Where AI video is headed next, with Synthesia's Victor Riparbelli

15:30Victor Riparbelli:Well, SpaceX is going to IPO. We know that Elon apparently wants to do it around his birthday in the spring.

15:38Ellis:I mean, truly, truly a little boy.

15:41Victor Riparbelli:Yeah, I mean, it's a great birthday present. We needed more evidence. And merging XAI with this thing is going to I mean, it was already going to be a meme stock, but this is going to be the biggest meme stock of the year unless maybe even still if open AI IPOs later this year, as they're rumored to maybe be considering. I mean, the to the moon memes on Reddit, on WallStreetBets, I mean, you can't write it. I mean, it's baked into the name. It's going to the moon.

16:08Ellis:So on the nose, shame on anybody for posting that.

16:14Victor Riparbelli:Lastly, I just want to mention a thing I wrote about in sources this week, which OpenAI released a Codex Mac app. It's kind of crazy to me how Anthropic has really run away with the narrative on coding, AI productivity and seems to be kicking OpenAI's ass on products in a way that you probably wouldn't have predicted a year ago. And I was on a press call with Sam Altman and the OpenAI execs last week and talking about this Codex app on Mac that they're launching. And it was funny. They were throwing digs at Anthropic basically saying, yeah, you know, like whatever, when people want to fix their code, they bring it into Codex.

16:54Victor Riparbelli:So I think they're recognizing that like the top of funnel awareness in AI coding has definitely gone to Anthropic, but they think that Anthropic is buggy-ear, which I don't know if that's a flex or not. But OpenAI is trying to get that coding mojo back, which I think is just an important little footnote in the AI race.

17:16Ellis:It certainly seems like they're doing a whole lot of things over at OpenAI. Do you get the sense that they're spreading themselves too thin?

17:22Victor Riparbelli:I think that's the concern. I think they're going a little more googly, but much earlier in the life cycle than when Google really started to become a conglomerate. I mean, the company as it exists as a commercial thing is only, you know, a few years old. And yeah, Sam's already kind of setting it up like a conglomerate with all these random things. And I do think Anthropix focus is really coming through, right? Where they They are really focused on the think boy and the coder. And, you know, I got the thinking cap, by the way, in Davos. And, you know, I think they're carving that lane out and the focus is helping them a lot.

18:04Victor Riparbelli:And I do think there's starting to be some chatter that maybe open eyes is a little too pulled apart at the edges.

18:12Ellis:I mean, having seen like in the early days of Snap when so many different bets were made, many of them successfully, I think these days, if all those teams, if the talent density is there and now it is way, way easier to spin up product, you don't necessarily need hundreds of engineers anymore. and you still have someone at the top applying a strong editorial point of view with vision over kind of the cohesion and the thesis of each product. I do think it's possible. I think the problem starts when the talent density and the talent quality goes down. And if you start spending too much time reading the news to try and place chess pieces on every square and stop doing your own thing, whatever it is that you want to do best.

18:58Victor Riparbelli:OpenAI's leadership is definitely spending a ton of time reading the news and X. There's no tech company on Earth that is more reactive to what is happening on X than OpenAI. It's pretty amazing.

19:09Ellis:I feel the product quality is good, though. I guess one of the ones that I feel like Atlas has kind of fallen out of the conversation, which is not unusual for any new product in tech to fall out of the conversation.

19:23Victor Riparbelli:Sora. I mean, God, remember Sora? We were freaking out about Sora.

19:26Ellis:Totally. Yeah. Yeah. But I thought the quality of Sora in particular was pretty impressive. We had a good weekend on there making videos. We did. We did. That kind of relates to that. I mean, if you look at why it dropped, I think you could argue that you're just simply not reminded of it, which is why that Alfie messaging app we talked about last week was interesting is that being in a group chat, once you're trained to turn inside jokes into media, that can start rolling. It took a while for people to build the habit of when they see something funny to snap a friend. That was brand new. Nobody was going to half the time open up their camera, take a picture, switch to the messages app, click the attachment button, add an attachment, hit send.

20:12Ellis:And so creating those new habits is very, very hard. So yeah, I think that's probably why Sora didn't happen as impressive as the results were.

20:20Victor Riparbelli:Well, luckily we can keep the AI video conversation going because we have a AI video CEO on the show today. You like that segue. I just came up with that. Let's throw it. Should we? Yeah, we should. Let's throw it to Victor Ripperbelly, the CEO and co-founder of Synthesia.

20:44Ellis:Support for the show comes from Amazon. There are the things you can plan for. A first birthday party, a movie marathon, a renter-friendly bathroom reno. And then there are the things you can never plan for. A surprise rainstorm, a Blu-ray player calling it quits, stick-on tiles that looked way better on the package. For all things planned and unplanned, Amazon has you covered. You'll find low prices on everyday essentials and last-minute lifesavers. shop Amazon and save on essentials save the everyday K-pop Demon Hunters Saja Boys Breakfast Meal and Huntrix Meal have just dropped at McDonald's they're calling this a battle for the fans what do you say to that Rumi?

21:31it's not a battle so glad the Saja Boys could take breakfast

21:34Victor Riparbelli:and give our meal the rest of the day it is an honor to share no it's our honor it is our larger honor no really, stop

21:43Ellis:You can really feel the respect in this battle. Pick a meal to pick a side. And participate in McDonald's while supplies last.

21:57Victor Riparbelli:How's it going, Victor? What's up? Welcome. Pretty good. Thank you. That's some nice bedroom apartment lighting you got going on. I've been spending like 10 minutes trying to find a proper setup. I'm in Utah right now in a hotel. Oh, Utah. Yeah. So one of our investors, Excel, has like a CIO summit thing that they do every year, which is basically like a bunch of CIOs. And they invite some founders as well. So it's usually like pretty fun, good for business. And we get to stay in nice hotel rooms.

22:29Ellis:That's rare that a summit is actually near to an actual summit. And it's not just a pure symbol. So part of the trip is actually that you go snowboarding or skiing, whatever you're into. It's also funny with what you were saying, Alex, I feel like a few years ago during COVID, everybody was like, oh, is your background real? Now it's like, are you real? That's how far we've come in just five years with AI and video gen.

