guess how do you you know, maybe two questions in one, you know, one is like the response when you do engage with people that are maybe stretching kind of your tools capabilities that do come from some like they're in charge of videos and the system. They they do have some video editing uh capabilities even if they're not the producers. um you know h how is kind of their response to it and then maybe we we kind of go from there a little bit to say like how do you see

51:16 this playing out? Do you do you do you still see a place for a role for kind of expert video creators uh or or is this really meant to be like quite

6 Highlight · Why domain expertise and taste still matter in AI generation

51:29 substantially offloaded uh from uh from from the world of you know of jobs? All right. So, what we definitely see is that people with video experience are much much better at creating Cynthia. They become power users much faster and they have a suite of tools they use depending on what they're doing, right? So, obviously, I mean, they're not trying to create Super Bowl ads inside Cynthia, but they really appreciate the tool for ease of use and for the speed at which they can kind of like churn out content, right? Um, I think what sets them apart is there's like some degree of technical ability like understanding how a video works and like you know how should you place things etc. But really I actually think it's a lot about something as esoteric as like taste, right? It's like how do you make a video that catches your attention the first 10 seconds, right? That's not that's not a science.

52:17 That's that's kind of like an art, right? Um what avatar would you pick for this market? Like well how do you write a script that's like fun and engaging? How do we make sure the video is not like way too long or like way too short? All this is like taste and that's stuff that that's taste that you build because you spend a lot of time thinking about videos and making videos. And I don't think uh the more videos you make, of course, you you'll get better at it also with the easier. But it's very clear to me that the people who understand the language of video storytelling part are much better at making videos. And I actually don't think that will go away.

52:45 I think that'll remain true for for like a very long time. the other end of the scale with like these kind of like new users. That's something that they had to learn, right? So, we helped them with tooling. We talk about this AI agent we have that can kind of you know make content on behalf of you and we've come up with our kind of own framework that we call foc some some good ground rules for like doing that. Um, but I'm very much of the opinion that I think uh people with with video skills, right? Of course, you know, assuming that they kind of migrate onto this new tool stack, no matter if it's easier, if it's like more creative, so they don't resist it. It's too much.

53:22 Yeah, exactly. They will be they would be the best at doing this stuff. They would have to learn new skills such as like what does it mean when a video is not just like broadcasting? How do you personalize a video meaningfully to different people? Just like slot filling with like saying, "Hey guy, hey Victor, love Cesia, um, you know, love Tesla, etc." like that's not going to cut it. How what can you do that's creative and using LLM and agents to kind of um create these experience, right? Because there's so many open-ended questions

53:47 there. I think video editors will be will much better at at doing that at like kind of evolving into that. For the casual users, it's uh it's more journey of like getting up to speed with this and in designing the app. I love this design principle called the Maya which is most advanced yet applicable which essentially that you kind of have this sort of window right of like the things that are like too scary for people to do and then there's things people like very familiar with and you want to like figure out where can you where can you sit in on this kind of spectrum right where you can in our case this is said many times now we're kind of anchored on PowerPoint users essentially so like okay make it 75% like PowerPoint because that makes people feel familiar but then we also push the boundaries of what you can do right like there's some new functionalities you have to learn and then slowly as more and more people on board more people try something this will be moved that window right so I think in in probably in in like five years time or something like that maybe maybe less probably the app will look very different than does today it won't be as as anchored on on PowerPoint but we've taken the user base from something that we're familiar with to push it into

54:47 something that's new and exciting and and it it almost for the people who've been with us for a long time feel so seamless that you don't really think about it it's like all of a sudden now you're just making conversational videos that are interactive that are personalized by default um and so on and so forth. So both such uses are really really really important. I think I love the May principle as well and I think probably the best uh uh kind of example of that is the gestures on the on the smartphones right where initially it's like what's the gesture and at some point there were some some big changes where you can actually start being demanding there and and people want some form of like you know funky way you would twist three fingers to

The evolution of the video creation craft

55:23 sort of achieve some specific examples and that might be okay. Um I think so so with that evolution I guess do do you still see uh kind of I guess this remaining the notion of like video creation it sounds like you still see that as a craft uh and it's just a different craft than than it is today. What about the the the production? I guess sort of the whatever the person with a camera sort of shooting uh shooting a video. Do you see that mostly disappearing? Take a longer timeline, right? 10 years out. I I do think that's that's probably like uh a more at risk for sure.

