Victor Riparbelli: Why the Future of Video is Synthetic

24:13Maybe I'm the person that created the content. What do I do? So when you start off, you have a choice if you want to make yourself as an AI avatar or if you want to use one of our stock avatars. It depends quite a lot of the use case. And then on the platform, it's really as simple as just like selecting your avatar, typing up the script, putting the video editing around it. In terms of the use cases, I would say where we are today, it's pretty much instructional video content. The avatar isn't really emotive or expressive enough to make something that makes you like happy or sad or, you know, kind of makes you feel something.

24:47It's great for kind of like teaching you something. And that's very much what it's used for today. So it could be around like onboarding. It could also be, you know, one of the world's biggest tech companies with lots of distributors around the world. They used to send them out like a monthly update with what's changed in our products, new policies, whatever. Nobody reads that PDF document. Now they can make a video instead, which is way more powerful. and they have to use the same amount of effort on creating it because again, it's the same people who wrote the PDF who make the videos, right?

25:16Could also be another big tech company that have 4 ,000 A's around the world. So massive Salesforce need to keep these folks updated on what's happening in the competitive landscape, new product launches, pricing changes. And again, they used to do this with lots of text, right? But information retention of text is like 11%. So if you scale that out to 4 ,000 people, then switching that text out with video, which generally has around 80 % information retention, it's just absolutely massive, right? So they make these small like micro-learning videos with, hey, this new feature just launched, this is how it works.

25:48Our competitors just did this, whatever. So it's a lot about this sort of, you know, it's not like the big flash, interesting videos that most people think about when they think of videos. It's much more around kind of like this instructional video inside, especially the enterprise that we work with. So you had this, you and your co-founders had this vision to make it really easy for people to create video. But did you have, how did you get to these, like this use case and this customer, that, you know, ultimately finding product market fit? Did you always have this vision in mind for the use cases?

26:22Or I guess walk us through the journey from starting to where you are now. Yeah, well, I think to some extent, we did what you're supposed to not do. We kind of started with the technology, right? Right. We saw this thing - From the paper, the research paper. From the research paper, right? We're like, this thing is going to change the world. how do we sequence a company from where we are today until all the really cool things are going to be possible, right? And so the initial version of the tech, which is all the way back in like 17, 18, was very different than what it is today. The first use case we kind of started to tackle was AI dubbing, which is actually like making a comeback now.

26:56AI dubbing, sorry. What did you say? AI dubbing? AI dubbing. Like voice dubbing. Essentially, the idea is like you take a video in English, say, and then you dub it over with a German or French voice track, but then you also reanimate the face so that it looks like it was recorded in French or German or whatever. Got it. And this is like the first product that we built and we went out and we sold this to like advertising agencies, basically video production professionals that knew what they were doing and already producing lots of video. And what we found back then was like, it was actually not a terrible business.

27:29Like we were 12 people and I think we did like 800, 900K in revenue in the year we were selling this. But what we figured out was this is a vitamin, not a painkiller. And we are a very, very small part of a much, much larger process that we have no influence over. So to go to market here is just very difficult, right? Because we can sell it into an agency who then have to sell it into their client. Then the client has to agree on the whole campaign as a thing. It takes nine months before we even film something. Then it takes six months for the post-production. It's just like not a great market at all.

28:03But what we learned also in this process was that, you know, me and Steph, my co-founder, were just like, I think we talked like 2 ,000 people in the span of like one and a half years just to build like a mental model around video production from first principles. And what we figured in that process was there are billions of people in the world who are desperate to make video content, but they can't because they don't have the budget. They don't have the skills. They have no clue where to start. And for these people, even if the quality would be lower than with a real camera, if we can make it a thousand times more affordable and a thousand times easier, that's a huge win for them.

28:42And so that kind of was the insight that led us to saying, okay, what if instead of working with these super high quality videos that we'll get from an agency, what if we could build this simple AI spokesperson kind of thing where you just give us the script and then we'll give you a video back of the person reading that out to the camera. And so we built this like MVP actually where we built this like very thin platform where you could submit a script and click generate video. What actually happened then was that we submitted that script, everybody in the office panicked because now we have to go out and try to figure out a way to like actually make that video.

29:18And it was using AI, but it was like, it took like a PhD, you know, two days to make one video. But we validated that this was really interesting for people. And so that led us towards this idea of building a product, not for video production professionals, but for everyone else. And that kind of all came together in the end of summer 2020, when we launched our first SaaS product studio, which is what we're known for today. And honestly, since then, it was one of those things where the day we launched it just went viral. And it's just been an explosive growth trajectory since then. And I think, you know, looking bad at it now, probably one of those things where we're all just like, holy F, there's something here.