22:58Victor Riparbelli:Yeah. And Victor, in case you don't know, we roll, we go when you get on. So we're going here. It's good to see you, man. It's been a minute. I think last time I saw you was in Vegas where all good relationships get formed. So good to be seeing you again. A few hundred million dollars raised later. I was looking on X actually today, and it looked like you guys had a little problem with spoiling your surprise internally. There's something with a cake. Can you fill us in what happened here? Well, I mean, as you know, like when these things happen, especially at scale, they unfortunately very often leak to the press.

23:34And then you're in this like awkward situation where everybody kind of knows what's going on. but you can't really confirm what's going on. And so we had a lot of internal chatter with me and Stefan just being like, yeah, maybe the rumors are true, maybe they're not. And we had some cake in kind of pre-celebration, I guess, of the round.

23:57Ellis:You say we're trying to do founder-led content that's super transparent, but we don't want to be that transparent. Well, it's really difficult, right? Because with these things, I think our philosophy has always been that no round is closed before the money is in the bank. And these things take a long time. You get to an agreement with a bunch of investors, then there's a lot of paperwork, a lot of stuff happening. Luckily, we've never had any kind of bumps in the road. But it has definitely happened before in history that something goes down. There's a huge down. Anything can happen. And then all of a sudden, you're standing there looking not so great.

24:32So I think maybe it's very European of us. Maybe we're just very conservative. But I think the way we've done these things, historically, it's always been to, you know, before the round is actually like sipped up, we don't really talk about.

24:42Victor Riparbelli:What's it like being on the other side, having it out now? Do you feel like you can puff your chest a little bit? Yeah, it's great. But kind of on that point, right? Like, you know, it's kind of interesting because these funding rounds usually take place like many months before the actual announcement. So in our case, you know, sort of October, November is when this round was kind of sipped up. then a bunch of diligence, make it all go around. You want to figure out when you think it's a good time to announce it. So it's obviously amazing and it's awesome, but it also almost feels like old news internally when you go out, right?

25:15So it's kind of a strange feeling.

25:17Ellis:Well, what's cool these days, Victor, you got to raise the next round while you're raising this round. And so it's just kind of a cascade of money that never ends. That's what the AI world seems like right now. I was going to say, yeah, I think half the companies, I work with now, they're working toward a raise and they pivot during it or they pivot right after it and the investors are like, wait a second. It seems very hard to plant your foot in the ground on almost anything firm in the AI era. For sure. I mean, looking back to 2017, when we found the contract, the first three years, you'd have like kind of a breakthrough of some sort in what we back then called synthetic media before it was renamed to generative AI.

26:03Maybe like once every year, once every one and a half year, there's a piece of like big breakthroughs. Someone writes like an interesting paper that kind of, you know, increased the quality of the capabilities by some, you know, margin. Now it feels like every three weeks, there's like a new bomb that gets dropped, right? And that's across all modalities. So I think both investing in and also running AI companies right now is very exciting and very thrilling. But it's also hard, right? There's just so much uncertainty. You don't really know what's around the corner.

26:32Victor Riparbelli:I didn't appreciate your all story when we met in Vegas at HumanX. Just how much like pitching you guys had to do in the early days, how much you were rejected. The Mark Cuban story, which we have to talk about, which is great. And then I saw somewhere that one of your co-founders was what? He worked on like a pig farm, something like this? he did he did private equity in africa before we teamed up private equity same thing yeah but they were they were running like you know sand mines and chicken farms and that kind of stuff that's pretty funny best possible roll-up for private equity

27:12Ellis:not swimming pools and lawn mowing businesses but chicken farms so you you have this kind of

27:18Victor Riparbelli:famous story now where you all cold emailed cuban and he wired you a million and now he He's very happy with that investment, obviously. He owns a big chunk and your valuation is huge now. Is he active at all? Do you lean on him for advice or intros or anything? We speak occasionally, but I think his style of investing is very different from a VC, right? He's much more like an angel. So I think the way most business angels work, including myself, now that I also invest, is more kind of on a sort of reactive basis, right? Like we're always open for someone has a problem, jump on a call, discuss it out live.

27:55I think it's been the same thing with Cuban. Cuban doesn't do calls. He only does email. He only does court sides.

28:02Ellis:Who needs desk sides or fire sides? We got court sides. Exactly. Yeah. How many games have you been to? We've actually only been to one game. None of us are like really big into sports. But we went there in the early days. We also did a bunch of the first demos we did back when we were doing dubbing, which was the first three years of the company. We went there and we did a bunch of cool content with the players. And I think his main thing, his help has always been access to people. So in the early days, you know, getting access to like sports stars, like his own team, for example, to do cool demos.

28:36He knows a bunch of people, both from like investors to operators. I think that's always been his main contribution, more than it's been kind of, you know, like a weekly sort of catch up.

28:48Victor Riparbelli:It's good to have a shark in your corner. It is. Actually, it's very interesting. I think Mark Cuban, he's obviously known for Shark Tank. I think people's perception of him is very anchored in him being on Shark Tank, investing in all these sort of like kind of wacky businesses with someone who made some fun invention that they'll do well on TV. And I think a lot of people actually don't know that he's very deep technically. His company Broadcast.com was the first company to take radio to the internet. and he went through a similar kind of story that we did where everyone was laughing at him, right?

29:20Like, why the heck would you want radio on the internet? You have a radio in your kitchen and your bedroom. And he kind of like pushed through that and ended up selling the company for three point something billion dollars to Yahoo. And when we met Mark Cuban, you know, way back in the day, and this is at a point where AI video was very French, right? Like no one was really talking about this. He had actually implemented the research paper that Synthesia kind of in spirit originates from, which was actually a real-time system for doing avatars. He had implemented that at home with a kind of a lecturer of some sort.

29:52So he knew everything about the tech. And what I always say is that I think the reason he invested was that we didn't have to convince him of the vision that in 10 years' time you can make a Hollywood film on your laptop with just your imagination, which was kind of like what we were trying to tell people back in 2017, and no one believed us. He already believed that. And so when he was evaluating the investment, He was more looking at us and he asked a lot of tactical questions. How are we going to sequence it? What's the first use case? Who are you as people? And invested on the basis of that.

30:24And I think where we fell flat with everyone else was everyone was looking at us and going like, Hollywood films on laptops in 10 years. I don't really care about that. Sell me some fintech. That was what everybody wanted. Fintech or machine learning for doing predictive data analysis on corporates, big data sets. That was what everybody wanted to invest in back then.

