Next Unicorns: Building a $1B AI Avatar Business with Synthesia CEO Victor Riparbelli | E1776

12 Jul 2023 · 1 h 6 min

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

Podcast Summary: This Week in Startups - Episode E1776

Episode Title

Next Unicorns

Building a $1B AI Avatar Business with Synthesia CEO Victor Riparbelli

Episode Description In this episode, Jason Calacanis interviews Victor Riparbelli, CEO of Synthesia, a company specializing in AI-generated video avatars. They discuss the journey of Synthesia, challenges in the AI industry, job disruption due to AI, and measures to prevent misuse of deepfake technology.

Key Themes and Discussions

Introduction to Synthesia

  • Company Overview: Victor Riparbelli discusses the foundation of Synthesia in 2017 and the initial focus on creating virtual avatars for customer service applications.
  • Funding and Growth: The company recently achieved a unicorn valuation of $1 billion with backing from investors like Nvidia.

Product Demo

  • AI-driven Video Creation: Victor demonstrates Synthesia's platform, which allows users to create videos with AI-generated avatars that can speak in multiple languages.
  • User Accessibility: The platform is designed to be user-friendly, allowing individuals with no video editing skills to create engaging content.

Business Applications

  • Enterprise Use Cases: Highlighted examples include major fast-food chains utilizing the platform to train employees efficiently without the traditional costs associated with video production (e.g., hiring actors, studios).
  • Cost Efficiency: The platform significantly reduces costs associated with traditional video creation, enabling companies to reach a broader audience.

Job Disruption and AI

  • Impact on Employment: The discussion touches on the dual-edged nature of AI technology, with concerns about job loss in traditional roles (e.g., camera operators, translators) versus the potential for new job creation in tech.
  • Creativity and Productivity: Victor argues that while certain jobs may disappear, AI will enhance human productivity and creativity across various fields.

Ethical Considerations in AI

  • Preventing Deepfake Misuse: Victor emphasizes the company's commitment to ethical practices in AI, including strict content moderation and obtaining consent for avatar creation.
  • Responsibility of Technology Providers: The conversation delves into the ethical implications of AI technology and the responsibility of companies like Synthesia to prevent misuse.

Future of AI and Content Creation

  • Evolving Media Landscape: Victor discusses the potential transformation in how media is created and consumed, predicting that video content will become as common as written content in professional settings.
  • Integration with Other Technologies: The potential for integrating Synthesia's technology with other AI tools is explored, suggesting a future where personalized, interactive content becomes the norm.

Conclusion

  • Optimism for the Future: Both Jason and Victor express excitement about the future of technology, emphasizing that AI can democratize content creation and open doors for new forms of expression and communication.

Key Takeaways

  • User-Friendly AI Video Creation: Synthesia enables anyone to create professional-quality videos without needing extensive technical skills.
  • Cost-Effective Training Solutions: Companies can significantly reduce training costs while improving employee retention and understanding through video.
  • Ethical Responsibility: Technology creators must prioritize ethical considerations to mitigate the risks of misuse and ensure responsible use of AI.
  • Future Potential: The landscape of content creation is set to evolve, with AI enhancing both productivity and creativity across industries.

Time Stamps

  • (0:00) Introduction of Victor Riparbelli
  • (3:47) Unicorn status and funding details
  • (12:25) Crowdbotics advertisement
  • (13:54) Pricing strategies for Synthesia
  • (17:19) Live product demo
  • (31:43) Discussion on job disruption
  • (39:38) LinkedIn Marketing advertisement
  • (41:06) AI technology discussions
  • (45:18) Deepfake misuse prevention
  • (52:40) OpenAI lawsuit discussion

Additional Resources

  • [Synthesia Website](https://www.synthesia.io/)
  • [Follow Victor Riparbelli on Twitter](https://twitter.com/vriparbelli)
  • [Check Out Crowdbotics](http://crowdbotics.com/twist)
  • [Explore Coda](https://coda.io/twist)
  • [LinkedIn Marketing](http://linkedin.com/thisweekinstartups)

This episode of "This Week in Startups" provides a comprehensive look at the intersection of AI technology, ethical considerations, and the future of content creation.

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:00If you look at the shape of the music industry, right? back in 85 versus today, how it's been flattened out, talent can come from anywhere. It's about making great music less than it's about knowing the right people. Like all those things, I think it's going to come to the tech industry as well, right? We've seen some of that over the last five years, but I think as the kind of art of coding becomes more accessible, it's like, you know what? Why couldn't you be this eight person company sitting somewhere in, you know, Norway or Denmark or something like that and create something absolutely amazing.

0:27This Week in Startups is brought to you by Crowdbotics. Great ideas can change the world. And Crowdbotics is the fastest way to turn those ideas into code. Get a free scoping session for your next big app idea at crowdbotics.com slash twist. Coda is the all-in-one doc for teams. And they just introduced an AI-powered assistant to take the busy out of the work. Sign up to use Coda AI today at coda.io slash twist. and LinkedIn marketing. To redeem a free$100 LinkedIn ad credit and launch your first campaign, go to linkedin.com slash next unicorn. Hey, everybody, welcome back to this week in startups.

1:11Big, big surprise for you. Today is the start of the fifth season of a series we call the next unicorns. And what is the next unicorns? This is where we find companies that have raised their series B round, let's say they're got some really great traction, maybe they have market pull. And we just think as an editorial team, the producers here, some of the investment team might make suggestions. This company's got a shot at becoming a unicorn. What's a unicorn in Silicon Valley, it's a company that's worth over a billion dollars. Now you probably heard that we went a little crazy during the boom period of 2019, 2019 2020 and 2021 and just gave a lot of companies very high valuations some of them are now zombie corns as we've talked about on the all in podcast a couple times maybe the companies didn't actually have product market fit in other words the product and the customers were in sync and there was a market for this product or a market pull when customers are searching to get themselves a tesla and uber or an airbnb because they heard about it from their friends and the product is so good that you have market pull.

2:17In other words, you don't even have to do marketing, people are just calling you up and you're taking orders as a company. Putting all that aside, I'm really excited about today's guest because he's been working on making virtual AI based versions of humans. You could cynically call it deep fakes. That is the negative framing of this. Or you could say these are alternative versions of yourself. And there's a ton of upside and we're going to get into the downsides the upsides in uh this dual use technology could be used a dual use technology could be used for good or bad just like humans can choose the path of darkness the path of light you could be darth vader you could be luke skywalker you could be a super villain lex luther you could be superman uh and uh the technology of ai is no different it is a dual use dual fate technology if you will and so we uh have a great guest victor ripper belly is on the program.

3:09He's the co-founder and CEO of Synthesia, spelled S-N-Y-T-H-E-S-I-A. Welcome to the program, Victor. Thank you, Jason. Excited to be here. Synthesia is how it's pronounced, correct? Synthesia. Synthesia, okay. And paradoxically, we invite you on the program to come on the next Unicorns. And then you, of course, make us one for one this season, because we like to track how we do. This is our fifth season. you can see all of the episodes at this week in startups.com slash next unicorns. This week in startups.com slash next unicorns to see the the other four seasons, but you're the first guest on the fifth season.

3:46And since we invited you, you actually closed around at a billion dollar valuation,$90 million seriously at a$1 billion valuation. Congratulations on that who led the round. Thank you. Excel led the round with participation from Nvidia. Oh, okay. So Nvidia is investing in startups that are in the AI space because they're making the chips, they're making the infrastructure. Do you need to buy these chips yourself? Or are you able to just use a cloud provider who has those chips? I'm curious. We definitely use cloud providers. I don't think for us as a business, it makes sense to go that deep in the stack.

