The Exchanges

Q Wow. Okay. And then you went out and wanted to raise like seven, no?

A So that was the later round. So that was back in 2017. We did a million at five. That kind of got us started. We went like 12, 18 months, as you do. Went out to the market, felt like now we had a working technology. We were back then focused in on, on AI dubbing, not the kind of avatar tech that we, that we mostly have today. And, um, again, learned a hard lesson. You know, we went out and we're like, okay, we built this great technology. We're a great team. Um, let's raise eight million. That completely failed. And for like nine months, which is like dragging our feet, I made all the mistakes you can make as a founder. You know, I dragged out the funding process over nine months, like different data points, different investors. It was a big shit show. And we actually had to rewind because we were running out of money and we ended up raising 3.1. And that kind of took us through to, um, the series A, which is when we had actually found product market fit and had like a sustainable business. But those two first rounds were very much rounds we raised based on story. And it was a story back then that just didn't resonate that much because it was very hard for people to see what we had, how that could extrapolate that into, you know, everything that we came today.

AI assessment note: “let's raise eight million. That completely failed. And... ended up raising 3.1.”

Q And like, bluntly, is Synthesia liable? Are you the arbiter of justice on what is good or what isn't? Also, if something is an opinionated rant that I put out there, maybe it's right and fair, but it upsets a lot of people. Do you see what I mean? Are you the arbiter of right and wrong?

A Yeah. So today we are, um, and that's a decision that we have made. We had a lot of discussion about this in terms of like, how do you, what's our approach to this problem? And for me, again, comes back to the customers, right? We're an enterprise product. Um, it's important that the avatars that we have are not seen to be used in all sorts of like wacky content online. We don't, as a company have, we don't see ourselves as having to uphold any kind of right of free speech. And, uh, frankly, from a business perspective, having someone pay me 30 dollars a month to create very questionable conspiracy content, it's just not good business. So for me, all of those things kind of lined up, right? Our, our enterprise clients, they don't want, uh, you know, ever to us to be affiliated with content that doesn't kind of like match the brand, um, and economically for us, it just doesn't make sense. So we've taken a very strict approach, uh, which means that we are actually going and being the arbiters of truth. And we definitely have people who are unhappy with that.

AI assessment note: “we are actually going and being the arbiters of truth”

Q Why do computer games make for better entrepreneurs? Because Toby at Shopify said that if you can run a clan, it's better than university.

A That's exactly what I did in World of Warcraft. No, actually what I think it is, is that we were probably the first generation that really had access to like, especially a bit more complex strategy games. And basically what you learn throughout life is like decision making to some extent, right? And computer games is a microcosm of the universe, right? And you can run so many simulations and it may not look like it for parents if you're playing like Warcraft III or Red Alert or whatever game that you're playing, World of Warcraft that you're playing back then. But actually what you're doing is you're, you're, you're, You're training yourself to make a lot of decisions really rapidly and to understand what the implications of each decision that you make are in some kind of like game that you're playing. But life is a game. Careers is a game. Building companies is a game to some extent, right? And I think we're just the first generation where you actually can sit by yourself and get so many iterations in that would have otherwise been impossible, right? Like you can't sit down before the computer era and like simulate lots of small decisions in how to run a business. Uh, because you just couldn't do it right. You would have to actually build a business and run a business here. You can play rollercoaster tycoon. Um, and of course rollercoaster tycoon is not like running a real the…

AI assessment note: “You're training yourself to make a lot of decisions really rapidly”

Q Wow. Okay. And then you went out and wanted to raise like seven, no?

A So that was the later round. So that was back in 2017. We did a million at five. That kind of got us started. We went like 12, 18 months, as you do. Went out to the market, felt like now we had a working technology. We were back then focused in on, on AI dubbing, not the kind of avatar tech that we, that we mostly have today. And, um, again, learned a hard lesson. You know, we went out and we're like, okay, we built this great technology. We're a great team. Um, let's raise eight million. That completely failed. And for like nine months, which is like dragging our feet, I made all the mistakes you can make as a founder. You know, I dragged out the funding process over nine months, like different data points, different investors. It was a big shit show. And we actually had to rewind because we were running out of money and we ended up raising 3.1. And that kind of took us through to, um, the series A, which is when we had actually found product market fit and had like a sustainable business. But those two first rounds were very much rounds we raised based on story. And it was a story back then that just didn't resonate that much because it was very hard for people to see what we had, how that could extrapolate that into, you know, everything that we came today.