30:45Ellis:Well, I think there's an opportunity for differentiation here. If we can get Cuban in the model itself, his likeness, his voice. I see a moat there for you, Victor. Maybe we can make the first AI Cuban that can automatically make investment decisions on behalf of him. AI Cuban. I'm curious, though. I mean, when you talk about the access to sports stars and this and that, I mean, I feel like being in B2B world, those connections can kind of mean everything. You know, the CTOs you get to meet, the founders you get to meet who decide to like take a chance on you, which is, you know, a very different type of path, I would say, than in consumer.

31:25Ellis:Have those relationships been a big lever for you? Bringing in will.i.am to close a deal. The ultimate trump card. I think they have, especially a technology like ours, which is, I mean, we're a very B2B enterprise focused company, but there's a lot of like consumer elements to it, right? Like video is mostly consumers watching it. And I think, you know, like media in general is obviously like a very consumer, it's a consumer world, right? So I think some of the, where it has the most leverage, back in 2018, we did a campaign with David Beckham, where we basically had him speak in nine different languages using our AI dubbing tech, which, you know, I mean, back then the tech like barely worked.

32:10It didn't work with AI voice. So it's actually like a real person who spoke in a different language. And we would take that person's voice clip, and then we use that voice clip to animate David Beckham. So it wasn't actually David Beckham's voice. It just looked like it was him, right? So very, very manual compared to what we can do today. And that campaign was the first time I think the world saw deepfake technology used for something good, something productive. And that campaign ended up being kind of a pretty big turning point for us. Because I think it just, first of all, it was like a big viral moment.

32:44And I think it kind of got people thinking about this technology in a slightly different way. and that would have been much more difficult to do if it wasn't David Beckham who had kind of consented to be deep faked as they would call it back in the day and kind of standing up for the technology. So I think that has had some impact for sure but I think since then we haven't done a ton of celebrity based content. We've done some and I think it definitely helps to bring up the brand of the company but I wouldn't say it's been within top five things that have helped the company grow faster.

33:21Victor Riparbelli:Since the last time I saw you, something else happened where you did a TED Talk. And I was wondering what that was like doing a TED Talk. I've always wondered. I'm sure there was a lot of pressure or maybe not. Maybe you're like totally cool with that and it wasn't any pressure, but you had like a provocative title, why I make us the last generation to read and write. How did it go? What was it like being on the TED stage? It's actually funny because I'm generally not like nervous anymore when I have to do like public speaking and doing podcasts with your lovely people and so on. But the TED Talk felt different in a way because it's a big thing.

33:58I remember like 10, 15 years ago when TED Talk was launched and it was like this huge thing and you would sit and watch a TED Talk. So I definitely felt like there was a lot more pressure. I also think you, I felt like it has to be like an interesting idea. It has to be a provocative idea. And I think I managed to deliver on that when you look at the comments. Like 90 % of the people disagree with me, which was kind of also the point of it. And I think I hit a hit a nerve there. But it was really fun. You work with the tech team. Of course, you come with your idea and then they kind of help you shape it into like the right format.

34:29And I think it's probably the talk I've done that I practiced the most before actually doing it because you do it in front of a live audience. Right. So you get one take and there was a really, really, really fun experience.

34:39Ellis:Well, Victor, did you deliver and did you answer that question or did you do a whole TED talk and then walk off the stage, as you say, and only time will tell. I think I put together like a pretty convincing case for why on a long enough timeline this is going to happen. Of course, I'm being a bit provocative by saying that we're never going to read and write anymore. I don't think that necessarily is going to be the case. But I do think that within the next 10, 15, 20 years, reading and writing will be something that we do either because we enjoy it for fiction, for example, or because there's going to be like a small subset of use cases where voice and video isn't better.

35:15So I think if you watch the talk, I would say give a compelling case why it's going to happen, but I'll let you be the judge.

35:22Ellis:Well, so how's Alex's newsletter going to be in 20 years?

35:25Victor Riparbelli:A video version? I mean, I sound kind of screwed in this scenario. I don't think so. I don't know what you guys' media habits are, but I think that when I look at my own way of consuming media over the last many years, there's a time where everyone's reading blog posts all the time. That was like when I grew up in tech, that was the kind of currency, right, of building a brand for yourself and sharing your knowledge. That has changed a lot since then. Of course, there's still, you know, lots of written content. A lot of it has moved to Twitter. People make threads where I'd argue you can compress or you force yourself to compress your thinking into less characters, which I think is generally useful for people.

36:04When you look at, I would say, even things like how I discover music, which is a big hobby of mine. Like that also used to be on blogs. you would like read through things you like listen to tracks a two-minute tiktok video of someone just showing you the music is is a much better format right and i don't think i don't think in every format necessarily is better as video but i think you can i think you'll be you'll do exactly the same that you do today you'll do great research you'll do great interviews then you'll condense those findings into i mean i don't know if you publish this as video but then

36:32Victor Riparbelli:you're already then you're already doing video right oh yeah we're video first great love it then you're already video first, right? And probably at some point, if you aren't doing it already, you'll do like five or 10 minute summaries of the interviews that you do with the most valuable points, right? And then you're already in the world where like no one necessarily needs to read the newsletter.

36:51Ellis:I mean, there's certainly fewer advantages to the writing as there used to be. I mean, I feel like one of the advantages with Twitter or with blogs was always that when you hear something or when something happens, you can get it up on the internet very, very quickly, be one of the first, reach the algorithm before others do. But now it seems pretty clear that in some ways it's going to be even faster to create video. I mean, I do feel like you're always going to need the writing up front to establish what you want to say, what story you want to tell, what the arc is going to be, what the outline of your arguments are.

37:25Ellis:Yeah, I guess one of the advantages still with writing, you can read at your own pace, which I feel like is not easy at all with video, hitting the space bar over and over again. Obviously, there are a lot of places where you don't have headphones on necessarily. But yeah, the speed benefit to writing seems like it's all but dissipating. Yeah, I agree. But also, again, I think it depends a ton on the kind of content you're creating, right? Like if I am, let's say I want to share an update to my team of like a project that's going on, is it then faster for me to sit down and write like one page?

37:59I'm probably going to be a lot more self-conscious about like how I sound, They're going to use the same word twice in the same sentence. I want to make sure my punctuation is on par and all this stuff. And of course, ChatGPT can help me a lot with this. But in a lot of instances, I find doing a quick Loom video is actually much faster than writing. On the other hand, if you're doing something that's very polished, then you're probably going to spend a lot more time on making the video look and feel really, really good versus writing a one-page document. So I don't think there's any one answer to this.

