Your Videos Are Now AI Agents

Hey, welcome to the Next Wave podcast. I'm Matt Wolf and I'm here once again with Maria Garib. And recently we sent Maria over to London to chat with Victor, the CEO of Synthesia. So we're going to play that interview in a minute, but first I wanted to just chat really quickly with Maria. What did you want to go and talk with Victor about? Related to Synthesia.

They put me in front of their cameras and like, we had to capture everything so that I can make an AI avatar of me. And there's one in the wild right now roaming around and she's funnier than me and she speaks 140 languages. I speak three only, which is wild. I saw her on screen, I was like, is she my co worker? Is she my intern? Is she another version? Is she my replacement? Should I be writing right now? But yeah, I liked it so much. And after we did the AI avatar, I sat down with Victor Repavelli, who.

Is the CEO of Synthesia.

And like, we got into the real stuff, you know, Express to Progress, which is one of their products, and like multilingual tone matching and like real business workflow and the future of video. And he has a whole TED Talk talking about the future of video and how, you know everything's going to be video rather than you kind of having piles and piles of memos and trainings, you know, like, you don't, no one wants to read that. And having a whole video actually makes every task 10 times better. So, yeah, I went in, like, extremely curious and I went out and I was like, my mind was blown by the amount of stuff that is in there.

But yeah, that's their main thing, right, is they sort of make like an AI avatar clone of you that you can then use in your videos. One of the cool things about tools like Synthesia, in my opinion, is the ability to make videos of you saying things at scale. One of the really cool use cases I've seen is let's say you have like an email newsletter, which I know you do. You can actually, when somebody signs up to your newsletter, you can actually send them a video that's like customized to them. That's like, hey, Mark, thanks so much for signing up for Mindstream and it looks like a video of you. Welcome to the newsletter.

You know, yeah, that would be so cool. I want to do these kinds of skits eventually. Just like every single time someone signs up to our newsletter, like send something kind of like personalized and like have a coffee with them without Even realizing, you know, that they're not seeing the real me.

Well, awesome. I'm super excited to listen to this interview of you and Victor. So let's go ahead and listen to the interview.

Yes.

Cutting your sales cycle in half sounds pretty impossible, but that's exactly what Sandler Training did with HubSpot. They used breeze, HubSpot's AI tools to tailor every customer interaction without losing their personal touch. And the results were incredible. Click through rates jumped 25%, qualified leads quadrupled, and people spent three times longer on their landing pages. Go to HubSpot.com to see how Breeze can help your business grow.

Hi, nice to meet you. Thanks for having us in the next wave. I wanted to actually catch your attention on something because I was watching your interviews yesterday and I saw that you liked sci fi a lot, and I wanted to see your perspective over the future of video, because everyone talks about it, even me, who reads a lot. I don't want to have stacks and stacks of books, and everything has been changing. Like people that used to read books now listen to books. So I want to see your perspective about the future of videos in general.

Yeah. Well, we found the company in 2017. There was like two big ideas that the company thesis revolved around. The first one was that with AI, we increasingly can generate content with AI, not just analyze data with AI, which a lot of people forget that that's how most people thought about AI not that long ago. That would drive the marginal cost of producing content, specifically video, for us, to zero. And what that means is not just in dollars, how much does it cost to create a video, but also the time required and the skill required to create video. That, of course, has huge implications for the media ecosystem. Right. We've seen what happened when text and writing became completely democratized, when we had computers and Internet. You can sit in your bedroom and you can write a book on your computer, you can publish your computer. You never need to actually leave your house. Right?

100%.

But there was a time where actually to publish something in writing, you have to have, like a typewriter, you need to have a printing press. And it's a very, very long, laborious process.

And loud.