AI assessment note: “let's raise eight million. That completely failed... we ended up raising 3.1”

Q Today, that would have been a five to ten million round on a 40 to 50 given team and given vision. Would you have been as successful as you have been had you raised that round instead of the ones that you raised, which were leaner and smaller?

A So we talk a lot about this, and I don't think we would. Um, I think if we'd raised eight, we would have a lot, we could, we could have done a lot more things, right? And we would have done more things. We would have built deep fake detection, which everybody wanted us to build. Remember, this is like, this is like in 2018, right? At this point, AI video to everyone outside Zendizia, it was just deep fakes. People thought that that was going to be like the big thing. And that is, of course, it's a problem and it's a real thing, but we always thought this is a subset of AI video. Most people are going to use these technologies to create Awesome creative concept. Um, but if we got that money, I think we would have built a team to do that. And that would have meant we've lost our focus. And I actually think that operating on the constraints that we had at the time really just focused us on our customers and selling the product. We were like ferocious about charging people from day one, even if it was like 500 pounds. Right. And I think that focus and working on those constraints definitely helped us to, to get to the point where actually found product market fit. So I'm a, I'm a big fan of like, uh, I'm a big fan of like working under constraints. I really do think it forces a lot of discipline that's easily gets lost the more money you have because you have more options, right?

AI assessment note: “So we talk a lot about this, and I don't think we would.”

Q And like, bluntly, is Synthesia liable? Are you the arbiter of justice on what is good or what isn't? Also, if something is an opinionated rant that I put out there, maybe it's right and fair, but it upsets a lot of people. Do you see what I mean? Are you the arbiter of right and wrong?

A Yeah. So today we are, um, and that's a decision that we have made. We had a lot of discussion about this in terms of like, how do you, what's our approach to this problem? And for me, again, comes back to the customers, right? We're an enterprise product. Um, it's important that the avatars that we have are not seen to be used in all sorts of like wacky content online. We don't, as a company have, we don't see ourselves as having to uphold any kind of right of free speech. And, uh, frankly, from a business perspective, having someone pay me 30 dollars a month to create very questionable conspiracy content, it's just not good business. So for me, all of those things kind of lined up, right? Our, our enterprise clients, they don't want, uh, you know, ever to us to be affiliated with content that doesn't kind of like match the brand, um, and economically for us, it just doesn't make sense. So we've taken a very strict approach, uh, which means that we are actually going and being the arbiters of truth. And we definitely have people who are unhappy with that.

AI assessment note: “Yeah. So today we are, um, and that's a decision that we have made.”

Q first amazing news. Congratulations. Thank you again for rubbing it in my face. Um, my, my question to you is we were talking beforehand actually about the kind of funding story of the business. Can you take me, and I know it's not what the schedule, but I so enjoyed this. Can you take me to the seed round? People didn't get it. Just what happened and how was that?

A So this is back in the seventies. That's a long time ago now. And, um, the very short story is that we were a bunch of people, uh, myself, my co-founder Stefan, my co-founder Professor Matthias Niesner, and we had this sort of idea that, uh, you know, Generals of AI, which back then wasn't really a term that most people thought about, but terms of AI would change how we create content. And the big shift was that back in 2017 when most people thought about AI was about analyzing data, right? Making decisions that that's kind of like that era of AI. Um, but there was this early kind of GANs, uh, which was basically a neural network that could kind of produce new data instead of just analyzing existing data. Um, and, uh, we thought that this was going to be a world changing technology. We thought it was going to change everything we know about how we create content from Video, speech, audio, music, whatever, but we were focusing on video. I went out to the world with a great PowerPoint deck, uh, and we think a great vision, but, um, I think understandably most people thought we were pretty crazy, right? We basically went out and said, look, in 10 years, you're gonna be able to make a Hollywood film from your laptop, needing nothing else than your imagination. And, um, that wasn't a pitch that landed particularly well, especially not in Europe. We were based in London at the time.

AI assessment note: “wasn't a pitch that landed particularly well, especially not in Europe.”

Q You said about kind of the importance of capital constraints, cockroach mode. I think you hadn't touched your A when you raised your B, you hadn't touched your B when you raised your C, and now you haven't touched your C when you've raised your D. Well, then why raise it?