4:23And NVIDIA is a great partner. I think most people think of NVIDIA as the ones that make the GPUs and the chips, which of course, is their main line of business for sure. I think what's less well-known is actually that NVIDIA does a lot of fundamental research, especially around media technologies, so generative video, speech, text, for that matter. And they're one of the world's best research teams. And a big part of motivation for us to get closer to NVIDIA is actually less on just the GPU and the chips, but these guys are doing really amazing fundamental work in a lot of the technologies that we're also developing, right?

4:53So I think that's one of the big angles for us to get closer to NVIDIA. Yeah. And so you started this back in March of 2017. And the business was, I believe, in the early days, your goal was to make a virtual avatar, not necessarily of a real person, but in order to do maybe say customer support. So you go to a website, you go to a kiosk. Is that correct? Is that where you started? Was that the first year's case? Well, sort of, I think, I think just we started the company back in 2017. Right. And I think we kind of did what you're supposed to not do, we started with a technology and then we tried to find what we could solve with this technology.

5:32This is very, very early in what's now called Genitive AI, right? But the key idea was the same as it is today. We want to make it easy for people to make video content, not by making smaller cameras, not by making slightly better video editing apps, but by actually building technology to eventually replace the entire physical production process, right? So that's the vision. That's where we want to go. It goes much further than just the avatars we have today. The first three years of the company i'd say it's like the wilderness period we're kind of building the tech figuring out if we worked or not we kind of got into the service business of doing ai dubbing so some of those who have known us for a while may have seen the david beckham thing the snoop club thing we did where we made essentially celebrities speak in different languages and this was a great learning path for us but what we figured out was that that was a service business it was a visual effects type of tool but was actually much more interesting than selling to hollywood and creative people was like there's billions of people in the world who are desperate to make video content, but they can't today, right?

6:26They don't have the skills. They don't have the budget for it. And all these people, they're actually okay with the quality threshold going down a little bit from a real camera. If it's a thousand times easier and a thousand times more affordable, right? And that sort of, that insight is what led us to then launch our self-service text-to-video platform that we operate today with the world's biggest. We launched that in the summer of 2020. And since then, it's just been an absolute fairy tale of going really quickly and just delivering tons of value to our customers. And again, it's less about this sort of video as you know it.

6:56It's actually more about replacing text for our customers, right? That's what they're primarily used for. So in that case, it's not necessarily to make a young Indiana Jones or Luke Skywalker and have them do a new adventure. You're not trying to compete with Disney. The goal here is, hey, I was going to write some materials here, some training materials, or i was a professor doing a lecture or i'm uh writing some um message to all of the i don't know uh franchisees in my jiffy lube franchise network and i can type it in text but have a human a synthetic human a replicant speak to the people on the other end so if i was the manager and the franchisee of this jiffy lube and you said hey here's our new pricing instead of just sending them a text document, they could watch it?

7:48Is that the customer base, those kind of applications that I described? 100%. That's very enterprise focused. We work with 36 % of the Fortune 100 today. It gives you two very real examples, right? So let's say you're one of the world's largest fast food companies, you're onboarding and training millions of people to work in your restaurants every single year. What they used to do, they'd have a 40 page handbook, you'd send that to the employee, they'd sort of attempt to read it, remember absolutely nothing and be much less prepared to go into that job where they actually hit a restaurant. Now what they can do is they can make that into a video, which is in the native language of whoever's watching that video.

8:20So it works in 120 different languages. So that in itself is a massive unlock. But then the really, really big unlock that's kind of the false multiplier is that our product is extremely easy to use, right? So if you know how to make a PowerPoint slide, you'll be able to use Synthesia. So now the same people who used to write that 40-page handbook, they can make the videos, right? Got it. So you don't need a video editor. So if we just pause there for a second, because what you said is profound i am this you know uh international fast food company and i got people in france i got people in japan and uh if i was going to make this 40 page training manual i would then send it to a company that would translate it make a local copy of it and then people would pretend to read it and then the french fries uh or the milkshakes would come out terrible because people didn't pick up those points instead you can take that manual the same person can write a script and the ceo of that company or the founder even could come on and speak to the person and then there's all kinds of technology to make sure there's compliance that people are watching if they open another window i know this like if you ever took a driver's test or any kind of test i recently went through this for like um joining um scouts previously known as boy scouts of america had to do background um tests and it was like a three-hour course and if you at one point i accidentally put another browser window on top deposit so here you have an enterprise thing where any new person can hear directly from the ceo why it's important to get the fries right and how to do that and you can make sure there's compliance and that's one person's job it reaches millions of people i mean that's just insane when you think of the leverage there that would normally have cost them.

10:06God, it would have to cost them 10, 10s of 1000s of dollars per region across 40 languages, you'd be talking about a couple million bucks. And now it's going to cost$10 ,000. And let me give you, I think one one of the really core insights we had early on with this was that what you just described here is kind of the front end of producing content, right? It's like cameras, actors, studios, executives don't know how to perform in front of a camera, all that stuff in itself is incredibly time consuming and expensive, right? But then what we figured out is actually all to that. Okay, so let's say you actually did that, right?

10:36You want to make this into a video because you know it's much more effective. Then what happens in every single business is two months later, there's a business policy that changes, the pricing changes, you change your POS system, something like that. Now you're screwed, right? Because all your video material is completely out of date. What's beautiful about this type of video production, right? Is that you can just open up the file again, edit it, change the script, and you can feed that into all the other languages. So I really think that it's very early, right? But this is as profound as going from using a typewriter, where you're producing something that's sort of hard-coded, hard-baked, like you would with a camera or a microphone, into something that's more like a Word document, right?

11:10Open it up, edit it, duplicate it, send it to someone else. Our platform is collaborative, so we have real-time collaborations coming out next month or so, Teams, workspaces. Kind of like using PowerPoint or using Myra or something like that, but the output is just video, right? And I think that's what we're seeing is just that the value in the enterprise is just like through the roof. I'll give you another cool example. Let's say you're one of the world's biggest software companies. You have 4 ,000 AEs all around the world. You need to keep these folks updated on... Account executives, salespeople, yeah.

11:38Salespeople, exactly, right? You have to keep these folks updated on what's happening with the product, what's happening with the competitors' product, pricing, so on. Let's say you have one minute to talk to these people every single day about these things. You send them a document or something that they have to read on Slack. Most of them will skim it. They'll forget it, right? What you can do instead is you can make a one-minute video. we know that information retention is like six, seven, eight times as high when you watch a video as compared to when you read something. So if you take just a simple use case like this, right, the sales enablement team creates video content instead of only written materials for these folks.

12:10And you do this for 4 ,000 AEs. You really just imagine like just how much value that has, right? Because you get a much better informed, much better trained sales force. And I think we're still early in these kind of use cases, but this is really just what's... I think that's the reason we've grown so fast over the last three years. All right, we all know the one thing that separates great startups from the good ones is product velocity. What does it mean? Product velocity? Fancy term, right? You got your product and your velocity speed, the speed in which your product improves. So can you ship updates?