38:27I think it really just depends a lot on what you're doing. And again, I think for some things, again, I think reading makes a lot of sense. For other things, video makes a lot of sense. But I think the world is clearly moving towards video as being the dominant format in the way that most people prefer to consume their content.

38:48Ellis:It's funny you bring up Loom as well, because I think when I talk to founders in our startup learning session, and I do storytelling when I'm not podcasting, Victor, I could surely just ask founders to answer all my learning session questions in a voice note or talk to my AI or whatever. But I find that when people are telling me something in person, they say it differently than if you are going to write it. I think it puts you in the mode of trying to keep someone interested, which is actually what content is about, right? Whereas for some reason, when a lot of people write out long answers in written word, they tend to do it chronologically or they start from the beginning.

39:28Ellis:And so I do think there is something nice about just asking people to start spitting when they're on video. A hundred percent. And I think, I mean, podcasts as a medium, I think it actually is like goes opposite the trend of like everything has to be like short, like three second TikTok clips because our attention span is like disappearing. I think that when the format is right, long form is really good. And I think the reason that my, I mean, for myself, do I prefer to, if I want to get to know about a person and their points of view, who they are, whatever? Do I prefer to read like a well-researched article in a big newspaper with some journalists who sat down with that person and spoke with them for two or three hours and then cook together their own version of what that person is like?

40:09Or do I prefer to just hear that person speak for two hours with one of you guys? I think definitely the latter. And I think most people, young people today, feel the same way. I think in a podcast, who you actually are is what you're going to be represented as. It feels much more authentic. It feels much less edited. And I think that that is something that most, a lot of people value really highly, which is also why one of the reasons I think that, you know, readership and the general interest and kind of legacy media is going down because it's, and I mean, I have the same experience, right? I think it's, I think it's so much more powerful to listen to someone just speak in a almost unedited way, which is what you get to do in a podcast.

40:49And as to your point, right? Like if we were to do this with written answers, my answers would probably be much more. PR approved, Victor, is that what you're trying to say? PR approved and with less nuance. And so, yeah, I totally agree. That's also why I think, you know, AI podcasts, like if this conversation was just generated, of course, I can't say that that's like never going to happen. But I don't think that's going to happen anytime soon. I think people, even if you could do it to the point where you would not be able to tell that was AI generated, I still think that people would value the fact that it's a real conversation, right?

41:24So I still think there's going to be lots of space online for real authentic content. I think it's actually the more AI content we generate, the more kind of real authentic content in some use cases is going to be valued, right? But I think it's all about the use case. I think that's the key word here.

41:41Victor Riparbelli:So podcaster will be the last safe job from AGI. That's good to know. I mean, do you, when I think of Synthesia, I think of, I mean, you're a B2B company, you're doing like, I mean, you're replacing HR, you know, systems internally, those boring videos that everyone skipped through anyway, like you're making them AI, that kind of thing. Do you think your company and the tech you have is, has a role to play in what we're talking about more broadly in this cultural shift? Like, cause, cause you're still B2B, right? You don't have like a consumer touch point. Yeah, for sure. I think the best way of thinking about what people use this year for and why we're growing as fast as we are is that essentially we're like the next generation of PowerPoint.

42:24If you think of PowerPoints, I think there's around 30 million PowerPoints that's made every single day. And a lot of those PowerPoints are internal communications, trainings, HR onboardings, like a very wide variety of things that, you know, to most people, you wouldn't classify that as being super exciting content. But it's content that serves a very real purpose for that company and people in that company who's consuming it, right? And it's a very, very high volume of that content, which I think one of the things people have always underestimated when they assess Synthesia from the outside.

42:57Like, oh, they're just doing training videos and some HR stuff. That's not really true.

43:01Ellis:And I love that this is true to you, Victor. I heard that since you were a child, you loved HR information sessions. That's right. That's right. When I grew up, I would go to the library and just find as many old PowerPoint things. HR onboardings.

43:16Victor Riparbelli:It's not what I would expect. I mean, you say when you're like, you're building like this crazy deep fake or whatever you want to call it technology at scale, I wouldn't expect you to say I'm building the next, what replaces PowerPoint. But I guess I do see that logic. It's all about the use case, right? It's about communicating something to someone. And in the world we live in now, most people in their private lives consume video and audio content as a default, especially, of course, the young generation, right? The problem is then that they go to work and then at work is like, hey, please read this 10 page PDF document I just sent you.

43:45Please go through this 65 page PowerPoint to train you on something or to sell you a product. And that's just like not a great way of communication. I think what we have found is just if you lower the barriers to create video by enough, which is basically in our world by making it as easy as it is to create a PowerPoint. then what you will see in every big company is that the amount of videos that they generate will explode by 100 to 200x, right? It's like, it's not just that people take, oh, we used to create video. We used to create 100 videos every year and they used to cost us X amount. Now we're using Synthesia and now we're saving a lot of money.

44:24Like it's, we used to create 100 videos every year. Now we create a thousand videos every year or even more than that. So it is just like expansive thing, right? Where the classical example of this is always Uber, where yes, Uber replaced part of the taxi market, but it also massively expanded the size of the taxi market because if you lowered the prices enough, all of a sudden a lot more people would take taxis than they did before. And it's exactly the same effect that's going on here. And I think what we've seen over the last, especially the last 12 months, when the video models really have begun getting very, very good, is that now you get to more and more use cases are being enabled because the quality of these videos just go up and up and up and up, right?

45:04With Vio, Sora, our own kind of latest avatar models. It actually turns out that when you give the right tools to people who make these internal things, they actually become more creative. They create more fun content, right? Which is pretty awesome. So an HR video, maybe not yet, but in a couple of years, an HR onboarding video could look like a Spielberg production, right? There's not going to be any technical barriers to that. There's still got to be a creativity aspect of the people making that video, but they are going to be a lot more interesting than they are today. And also, of course, the higher the quality of the videos, the more external-facing use cases they unlock.

45:40I think people always think of AI videos like, when are you going to make Super Bowl ads and performance marketing and all that? I think that's obviously going to happen. But the quality of the videos we can generate needs to kind of get up there before those use cases really get so locked.