A Well, I think when you have the money, you do spend it, right? I think for us, it's, it's been more a matter of just We've always been obsessed about actually building a business, um, which, uh, which sounds a bit wacky, but having great unit economics, you know, making sure that we're a business that generates money and revenue and that we're always in control of our own destiny rather than being like tied to like a VC parachute. Um, and so I think we've always just spent conservatively, but it's also very clear, right? That if you know where to spend the capital, it is, uh, it is an amazing asset, right? So building a great go to marketing, for example, I mean, that does cost money and that, It means you have to, there's like a cash flow thing of like, you have to hire like a bunch of people that are really great. They have to train. They take nine months before they wrap up and then the investment is worth it. So I think you want to have capital, right? We want to have a really healthy balance sheet so that you can chase any opportunity that comes ahead of you. And for us, like we want to build like a really, really big company. I think there is easily a 5000 dollar company could be built in the space that we're in and we want to win that. And you're not gonna win that by just bootstrapping all the way, right? I think that's, that's, uh, that's a myth.

AI assessment note: “want to have a really healthy balance sheet so that you can chase any opportunity”

Q Today, that would have been a five to ten million round on a 40 to 50 given team and given vision. Would you have been as successful as you have been had you raised that round instead of the ones that you raised, which were leaner and smaller?

A So we talk a lot about this, and I don't think we would. Um, I think if we'd raised eight, we would have a lot, we could, we could have done a lot more things, right? And we would have done more things. We would have built deep fake detection, which everybody wanted us to build. Remember, this is like, this is like in 2018, right? At this point, AI video to everyone outside Zendizia, it was just deep fakes. People thought that that was going to be like the big thing. And that is, of course, it's a problem and it's a real thing, but we always thought this is a subset of AI video. Most people are going to use these technologies to create Awesome creative concept. Um, but if we got that money, I think we would have built a team to do that. And that would have meant we've lost our focus. And I actually think that operating on the constraints that we had at the time really just focused us on our customers and selling the product. We were like ferocious about charging people from day one, even if it was like 500 pounds. Right. And I think that focus and working on those constraints definitely helped us to, to get to the point where actually found product market fit. So I'm a, I'm a big fan of like, uh, I'm a big fan of like working under constraints. I really do think it forces a lot of discipline that's easily gets lost the more money you have because you have more options, right?

AI assessment note: “So we talk a lot about this, and I don't think we would.”

Q Are we seeing that wall of churn moment now?

A I think a lot of companies are seeing that big time, right? And what I think what's very unique about, uh, about AI the last couple of years is that people are extremely willing to part ways with their money. People don't mind, people don't mind like paying, On a consumer level, like, 30 dollars a month to, like, try out something that looks cool. Enterprise level, like, sign for a fifty-k pilot to do something. Um, but the real signal is not that you sign a contract. The real signal is renewal. And I think there's too many AI startups who optimize or have optimized too much for, like, closing new contracts, not for the renewal, right? And if you optimize for the new contracts, not the renewals, unless you've hit the right thing, which of course some people do, then you're in for a whole, whole bunch of trouble.

AI assessment note: “I think a lot of companies are seeing that big time, right?”

Q Do you think that made the right decision to pull back on moderation?

A Well, it's in the statement. I don't have the actual details of it, but I think it is directionally correct that having humans sitting down and evaluating content is not the right way of doing it. I think what we've seen with products like Wikipedia, for example, is that the collective power of people working together to arrive at some sort of truth Is really, really powerful. And I think community notes is kind of like taking that Wikipedia way of thinking about the world and trying to implement that into every single piece of content. And it's not easy and it's not solved yet, but I do think that is the, that, that is the right way for us to, um, to, to, to kind of have some degree of control of like what people do and say, where it gets really messy is, is this kind of like gray content. We, we have the problem at Synthesia as well, right? You know, we have a full, we have a basically a product internally for content moderation. Um, and it's been a journey for us because the hard thing is you have the, the, what we call the green content, the content that everyone agrees is great. That's 99.9% of the content. You have the red content, hate speech, violence. Most people will agree that that's bad as well. Then you have all the, the gray middle and that's where it gets difficult. I think this is the content that Sok is talking about here, where he's gonna like let off controls…

AI assessment note: “I think it is directionally correct that having humans sitting down and evaluating content”

Q What traits are you slightly ashamed of, but that has contributed to your success?