30:04If in this state of the product and the AI technology, people are willing to pay for it, like there's something big here, right? And even though we've made a lot of progress, I think the feeling here is, I think for everyone working in Gen AI, we're still so early, right? And this is going to be an entirely new industry. It's fascinating. I mean, it sounds like you went through effectively three years. I mean, you had a product, you were doing the dubbing, but it sounds like it wasn't totally working. So you kind of went through this three years of, of being in the wilderness, as they say. And it feels like all of the great startup stories have this period where you're kind of searching and it might be for years.

30:40And then, and then you tap into this vein and, and boom, just like that, everything, everything changes. I mean, what were those three years like? I mean, did you feel like you were making progress or did it feel like you're sort of running through mud? Like, what was that like? That definitely felt like running through mud. I think anyone who's been through those, like, especially if it's free, if it's six months, it's like, man, for three years is a long time to keep a team motivated, to keep your own motivation high. It was not fun. And we had several near-death moments. We obviously had to raise capital to build the company.

31:11This is a fairly expensive technology to develop, right? And it was just... The first rounds were really, really difficult to raise. it was an absolute shit show to be completely honest with you and so they were not fun but we learned a lot and i think now looking back at back at it i'm actually really happy that we had those three years because the amount of insights we got from first principles about video production was just so strong i think still think we lean on a lot of those today um but uh but yeah i mean it's it's it's rough right yeah it's really rough when you just feel like you know you're making a little bit of progress, but not really towards building an actually kind of scalable and investing company.

31:58How do you keep motivating the team? Like, how do you, you know, you're running through mud for three years. You want to hold on to your team. You want to make sure they don't leave to go do something else. Like, how do you keep the team motivated? Well, I actually think that despite it being an uphill battle, we all fundamentally truly believe that the moment that we're in right now was going to come. I think all the early signs were there. In 2017, 18, I remember around the conferences, I had this slide of an image of a human face generated by AI. And it was like 64 by 64 pixels, some of that, black and white.

32:36If you see today, you'd be like, oh, whatever. But back then, that blew people's minds, absolutely blew people's minds, including our own, right? Because how insane is it that a computer automatically can just generate a picture that, you know, more or less like looks real and seeing the progress in our own technology and those other technologies only speeding up, not slowing down. I think everyone's team just truly had confidence, like this is going to happen. And even if Synthesi as a company didn't work out, we were just too early. And I think that's really, that's really strong, right? So I think that played a huge part in it.

33:13And I think also just having a strong thesis, I think we've always had a strong thesis internally of how this space and how this market's going to evolve. And even if it went for something slower than we had anticipated and would want, we did see the signs that we were moving in that direction. And I think that helped a lot in everyone's motivation. Let's talk about the technology. You just touched on it a little bit here about the face. Everyone, most of the people in this world first learned about generative AI when ChatGPT launched or maybe when DALI launched, you know, about a year ago. And that was obviously like a huge moment for so many companies.

33:53Was that moment consequential for your company? Or, I mean, was your technology sort of progressing on its own independently of what else was happening? I mean, I think everyone who runs an AI company will say that has definitely been consequential and has been absolutely like crazy. nine, 10 months since the chat GPT moment happened. For us in terms of the technology, definitely. And I think a lot of the things we've seen working in the last 12 months is just big models work really well. And that magnitude of data really does make a difference, which is super exciting. For you as well, that's affected you as well, or more just like the large language models?

34:32No, that's affected us as well. That's also just something like a DALI, for example. It is also to some extent a case of like, If you throw enough data at these things, they will eventually figure out how to make something that looks real. Whereas if you go sort of before that whole moment in time, most people are still trying to build smaller models with high quality data, spend a lot of time on fine-tune the algorithms, figuring out new ways of doing things instead of just throwing massive amounts of data at it. And I think throwing massive amounts of data has to some extent really proven that that was the step change that the industry kind of needed and was waiting for.

35:10Back when we started the company, everything was about predictive AI more or less, right? So it was like CNNs, it was like machine learning, a lot of stuff people were talking about as being AI. Gans were the first real big technology that started to be able to generate especially visual content. The hard thing I think as a founder, right, is that when you're in a space like this that evolves so quickly, then you have these step changes to sort of come by once a year or something like that. And it's always difficult to be like, how much should you chase that shiny new thing that looks really exciting versus continue on your existing roadmap?

35:43And I think that's something everyone is struggling with that runs a big AI company, right? It's the kind of exploitation versus exploration is always really hard. But I think if we look at where the world is today, look at some technology like text to speech, for example. It's also one of those things where like someone figured out that if you just throw 100 ,000 hours of data at these models instead of like, you know, 500 hours, they just become exponentially better than what you could have before. And I think anyone who is building AI technology today who's not thinking in line of those terms, I think it's going to lose out in the long term.

36:18Yeah. So what what has, I would say, this realization and really the last year done maybe both to the product and the actual capabilities of the product and also the growth? Like it sounds like 2020 is when you found that product market fit and things really took off. What did that look like post 2020, you know, October or whatever it was or whenever ChatGPT launched about eight months ago? What was the impact of that on the product and also the growth? So I'd say there's been a few things. First of all, we've been growing really fast since we launched in 2020. Much more under the radar, though.