12:41Can you release new features? Can you do bug fixes? Can you iterate on the interface? Can you solve problems for your customers? And can you do it quickly? Because you're not alone, you have competitors and your customers have choices. They may solve their problems by writing their own custom code, or they might use your solution. This is what startups are about. How fast can you get that product velocity going? And so, you know, how do you supercharge it? Everybody says, yeah, we want to go faster, but you got to go faster intelligently. And Crowdbotics is going to help you do that. They're your CTO as a service.

13:14Basically, they provide you with the most optimal architecture to get your product to market as fast as possible. You'll have access to an on-demand product manager and developer talent, and they will help get your app into production 10 times faster than conventional development. Crowdbotics can work with your in-house dev team, or you can just have them work independently. And you own all the IP. You own all the source code. Let the folks at Crowdbotics supercharge your product velocity today. No more waiting. Get a free build plan at crowdbotics.com slash twist. That's a$4.99 value just for the twist listeners you get that for free that's c-r-o-w-d-b-o-t-i-c-s.com slash twist for a free build plan so just to recap there you've eliminated the need to hire a media training person that costs 20 10 20 000 a pop to train the host you've eliminated the host that costs 10 20 000 to hire some person you've hired the lighting the camera operators and the set that's another 50 to 100 grand in my estimation because you gotta have a location to shoot in as well it's another 10 grand a day five grand a day you got rid of all the translators and you got rid of reshoots while increasing compliance and retention and you've only given up perhaps if this was done in a really really by a very high quality team that spent the quarter million dollars to do it that i just described is what i would put this kind of shoot at uh if it was i'm talking about like a full employee handbook you know a couple of hours of content maybe 40 segments or something one per page all you've given up is a little bit of polish on the edges of a great host that might be slightly more compelling um i wonder in this do you have the ability to do you make the full tool of like here's the quizzes and compliance and making sure that people watch the video or do you use like third party educational tools and they just export their videos into those yeah so at this point in time we're not uh we're not like doing the kind of last mile delivery uh which also means you don't do the quiz so you put that in lms system something like that a learning management system lms learning management system exactly so yeah that's incredible can you imagine like microsoft or salesforce having all these sales people around the world and every day they just get a video and then it you know the learning management system just knows if everybody watched it, if people didn't watch it, they can't do work the next day and their manager gets an email.

15:40Hey, this person still hasn't watched the video. This person's three videos behind. That's incredible. And what do you what would you charge in this situation if without mentioning a specific client, but if you did have that at scale use case, tens of thousands of AEs, how do you charge? How much do you charge? What does this cost to do this 40 page manual? If you were running, you know, a burger or a fast food chain or something? So really, our thesis revolves around the idea that in the future, we're going to be reading much less and we're going to watch and we're going to listen way more. And we're basically building our product and pricing structure around that.

16:16This is, as I said before, this is really about replacing text in the enterprise. So the way we can price the product is per seats. We really view this as having so many analogies with something like PowerPoint in the early days. So it's seat-based. That, of course, depends on how many seats that you decide to buy. account enterprise plans on the person creating it or seats on the people receiving it so right now it's one type of seat um that you buy so basically it's kind of like a model you know from like my row or figma things like that right and go in you have a workspace together you can invite your colleagues you can leave comments on the videos and all that sort of stuff so if you had 20 people in the training organization hey you know you you pay what three hundred dollars a year four hundred dollars a year per person something like that 25 bucks a a bit more than that i think i think i would probably land at around 40 50k dollars a year oh okay great so it's a couple thousand per person so this is high-end software in the in like the salesforce category a couple thousand per person per year uh so for 40 50 whatever for 50k a year you can not have to do all these shoots and everything all right let's let's do a demo here let's do a demo here let's do it let's do it and just to not scare people off if you had the personal plan on the website says 30 bucks a month 270 years so i'm assuming yeah yeah yeah so sorry to lift that out right so it's a traditional kind of you know product-led motion so you can go in you can even make a video for free on the website or you can sign up for a 30 dollar plan which you'll kind of get to get familiar with um with the platform and then enterprise offers a bunch of other features so here we are we're looking at like a powerpoint-ish notion-ish interface with a bunch of projects and templates that you can pick from an office interior, an inspirational talk, whatever it is.

17:56And I see you have a couple of This Week in Startups videos you've created. Oh, no. Oh, no. Exactly. I'm coming well prepared, Jason. Uh-oh. So what you're seeing here is the homepage, as Jason just described it, is a traditional creator tool, as people probably have seen it before. I'm going to go to the avatars out here in the left-hand side. And what we're seeing now is a mosaic of all the different avatars that you can use to create your videos. So if I scroll down here, there's around 150 the avatars they come in all different um ages genders ethnicities are these computer generated or are they um made by uh like having partly computer generated they're based on real actors but but they are sort of ai generated um and there's a lot of people there you got a hipster you got a four person not a four man we have to use gender neutral language uh i can see you have like a construction worker uh you know a police officer type thing doctors or a fitness person and you've got all the colors of the rainbow genders body sizes so you can be super woke or super diverse here you can pick whatever you like or you can actually do it i'm assuming by region so uh you know not to uh make light of it but it might actually the information might be received better by somebody who's wearing clothes or has the accent of a japanese person in japan rather or an indian person india correct it's funny you say that because we have a lot of customers in japan we work with a very high percentage of like the hundred biggest companies there and actually what a lot of those they use it for is to create partner materials for english speaking or other countries speak different languages it's a huge line of business for them right because it really just breaks down the communication barriers and that goes both ways we also work with a north american companies who want to produce materials for the japanese market um and this comes in really handy not just on like the avatar appearance of course but also with you know the 120 languages that's so i mean this actually makes a lot of sense english is the world's language now or the language of international business um and i'm not saying american i'm saying english right uh queen's english here and so if you're a japanese company and you want to reach the maximum number of partners if it's in english and the person is localized yeah that's going to be a lot better than perhaps a japanese person with english as a second language who's a rep locally in that market uh whose english might be whatever an 8.5 out of 10 here you're going to have 10 out of 10 exactly and that's fantastic my japanese by the way is a zero out of 10 mine is true i can assure you yeah but what we're also seeing with this right is like you know let's say you have a lot of technical documentation integration videos which can be a lot of screen recordings and voiceovers and things like that right it's again just like and it's so moldable because you just make it in japanese and then you could go back and edit it every time you don't have to go to the whole voiceover actor and kind of process right but if we're out to make a video here let's say i pick alex got a click on her she's one of our most popular avatars.

20:56You've probably seen her before. Then we get set into the video creation screen here, which is where the magic happens. So what I'm seeing here is a thumbnail of Alex, who's our avatar. And then under her, there's a script box. And the script box is really where you just type in the script of your video, right? It's kind of like that easy. So let me just try and type in something here. I'll type in, hello, Jason, this is a test from Alex. And what you'll see here is that we automatically detect that you're typing in English. So there's this little drop down here, which says English US professional, I can click on this drop down.

21:29And then I'll get a bunch of different types of voices that could just be the style, but it's also different accents. So let's say you want a British voice, or Irish, Australian, whatever, you can select whatever you prefer. And so that's sort of like how that works. But then the magic here is that you can type in any language, right? So I'm Danish originally, for anyone wondering where my accent is from, let me type in I am now Alex, as you'll see here, it just automatically detects that I'm typing in Danish, now the flag has changed to Danish, and so on and so forth, right? So you just put in whatever language that you want the videos to come out in.