A I am a generalist in the, like, pure sense of, of the word, and, um, you know, when I was younger, I wanted to be, wanted to be an artist, wanted to be, like, a great computer, uh, Programmer, developer, and I just, like, never really excelled at any of them. I still have a dream of, like, making the music improve one day, but I think what I'm really good at, I know a lot of things about lots of, like, very random things, and I'm just extremely curious, so I'm not, like, an expert in anything specific, but I have a very, very wide range of interests, and I think, like, my neural network's been trained on, like, so many random things, like how elevators work, to black metal in Norway, to, like, some, you know, How people make technology, like, all these kind of different things, and I kind of used to be a bit annoyed at myself, like, why can't you just sit down and become, like, you know, the world's best music producer, like, the world's best programmer, AI research, or something like that, and I think now I've realized that my power is actually the fact that I'm, like, decent at, like, a lot of things, um, because I think especially in my role, right, what you do is you make decisions, and then you get ideas, and you can't force those things. I think, I really truly believe that those companies are subconscious, And the more food and diverse things you put into your brain, the…

AI assessment note: “I kind of used to be a bit annoyed at myself, like, why can't you”

Q So you've got ten million dollars, okay, and you can put it in OpenAI at one 60, Anthropic at 60, or Axe at 50. Which one would you do?

A I actually think I would do X. I think they're all great companies. I would do X because I think there, there's the most asymmetric upside if Elon delivers what he usually does. And I mean, we've all heard of some of the, like, that data set that he built, like, 10 days something, right? I think I would just never bet against Elon, and I think the upside potential there is, is, I mean, it's huge. And I also think that it's, Under, um, I think the fact that he owns X is really, really powerful. OpenAI clearly managed to capture, like, the consumer as in, like, the destination that you go to to use an LLM. I think we'll see LLMs being a part of, like, many different apps, and I think owning X along with building the LLMs is actually really powerful, especially for the real-time information that will be able to feed into the models directly from X, and the fact that X, of course, already has, like, hundreds of million users that, um, that In theory, at least could start using, uh, their, their LLMs rather than going to.

AI assessment note: “I actually think I would do X. I think they're all great companies.”

Q speak to many investors over the last 18 months, and when Synthesia's come up, bluntly they've said, like, amazing and amazing what they've done, but OpenAI are gonna move into this. This is the most obvious play for OpenAI to move into. This and customer service is always kind of the two. How far do you think model providers go into the application layer, and how do you assess that?

A I think this is like classical VC brain, um, which is again, too much focus on the technologies, right? Like, you know, you speak to, I think there's a big misconception about Synthesia is that we're like an avatar company. Um, and it's kind of a fair assumption to make because you go to a website, if you like, you know, see about us online or like, avatars is a headline feature for the company, right? It's one of the things that made us where we are today. If you go and talk to our customers, they don't buy us with the avatar model. They buy us for the workflow. Like the way we've taken the entire video value chain, we get an idea. So you make a draft for that video. You don't have to use a camera or a voiceover artist, right? Because we have the AI models to that. We give you a great editor, like using PowerPoint or Canva. We give you a distribution mechanic with an AI video player that's made to serve multilingual content. This could go on and on, right? Talk to our customers today. The reason we have, you know, million dollar plus contracts is not because people are like, oh, I want to make an avatar video. It's because they want to convey a message to someone and they want to do that with the highest, um, efficiency and engagement and in the best workflow, frankly speaking. And that is so much more than just the models. So of course we want to win on the models and we are …

AI assessment note: “They buy us for the workflow... that is so much more than just the models.”

Q You're also an angel investor now. How do you feel about the five on 25, 10 on 50 rounds with pedigreed founders coming from your massive names? How do you feel about them? And have you seen a trend in terms of how they perform?

A Too much money too early is not healthy. Maybe some people are good at really having that discipline, maybe like second time founders, but it's very tempting to spend on things that you shouldn't be spending money on, especially when things are not working, right? I think once you clearly have product market fit, then a lot of things, a lot of decisions becomes a lot easier because they're more obvious. But before product market fit, it's really dangerous, I think, to be developing two, three, four things at a time, having 15 people working on your team when you're still at ideal stage. And I think a lot of people make the mistake of raising the money and then using it too quickly. Um, one thing you cannot use money for a buy away with is product market fit, right? I think learning about your market, about your customers, it just takes time and having a team of 20 people instead of five people trying to learn that, I think actually slows you down. You have to own that as a founder. And it's just, I don't think we could have learned the same amount about our customers, um, faster with more money back then. It really did take us like two years to just get into the minds of customers, understand video from like first principles. And I think people try and then you hire like a product manager to help you fix the product market fit problem because they're a product person, um, or yo…

AI assessment note: “Too much money too early is not healthy.”