56:00 Um but I I I would still say that something like video, I mean I came to video having never done video before, right? So I learned everything from kind of starting company. I think it is one of those disciplines that when you get into the weeds of it, most people don't understand like how much work it actually takes to create like a even a good looking corporate video of someone just talks to the camera. When you watch the video back on LinkedIn, it looks like oh whatever someone put up their smartphone and hit play and something in reality like there's how does the light fall in right? How does the camera move?

56:29 What should the angle be? What should the framing be? um like there's all these kind of like stylistic decisions that you make and of course those are not usually made by camera operators but there are like degree of like also choices a camera operator has still say I think a lot of these people are like you know have a much better foundation for being the ones who pick up these AI tools and then become the experts but I think it will require a migration of their skill set and I definitely think that the job of like uh carrying a camera and filming things in the real world that will probably be a less attractive uh business to be in in in five or 10 years time, right? Just like being a typewriter um is also not very um a very true business to be in today.

57:10 It's always very hard to I think you know predict exactly what's going to happen. And I actually don't think that like real video is going to die and all of a sudden everything is going to be generated. I think we'll put a lot of value on real video in in in in some sense just like we still like to see people playing piano or guitar or um if we listen to vinyl like there's there's an emotional part of it and there's also something I think we just as humans we we we ascribe some some value to that right we also still read physical books despite the fact you could read everything on a on an ebook so I think it's going to go away I think it's going to coexist but um probably what is is going to be more at risk is a lot of like the business content you know where that that's sort of per definition there's less romanticism, nostalgia, creativity around it. Um, so I think it is a market that that will shrink, but I don't think it's going to like disappear completely. Yeah. You know, I love that sort of super succinct on it and I uh um I love how you keep coming back to kind of I guess historical references of music here as a great analogy of you know what what has happened there and I do that for a bit and I think clearly we

58:11 kind of face that question in the world of software development of what will happen to to software developers and I think it's hard to predict the specificity but like software development as a craft hard to sort of see that going away or becoming any less important. uh coding as a skill as a sort of the means of doing it probably changes uh or or diminishes and maybe becomes a bit more the you know artisal carpentry versus the factory line uh production uh example. I think also one thing we we see that we did an analysis recently of how people prompt like our AI agent to help them make videos, right? And I don't think it's a surprise and I'm I'm curious what you're seeing, but I'm pretty sure it's probably the same in software development, right?

58:53 which is just um maybe an expression of like this idea of like taste. It's just like how good people are at prompting these things, right? Like if you take someone who's never made a video before and you ask them to like prompt a video, describe what you want in the video, they're clueless time, right? Like make me a video about our vacation policy. It's like if you get someone who knows video and who's an expert on video, right? You get this beautiful pros. they clearly have a vision in their mind like what they want and what they think the video should look like and so they can help guide the system to actually create that for them right and I think that's the thing that's that's pretty hard to automate right like we still can't go into brains and like uh you know kind of hijack internal LLM to like come up with the the best idea for like how they should put something out in the real world and if you give a very limited prompt like that is that is what the system has to work with most I mean you can add a lot of context around it right but I think that that's definitely something that's pretty interesting right like you you see is in in kind of people's ability to prompt um that the more taste they have, the more experts they are, the much better

59:54 at like prompting these systems. I have a lot of friends that are artists and illustrators and like when I go in and try and use mid journey for example, I face the same problem like I don't know what to write like I want like a picture in a jungle. When I have my friends illustrator do this, right, it's like a it's like a free paragraph thing and that's the images he generates is like 10 times cooler than whatever I can do. Yeah. Yeah. as well as their assessment of what gets created and I think that's a really good spot and I think in the world of software that translates to kind of architectural and product skills uh probably more so they're still kind of elevated I guess kind of a at a higher resolution that you know use this library or write that line of code uh much more about you know the priorities and the bigger picture. Um, so just Victor before I uh again so many more questions so if I sort of to ask but I think we're running out of time maybe just to uh uh move out of all of these and really talk a bit more society.