36:54But we were growing at a really, really fast clip. When the moment with ChatGPT sort of happened, there's a lot of things that were exciting. The obvious one is just top funnel just exploded like it did for every other AI company. Why is that? Why are all the AI companies just getting crazy top of the funnel growth? I've seen that as well. Why is that? I honestly think it's just, you know, ChatGPG had such a big wow factor. It kind of pushed through. It's a bit like what we talked about earlier with VIAR, right? That was the moment for like chatbots. Like, here's the thing. You're going to ask it a question and I'll respond back to you in an almost like magical fashion, right?

37:30So that happened. And I think everything else about AI and the slipstream of that, including like image generation, video generation, all these technologies has kind of been developed also before that. Everyone does this cultural moment, right? Where everyone is now thinking about AI. from like the C-level all the way down to like, you know, primary school pupils, right? Everyone is like obsessed with this AI thing. It just creates all this momentum. Then you have all these AI inferntures making lists of like, you know, these five tools you need to try. Like there's just the whole thing that kind of like happens.

38:02It's all over TikTok. And that gave a lot of top fun to like every company out there, which is really awesome. Another thing I would say that happened for us is that as these technologies really matured, For us, we're very conscious about what technology do we want to be the best in the world at and develop in-house, and what other technologies are false multipliers in the product that we're building. So you take something like LLMs, for example, for text generation, amazing technology. We're never going to be going into building foundational text models. That would make absolutely no sense for us.

38:33But what a gift it is to now have an API where you can help our customers write the script for their videos, for example, which is actually quite hard. So now we have a feature like many other creative tools where you can just say, hey, make me a video about fly fishing in Oregon or whatever thing. We'll write that video for you. Amazing. With image generation, you have something like the same. Great. You can now generate backgrounds for your avatars. You can even use it to maybe generate new outfits for your avatars. There's all these false multipliers. And I think it just, you know, seeing all those technologies progress so quickly.

39:06and change our mindset around how to build an AI-first product. So that'd be the second thing. And then I think the third thing from a GTM perspective, I think is quite important, is this is now board-level topic, right? In every Fortune 2000 company, what is our generative AI strategy? And that has elevated a lot of the conversations that we have and that I'm sure other of our peers are having, because this is no longer this sort of little weird fringe thing. Now everyone is thinking about how's Jet AI going to impact our company. And that could be anything from LLMs to making your internal communications more efficient.

39:46But that's definitely a great tailwind, right? Now that we're sort of a bit on the other side of the hype cycle, I think the top funnel has gone down for everyone, which is quite natural. Now the tourists are slowly kind of like disappearing. The tourist, yeah. You had a great tweet on this, by the way, just to quote it really quick. You said, product market fit and longevity test for generative AI companies. Do your customers overwhelmingly present what they make as AI generated? If the answer is overwhelmingly yes, you are likely monetizing novelty, not utility. I think what you're saying, but I would love for you to clarify it, is are these people joining because it's generative AI or because it's providing value?

40:30It feels like the former are starting to get flushed out of the system. Is that right? Yeah, 100%. And I think it's also one of the things that I think is important that you as a company, whether you're building an AI, you'll have all these hype cycles. We've also had many hype cycles before the chat TV moment, right? And I think it's just this, even though you're doing cool things with AI and everyone's excited about AI and everybody wants to talk about AI, it's very important to remember that the core principles of building a great product and a great company as a result of that is that you provide actual real utility for people.

41:05right? You can do a pass grab, of course, with something that's exciting for just a few months. But if you want to build a truly lasting company that has happy customers and changes how they work, it has to be about utility, not just look at this cool new thing I made. And my point of that tweet was just when we were in that hype cycle, everyone was like, now you can make videos with AI. Now you can do all these things with AI. But everyone always presented it as AI, right? Like Coca-Cola, I think that was like, for me, like top of the bubble. Coca-Cola did an ad where they were like, it was like an AI generated ad, right?

41:38Using image generation, all this stuff. And it's a great ad. And they said it. They're like, this is AI generated. The whole point of that is AI generated. Everybody who knows something about AI here knows that it was like 5 % AI generated and 95 % visual effects generated. But it served well, right? It had that novelty factor to it. and it probably, for their business objective, did really well. But if there's a company that's been betting the business on great, Coca-Cola want to make AI video now, then you'd be disappointed, right? Because once they've done your initial contract of making that one really cool video, then next time they have to shoot an ad and there's something else that's cool, they're not going to need your tool anymore, right?

42:19So we just always have this mindset of being maniacally focused on finding use case that have real utility and being very honest with ourselves around how good is the technology actually today, right? And I think that is the key question because there are lots of people who are excited about using AI for all different kinds of things. And a lot of those things will probably happen in the future. But if you're trying to build a product today, it's really important, like what is the technology actually good enough for, right? And that's why, you know, when I, what gets me most excited is like a cement mixing company in France with 10 ,000 employees that you've never heard of before, where we have 30 people using Synthesia, not because they care about AI, I think it's cool, but because it makes their day-to-day job 10 times easier, right?