45:52Ellis:We got to get mnemonic jingles into the product. Here's how to use your YubiKey. It's so easy. One, two, three. I mean, because that's the real use case, right? You actually want the employees to retain something. Like I could see a world where making it more entertaining makes it more retentive. But I mean, people have been saying forever that even your TED Talk, people are only going to remember one idea or 25 % of what you say. Have you done any research on that in particular to say that, you know, the avatars are creating more information retention than otherwise? Yes, I think to address the first part of your question, I think there is a huge part of this, which is just what's the table stakes to communicate something, right?

46:38And I think that's kind of a bit the point I touched upon before. Or when people are used to consuming TikTok videos and podcasts and YouTube videos and all these like bits of content, which generally are like very engaging because the way the internet works is that you get like a gazillion new videos on YouTube every single day. And a very select few of them that are the most engaging will rise to the top. And those are the ones you're going to see, right? That's how all the social media platforms work. And then when you go to work and you're presented with these long documents, that kind of sucks, right?

47:06So what's happening is that the table stakes, if you want to be heard, is just right now, it's like that is an audio and video. But pretty soon, it might just also be that even if it's an HR onboarding video, that it's an engaging video. It's an exciting video to watch, right? Because that's what we're going to be used to consuming. In terms of the avatars, we did some research with UCL a few years back. And we found some interesting things for the type of use case we're testing for here, right? Which is more kind of like in the education, training, learning. It's not about engaging you and make you feel deeply emotionally connected to a product or a story.

47:39The avatars actually did a better job than the humans because the scripts are written. They're much more to the point. What can often happen if you're doing a Loom video or something is that you ramble a bit too much. You could probably have delivered whatever it is you wanted to deliver in 30 seconds, but you ended up delivering in two minutes. and where there's an avatar and it's like a written script, you kind of, there's been a bit more kind of editing on what they say. But I think, again, it goes to the use case, which is so important here, right? And that's what a lot of people, I think, misunderstand about Synthesia.

48:10If you're trying to like create a video that deeply connects with someone over something, you want them to be really, really, really inspired about like your new vision. It's not the right tool for that, right? But if you're trying to teach people, hey, this is like our sales methodology. It has these five steps. and you need to learn each of them, it works phenomenally well, right? Because that's not an emotional, that's not a storytelling use case. It's information or knowledge sharing. And I think that's the, that's the thing which we always talk about internally. It's like knowledge sharing versus storytelling.

48:38Those are two very, very different types of videos and communications.

48:47Ellis:This episode is brought to you by Indeed. Stop waiting around for the perfect candidate. Instead, use Indeed Sponsored Jobs to find the right people with the right skills fast. It's a simple way to make sure your listing is the first candidate C. According to Indeed data, sponsored jobs have four times more applicants than non-sponsored jobs. So go build your dream team today with Indeed. Get a$75 sponsored job credit at indeed.com slash podcast. Terms and conditions apply.

49:15Victor Riparbelli:No one goes to Hank's for his spreadsheets. They go for a darn good pizza. Lately though, the shop's been quiet. So Hank decides to bring back the$1 slice. He asks Copilot in Microsoft Excel to look at his sales and costs to help him see if he can afford it. Copilot shows Hank where the money's going and which little extras make the dollar slice work. Now Hank's has a line out the door. Hank makes the pizza. Copilot handles the spreadsheets. Learn more at m365copilot.com slash work.

49:50Ellis:one of the things that's jumping out at me too with generative video in terms of the use case obviously you could talk about its authenticity or whatnot but it always helps as you've done to hone in on the icp what's the use case and i think one of the things with this too is that internally uh information changes often right and having to like reshoot the entire video on some set in a specific part of the office, like is not going to happen every time a piece of information changes. I mean, it's funny. I was told the other day that people are still watching my Snapchat philosophy video I used to do in person during onboarding, like years and years later, even though the product is completely different.

50:30Ellis:I think, you know, for sure, a lot of those fundamental truths and origin story things are the same, but might be cool if we could update some of that stuff, right? As things change. and I think generative AI definitely, definitely affords that ability. This was one of the really big insights we had early on, which was that in the early days of Synthesia, we were very focused on like the kind of front end of creating a video as in like, you know, get a camera, get an actor, the microphones, all that stuff, which is very time intensive, time consuming, expensive and so on. And I think what we quickly learned was that while of course like replacing that part is valuable, actually what most people were struggling the most with was kind of like, as you said, after you've shot the content and you need to edit it.

51:13That is such a huge problem for people. And so people were like, it's okay if we have to record things once. If you can just allow us to edit in an easy way, then a lot of our problems are actually solved. Of course, doing both of them is a really powerful thing. But it is, and it makes sense, right? It's the difference between using a typewriter and writing something in a Google Doc, right? it's actually even worse than that because typing something again on a typewriter is a lot easier than recording it again in video but the editability and taking video from essentially being an analog type of media to a digital editable type of media is very, very powerful both because you want to go back and change a video something in a video that you shot a year ago created a year ago but also for things like you want to translate it into a different language and you don't have to usually when you translate a video into a different language thing is a huge one isn't it yeah and then you have like french french might be 10 longer than italian and so even if you just translate it as it is right the elements will be out of timing and it's just a huge hassle to do things like translation and so when it's this fully editable type of media you kind of leapfrog over those problems and makes essentially video editing like making a PowerPoint, right?

52:34A PowerPoint file, you can always, you know, open it up, change something, duplicate it, send it to a friend, translate it. Like it's very, very, very easy to work with.

52:43Ellis:Well, is it, can you do internal folks or is it just actors in your system? I'm imagining the head of sales at some company who is there long after he dies, still giving completely up-to-date presentations. You're going to have to watch our contracts, join SAG-AFTRA or something.

53:00Victor Riparbelli:Yeah, Ellis is going to be doing internal snap meetings in 20 years. Well, actually, this is a real problem with a lot of companies that Synthesia is a product to help solve. We have had many customers over the years who invested a lot in having their head of sales, a head of sales table or something like that, do a great 20-part video series about how to sell a thing, right? And then the person leaves the company and joins the competitor. What do you do now? Now you have 20 videos that are useless, basically. with Synthesia, right? Go in, change the avatar, and you're good to go. Maybe someone else said it.

53:37You're not beholden to, exactly. Oh my God. But creating your own avatar is a very popular feature. So you can do that now with resume launching kind of single image avatar. So you just upload one image of yourself and then you can prompt yourself to be in different environments, have different outfits on. And essentially, it's like a really, really scalable product.