Q Dude, before we kind of dive into kind of journey and elements there, I do just want to discuss AI landscape today. Um, a lot of, I, I speak to so many CEOs and they're like, Harry, you and your tech bros, you sell me the ROI to enterprise. We're still kind of waiting. Where are we at in the AI hype cycle cycle today?

A I think that, I think that's pretty spot on. What I see a lot in the enterprise is, uh, buyers don't really know what they want. A lot of people have been told that they need to have an AI strategy. They need to execute on AI strategy, which means they were very willing to have conversations. They're also very willing to spend their innovation budgets on doing things, but they don't really know what they actually need and want. They don't understand the technologies well enough, um, to, to, to kind of themselves figure out what could they need for their business. And that is, uh, both an opportunity, but I think it's also a problem for a lot of AI startups that doesn't have that customer obsession. It's great because you have a lot of budget available and people are extremely willing to do things and, you know, sign up for pilots and POCs because they want to deliver to their boss that AI strategy. But if, when they don't know what they actually want, it's very difficult to, to prove, to prove the ROI to them, right?

AI assessment note: “they don't really know what they actually want, it's very difficult to prove the ROI”

Q Are we seeing that wall of churn moment now?

A I think a lot of companies are seeing that big time, right? And what I think what's very unique about, uh, about AI the last couple of years is that people are extremely willing to part ways with their money. People don't mind, people don't mind like paying, On a consumer level, like, 30 dollars a month to, like, try out something that looks cool. Enterprise level, like, sign for a fifty-k pilot to do something. Um, but the real signal is not that you sign a contract. The real signal is renewal. And I think there's too many AI startups who optimize or have optimized too much for, like, closing new contracts, not for the renewal, right? And if you optimize for the new contracts, not the renewals, unless you've hit the right thing, which of course some people do, then you're in for a whole, whole bunch of trouble.

AI assessment note: “I think a lot of companies are seeing that big time, right?”

Q Why do computer games make for better entrepreneurs? Because Toby at Shopify said that if you can run a clan, it's better than university.

A That's exactly what I did in World of Warcraft. No, actually what I think it is, is that we were probably the first generation that really had access to like, especially a bit more complex strategy games. And basically what you learn throughout life is like decision making to some extent, right? And computer games is a microcosm of the universe, right? And you can run so many simulations and it may not look like it for parents if you're playing like Warcraft III or Red Alert or whatever game that you're playing, World of Warcraft that you're playing back then. But actually what you're doing is you're, you're, you're, You're training yourself to make a lot of decisions really rapidly and to understand what the implications of each decision that you make are in some kind of like game that you're playing. But life is a game. Careers is a game. Building companies is a game to some extent, right? And I think we're just the first generation where you actually can sit by yourself and get so many iterations in that would have otherwise been impossible, right? Like you can't sit down before the computer era and like simulate lots of small decisions in how to run a business. Uh, because you just couldn't do it right. You would have to actually build a business and run a business here. You can play rollercoaster tycoon. Um, and of course rollercoaster tycoon is not like running a real the…

AI assessment note: “training yourself to make a lot of decisions really rapidly and to understand what the implications”

Q speak to many investors over the last 18 months, and when Synthesia's come up, bluntly they've said, like, amazing and amazing what they've done, but OpenAI are gonna move into this. This is the most obvious play for OpenAI to move into. This and customer service is always kind of the two. How far do you think model providers go into the application layer, and how do you assess that?