22:01And that's really as simple as it is. Now in this text box, we can do a bunch of other things as well. If I want to have a different gesture, for example, let's say at some point in video, I want the avatar to hit not yes, or raise the eyebrows, I can put that in with this little kind of pill. How many gestures are there? there's five right now it's a little better feature but uh but we're adding a lot uh in the coming ones are the gestures made by computer or did you have the actors actually make the gesture and then they're made by computers and so that's a big part of the fundamental tech platform and that we're switching into is essentially that the avatars become much more live right they look much more realistic they can do much more things disappointed in there like just a completely i'm just completely disappointed in you you didn't get the answer correct you can have a lot of fun with the gestures like um putting like your hand on your what is that like face palm you just do total face palm oh my god and and this is just not to be dystopian but um in a black mirror episode this is how you'll be laid off tech layoffs personalized because i assume you could personalize these too like i could put in you know percentage first name percentage and have like a custom one made for each employee I'm laying off or giving a raise to.

23:11Exactly. So I just put in an arrival here. For those who can't see, I just put in a arrival. So I can put in like first name. You'll be familiar with this from like mail merge and similar styles of things, right? I'm creating a template where you can also switch out the background images, videos, and all that sort of stuff. But what happens from here on is like I've typed in some text. That's great. That's a script of my video. But then, you know, one of the big unlocks and the thing that we discovered early on was for this to really be something that can be used by anyone, you have to be able to make a video end-to-end in here right so we have this sort of tool that sits around it and it'll be you know something like a powerpoint or a camera i'm just going to open up a template as you put in you know your text you can upload your own custom funds right yeah so you have your own branding host on the left and then just right behind them like they're a weather caster or you know a weather person yeah exactly exactly you could have the video behind them you know saying hey welcome to your first day at stripe uh today we're going to go over are HR policies.

24:10Here's what you need to know. And the HR policies could be done here. And like you're saying, Hey, you know, we have a new HR policy, we're going to add June 19, that's a holiday, and we're going to take away one of the floating holidays. And instead of having two floating holidays, you now have one, boom, you just put it in there. Every new person understands that. And you could send that just that one slide to everybody in the company announcing like we did, hey, June 19 is now our, you know, whatever ninth official holiday of the year, you have one floating holiday to choose from. Exactly. For many knowledge workers, over 50 % of their day is filled with doing tedious, repeatable tasks.

24:46We all know this. They're technically working, right? But are they being productive? Are they driving the business forward in any way? Or is it busy work? Imagine if you recaptured 50 % of that person's time, think about all the things you could redeploy and all the projects you could finish. Well, now you can with Coda. Coda is the all-in-one platform that changes how your teams work together. And they just introduced an AI-powered assistant that is brilliant. It will take the busy out of work. With Coda, your workflows and content, they're already living in one place, right? But Coda AI, this will help your team focus on the highest priority work, even as your priorities shift, which in startups and dynamic bigger companies, they always do.

25:27This is going to empower you to prioritize work that's long term and strategic, that pays off massively. That's what we should be doing with AI. And you need to get your entire team on Coda and using Coda AI every single day like me. If you're watching the video right now, just look at this powerful and awesome jtrading.com website. This was built to track all my stock trades, and I am crushing it. I was able to do this and track all the stock prices live and build tables and formulas all in Coda. Get a competitive advantage if you want an assistant that lets you get back to work and does all the busy work get started with coda ai today for free how's that for a price head over to coda.io twist coda.io twist to get started for free so you have like all the creative tools that you need here that you could have in powerpoint or canvas and like that you can record your screen you can use shutterstock assets do different shapes all this sort of stuff love it where i think this is interesting right so i think up until now this is like video as you know it right it's a smarter way of training video but what's exciting for me about not just AI video, but any kind of like generative AI tech right now is it opens up an entirely new world, right?

26:33And we're just dipping our feet in this. Right now, I think all these technologies are kind of like the first websites that look like newspapers on the screen. That's kind of what people could imagine back then, right? Today, websites are very different. They have links, there's audio video content, we have news feeds that are personalized to everyone, a website is very different than a paper magazine today. Our bet is that the same thing is going to happen to AI video. So right now, we just understand there's like an MP4 file, right? But we've eliminated the camera from the equation. But really, this is going to be an entirely new kind of media object moving forward.

27:02Maybe we won't even call it video in a few years. And a lot of what that I think is about is like the UX of how you create these things. So I'll give you a very simple example of how when you work with AI, and you don't work with camera footage, you can do cool things. So let's say you have my script here. So it says this is the same title screen layout. In here, I want to have an animation. I want to have this like title text element in my video come in right here. What I'll do here is just add a marker. This is marker one here. I'm going to click on my text element on the canvas. I'm going to scroll down.

27:31And I'm just going to tell the app that I want this particular element to fade in at marker number one. Pretty simple. But when I didn't play the video, then what's going to happen is that this text title in my video is going to come in at the point in the script that I want to come in that in itself is of course i mean that's cool that's easier than doing it in adobe premiere we have to sit with different layers and a timeline and things like that but where this becomes really interesting right is if i type something in front of this then it'll just move down so it's completely responsive right right now imagine that you're translating the video which is a massive headache today right then french is 20 longer than german who's a bit shorter than japanese so if you're automatically yeah exactly all of this stuff you just translate the text, it just works in different language, you don't really have to do anything else.

28:17In the future, you know, all this stuff, I think will be much more personalized. Maybe it's interactive, you can talk to the avatar. I really think we're just dipping our feet in this stuff. But we're really building for a new type of media format. We're not just building to make kind of a better tool for making MP4s, right? And we're dipping our feet in here. I think this industry is just so early. When you generate this, it just generates you an MP4. And then you would copy this into your learning management system or whatever it is. Exactly. But our bet is that over time, it will be something different, right?

28:45And that will probably also require a new video infrastructure where every single video might be personalized to whoever's watching it. Maybe you can just ask the avatar, hey, sorry, I didn't really understand that. Could you explain it in a different manner? We have all these amazing technologies that are false multipliers. GPT is an obvious one here, right? That can help you write the script. We also have that functionality. Stable diffusion can generate the images for you. I think when you think of all these tools sort of working together in the next three to four or five years, I think it's going to be absolutely amazing the things that we can build.

29:13Yeah, I mean, it's obvious you will not need to export these videos eventually. They'll just live at a URL on your service or a white label URL. And then you don't need to export it because that then leads to all kinds of problems. If you're having to upload it to another system and then compliance, maybe you forgot to upload something. It just puts a human into the process. us and i guess your idea here is if you had all 40 uh of the lessons to be an employee at a fast food restaurant you could then ask you questions and just you could have this on an ipad or somewhere in the you know at the workstation and say hey remind me how to make a big mac or do you know how does uh what is a uh double double animal style and it'll just be like oh yeah double double animal style just play the video you know how to make a double double animal style and the person doesn't have to feel bad about not remembering and the chat gpt interface will be able to interpret what they're asking i mean it's just exactly you're a salesperson you're on a call after the call you just get a two or three minute long kind of feedback video right because the transcript has been taken out gpt for whatever maybe a different tool than us has analyzed you know what can you do better talk a bit less remember to disobjection face it in this way like there's so many things you can do with this and i think it's just kind of really exciting to see what people will build once.

30:32We have an API today, we have a bunch of users using that, but I think as the technology evolves, we'll really see a lot of very interesting use cases evolve. They'll just fundamentally be very different from like video as we know video today. If you think about this, and great demo, by the way, anything else you want to show in the demo? And for folks who are listening, you can go to youtube.com slash this weekend and you can find the demo of this. Yeah, for sure. I mean, maybe just give you a cool real life example. This is one of our amazing customers, GoCardless, right? They've built the entire university using Synthesia videos.