53:58Victor Riparbelli:Victor, remind me when we were catching up in Vegas, did you decide to do enterprise because you correctly identified that you need to have a business and make money? Or was it also because there was something about putting this tech out into the world in like a consumer easy to access way that you didn't feel comfortable about? Am I remembering correctly? Like how much of this is about we need to just be a real business and have, you know, be selling AI and not just like giving it out for free and figuring it out later. and how much of it is the world is not ready for this kind of stuff at scale?

54:29I mean, there's an element of the second one, but I don't think that negates consumer use cases. I think that's just about, you know, we've always invested a lot in trust and safety and content moderation and those types of things. It's definitely more the first part of like building a real business. But if you look at most of the AI video companies that spawned back in like 2017 to, I don't know, 2022 or something like that, almost all of them made the same mistake. They would kind of sit down and think about AI video. Where could this be valuable? Well, it'd be great if you could do personalized sales videos.

54:59That seems like something that's very valuable. It'd be great if you could do performance marketing, generate a thousand versions of a video and test which one works the best. Both of those things are totally true. But I think with any type of technology, but I think specifically with something like AI video, it was just very important to us to look at ourselves and be like, what is this technology actually ready for? What is a real use case? And not just like, we would love this to be true use case. And I think most of those companies made the mistake of thinking that, you know, performance marketing videos with AI would be great back in 2021.

55:37It's, you look at those videos and they're crap, right? I think what we identified very early on was just that when you look inside these companies, in general, the content that lives inside companies and is not external facing is of lower quality, right? Which makes sense. It doesn't make sense.

55:53Victor Riparbelli:Yeah, I was going to say, like, you correctly identified that the only thing worse than 2021 video AI was, like, internal HR training videos. But it's not even that, right? Because it's more like a 15-page PDF document. How can we make that better? That's a very different thing than saying, can we take a performance marketing video and make it with AI, right? Like, what you compare the AI video to is very, very important. and all those people were basically picking up use cases where the comparison would be to a real video, which you're always going to lose. Whereas what we did, we picked a use case where the comparison would not be a real video.

56:31The comparison would be a 15-page PDF document. And that's a very, very important distinction, right? So I think what we did early on was just we figured out that the use cases that are mature today are internal-facing comms in companies because in general, in companies, which makes sense, Why spend time polishing a piece of internal communications the same way you would polish an ad you're going to put on TV? It just doesn't make any business sense to do that. And so that's the kind of reason that we went down that path. Because we could just see that we were very clearly delivering real utility.

57:05Not just like a cool demo which we could sell to dreamers who wanted to believe that this stuff would work. But when, you know, at some point kind of you had to test it out, it just wouldn't perform right, which is what happened in a lot of these cases.

57:19Victor Riparbelli:Do you see it broadening, though? Like, do you want to be a consumer company? I mean, so we've seen this, like, shift over the years where the better the technology gets, the more use cases we unlock. So right now it's about 50 % of our use cases which is internal and 50 % which is external. And that is already a big shift from just a couple years back. And I think we'll continue to see those. Are we ever going to be the company where you're going to go in and want to make like performance marketing ads or really cool storytelling, creative content? Definitely not. Like there are other companies who are chasing that.

57:51We will always be focused on the PowerPoint user more than anything else. And that is just, again, more in this category of like knowledge sharing, information sharing than in storytelling. So we have a consumer use case in the sense that we have a lot of individuals who sign up to Zynthesia and make content for it. But I don't see Synthesia going into the world of creating more storytelling, social media type of content. I think we're very squarely focused on the B2B side of things.

58:20Ellis:So funny question. You obviously can't control what your head of sales looks like. But when it's an avatar, you have the ability to have them look however you want. And it's interesting. I think we've certainly learned from TikTok reels and everything else that cuties convert better. How are you casting your AI avatars? How much are you giving in to their appearance as a way to get attention from internal audiences? That's a legit question, right? That's a choice. Well, I think what you see in the enterprise is actually kind of a bit the opposite because people care a lot about inclusion and diversity and having like representative voices in their videos.

59:01I mean, of course, it matters. And I'm sure people think about it. But you have to, again, remember that most of the content we create is not made to compete in news feeds with other people's content, right? Like, it's a video that a big company creates for like a very specific thing. And the person watching that video is not going to have like 15 ,000 other alternatives to what they could spend their time on. So I think it matters a bit less in our kind of use cases. But of course, like how the avatar looks is a part of the engagement.

59:28Victor Riparbelli:It helps to have a captive audience. You will watch this video or not get your next paycheck. That's true. That's true.

59:35Ellis:This appeals to me, though, as a former corpo. I mean, I was always struggling with writing teams or brand teams or marketing teams to have a source of truth that actually remains up to date. And I think you could argue that's one of the biggest problems that still remains in corporate life. And I think there are a lot of AIs trying to help solve this and say, hey, like, here's the source of truth. I'm going to update the source of truth brand doc or the voice doc or the key testimonials from our partners doc. It seems like when it's a video, having that be the source of truth, having it be updatable at any time, seems like it could actually create a higher level of understanding and cohesion within companies.

1:00:16Ellis:I don't know if you've heard that from anybody yet, but it does seem like when there's a video demo of the thing that people can share and bookmark and that's always up to date, it's going to be used more. For sure. And I think because these videos are programmed, right, they can be updated manually, but also automatically. So I think what we're seeing right now is, and we're launching this new kind of product around a Gentic video, which we should probably also talk a bit about. But I think with LLMs, video models, real-time video models, speech models, like there's so much interesting tech going on right now, which I think is going to be a big component of solving that specific problem, right?

1:00:55Because even if you have videos you can edit with AI, like as you said, you still need to have a knowledge base somewhere that's up to date before you can update the video to sort of match that. And I think right now there's just a lot of exciting stuff going on. You can build agents that can kind of figure out what are the discrepancies in the information, like what do we say on our website versus what do we say on our internal documents? What does person X slide deck say versus what's actually in the mission statement? Like all these things we can actually begin to kind of map out and work on having a real source of truth, which in a way I think has just been very, very difficult to kind of do before.