A I think this is like classical VC brain, um, which is again, too much focus on the technologies, right? Like, you know, you speak to, I think there's a big misconception about Synthesia is that we're like an avatar company. Um, and it's kind of a fair assumption to make because you go to a website, if you like, you know, see about us online or like, avatars is a headline feature for the company, right? It's one of the things that made us where we are today. If you go and talk to our customers, they don't buy us with the avatar model. They buy us for the workflow. Like the way we've taken the entire video value chain, we get an idea. So you make a draft for that video. You don't have to use a camera or a voiceover artist, right? Because we have the AI models to that. We give you a great editor, like using PowerPoint or Canva. We give you a distribution mechanic with an AI video player that's made to serve multilingual content. This could go on and on, right? Talk to our customers today. The reason we have, you know, million dollar plus contracts is not because people are like, oh, I want to make an avatar video. It's because they want to convey a message to someone and they want to do that with the highest, um, efficiency and engagement and in the best workflow, frankly speaking. And that is so much more than just the models. So of course we want to win on the models and we are …

AI assessment note: “They buy us for the workflow... That is so much more than just the models.”

Q What have you changed your mind on in the last 12 months?

A I think that one of the things I've changed my mind on in this whole AI hype cycle, I think when you're in the middle of these cycles and you see, like, all these things getting funded, all these, like, you know, competitors pop up, you, you get a lot of, like, noise and a lot of, like, oh, should we do these things? Should we, like, listen too much to what the market says? Like, someone is doing this over here, and I've really learned the lesson again, although I think we've stayed the course really well, to, like, never do things out of, like, guilt of other people doing it, or, like, looking too much at, like, competitors. Or, you know, what the broader kind of AI landscape is doing. Like, listen to the customer. They're the ones who pay your bills, and ultimately, they're the one that you need to, to, um, to make, to make happy. I had a period where maybe overindexed a little bit too much on, like, what other companies are doing as, like, the AI space in general, like, what's happening there as, like, the most valuable signal as opposed to the customers. Maybe more like a relearning of a fundamental lesson in building startups.

AI assessment note: “overindexed a little bit too much on, like, what other companies are doing”

Q Dude, before we kind of dive into kind of journey and elements there, I do just want to discuss AI landscape today. Um, a lot of, I, I speak to so many CEOs and they're like, Harry, you and your tech bros, you sell me the ROI to enterprise. We're still kind of waiting. Where are we at in the AI hype cycle cycle today?

A I think that, I think that's pretty spot on. What I see a lot in the enterprise is, uh, buyers don't really know what they want. A lot of people have been told that they need to have an AI strategy. They need to execute on AI strategy, which means they were very willing to have conversations. They're also very willing to spend their innovation budgets on doing things, but they don't really know what they actually need and want. They don't understand the technologies well enough, um, to, to, to kind of themselves figure out what could they need for their business. And that is, uh, both an opportunity, but I think it's also a problem for a lot of AI startups that doesn't have that customer obsession. It's great because you have a lot of budget available and people are extremely willing to do things and, you know, sign up for pilots and POCs because they want to deliver to their boss that AI strategy. But if, when they don't know what they actually want, it's very difficult to, to prove, to prove the ROI to them, right?

AI assessment note: “I think that's pretty spot on. What I see a lot in the enterprise is”

Q What is getting money today? That people think will be very valuable that you don't think will be.

A Like, buzzwords always kind of, like, take me out a bit. People say that, like, building AI agents to do all sorts of different things. That always, like, lights up my bullshit detector a little bit. But I just feel like when you're, like, overly obsessed about, like, the technology and, like, the latest buzzword, that's usually, like, a yellow flag for me. Maybe I'll give you one concrete example. I really, really hate it when people call, like, AI employees. I think it's, like, I think it's so dumb. I think it's, it's not helpful to build useful technologies that people want to adopt. And I just think it's the wrong way of thinking about that. You're going to have like AI employees doing all sorts of different things for you. Like these are algorithms. It's a piece of software. Like you wouldn't say that like Miro or Figma is like an AI employee that like sits and takes people's design and put it onto something, right? I understand that it's because people think these things are going to be making decisions autonomously, but I just don't think it's that different from software that we already know.

AI assessment note: “I really, really hate it when people call, like, AI employees.”

Q So you've got ten million dollars, okay, and you can put it in OpenAI at one 60, Anthropic at 60, or Axe at 50. Which one would you do?

A I actually think I would do X. I think they're all great companies. I would do X because I think there, there's the most asymmetric upside if Elon delivers what he usually does. And I mean, we've all heard of some of the, like, that data set that he built, like, 10 days something, right? I think I would just never bet against Elon, and I think the upside potential there is, is, I mean, it's huge. And I also think that it's, Under, um, I think the fact that he owns X is really, really powerful. OpenAI clearly managed to capture, like, the consumer as in, like, the destination that you go to to use an LLM. I think we'll see LLMs being a part of, like, many different apps, and I think owning X along with building the LLMs is actually really powerful, especially for the real-time information that will be able to feed into the models directly from X, and the fact that X, of course, already has, like, hundreds of million users that, um, that In theory, at least could start using, uh, their, their LLMs rather than going to.