31:02So what you're seeing on screen here is sort of their hub for learning how to use GoCartless as one of their customers. I mean, it's all the kind of this entire value prop we've talked about already, right? Like, video is a much better way to communicate with people in 2023. These videos, they just go back in, edit it. All these are created entirely in the platform, right? And then they're served by Wistia, probably, or something. Exactly. That's an interesting thing. I mean, if you look at Wistia's business, which I love, they do really great compliance things where you can stop for a prompt or a question or capture an email address, you know, whatever.

31:38And so I think there's some great API uses there. Amazing job. Let's sort of then level up this discussion. I know the company's high growth, tripling revenue year over year. You raised about 100, just over 150 million, got 200 employees doing this or so. um and so let me ask you a question about how you think about running your own company i have noticed i'm forcing my employees to use chat gpt for every day barred etc let's say about 60 i got over the line the other 40 uh basically got the summer to get their butts in gear if they don't they're out i'm dead serious like if you can't use chat gpt for effectively can't work for me starting september 1st so for people hearing this i've been pretty gentle about it but now it's going to be a mandate i'm going to start having people tested on their chat gpt4 ability why because once people start using it what i'm finding is they become maybe between one percent and ten percent more effective at their job every month if you compound those numbers at but let's call it i don't know three percent that means uh in every three percent divided let's just say 2 % that means every 36 months you'd be twice as good at your job if it was 4 % you know every 18 months you'd be twice as good at your job and that means I don't ever have to hire anybody because the pace of our growth if everybody just commits to being 5 % better at their job every month they become twice as good at their job every 14 months so do you think you can just keep growing your company without hiring people and just making the existing team more effective through using these tools?

33:21I mean, we use GPT everywhere. I think I'm very aligned with you. It's something that I don't think there's any function that can't use GPT. We use it in the most random of tasks, like if you just want fun example, right? But when you have to record your avatar, that's like a four or five minute process. You speak to the camera. We want to get like a nice variation of things that you say. We use GPT-4 to just generate infinite amount of scripts, right? Otherwise, we have to sit down, write a different script for every single person. Now we get a great variety of what people are reading out to the camera, which is great for our algorithms.

Read the full transcript

33:51I think for us, I mean, we'll continue to grow the team. I think people still want to talk to a salesperson. I think in our case, R &D, we do very fundamental R &D here, right? So we publish papers. We have a team around 40, 45, primarily PhDs, solving some really hard things here. And we'll need more headcounts with that. GPT-4 is not going to do that yet. I really see it more as just kind of like up-leveling everyone. I don't think we'll stop hiring people because of GPT-4. but I definitely think it changes the shape of a lot of functions right um and when you think about it if you have 40 people in the development team how much better are your developers yet you have how many developers how many engineers you got ballpark 80 I think okay so we got 80 engineers how much better did they get this year using copilot would you guess if you had to guess on average how much more effective are they at writing code probably 20 30 percent okay like means you hired eight.

34:47That means 20 % of 80 hired 16 people. Absolutely. I mean, and then do you think this is accelerating, staying the same or slowing down? Do you think the gains in 2024 will be faster? Or do you think it'll be 20 % in 2024? Do you think it'll be 40 % or 10 % if you had to guess one of those three numbers? 1020 30 30. I think I mean, there's no doubt But all the technologies are just still incredibly early, right? So they're accelerating. So that means next year, you got a whatever it is 96, we'll round it up to 100 developers. And you think it's gonna get 30%. That means you're hiring 30 new developers next year.

35:27Without having them. This is the profound thing. When I walk people through this, and I reflect back to founders, what they're telling me, I think the obvious thing to do is to focus on the productivity of the existing team. because when you add people it doesn't make an organization faster it actually slows it down because you got to onboard those people at culture this and that i think companies the really great companies are going to focus on individual performance getting those 30 gains 10 20 gains whatever it is consistently for the next three or four years and not add headcount and it's requires a brain shift of writing a job rec versus and doing a search what does that typically take 30 days 60 days you know probably 60 days worth of time and probably four weeks of effort right uh definitely i mean that's a lot more than making a just you know doing some professional development with existing people in my mind yeah for sure i think the way we think about it is i I think we've always been very conservative on hiring.

36:34So we run a very, very sustainable business. And I'm obsessed with how much revenue do we bring in per employee as a metric for how much are we leaning forward into growth. And I think especially if you compare that kind of number to what's the average salary. For me, that's a great benchmark for how out of touch with reality you are. When we started the company, we found the company. We've always run the company like a business. I think there's a big problem in Silicon Valley with people with overhiring. I think obviously now that story is very different, but in 2021, right? It was just crazy.

37:06I mean, looking back at companies hiring 600 engineers in one year and thinking that that's going to make you more productive, you're probably way better off just sticking with 30 people, right? So I think the productivity question is very good. I think the best company has got to be the ones who utilize these tools, not just as developers, but in customer support, in content marketing, in every single facet of the business. I mean, that's what you're doing. your your literal the purpose of your tool is to not have to hire an av production company and the fact is you know people used to hire a production company to do even the most you know short video and now this you literally had to have a studio in your office like people would build a broadcast studio for a quarter million dollars at yahoo they actually they had probably a 10 million dollar broadcast studio so that their ceo so marissa or before her whoever could then address the entire company and they said yeah you know we have this 10 million dollar studio we do yahoo finance every day the the ceo or whoever can use this to talk to a thousand people to 10 000 people and now it's like or they could turn on zoom and for 1995 a month they could talk to everybody and i'll tell you i think if you one of the stories we talk a lot about in uh in synthesis right so we go even further back let's say we go back to like the 1960s 70s something like that right producing text was literally someone's job most people did not go to work and type out a lot of things right you don't have a secretary who would like write all things you have copywriters and if you look at today like writing is a part of every single white collar job right there's not a job where you're not like doing some sort of writing and that's just normal now we don't really think about that because you know why would you not i think it's going to be the same for video content right in 10 years i think just as everyone today writes things all the time everyone will create video content all the time the reason we're writing is because it's the only really truly scalable way of communicating with one another, right?

38:55It's really easy writing on a keyboard. But as it becomes as easy to create audio, video, maybe imagery, I think we'll just really see a lift in how we communicate. We already see this to some extent, right? Look at TikTok, for example. It's like the interface is almost only video. You're literally just scrolling through videos. There's almost no text left in the interface, which is very different from like a Facebook, right? Where it's all about text and Instagram is maybe halfway text, halfway images. I think if you just look at history and kind of, you know, the trajectory of how we communicate, it's always about being more visual, more interactive.

39:26And these technologies are going to be the things that truly unlock this at scale. And I think that's the really exciting part for me, right? It's just, you know, 10 years, everyone will be creating video as part of whatever job that they're in. Yeah, it's amazing. When you're selling to B2B buyers, you really need to get your pitch in front of decision makers, the people who control the purse strings out of business. Why? The upper level execs. They're the ones who eventually make the purchasing decision. At some point, you need to have the CFO, the VP of Ops, the CTO, the CEO, the president themselves approving any kind of big spend.

40:04But here's a problem. How do you find those people, especially on social media? You don't unless you use LinkedIn. and LinkedIn is about to hit a billion users, 930 million members. In that 930 million members are 180 million senior level executives. And I don't want you to know this, but there's 10 million C level executives like me on there. Those are the ones you need to build consensus with to get your product integrated into these companies. They have the purchasing power and LinkedIn ads is specifically meant for B2B marketers. No other platform can offer you these people and all the data inside of LinkedIn exists only there and nowhere else.