1:01:29And when you have that, that enables a lot of exciting things on top of it. You can take that knowledge and you can generate content on the fly essentially, right? You could personalize a product demo. So, you know, you go to a website, you want to see a software product, you put in who you are, your role, your title, your company, et cetera. and based off that you can just generate an experience that's completely personalized to you the features you want to see you can talk to the video ask it a question in a way that hasn't really been been possible before right so i think we're about to see an entirely new type of media emerge based on all these like cool new ai models and it's always been a big thing for us that we i think we had these two big ideas we started the company the first one was just with ai you're going to be able to create content at a very low cost with much less skill required and therefore also much less time required to create something.

1:02:23The marginal cost of creating video and other types of content is going to go to zero essentially. And that's a big business opportunity. And I think that's roughly where the Gen AI content industry has been so far. The second big idea was this idea that when we create new media technologies as humans, we always create new media formats native to those technologies. And so the big question here is like, what does truly AI native video or whatever we end up calling it actually look like, right? Right now we're using...

1:02:50Ellis:So we're talking about the interactivity, the two-way side of it. Exactly. Right now we're... If I'm hearing you correctly, we're talking about choose your own adventure HR info session. That's what I'm talking about. Exactly. It's got to be this sort of interesting halfway house between a video game and a video, most likely, right? Where you can personalize the experience, you can choose your own adventure. We're launching a product very soon around training and role-playing. So if you're a salesperson, rather than just consuming, let's say, a 10-minute piece of content on a new product that you're launching.

1:03:25Ellis:So it's almost a video game at that point. Yeah, it is. And we know that video games are great ways of learning things. So after you go through a 10-minute, maybe just a normal Synthesia video, you can insert these agents at any point in your video. That could, for example, be a role play, where by the end of the video, as a seller, you have to spend 15 minutes talking to an AI agent pretending to be a customer. And you have to overcome objections, show you've actually understood the content. For you as a learner, that means that, you know, you learn it. And for a manager and C-level folks, what we can actually begin to do with this is to give you a map of your talent.

1:03:59How enabled are they? You can go and say, well, people are really good at talking about like product A, but the product we launched three months ago, nobody's any clue how to actually sell this. That's a big problem. Let's use Synthesia to then train you further on this. It could be videos where you just have a simple Q &A, right? So you watch a piece of content, you have a question, just like if you had a real tutor sitting by your side, you can say, hey, can you explain this again in different terms? What does this mean for my role or whatever else thing you want to do? So there's got to be a lot of interesting things with what we call video agents, where we take video from being this kind of static one-way communication to essentially being a video you can talk to.

1:04:35It could be that you're role-playing with it. It could be that it's interviewing you, asking a customer how they like the product, and doing patient intake for a doc. There's a million different use cases you can imagine where these types of videos becomes less like a video and almost more like kind of a small app that does something for you. It could be a form. It could be a role play. It could be a small game. And I think that's the$100 million opportunity in this space. It's figuring out, like, what does this new media format look like? Where is it really valuable? scaling on that and then becoming the kind of platform for people to create these experiences.

1:05:08Victor Riparbelli:Well, it sounds like maybe podcasting is going to get replaced then. I mean, do you see that extending to more creative interactions or do you think that's always going to be something that is the more mundane, like taking a PDF and making it a video? Actually, I know quite the opposite. I think that if you think about like, how do you best learn something, right? The best way of learning something is rarely just to like go to a room with someone and just start asking them questions, which would be like just going chat to chat GPT, right? Most places, and if we take the enterprise version of this, is that you have some curriculum.

1:05:42There's someone in the company who sat down and said, this is our sales methodology. And they've defined what it is and how we talk about it. And they've probably put some visuals to it in our PowerPoint slides, for example. And then you consume that. And then you also have a manager or tutor or something that you ask questions. You learn by talking to your colleagues next to you, by kind of seeing what other people is doing. And I think it's exactly the same thing. So the way we're building this product is with a very big emphasis on combining curriculum, static content, video you've created in Synthesia, with these agents.

1:06:14And so what that means is that you, as someone who's training someone to do something, you kind of get the both of both worlds, right? You can have both the static content and you can have the interactive versions of it as well. And I think with podcasts, I don't think – Netflix did this like Bandersnatch series a while back, which was an interactive series, right? It's flopped pretty hard. And I think it's in some sense kind of obvious why it flopped. Because when people come home from work, they want to sit down and they just want to watch something. You don't want to make a decision every 10 minutes on what direction to go, right?

1:06:46And we just do different things as humans. When I'm walking to work and listen to a podcast, I want to listen to a podcast that you guys, you've sat down with interesting guests, you've edited out the most interesting parts of me, and I want to listen to that. That's very different from me coming home and wanting to Google things and ask specific questions and find specific information. We just have different modes. And maybe in the future, there's going to be a way where you have an AI version that I can ask you a question of what you think about what this guest said or something. But I still think there's a big difference between I'm watching a film and playing a computer game.

1:07:20It's two very different things, right? They haven't replaced each other. We still most people both watch films and they also play computer games. Computer games is like you're making decisions every single moment. And a film is something you're much more like relaxed, sitting back and just consuming something.

1:07:34Ellis:I think that point stands. And for those who don't know, Bandersnatch was kind of like a Black Mirror mini episode where you could kind of choose your own adventure for the main character. And he eventually becomes self-aware. and it rules. But to my knowledge, the reason it flopped was because you could only do it on a laptop because you had to click. I don't think you could do it on Apple TV. That's why I waited so long to watch it. Maybe you could on Roku or otherwise, or does someone know better than I do? Yeah, I definitely watched that on TV. You did, and you were able to click and whatnot.

1:08:06Ellis:Well, I think this is a good segue to our next point. I know we don't have a ton of time left, But I know you're focusing very much on your use case and your ICP right now. And I think there's a pretty big difference between creating a video that is not true or that misrepresents reality. If you share that with the world, your boss fires you. Whereas if it's just making up a video of a politician or somebody like that, maybe you actually start a movement instead of getting fired. I feel like you're on just the absolute cutting edge of this stuff. In five years, when we see political videos online or just videos of anybody, entertainers, celebrities, whatever, are we going to by default assume that it's not real?

1:08:45Ellis:And what do you think is going to happen? I think there's a pretty high probability that we, I think we probably should expect that something is not real unless we're told it's real. So today you have this system, right, where kind of change on X now, but we use that like of like verified accounts, which is essentially like if you're a celebrity, you get like a blue checkmark. Can't believe they dumped that. It was so valuable. It actually was pretty valuable. But I think a better version of the world would be that, you know, when you watch a piece of content, rather than having it verified, if it's actually true, you should have, it should stand out in the interface when something is not verified.