43:05That's the things you want to find and focus on and build your product roadmap around. You can also, of course, spice it up with doing cool things that gets a lot of attention and gets a lot of buzz. But I think there are many great companies who've kind of fallen into those kind of potholes and all of a sudden like, oh, like we're doing a web-free company. We're launching a token, right? because that's like the new cool thing to do or the IR. There's so many of these different cycles. And I think all of them presents both an opportunity, but also a huge risk to founders, especially if you're doing deep tech, where I think you're naturally more, you'll be pulled in those directions by lots of people who might be more interested in the novelty rather than the utility side of what you're doing.

43:49Totally. I totally agree with everything you just said. It feels like this technology is progressing so quickly, and it sounds like it even has for you just in the past year. If we play this out, what does this end up doing to traditional means of video creation? I know you said you're not trying to make Hollywood films, but maybe other companies are. What happens to cameras? What happens to timeline-based video editing? In Synthesia, you don't even need a timeline, right? You just type a script. You tap a button. it's done. So what is, give us a, give us a peek into the future five, 10 years from now.

44:29So I think to a lot of people, you know, this sounds kind of crazy. I think maybe less so today than it did definitely like six years ago, but even a couple of years ago, it's like, how's it even going to work? And like, who would want to watch AI generated videos and all those things, right? The best way I think to think about it is just looking at the history of media technology and technology in general, right? Because I think history repeats itself all the time. And even if it feels new, now that we're in the middle of a cycle, I think we're going to just see the same thing happening again, right?

44:59When there was a time, you know, when to make music, you had to have real instruments, you had to have a huge recording studio, and there's very few people in the world who would afford to do that and knew how to do that. Then something like a drum machine was invented. And the drum machine was actually originally invented to replace drums, right? That was the intent of the drumming machine. But it turned out the drumming machine, like, is just very far away from being something that you can replay the drum kit with. But what it did open up was you can buy this little machine, and you don't need a drum kit anymore, and you can use it with headphones.

45:32And all of a sudden, we get, like, house and techno electronic music, right? Along with synthesizers and all these other things. And what slowly has happened is that what used to be an entirely kind of analog workflow with real instruments now is completely compressed in a MacBook, right? where you can, as we spoke about earlier, you can use software instruments, you can use samples, you can play an entire orchestra on your keyboard if you want to, right? You're not really restricted by having to go out in the physical world. But of course, people still play piano, people still play guitars, and they have some soul you can't get out of a machine.

46:04We like it as humans to do those things. I think the same thing is going to happen to video production. I think we're going to have a very real alternative to recording in the real world. And I think a lot of people are going to use that. But I also think we'll have normal video that's recorded with a camera. I think people will like that and it'll probably just end up being a different genre than the content you can create with AI. And I think the shift is, I often think of as the shift from typewriting to computer. I think it's ahead of a lot of those same elements. If you look at it from an office and workplace perspective rather than a creative perspective, the act of writing, for example, that used to be someone's job.

46:48You had dedicated secretaries whose job it was to write things down and write letters for the CEO or whoever is in a company. And you had typewriters, and there was a whole industry around that. Then once keyboards and computers came, everyone could write. Now this is a part of everyone's job to write. And we don't even think twice about that. We write SMS, we write WhatsApp messages, we write emails. We've even elevated ourselves. Most people now are also designers. Most people can use PowerPoint if you have a white-collar job. You may not be an expert. Canva. Canva's helped. Exactly. Canva is a great example of that.

47:25I think the same thing is going to happen to video and audio. I think as humans, we want communications that are as close to a real conversation as we can. The reason there's so much text in the world, not because we prefer text, is because it's the only really scalable way of communicating we've had. And once it's going to be as scalable to create podcasts, audio and video. I think we're just going to move towards a world in which visual and audio content and maybe someday VR will just be the default way to consume content, right? Yeah, I wonder, you made some really good analogies. I wonder if vinyl is another analogy where 90 % of the world, 95 % of the world listens on Spotify, right?

48:07But some people like to listen on vinyl or filmmaking. So many of the films we all see today are shot digitally, but every once in a while, Christopher Nolan makes a film like Oppenheimer on classic analog film, right? Maybe it's like that, right? Where the efficiency gains really are, people will use AI to generate content in the ways that are most efficient and valuable. But there will be the purists, right? That want to do something like something old school. For sure. And I also think that the best content we'll consume will be made by experts, right? And the creatives that we have today, like those who pick up all these new tools.