40:38You want to be where business is done. LinkedIn equals business business equals LinkedIn. It's that simple. Whenever we post content on LinkedIn, we get really thoughtful responses from business people make b2b marketing everything it can be and get a hundy$100 credit on your next campaign from your pal jacal and the fine people over at LinkedIn, go to linkedin.com slash next unicorn to claim your credit. That's linkedin.com slash next unicorn in terms and conditions apply because hey, they're giving you a hundy i'm curious when you look at this technology as a technologist who's been grinding at this for you know getting close to a decade um and i mentioned at the top of the show it's it's a dual fate technology there's a group of people who believe this could uh lead to massive job destruction at a pace in which we will not replace jobs uh you and i just had this conversation about how efficient more efficient humans can be we both put it at 30 a year i'm i think we're exactly in sync and when i had aaron levy brian chesky and reed hoffman i asked them how much on average are people becoming uh more efficient already they actually put the number at 20 to 40 either 20 30 or 40 i think they two of them picked 30 and one of them picked 20 and you and i just picked 20 or 30 and i picked 30 so let's put the number at 30 a year uh that means everybody doubles their productivity every two years so are you in the camp that let's just start with jobs not terminator and it's the end of humanity let's just talk about jobs in the in you're in europe i'm in the united states europe's got a very protectionist attitude towards uh employment pretty hard to fire people and uh eu's you know really concerned about ai jobs are going to uh be eliminated at a massive rate your technology eliminates the need for camera operators um a studio a location uh translators i mean dozens and i mean if you're if you keep succeeding at this rate you will eliminate millions of jobs just with your own piece of software so do you are you concerned that those jobs will not get refilled are you in the camp of we will have universal basic income this is going to be a tragedy uh and in the dual fate uh this is going to be uh cause riots in the street because people are going to be 30 percent unemployed or are you in the camp of humans uh are creative beings and just like podcasting and being an influencer uh or uh being a podcast producer whatever jobs didn't exist previously will now exist and so humans will find something to do with their time where do you land i'm definitely long-winded questions no that's great i'm definitely the latter camp i think in 10 years being a camera operator is is probably not like um you know the most desirable industry job to be in but i think what can easily be missed here is also the kind of expectations of what what in my world of communications right so it's like okay all these people who are doing all this this content today they'll be disrupted to some extent but i think you know the expectation would just be that everything is video in 10 years.

43:47So we will all need to be able to produce video content. I also think that there's going to be, I think, human creativity. I think the absolute best films, music, like whatever in the creative sphere that you're kind of doing, right? I think that's not just about production. That's very much what we are focused on. I think a lot of these tools are today. There's already lots of people who make video content. There's already lots of people who make music, right? But it's still only like a very small percentage of those people really make it to the top. And I don't think that's because they have better production tools.

44:17I think that's about creativity, understanding cultural trends, remixing things that haven't been remixed before. I think that's like the pace of AI right now. It's hard to claim that it's never going to be like as creative as humans will be. But I definitely think that this is about lifting all boats and ultimately making everyone more productive. But of course, there's got to be job disruption. and it would be, you know, we don't have people making, we don't have that many blacksmiths anymore after horses got replaced by cars. The typing pool you referenced earlier is gone. There's also no more horses and buggies and we don't have the people who clean up the horseshit behind the horses and buggies, which was a job for humans.

44:59In New York, you literally had people with shovels who shoveled horseshit for a living. People had to walk with a big red flag in front of a car, right, when they were driving in the beginning in the early days of cars so let's go to the next one which uh you have to own a bit of which is defix um how do and i'm asking just for the audience's sake if there's any mids in the audience when i ask a dumb question or seemingly dumb basic question i'm doing it to have a open-minded conversation about that conversation not because i don't have my own views on the answer so here's the question um deep fakes are going to be confusing two people uh there was just an example of a fake person on twitter writing all kinds of racist stuff to create uh animosity in the united states with our original sin and the most painful part of our history here in america slavery and race relations it's the wound that you know keeps getting ripped open and guess what people want to keep ripping it open so they make fake videos uh and uh they incite racial divide or they incite divide uh with fake videos your tool perfect tool to create those uh so what is your responsibility as the ceo of a company that allows you to create fake avatars or synthetic avatars in combating this or do you have any responsibility and is the responsibility on the individual to not trust everything they see uh and to be questioning how do you look at this issue i think we absolutely have a responsibility and and we do a lot I think as with anything, there's no one single answer to like, what can we do to alleviate this problem?

46:39I view this primarily through two lenses. One for me as a CEO of Synthesia, how do I make sure that people are not misusing our tool? I'll talk a bit more about that. And secondly, I also spend a lot of my time with NGOs, governments, media organizations, and like, what can we as society do to alleviate the misuse of these tools, right? Because no matter what I do, there's a lot of, got to be a lot of open source stuff. There'll be a lot of other players who may have less ethical standards than we do. and I think a big part of this is going to be about education as you propose yourself. In terms of what we're doing, so for us, AI safety is incredibly important.

47:09We found the company on an ethical framework, consent, control, collaboration. I think the first one, consent, is fairly obvious. You can't make an avatar of someone without the explicit consent. You can't just go in and upload a video of a politician or something like that. It goes through a KYC style process when you actually make an avatar. On top of that, we do very, very heavy content moderation so we definitely have a view on what kind of videos you can create and what kind of videos you can't create. We have an entire team who sits with that. It's a mix of algorithms trying to figure out content that we don't want to generate and also human moderators looking through content.

47:42It's really hard and we're not perfect. I know you've been involved with the Twitter team, so you know that moderation at scale is a very, very difficult task and we are creating a lot of videos every single day. For us, I think what's interesting is that we have what we call red content, which is a country you obviously don't want, right? to like hate speech, you know, swear words, things like that. Those things are actually sort of fairly easy to detect. What gets much harder is in the kind of gray area content. So let's say you're creating content about cryptocurrencies, for example, right?

48:12Are you explaining how blockchain works? Are you trying to get me to throw all my savings into a get-rich-kick scheme that ultimately is going to lose me all my money? Are you creating health content that's about a great diet? Or are you trying to sell me pills that promise that I'll lose 10 kilos in a week or whatever? oh my lord you have to become do you feel as the platform it's your job to police that 100 so we police that very heavily yeah very heavily so if you try to create content even news content we've also uh you can only create news content if we if you're an enterprise client and we know who you are basically right so if you just sign up a third dollar plan and you try to create content around politicians for example that's not going to make it through content moderation to say that's really um that you're really taking it seriously because if nobody in their right mind would expect microsoft office or google doc suite or zoho whatever suite of office and any office suite nobody would expect them to police the output of powerpoint if somebody some nazi group in the united states wanted to create a nazi powerpoint there's no way microsoft or google slides is going you know what we need to have somebody looking at the output here and when the keyword Nazi comes up, or white power comes up, we need to get in there and stop them from using it.

49:27So kudos to you. I mean, I think some people might. Yeah, I love you made that connection. Because that's, that's what we always talk about. In terms of what I usually explain to people, right, is that because these content creation tools, and I think what's happening right now is that content moderation sort of traditionally has been very much at the point of distribution, right? So as you said, you make a PowerPoint about something horrible, or maybe not a PowerPoint video about something horrible, and then it's up to YouTube to figure that out and take it down or Twitter or whatever. And I think what's going to happen with more and more of these AI tools is that at the point of creation, we're going to monitor what kind of content you're creating and making sure that people are creating horrible content.