1:09:27So I think what we'll see is that politicians, celebrities, all these folks will essentially have ways of confirming what content is real and what is not. And this could be anything from you kind of fingerprint the content. You have to kind of, I usually explain this as like Shazam, but for all the content in the world, you know, the thing that can recognize songs for you. If you imagine that every time someone creates a piece of content and it gets uploaded to a global database somewhere and it's searchable, we could tell you, hey, what you're watching right now is an edited version of the original, which you can view by clicking here, for example.

1:10:06Or like this pretends to be from the BBC, but the BBC has no record of ever actually uploading this. And so we should tell you that this is probably like an unsafe or an unreal piece of content. There's a lot of challenges for sure in all this stuff, but I think what we will move to eventually is almost like something like SSL for video and for content online, where there will be the majority of the content on the internet will probably be, you know, not verified, but then there will be some content that's verified. And I think there'll be a default way for like well-known people to push out content.

1:10:38Now, I think I'm actually less worried about like politicians and celebrities. I think what's happening today is a bit like the Barbra Streisand effect of as soon as someone creates something outrageous, people in the comments very quickly go to like, this is AI. And I know that's not a perfect system, but maybe a controversial point of view where I think it works like pretty well. We have things like community nodes, which I think is an amazing system that I hope is going to scale. Where essentially you have, the way community nodes works, right, is that you have, you basically have a lot of people and you figure out what the people who generally disagree the most can agree on with this piece of content, right?

1:11:14So what can like hardcore Republicans and hardcore Democrats agree on? Well, they can agree that this clip is not real for some reason. That's a pretty cool way of like getting to the truth and it's a bit the same way that Wikipedia works. And I think we'll see this scaled up pretty significantly.

1:11:30Ellis:Never thought I'd say this, but sounds like a good use for blockchain. Or is that not true? It is, I think, a good use for blockchain because it would be immutable, right? But I mean, there's a lot of scale problems still, but I think the theory of having this on a blockchain where nobody can edit it after the fact, I think is very powerful. And maybe to finish my thought before, I think celebrities and politicians, I think we'll have less problems with that. I think where the big problem of unfortunate will be is where we're seeing it today, which is non-consensual deepfakes of pornography, for example, of everyday people, unfortunately, especially women, right?

1:12:09We'll see this with scammers creating videos, voice clones, et cetera, of family members. This is much, much harder to detect in many ways, right? Because you're not going to have community notes on a video that someone sends to you personally. You're not going to have... You need a family code word instead. Things like that might actually be something we have to begin thinking about, right? So I'm more worried about the long tail of like individuals for things like scamming than I am of, I think, what has historically always been the biggest worry, which is like, what if someone makes a deep fake of a politician saying something they didn't say, and it kicks off like some crazy chain of events?

1:12:44I just, of course, that can happen and it will probably impact the political landscape. But I just think we've seen it enough times now that, you know, it's, yeah, I think I'm less worried about that and more worried about like on an individual level.

1:13:00Victor Riparbelli:Kind of a bummer note to end it on. Is there anything positive we can end it on? Do you have any like positive views of what this is going to do? I mean, I do think maybe it leads us to all kind of band together in a way, right? To try to figure out what's real, like you were saying about community notes. But it is kind of refreshing when you see a TikTok and it's like, oh, is that fake? And then you open the comments and it's like fake, fake, fake, fake, fake, right? Like that kind of gives me hope for humanity. Look, I think there's a lot of tough challenges ahead. But I actually do think that the combination of community notes, figuring out this way of like proving media provenance by fingerprinting content or doing a Shazam style database.

1:13:41I think there is a lot of things we can do, which I think will make the media information ecosystem a lot better and actually also a lot better than it is today. That said, like there's going to be a lot of AI content and there's going to be a lot of slop like we're seeing right now. There's going to also be a lot of amazing art. And I feel like that part of it doesn't get nearly as much attention as all the slop. There's so much cool stuff. I have so many friends who are musicians who are making awesome music videos. You have filmmakers experiment with this stuff, creating really cool things.

1:14:11So I think there is as many pros as there are cons. I also think it will push us to be, I think it will push the digital world to be even more artificial than it is today. And I think in some ways, maybe that's actually a good thing, right? As you said earlier, we should probably be more skeptical of the stuff that we see online by default. and we should probably get more off our phones and out into the real world. And I think there will be a movement towards both economically in terms of like, what do we value as society? What do we, who makes the most money? Where does the value accrue to being potentially in, I don't know the timeline, but let's say 10 years, five or 10 years, 15 years, to be less about people who have like highly analytical numerical jobs of like looking at Excel spreadsheets, coding, all the stuff that's high paid today, to be more about the things that drive real human connection, right?

1:15:02Because as much as we can get Optimus robots and we can get, I mean, all this crazy AI stuff, I think what we will not be able to replace with AI is the feeling of having drinks with your good friends. It's going to be an amazing musical experience somewhere. It's going to be seeing a piece of art. It's going to be walking out in nature and enjoying the sun. And so I think there is a version of the world where right now the digital world, software, tech, all this stuff is like the pinnacle of society in the sense that that's where all the wealth is kind of accumulated. And maybe in the future that will actually change as all those things just become like very, very, very commoditized.

1:15:41Victor Riparbelli:Hear, hear. All right, Victor. It was great catching up with you. We can let you get back to your nice retreat in Utah. Thank you. I hope it's good. Thanks for joining us.

1:15:50Ellis:Enjoy the summit. Great to meet you.

1:15:52Victor Riparbelli:Yeah.

1:16:19Victor Riparbelli:$45 for three-month plan, equivalent to$15 per month required. Intro rate first three months only. Then full price plan options available. Taxes and fees extra. See full terms at mintmobile.com. Stitch Fix. Shopping is hard.

1:16:29Ellis:Let's talk about it. I don't have time to shop, so I buy all my clothes where I buy my seafood. I just want someone to tell me what shirt goes with what pants. I just want jeans to fit. Stitch Fix makes shopping easy. Just show your size, style, and budget. And your stylist sends personalized looks right to your door. No subscription required, plus free shipping and returns. Man, that was easy. That looked good. Stitch Fix, online personal styling for everyone. Take your style quiz today at stitchfix.com.