AI assessment note: “I actually think I would do X.”

Q How does Synthesia become a 50 to a hundred billion dollar company? Can you just paint that picture for me?

A So I think we're in the early stages of a shift in how we communicate. If you think of most communication today, it's, it's text-based, right? We send emails, texts, we read things, and text is a great technology. It's amazing technology. It's kind of built the world up to where it is today, but it's, it's actually like a, a, a pretty bad way of compressing information. You lose a lot of context when you transform like your thoughts into something that's written down in a document, right? Uh, as humans, we're much better at consuming visual content. We like to hear things. We like to see things. We like to feel things in the physical world. We can't do that yet. Um, but it's very clear that like higher fidelity content, like video and audio is a better way of training, informing, and entertaining people. The reason that we're using that much text today is because text is the only scalable way we have of, um, essentially storing information and sharing information. Right. But that's changing now because The more we don't need cameras and microphones and capturing things in the physical world around us, the more we can get video creation, audio creation to be as scalable as text. And once that happens, there isn't really any reason for us to use text anymore. And this, this sounds a bit crazy. Um, but I actually do think that maybe not us, but maybe our kids, kids are going to be…

AI assessment note: “we can get video creation, audio creation to be as scalable as text”

Q you need to move into SMBs. Well, you don't have product market fit with them. You have it with creators who are independents, and then you need it with enterprise. And so I just see product market fitters is like ever moving chapter book, which is different. And so actually, can you ever afford to move away from customers? And does it not always become more impossible with bigger teams?

A I totally agree with you. And I think a successful company is a long series of like product market fits, but you need that initial spark, right? Because once you have that, you have something to build on. The bigger you get, the more, you know, bigger company you're building. You need new products and your new product market fit essentially. And I actually do think that that's one of the things as a founder that you should always be focusing on. What is the next market? What is the next, uh, product you're targeting? Right. But in our case, you know, we have, I would say we have like a bunch of product market fits already. And those, I think you can hire super smart people that can run with it, make it great lines of businesses, make it great products. But I think pushing to those next product markets and like where the company needs to be in two or three years, I think that remains to be the founder's job.

AI assessment note: “I think pushing to those next product markets... remains to be the founder's job.”

Q that you're going to continuously upgrade and improve them yourself. Can you imagine having to continuously upgrade and improve a 150 different internal tools? Unbelievable. Um, speaking kind of rational and clear mindset around where we are today. I think everyone acknowledges that we're seeing the complete commoditization of foundation models. Do you agree with that from what you see and how do you see the foundation model landscape evolving?

A For sure. Um, I think, I think what I'll come back to, as it always does, like distribution is king and great products are king. And of course there will be new LLMs that'll be better and more powerful as everyone else excited to see what GPT-V kind of has in store. But I do think we are seeing the commoditization of the text generation layer, right? For most of the use cases that, um, We'll see LLMs transform the world as we know it today. I think the current generation technologies are good enough. It's about, of course, improving the base models, but it's a lot about like building the product, the scaffolding around it. Um, but, um, but, but I, I, I, I definitely think we are seeing that, right? Like if, if, if we're seeing like X and Elon Musk catching up like pretty quickly and Frobek has a really great product. A lot of people prefer those models over open AIs. OpenAI has a huge distribution mode, and I think they've really managed to capture, like, the consumer version of the world here, and that's definitely gonna be really valuable, but, um, I think it's a lot of distribution of products going on.

AI assessment note: “I do think we are seeing the commoditization of the text generation layer”

Q So you agree that scaling laws will continue?