50:04Every business will, of course, be different. We've decided to take a very forward-leaning approach to this. We work with the Fortune 100 and I have no incentive to have someone pay me$30 a month to make questionable content. And I also really think that's what's right for the world right we can't keep this back forever but i think right now we're clearly technically way ahead of our competitors and i think it's probably a good thing that we wrote it out inside in this way and then you know down the line once open source libraries become good enough people have kind of seen avatar videos before and hopefully it'll be kind of like more on guard right but it's a really difficult question and it's going to be a problem just don't know about that i think victor you're you're you're taking it seriously um i think most people would say hey we're just a tool if people want to rob a bank with a volvo or a tesla or a bmw that's their choice uh it's not bmw or volvo's uh you know concern uh it's not their responsibility but you kind of fall more into hey this is a powerful tool this is not an average tool therefore it would be as if a gun manufacturer said hey we know in i don't know whatever texas you can just walk up and buy a gun but since our guns are pretty darn powerful uh we're going to ask that uh we have your name and you register your gun and you get insurance for it and uh you sign off on this code of conduct now in america that would be unconstitutional but i'm giving an example here that if you do sell to the public you have a choice of who you sell to and you might not want to the gun manufacturers have a religious crazy thing here in america where they believe everybody should be able to trade these things like you know uh they're trading old comic books and they have gun shows where people just trade these things willy-nilly but um you know it'd be quite refreshing if you know the person who sold the ar-15 or the most popular assault rifle said hey uh we're only gonna sell in these states uh under these circumstances and we want everybody registered and uh you're licensing the gun from us for 50 years and if you transfer it we want to know who you sold it to and they could actually build those laws you'd probably get fought but i do think being thoughtful is important and i actually think that like i find that in this industry like most companies are thinking along those lines like open ai of course also doing quite a lot of moderation like what you can make dbt for say which can be annoying as a user sometimes the trade-off is ux right like and it also is for us like also have users who complain that their videos is caught in content moderation but ultimately it's like it's a trade-off right and i'm really happy to see that at least the most powerful technologies right now are being kind of somewhat safeguarded i think did you see the open a uh did you see the open ai lawsuit last week by chance i saw that and that's going to be very interesting to see how this is going to pan out just to give the audience a little refresher here open ai um has been sued by a group of anonymous plaintiffs individuals lawsuit alleges open ai violated privacy laws by secretly scraping 300 billion words from the internet from quote books articles websites and posts including personal information obtained without consent the lawsuit is seeking three billion in potential damages based on a category of harmed individuals estimated to be in the millions plaintiffs are anonymous but some online information about their backgrounds revealed it's all over the map social workers minors uh all chat gpd users 16 total plaintiffs uh you know across the thing the training data is important here.

53:31It's very important. How do you look at what would be a logical and pragmatic and fair way for people building language models to get permission, and for people whose content is being used to get compensation or to be able to opt in or opt out? What's your solution, Victor? I think it's a really hard question. It's been a theme for many years, right? But the kind of emergence of these very, very large models, I think, have changed a little bit. So if you kind of go back one year, one and a half year, whatever, let's say before ChatGPT was released and we had the big cultural moment around that.

54:11A lot of companies, including ourselves, were like, should we scrape data online? Should we not do it? For us, we've always stayed within what we know is definitely okay. We're just getting permission to every single video that we use to train our algorithms, right? That's a lot of hard work. It's a lot of money. It's a lot of time to get that instead of just scraping it off the internet because we believe that that's the right thing to do i still think that's fundamentally my kind of belief but i think what has changed now is that these very large models gpt style um uh you know technologies i think with them is that they're literally impossible to create if you don't train an internet scale data and if you train an internet scale data it is almost like you know uh it's it's just going to be impossible to get consent from everyone like If you need to get permission from every single author of a tweet or a Facebook update or a blog post of that, that's not going to be feasible.

55:01I don't think I have the answer as what should happen here. I can tell you that from our perspective, we care a lot about having clean data. And I hope that's not going to make it harder for us to compete in the future because we're staying within those boundaries. Right now, we definitely have some set of companies who say, we'll ask for forgiveness down the line and just scrape the internet, train our models on whatever data we can get our hands off. There's other companies who are saying, we'll actually stay within the kind of boundaries where we definitely know is okay and do the hard work to get the data both in time and money.

55:37For me, the most important thing as the founder of Synthesia is just clarity. Right now, nobody has any idea of what's okay, what's not okay. and i hope that um that the that governments around the world will unite around some sort of guidance on what should you and what should you not do right but it's a really hard question i don't have to answer for it yeah i think um we as an industry uh should get ahead of this and a very simple solution i have is robots.txt uh is how google decides if they're going to scrape your site or I should say scrape. They're scraping it. They scrape your site and index it, you get the benefit of getting traffic from Google.

56:18But you also have the choice to opt out. There needs to be something that's not binary on or off. It should be you can scrape the site with permission, please ask first or two, you can scrape the site for a minimum of$1 million per year, or you can scrape the site for$1 million dollars a year or you can scrape the site for one million dollars a year for five hundred thousand dollars a year if you send five hundred thousand dollars in referral traffic back at a minimum uh etc etc you can you know you can come up with all kinds of licensing arrangements and those licensing should be since we have smart contract technology easily to enforce and so what open ai could do or let's say twitter and reddit twitter could say in order to do this you have to go through our API and get permission ready could say you know what you simply have to pay us$1 per user of your system per year with a minimum of a$1$1 million commitment and it's paid by buying these tokens here put in your credit card and then put in your you know hash so they have it i agree i think i think this is one of the areas where crypto is actually a really interesting solution where we could I think it's a great way for computers to transact with computers, right?

57:40I think that's a lot of interesting things you can do with that. It's also like, I'm in a lot of the you know, I spend a lot of time with regulators and I'm in both the EU AI Act which is very, very restrictive and also the UK legislation. And one thing that's interesting, right, is that there's this sort of debate around, and of course this is by the AI companies who say, look, we shouldn't moderate what you use to train your models. we should moderate the outcome of those models, right? So now let's think of this as like, you could train all the world's images, but you have to make sure that if you have an image generator, for example, if you type in, you know, create me this image in the style of Andy Warhol, for example, then you can, at a prompt level, you could block and say, Andy Warhol does not want to have anyone prompt images in his style.

58:25Brilliant. Good punch up. Yeah, I think that - Or, yeah, it was really, oh, you know what, just to build on yours. If you typed in Andy Warhol, say Andy Warhol photos need to be cleared for, you know, non commercial use at$100 per image for commercial use, click here for your license options. And it could just be automatic where you could go buy tokens. API calls, essentially, and open AI has this, you buy a bunch of calls, you say, Hey, listen, for this number of, you know, call searches, whatever, you know, just load up on your Andy Warhol tokens, or your Marvel tokens, or your Disney tokens.

59:02And it's all for non commercial. But yeah, if you want to make disney characters yeah just buy a hundred dollars worth of tokens or 50 bucks worth of tokens and have at it it's a brilliant idea so get at the prompt level anything that is ip protected we can make an ip this is such a good idea is are you the first person to did you hear this from somebody else or did you just come up with this oh i think i i came up with it one of these meetings i think it's just the most i think it's the most pragmatic way to ensure that what because a lot of the concern from people like artists today right which the world that i'm in is someone can just create 100 ,000 different images in the style of Banksy or Andy Wall or whatever.