48:48You could argue that AI at some point will become as good of a storyteller as a human being. And I definitely won't argue against it. But at least like here in the near term, right? We always, to some extent, have democratized video creation, right? Like we have cameras everywhere and you can record yourself without needing much more than just a smartphone. but yet, right, it is the greatest storytellers, the greatest musicians who will win out and make the most interesting content. And I think the same will be true in the AI world, right? Like, you can regurgitate AI music and AI images a million different times, and it'll just end up in kind of like the junk folder of the internet, which is already what like 99 % of YouTube videos are, right?

49:27And the few ones that are really good will rise to the top. So I think the human ingenuity, the human storytelling element, it's still going to be an incredible advantage. And the best artists, the best storytellers are generally also those who go against the stream. And at least right now with these models, what you are finding is that they're great at kind of spitting out reactively what's happening in the world and what kind of art people like and different kind of styles. And with music, it's going to be the same, right? Like make me a hip hop track in the style of like someone else. But great new art usually comes from people combining two things that should have been combined, right?

50:03And create something that's like fundamentally new and interesting to the world. And I mean, I can't, of course, say that like AI will never be able to do that. But I still think that that requires more than just understanding news. That requires a very deep cultural and human understanding that at least so far, I don't think AI is anywhere near being able to comprehend and react to. how close are we to being able to make our own avatars and our own voice models instantly, right? Like, like right now I know in, in, in the product, uh, I can make my own avatar. I think you also let me make my own voice model, but I think it requires a different process and more manual process.

50:45How far are we away from me just saying, you know, not training, not, not, not even needing maybe to train my own model, just typing a sentence, wanting to do it in my image and likeness and it's just happening? I mean, honestly, I think we're like more or less almost there. Really? Both us and like other companies, you have like what's called zero shot voice cloning now, which is essentially from like just a few sentences of your voice. It can generate like a pretty good approximation of what you sound like. It'll get better with more data. And you also have some other companies doing like with just a single image, they can create an avatar of you.

51:18Is that our technology? I think the quality there is still pretty far from what you can do with video. but for you to have an avatar of yourself kind of like you're sitting and talking right now I mean I think if we put the bar as like you know 10 minute time investment from you like in six to nine months until these at least you'll definitely be able to do that very rapidly and it'll be very very very difficult to tell that it's not a real video we're getting ready to release the next version of our avatars which is going to take them from this bit slightly kind of robotic very like corporate presenter style with a lot of emotions to giving them emotion and teaching them how to perform a script depending on the context, right?

51:59So if you put in the script of a Tolkien novel, for example, you probably want that to be read out kind of like an audio book, a little bit slow with like lots of intonation. Whereas if you put in the script of someone selling you like a used car, you want that to be more upbeat or the body language to match that. And I think it's like, as we go through these kind of iterations, we're just not far from being able to generate video and audio that is modest and distinguishable from the real thing. The implications of that are fascinating. You know, I know you've talked about how Synthesia, you know, the, the, the promise of it is really being able to turn text to video really, really easily.

52:38I mean, once we have the technology you're talking about now, it almost feels like anyone could do that for any content anytime they want, you know, again, using the, the Tolkien book as an example, maybe I go to read this book and I decide, you know what? I want, I want Tolkien to just read it to me directly and I can just spin that up on my own. That, I mean, the implications are so far ranging for that, both for good and obviously also for potential abuse. You've been really, you and the company in general have been really thoughtful about this, about, about the potential for abuse and AI safety overall.

53:14How do you think about that? Yeah. Totally agree. I think, you know, there are so many exciting opportunities, but there are definitely also some scary dimensions of these technologies. So I think, you know, when we found the company back almost six years ago and people were talking about deep fakes, I was like, that's the only thing people could talk about when they talked about this technology, right? We built the company on an ethical framework, which around what we call the three Cs, consent, control, collaboration, which is pretty simple. Really, it's just about like we don't create avatars or voices of anyone without their explicit consent.

53:48So you have to go through this KYC style process to actually make yourself an avatar. And Control is all about like we take a very strong stance on what kind of content you can create with Synthesia. So around 10 % of the company works on AI safety, moderation. And it's one of those problems that seems simple on the surface, right? But are really, really complex once you dig into it. you have the content you clearly don't want the content you clearly want and then the really difficult part is all the content in the middle if you're creating crypto content you're trying to scam someone out of their life savings or trying to explain how blockchain works those types of things it's still hard to do that automatically so we also employ a lot of humans to actually look for those scripts I think for us we take it very seriously that we should that we play a very important role in how these technologies are kind of integrated into society And I think every company should.

54:40And I think if you look at most of the big companies, at least, most people are taking that seriously. OpenAI does a lot of work on making sure their models don't spit out bias, disinformation, misinformation. So I'm actually really happy to see the industry kind of rally around this. And I think for me and for us, I think of it as like two problems. There's one, which is, how do we make sure Synthesia is not misused by anyone? That's what I just sort of described before. And I feel fairly good about that. With these things, content moderation is hard. We're never going to be perfect. But I think we're investing a lot of time in it.