A I think they will continue, but I don't think it's just going to be linear. Like whoever has the most compute is going to win. Uh, the world rarely works in those ways. Something will happen, right? Someone will come up with an algorithm that can, that is like 10, a hundred times as efficient as what it is today. Um, so I, I think compute is important, but I actually also think data is, uh, actually I would say compute algorithms and data maybe. I think in algorithms, what, what everyone is trying to build into models in AI system today is control, right? We've proven these things extremely capable at replicating the real world, producing video that looks real, audio that looks real, text that sounds real, and, but what we all really want to do is, like, get a deeper level of control over these things. In my world, right, it's like you have some of the big video generation models. I kind of delineate what we do versus, like, what Zora does or Runway or something like that. Zora and Runway, these models, extremely capable, extremely powerful, right? You type something in, you put out everything, you put out, like, Basically anything you type in will actually spit out and that's really, really powerful and it's a great demo, but to really be able to use this, you have to be able to say, I want this same character in a different scene. I want the character to say this particular, …

AI assessment note: “I think they will continue, but I don't think it's just going to be linear.”

Q Do you think that made the right decision to pull back on moderation?

A Well, it's in the statement. I don't have the actual details of it, but I think it is directionally correct that having humans sitting down and evaluating content is not the right way of doing it. I think what we've seen with products like Wikipedia, for example, is that the collective power of people working together to arrive at some sort of truth Is really, really powerful. And I think community notes is kind of like taking that Wikipedia way of thinking about the world and trying to implement that into every single piece of content. And it's not easy and it's not solved yet, but I do think that is the, that, that is the right way for us to, um, to, to, to kind of have some degree of control of like what people do and say, where it gets really messy is, is this kind of like gray content. We, we have the problem at Synthesia as well, right? You know, we have a full, we have a basically a product internally for content moderation. Um, and it's been a journey for us because the hard thing is you have the, the, what we call the green content, the content that everyone agrees is great. That's 99.9% of the content. You have the red content, hate speech, violence. Most people will agree that that's bad as well. Then you have all the, the gray middle and that's where it gets difficult. I think this is the content that Sok is talking about here, where he's gonna like let off controls…

AI assessment note: “I think it is directionally correct that having humans sitting down and evaluating content”

Q not necessarily what I am, clearly. But she said, great minds discuss ideas, average minds discuss events, and small minds discuss people. But I want to discuss the future across a couple of different segments with that in mind, and like discussing ideas. When we look at the future of content creation, this is my life, dude. What does the future of content creation look like in, say, five years?

A I think what has happened in very broad strokes is that we democratized distribution with the internet, right? Like, there are no gatekeepers. Anyone can make a website, everyone can share their content online. That's been really, really powerful. Then we saw, to some extent, the democratization of content creation because all of our smartphones have cameras in them, the price of camera equipment dropped, and all those things which, I mean, we're sitting here with, I mean, this setup, like, 30 years ago would probably have been, like, a hundred extra price what it is today. What's about to happen now is that we're truly gonna democratize content creation. We're gonna change the world of making content from something that you need to, where we capture things with sensors in the physical world, to being able to generate everything digitally, right? And that's gonna have huge implications. When you think about text, which we don't really think of as, that's just like a fabric of society. We don't really think that much about text, right? But text went through this journey of like, you know, the printing press is like one of the first really big inventions around text. Then we went to the keyboards and computers. And now it takes just everywhere. Everyone can create content online. It's entirely digital, right? And that's really powerful. With something like music, we've somewhat a…

AI assessment note: “change the world of making content... to being able to generate everything digitally”

Q that you're going to continuously upgrade and improve them yourself. Can you imagine having to continuously upgrade and improve a 150 different internal tools? Unbelievable. Um, speaking kind of rational and clear mindset around where we are today. I think everyone acknowledges that we're seeing the complete commoditization of foundation models. Do you agree with that from what you see and how do you see the foundation model landscape evolving?

A For sure. Um, I think, I think what I'll come back to, as it always does, like distribution is king and great products are king. And of course there will be new LLMs that'll be better and more powerful as everyone else excited to see what GPT-V kind of has in store. But I do think we are seeing the commoditization of the text generation layer, right? For most of the use cases that, um, We'll see LLMs transform the world as we know it today. I think the current generation technologies are good enough. It's about, of course, improving the base models, but it's a lot about like building the product, the scaffolding around it. Um, but, um, but, but I, I, I, I definitely think we are seeing that, right? Like if, if, if we're seeing like X and Elon Musk catching up like pretty quickly and Frobek has a really great product. A lot of people prefer those models over open AIs. OpenAI has a huge distribution mode, and I think they've really managed to capture, like, the consumer version of the world here, and that's definitely gonna be really valuable, but, um, I think it's a lot of distribution of products going on.

AI assessment note: “I do think we are seeing the commoditization of the text generation layer”