Predicting a future where video replaces text

1:00:48 you've had a an interesting uh uh sort of statement in a TED talk recently where you kind of made this bold claim where you think maybe even as a society as video creation becomes easier uh text will play a smaller and smaller role in our in our world and switch to video guess what's your what's your prediction

7 Highlight · The generational shift from text to video consumption

1:01:08 around the societal change that this type of sort of video generation will will yeah so I had this uh very provocative oneliner right which is that your kids kids are not going to be reading and writing. They're going to be watching and listening almost exclusively, which definitely like set people off and it definitely gets people going. Yeah, gets people going. Was also intention, right? But I think the core idea just is is is pretty simple. If you look at the history of media technology, right? It's very clear that we're always on this, you know, kind of trajectory towards more visual content and more interactive content, which essentially is content or like information consumption feels more like the real world, you know. um that has there's been a huge barrier to that historically speaking in the fact that like making a video for example is way more work in cost and you know time etc and skill than it is to like write a document. So most of the things we've kind of been consuming has just been text. Text is an amazing technology. Um text is sort of like a compression um algorithm in some sense, right? Like we have something that happens within the real world and

1:02:08 we can kind of like put that down and we can store that in a scalable way. And of course, you know, the world is built on text. We wouldn't have been there without without text. But the question is if the friction to storing information, ideas or knowledge in a video format versus a text format, if it literally is as easy as that, right? What would that mean for text? Will it mean we will still be reading and writing or will it mean that we'll just default consume almost everything is as as visual or kind of like real life content right it's it's a freaky idea but if you imagine that if you if you believe that AR and VR in 50 years or 100 years the form fact will be so that we're just wearing a length something and we have screens at all times I don't think it's a completely preposterous claim to say that we probably will be consuming most of information um as as as video and audio and and in

1:02:58 interactive formats. A lot of people have um very strong feelings about that and and I I kind of get that, right? I think we we we we kind of really value reading in society like sitting down with a book is like is it probably represents a lot more than the reading. It represents the the focused uh area the the specific attempt at learning or sort of you know immersing in a story. But I think we have to ask ourselves the question, right, of like if you're learning something. Um, I think few people would say, "Oh, we think it's better to sit down and like read a book than you're in a classroom with the world's best tutor who can talk to you and interact with you and also has a whiteboard where they can show you relevant pictures or images or graphs or flow diagrams, whatever they think you're learning, right? Like most people would probably agree that that's a better way of learning. Um, and that's probably going to be possible in the not too distant future, right? Um, and so I think it'll be interest to see how all this stuff pans out, right? And of course I'm being a bit tongue and cheek when I say that. I think it'll probably still be a use for text like texting, etc. But I do think a majority of the

1:03:58 way we consume information is going to move to being video driven. Um, and I kind of ended on this. I think the thought that I have a lot, which is it's it's so interesting how we put like so much more value on reading books, including myself, right? I I read less than I did five years ago for sure because I've just found that often I'll buy a book on Amazon and it's 200 pages and I'll spend like a day reading it and feeling half the time reading it feeling like I'm they're just rehashing the same examples over and over with so much filler in a lot of these book which makes sense because there's you can't you can't you can't publish a book that's 50 pages long, right? Yeah.

1:04:32 Exactly. You can't sell it for sort of any real amount. Yeah. You'd feel cheated. Yeah. Exactly. And so often I've just found like when I want to learn something new, right? I go on YouTube and I watch a 25minute video, but I feel almost guilty for not reading the book and watching the YouTube videos. Um, and that's so that's that's that's something like I think philosophy is kind of interesting like why clearly like if it's again if it's music or something, right? Obviously much better to learn that in a YouTube video we can hear something and see someone playing as opposed to like sitting and like reading sheet music or reading a book.

1:05:00 It's just it makes no sense to try and read music from a book, right? And I think a lot of cases like that. Um, and then maybe the last uh thought I had on this was I think there's this uh there's

Re-evaluating moral panic over new consumption habits

1:05:11 always this sort of moral panic about like new technologies that's been consistent throughout all world history. And one of the things we're hearing a lot right now is that young people they can't focus. Their attention spans are decreasing. Uh they just need like you know quick dopamine hits from like Tik Tok videos. Um, and you know, maybe that's right, but I have this progative idea like what if young people are just becoming so good at consuming information because they they consume at such high velocity all the time that they're actually just really bored by a 200page book. You can't we can't keep our attention because why would you, you know, you can watch a YouTube video that's super dense, super interesting, it's exciting, it's communicated extremely well, and then now you're forced to sit and read this like 200page book which is is is not very exciting, right? So the question is like are we getting better at consuming information?