55:16I'm actually really confident where we are at the moment. The other part of this is like outside of Syntesis, these technologies will develop, not just around video generation. And that's much harder to control. I do have a lot of open source that comes out. And I'm definitely pro open source. But if you look at the bad actors today that are using open source, because that's, of course, come to like no safeguards around it. Might be other companies and the less ethical actors. You'll have lots of other problems around these things. And without going too deep on it, because I can talk about this for hours.

55:50We'll do a separate episode on that. The tech that I find really interesting, the solution to this problem, the obvious ones is like education, for example, right? We need to expose people to this type of content as quickly as we can in good positive use cases so that people understand that this is now possible. That's like ground level. I really hope that schools start teaching people about generative AI, make them generate images, use chat, GPT, Synthezia, whatever. That goes without saying. Education is super important. There's been for many years this detection idea that's still floating around.

56:24And I think it might be one signal you can use in the future. But the problem with deep fake detection, for example, is that it's super unstable, provides a potentially false signal, and it's an internal cat and mouse game, right? Like the people who are motivated to try and bypass those systems will put a lot of effort into doing it. So I don't think that's the solution. What I think is really interesting is this idea of provenance. So we're working with Adobe on this. They have something called the C2PA, Content Authenticity Initiative. I've heard a bit about this. It's really, really interesting.

56:55It's super cool technology. The idea is basically that you want to kind of have a provenance chain for all content you consume online so that you as a user know who created it, know who it uploaded, how is it created. And this goes above, just outside of just like generative AI, right? This could also just be, you know, any kind of video clip where it's actually originated from. We don't know that today. And you could do that by essentially embedding this kind of invisible watermark slash metadata inside the content itself. And if you have that, I think we could move to an internet where the default would be that you're watching trusted content.

57:34And if you're seeing something where there's no provenance chain, you as a user would be alerted to this. That's kind of the opposite of how this trust system works today, right? So if you go to Instagram or Facebook or whatever, you will have celebrities that are kind of high-risk targets. They'll have like a verification thing so that you as a user know that they are verified and it's actually them. So the way it works today is that you are actively told if something is verified. We want to move to a world where you're told if something is not verified, right? And so I think that from profiles and social media networks all the way down to the individual piece of content, if we start tagging all this content, we can actually move into a world where you as a user will have probably, hopefully a fairly good way to assess and critique the content that you're watching online.

58:24It's a gigantic task from a technological perspective and it's definitely not going to be easy, but with music, this has kind of already happened, right? If you upload a video of yourself dancing to Michael Jackson in your bedroom, YouTube will analyze the video when you upload it. They'll detect that there's some copyrighted music in here and they'll slap an ad in the middle so they can pay the rights holders. And this is sort of like the system we want, right? I usually explain it as like Shazam, but for media content, where when you're consuming some content, you as a user can kind of like Shazam it and you'll be able to see roughly kind of where it originated from.

59:00Because this solves not just the deepfake problem, it also solves problems with you taking a video of an explosion that happened seven years ago somewhere and then saying that it happened yesterday, for example, which is most misinformation and disinformation today revolves around like what's called shallow fakes, right? So I'm really excited about that kind of path to the technology. And I think great strides have been made toward it, but it really requires that the whole ecosystem collaborates from content creation tools like us all the way to distribution platforms like YouTube and Meta's platform and so on.

59:34It's so interesting. Yeah, I was talking with Adobe's Scott Belsky recently about this exact problem that you're talking about. And we were discussing how music, as you just pointed out, has a similar structure that's already in place that people have been bad-mouthing for decades, right? It's a very antiquated system. It's extremely manual. But now that it was as we entered this world of AI, where everything that gets created is a derivative of a derivative of a derivative. It's actually like kind of a model for what the future should be. And so it's pretty cool to hear that companies like yours and Adobe are heading in that direction.

1:00:13Very, very cool. What do you think about for Synthesia? What do you think about other media formats? Will you always be super focused on video or are there other formats you want to tackle eventually? So the way we look at like AI video today is that it's still incredibly early, right? We're kind of like at the time in history when like the first websites came out, they looked like a newspaper on a screen, right? That was kind of like what people could imagine. They would like take their physical newspaper, put it onto a screen. And already that was a pretty good product, right? Like, it's pretty cool.

1:00:49You can just sit behind your computer and like go to the website and you can kind of read the newspaper. Since then, people then figured out that you can actually make links. That's pretty cool. You can make video and audio. You connect people together in like a commentary field or whatever. You can do so many things on a website that you obviously cannot do with a newspaper, right? And today, like the structure and the mental model on a website and a newspaper is just, it's so completely different. But they started out looking very much the same, right? And I think that's also where we are with AI video and AI content actually in general.

1:01:18That we're still in this phase where like what we're doing today is essentially offering you a better way of creating MP4 files because we've kind of taken the camera out of the equation. And that's really powerful in itself, as I think, you know, the growth of the company and everything we've done so far kind of proves. But as time goes on, I think we'll see AI video and AI content in general evolve into its own new type of media format. It's so hard to predict what that's going to look like. But once you remove all the constraints of working with hard-baked linear video footage, it's just obvious that this is going to happen.