59:38And this is what people are concerned about. And I think rightly so. And this, to me, is a very practical way that could be implemented tomorrow. I mean, OpenAI is already, to some extent, doing this. If you ask it about specific people, it won't give you an answer unless that person is a public persona. So I think there is something very interesting around limiting what these models can output to not include copyrighted works, for example. right so i feel like because i think the other solution with tokens is amazing but we all just we both know that that's going to take that's not going to be implemented in six months right that's a much bigger project to get all this stuff sort of running and and and i think there's a lot of inherent issues with it i think what's going to be interesting especially on the tech side of things with chat gpt right is okay so you have this travel blog and you've you publish a lot of content and that gpt has been trained on that and you you go and you ask it hey what should i see in the North Vietnam or something like that, right?

1:00:32And actually a lot of what is learned comes from that travel blog post. Also, you don't need to go visit a website anymore. I think that's, I don't think the technology necessarily is there yet, but it's going to be there very soon where maybe you don't really need to visit that many websites, which are just text. That's going to have a huge impact on those creators. I again, don't know what the solution is, but I think we're just very early, but that feels like one of the kind of inevitable things that's going to happen as these technologies get better and better. You have been an awesome, unexpectedly awesome guest well done i'd like to have you come back on the program and maybe do an ai roundtable would you be open to that yeah i'd love it okay producer nick are you on the call yes okay producer nick victor is very smart and insightful i'd love to pair him on one of our ai roundtables in the next 60 days uh you know maybe with sunny and uh just have him come on and we just wrap out uh since he's he's deep in this and he's got really great ideas everybody check out synthesia spelled s-n-y s-n-s-y-n-t-h-e-s-i-a are you the dot com or dot ai dot i o dot i o okay there you go uh synthesia and uh maybe he's hiring maybe he's not i don't know i may have convinced him to never hire again but i think he's got some careers his staff is watching this podcast are we gonna hire people again or you or did you drink the jay cal kool-aid of just keep having earnings go up i think this is what takes us out of the recession is more and i don't worry about people's jobs i think victor there's going to be 10 000 more startups that were not possible before ai that will be equally equally they're they're uh a ai companies bai companies there's before ai companies and there's after ai companies There's a whole swath of companies that are possible after AI, uh, you know, exists just like before Christ and after Christ, uh, you know, things were possible that weren't.

1:02:31All right. I mean, it's just going to be so much easier to build products, right? It's like English is the English is the new interface. Yeah, exactly. I totally agree with you. And it's very, very exciting. This is what's going to scare people, but should inspire them. if you speak english or any language any human language you're going to be able to produce videos produce illustrations produce incredible writing songs images uh code apps the the verticalization and the specialization is going to be democratized and everybody's going to be you know let's call it maybe not a great songwriter but okay an okay songwriter a good one everybody's going to be able to write a good blog post maybe not a great one an okay one a good one eventually may be great but if you thought hey i have to hire a blogger to write the this or i have to hire a pr person to write this press release the answer is you're not going to everybody in the organization is going to be able to write a press release that doesn't mean you might have some comms expert who's going to do the strategy but we're all leveling up and the idea that we would sit there and worry about the ink and the hammers in the typewriter anymore is over and i think you just have to say listen we took the typewriter and we threw it in the garbage and now everyone's on word processors okay we took some of these jobs that were tedious and we got rid of them my hobby is music production right and that's that's what i grew up being being interested in and it's just amazing how if you look at the shape of the music industry right back in 85 versus today, how it's been flattened out, talent can come from anywhere, it's about making great music lessons, it's about knowing the right people.

1:04:12Like all those things, I think, is going to come to the tech industry as well, right? We've seen some of that over the last five years, but I think as the kind of art of coding becomes more accessible, it's like, you know what, why couldn't you be this eight-person company sitting somewhere in Norway or Denmark or something like that and create something absolutely amazing? And I'm very excited for that, I think. I think I love technology that democratizes enables people to do more with that time. It's still going to require great ideas. You're still going to need to have a great product sense.

1:04:40You're still going to need to be a great songwriter, right? If you really want to build something cool, we can just remove all the barriers that exist to creation today, right? And that's going to be fantastic. All right. You've been awesome, dude. Thanks for giving me an hour of your time and sharing it with the This Week in Startups audience. This Week in Startups audience, Founder Fridays are coming and they're going to come to your city. You can go to founderfridays.tech I'm looking for some folks who want to host founder Fridays. It's only for founders. If you're not a founder, you can't come.

1:05:12And we just did the first one here in the Bay Area. 50 people showed up. It was awesome. So this week in startups founder Fridays are starting in September and we're looking for hosts. These used to be called twist meetups. We'll call them founder Fridays. We thought we'd come up with a fun brand every Friday, 5 to 8pm after work. Everybody gets together and shows what they're working on. And we'll see you all next time on this week in service. Bye bye.

From the publisher


Crowdbotics. Great ideas can change the world, and Crowdbotics is the fastest way to turn those ideas into code. Get a free scoping session for your next big app idea at crowdbotics.com/twist

Coda. A new doc that brings words, tables and teams together. All your valuable data, plans, objectives, and strategies in one place. Go to https://coda.io/twist to get a $1,000 credit!

LinkedIn Marketing. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to linkedin.com/thisweekinstartups

*

Today’s show:

Synthesia CEO Victor Riparbelli joins Jason to discuss how his startup uses AI to generate realistic video avatars (10:16) before demoing the product. Then, the two discuss job disruption (31:43), preventing deepfake misuse (45:18), and much more!

*

Time stamps:

(0:00) Synthesia CEO Victor Riparbelli joins Jason (3:47) Achieving unicorn status after Nvidia-backed fundraise (12:25) Crowdbotics - Get a free scoping session for your next big app idea at crowdbotics.com/twist (13:54) Determining pricing for the product (17:19) Victor demos Synthesia and its AI features (24:38) Coda - Get a $1,000 startup credit at https://coda.io/twist (31:43) Job Disruption: How AI will impact hiring and business functions (39:38) LinkedIn Marketing - Get a $100 LinkedIn ad credit at https://linkedin.com/thisweekinstartups (41:06) AI and dual fate technology (45:18) Preventing DeepFake misuse (52:40) OpenAI hit with a lawsuit

*

Check Out Synthesia: https://www.synthesia.io/ Follow Victor: https://twitter.com/vriparbelli *

Read LAUNCH Fund 4 Deal Memo: https://www.launch.co/four

Apply for Funding: https://www.launch.co/apply

Buy ANGEL: https://www.angelthebook.com

Great recent interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland, PrayingForExits, Jenny Lefcourt

Check out Jason’s suite of newsletters: https://substack.com/@calacanis

*

Follow Jason:

Twitter: https://twitter.com/jason

Instagram: https://www.instagram.com/jason

LinkedIn: https://www.linkedin.com/in/jasoncalacanis

*

Follow TWiST:

Substack: https://twistartups.substack.com

Twitter: https://twitter.com/TWiStartups

YouTube: https://www.youtube.com/thisweekin

*

Subscribe to the Founder University Podcast: https://www.founder.university/podcast

More from This Week in Startups

All 653 episodes
Next Unicorns: Building a $1B AI Avatar Business with Synthesia CEO Victor RiparbelliThis Week in Startups · 1 h 6 min
Listen in VO