1:05:56 and that's kind of reflected in the content that we consume or becoming lazier and our ADHD brains are just like taken over and we can't focus in anything. I don't have the answer, but I feel like that that's an angle that's very rarely discussed, right? Like everyone always goes to kind of like the the negative impact of saying like people can only watch like 30 second Tik Tok videos and I just I just I thought that was kind of an interesting idea to to try around with. First of all, like I love that you're sort of uh challenging the conversation. I think like as long as uh as long as you're sort of doing it to sort of facilitate a conversation versus of you know I don't think there's any like agenda in sort of you know sharing this it's more about sort of the the narrative and I you know I initially was dismissive when I sort of saw this like oh Victor's just being uh just sort of you know stirring uh stirring some mess here. Uh but that I did think about emojis you know are actually like a good example probably of a live uh a live example of something that that literally is happening. people are switching to a visual uh and and I even like an old person like me sort of still sort of appreciates sort of the value of conveying an emotion in that fashion

1:06:57 versus a bunch of words. Um and I guess the the part that I uh am still kind of uncomfortable with is is almost thinking about the degrees of freedom that we leave ourselves. Uh and so I think the difference between watching a video or having someone illustrate to you versus reading a book is that it leaves room for imagination. Right? when you read a book then then you actually get a lot less you get it takes more time and but to an extent you get a lot less information because there's only so much that can be described uh in the book especially when you're talking about uh uh uh maybe in business books it's a little bit different but like in a storytelling type setting and so it leaves you the degrees of freedom uh you know that but we're talking about living to the LLM uh and I guess the question also is like how do we how do we balance those on it but but I agree with you that society will move towards kind of the path of least resistance uh and you know what is uh what is of a means of sort of running and and scale um so it's interesting to sort of see how how things play out. I totally and I

1:07:58 think the argument around um your imagination right is the one that I met with the most often and I think that's very true for sure. Um yours in a video you're generally like served everything and there's less you have to do less imagining. Um but on the other hand, right like a lot of that book could be consumed as like a podcast for example or like audio get that imagination and and that that's an example many people listening to this are doing precisely that and you know I'm a big audio book fan and so yeah and and I mean I don't have kids right but but if I did I'm also pretty sure I wouldn't be like hey you know just just stay on TikTok you know just ditch the books it's not that I'm kind of like you know that I'm we will look back at this period in time and this evolution is very different as as we are right now because we always do, right? Like my favorite example of this is if you go back to like the 40s um when when radios kind of became mainstream, something everyone had, right? The parents you you could find all these like old newspaper clips, right? Where parents are outraged that their kids are like sitting in their rooms and they're listening to these like cheap lowquality radio

1:09:00 novels, right? They never leave their room, they don't want to go out, they just want to sit in their room and listen to these novels, which is exactly what people say about like Tik Tok videos or computer games, whatever today. And you know just in in my childhood I I grew up uh obviously spending a ton of time on my computer and playing lots of computer games. I remember like this when I was uh I guess was like around 10 11 12 13 years old something like that right and there was a very real fear that you become a school shooter if you play too much Counter Strike right or that you become like a violent person. Um there's famously a Senate hearing around Mortal Kombat, right? Like 2D Mortal Kombat, which um people were outraged because they were convinced that, you know, this would create a generation of young violent men that had learned to fight from like playing too much 2D Mortal Kombat. And it's like we look back at this stuff today, right? It just is absolutely like ridiculous. There's a kernel of truth to some of it for sure.

1:09:51 But I think we'll I think we I'm very curious to see like in in 30 40 50 years time when I'm going to be sitting scrolling TikTok in the retirement home and I'm going to be like I was right. Yeah. At a time in which we were cool but yeah exactly. We'll see if that survives. Uh Victor this has been excellent. Thanks a lot for uh coming on the show. Congrats on sort of the amazing journey and looking forward to sort of seeing it continue both uh society uh wise and sort of as the business evolves. Thanks. Thanks for coming on and thanks everybody for tuning in and I hope you join us for the next one.