1:01:55And that's really what I think is going to happen here. I think in our world, videos will be non-linear. Maybe you can talk to the avatar if it's something you don't understand in the video that you're watching. It can be interactive. We do so many things with these videos that we probably can't really comprehend yet. And with music, maybe it's like every time you play a song, it's played slightly different. if you wanted to, right? You can do like a slightly different grunge shuffle or slightly different melody. Like I think it's going to afford all these new types of media formats. And so for us, the way I think of Synthesia in the long-term is that we've only been really, really successful if we've managed to actually breed a new media format that's not an MP4 file, but it's something different, right?

1:02:34And that's very exciting. I think that's ultimately what I just can't wait to see. We know that the technology is going to get good enough to synthesize extremely lifelike photo-real video content. but what are people going to do with this right that's that to me is the exciting thing right and i think it's i'm excited about that because i feel like this it's like being in the middle of when electronic music was invented right or something like that where you have all these people experimenting and doing different things with it and in my lifetime we've already seen it you know i think when when um when live streaming for example i love this example in particular became you know a technology that we could put into our mobile phones and you could just live stream from anywhere, right?

1:03:14And people kind of linked it with broadcasting because that's the way you could think about that. All the McKinsey consultants was like, everybody's going to start TV stations, right? Everybody's going to do all these things that people are doing with live TV. And absolutely no one had guessed that the$1 billion use case is going to be watching other people play computer games on Twitch, right? For all these sort of things where there's something there about if you know the game, you can jump into a stream at any time and you'll have full context of what's going on. That's pretty well suited for like internet streaming, right?

1:03:46There's, you know, this whole community of, I mean, there's all these different things that just all kind of came together. And that was just like, okay, Twitch is the thing, right? That's the thing. That's what all everyone wants to do. And so I think, you know, that kind of made live streaming to a different format. Now YouTube has come after it, but that took YouTube like five years, I think, right before they had a good live streaming product. And so I think, we're going to see this play out again. And again, like my mental model, this is just history repeats itself over and over and over again.

1:04:15And we always think it's different this time, but it just rarely is, right? So I am even internally, we're very kind of like, we don't want to be prescriptive around what we think this is going to look like. We want to learn with our customers and sort of one feature at a time, build our way towards that future. But I can't wait to see where we are in three years' time. It's going to be very exciting. It's so true. and it's so interesting right like it almost feels like we're in this phase of ai right now where we're trying to map everything that gets done to existing formats or legacy formats right and to use your example from earlier it's like when the drum machine came out you know the first use cases were probably how do i replace the the kick drum in this rock song and it's like no no no this isn't actually this isn't for rock this is a whole new genre and it's called electronic music right like if you and if you take that example right it's like you can do things with a drum machine you could never do with a real drummer right like you you could literally have like 10 drummers on 10 different drum tracks um in in a machine that's like the size of like a vhs machine or something like that back then right you could play you know at speeds and like different patterns of overlapping ways you really couldn't do before and and that is the exciting thing about new technology i think right it's just absolutely it's and i i think it's so i think a good mental model for people who are building ai today really is like what is it you can do with this technology that you couldn't really do before.

1:05:37There's one element of this, which is just like, because you want to do something cool and exciting. But it's also, if you can anchor it to something, you know, that kind of, if you anchor it to this like newspaper on a screen sort of thing, if that's like what you're building for an incumbent, it's also going to be much easier to just add that as a feature to whatever product that they already have. Whereas if you're building something fundamentally new, that where everything revolves around, for us that would be AI voices and AI avatars. You can build something very new, very cool, very unique, as opposed to if you're just like vaulting AI onto a customer support chatbot or whatever, right?

1:06:15My favorite example right now is like Character AI. I don't know if you're familiar with that company. Yeah, absolutely. But it's like, if you told people one and a half year ago, two years ago, that one of the most retentive use cases for like chatbots was going to be people talking to LLMs for hours, right? Like this, that's just insane for hours on end, despite knowing that it's an AI chatbot they're talking to. People would have thought you were ludicrous, right? You wouldn't believe it. Yeah. You wouldn't believe it. But this is now one of the things where like, that is probably one of the most promising AI companies right now who did something completely new, right?

1:06:51And it's not about like creative writing. This is literally about like having an AI friend or whatever way to phrase it. And so I think we'll see so many more examples in the next like 12 to 18 months now that these technologies are kind of like more broadly available, easy to build with. And it's just going to be exciting to see what people build. Totally. Before we wrap, I have a couple of fun ones for you. So obviously, you know, Synthesia is easily one of the most hyped, one of the most exciting AI companies of the past year. What has been some of the fun kind of results of this? And I'll just throw out a couple examples.

1:07:24I saw that Messi made a video using the product. I saw that you recently went to 10 Downing Street. Like, talk about some of the fun things that have happened over the past year. I mean, there's so many good memories. I actually think one of the best ones is like all the way in the beginning when we started the company. This kind of sounds like one of those founding stories that I kind of made up after the fact. But everything here is actually 100 % true. So when we tried to raise the first round of funding, we thought we had like these amazing professors, great idea. We were like, I think it was 24, 25 at the time.

1:07:59So we went out into London where we're based, right? And we just got no, no, no, no. I think we got like a little bit under like a hundred no's, like nobody bought into it, right? It sounded crazy. And I think, I don't think we were like, we're not the typical founders that people were looking to back. And so after this very long, very painful period, close to giving up, we've like, you know, we liquidate all of our Bitcoin and crypto stuff, of like pay for like our first employee to like build the tech. Then Stefan, my co-founder, sends a cold email to Mark Cuban, which he's found his email in the Sony hack one few years before.

1:08:36Saying, hey, Mark, we're building this technology. Do you think it's interesting? He replies back five minutes later. I, of course, don't believe him. I think he's just like pulling a joke on me. But it was actually him. And then we have a 13-hour long email conversation with Mark Cuban asking insanely good questions. He knew everything about this technology. He had implemented himself in his free time with a lecturer. He already, it was one of those things where like the thesis we completely agreed on. He was like totally on board with that. He was just trying to understand how we were thinking about actually getting there.

1:09:12And after those 13 hours, he just said, I'll do the million dollars, assuming diligence goes through. That was like 4 a.m. for us. And, you know, had some champagne. and then we jumped on a plane to LA to see him in the days after. And that was just like an absolutely magical experience. And to this day, still so impressed by Mark. I think his public image in like Shark Tank is great, but he is like, he really gets technology and he understood Genitive AI like way before anyone else we met at least, which was awesome. And it's been fantastic to partner with him. That's super cool. That is an amazing story.

1:09:50Real quick, your company's based in London. There's a lot of buzz right now about Europe, London and Europe more broadly, as a very sort of growing tech hub. Give me your thoughts on Europe tech. I think we're in a good spot, right? I think when we started six, seven years ago, that's been a lot harder. I think there's been a few things that has helped significantly. I actually think COVID has been a huge factor here because in COVID, right, everybody had to raise money entirely via Zoom. Like I used to do the pilgrimages to the Bay Area before COVID to try and raise funds. But over COVID, right, the whole world just became more global.

1:10:32And I think that's been really helpful to European companies who all of a sudden had much better access to the American investors, American market, which, you know, my read of this is especially for the seed stage. it's been great because as much as I love Europe, I think European investors are still more of the kind of, you know, reduce the risks mindset rather than like seeing the upside. So I think a lot of money, of course, went into European startups in the last couple of years. That's really awesome to see. And it's with the remote culture, it's much less controversial now to be based in Europe.

1:11:05And in AI specifically, there's also just amazing talent, specifically in London. And that is just very important for an ecosystem to develop. So we have Oxford, UCL, King's Imperial here, and they are some of the world's leading AI labs. You take that and then you combine it with some earlier success stories. So PeepMind, for example, was founded in London at UCL. You really start to get this very healthy ecosystem of people wanting to not just work in FANG and in research, but also start companies. So I feel like Europe is a great place to be. There's pros and cons to everything. but I think as a founder, I think when you're doing something that works, there's less competition because there's less great companies in Europe, as much as I hate to say it.

1:11:52It's early. It's early. I think that's actually quite a good competitive advantage and I'm personally really excited to see where Yelp will go. One of the sort of scary things with Europe right now is the AI Act. I don't know how much you've been following that But AI regulation is, of course, a huge topic around the world. I'm very pro-regulation. I think if you have the right safeguards, the company should be held responsible. But some of the suggestions that are up at the moment, I don't think it's going to be beneficial to Europe's growth in the long term from a startup and AI perspective, fortunately.

1:12:26This has been awesome. Thank you so much, Victor. Really appreciate you coming on and looking forward to talking to you again soon. Thanks for having me, Mike. This was fun. Thanks, Tom. Thank you so much for tuning into my conversation with Victor. Follow Generative Now, this podcast, wherever you get your podcasts, whether that's Spotify or YouTube or Apple Podcasts. Also, if you enjoyed the episode, please do us a favor and rate and review it. That really helps. If you want to hear more from me or Lightspeed, you can follow me at Mignano on X slash Twitter, LinkedIn, Instagram, everywhere else.

1:13:02And if you want to follow Lightspeed, you can follow them at LightspeedVP on all those same platforms. Thanks again to Victor. And Generative Now is produced by Lightspeed in partnership with Pod People. Special thanks to our production team, including Rebecca Chaison, Brian Rivers, Grace Pena, Amy Machado, and Rachel King. We will be back next week with another awesome conversation. We hope you tune in. Don't miss it. See ya.

1:13:31Thank you.