Yeah, yeah. To create something. Right. And that's kind of what video is like today. We still need to run around the real world. We have actors, studios, cameras, all these kind of things. I think we're in the early days of this, but we're already seeing the immense power of that. Right. Like you're seeing people create amazing things that would never have been possible before without multimillion dollar budgets. And that's really going to accelerate. So that in itself, I think is a big idea. The other big idea which we're kind of getting into now is where it gets probably like a little bit more sci fi in some ways, which is when we invent new media technologies as humans, we always invent media formats which are native to those technologies. So the really interesting question is like, what does video look like in the age of AI? Right now, most AI video is essentially using new technologies to create old media formats. But that's kind of changing now as we're launching our new agent product, which is video you can talk to. So instead of you just passively sitting through, but it's essentially like a broadcast, you can engage with the video. It could be as simple as you can ask the video a question if there's something you haven't been answered in a marketing video, for example. But it can also be, if you're doing a training, for example, that you can actually go in and practice whatever you just learned with a kind of AI tutor or AI role play as we call it. Right. So if it's a sales training, you just learned about the new pricing policy in your region and instead of just hoping that you watch the video and you remember everything.

Yeah.

You could now go into like a 15 minute role play where the AI agent pretends to be a customer and you have to overcome objections, answer questions, and then after you're done with that session, it could go into coaching mode where you can say, hey, you did really well in Area xyz. But over here you maybe want to read up on these articles out here. So that's like the first foray. But I really think it's going to be a world in which video is going to be much more interactive, personalized and conversational, which is going to open up a very new way of in our world, a lot of communicating with customers and prospects, but also learning and training.

I would rather honestly watch a video like a tutorial. My car broke down in Lebanon and I had to watch a YouTube video on how to fix that because I'm not going to read a manual on how to do that. Also eventually, like you were saying something like your kids one day would be able to be taught by a video rather than you ripping your hair off when you're teaching your kids. That sounds more likable. So I want to start off also with Express 2, which is insane. Right now you guys are talking about it because you Know, full body gesture feels like you just leveled up when it comes to, you know, from good presenter to this person is about to keynote Sci Fi thing. As we said, what are people actually using it for outside?

So what we've seen is since we invented the first Avatar technology in 2020, back then the quality was way, way, way lower than what it is today. And we've of course gone through multiple kind of step changes in the quality of the avatar. And generally what we see is just that the better the quality of the avatars, the more creative people get with it and the more use case we unlock. So on the one end of the scale, people are not yet making super bold ads with avatars yet. That just doesn't make sense. The quality is not as good. You don't have as much control as you would have if you filmed it with a real thing. On the other hand, in 2020, when people launched the first iteration of the avatars that were only good enough for low stakes internal training content, very few customers would put it on their website. Then we've gone through this interesting transition where I think with Express too, most recently, most customers are now very comfortable using them in public facing materials. And it's just because it just looks and feels more natural. Right. And that just means that you can use it for more things that are more kind of high stakes in a general training video then generally speaking, you kind of have to watch it. Whereas if you're creating a marketing video, you're more competing for people's attention. Right, right. And so with Express tool, you can begin to actually do that in a big way.

Right. I love when tech behavior goes slightly out of the norm because that's what we really wanted to see in the hands of actual real people. But what's the most surprisingly weird thing that you've seen people use AI with?

I think we just really see the breadth of like people in all sizes of companies and individuals need to communicate a lot of things every single day. Right. And video is such a powerful medium to do it. And so one of my favorite examples, which is like very early on was like we just opened up the platform, there was like a barbershop in Brazil somewhere, really made like avatar videos about like the different hatchbacks you can get. Which is like very niche but super cool. Right. Because someone just found the project, someone just found like a product and used it to do that and post it on Facebook. This is back with Facebook pages and more of a thing. Wow. So that's one of the scale. Right. If you look at more in the corporate world than where we operate today. One of my recent favorite examples is UPS, big bank in the US we have, I think around 200 analysts who have their own avatar. And these analysts essentially communicate to their customers about the markets, investment opportunities, et cetera. And they used to do this with text, but now they switched to doing that with video. And the magical thing is, of course, that video is a better means of communication, but they can communicate with video at more or less the same effort as it would be to communicate in text. Right. And kind of like lowering that gap is really what massively expands the market size for these kind of technologies. If it's as easy to create something in video as it is to make a slide or to write an email, then a lot more people will naturally use video for many more things in the business.

Right, right. And what do you say to the people that say that our attention spans are getting shorter?

I had a controversial talk about this, right. Where I kind of proposed that maybe it's not just because our brains are turning into like brain rot, which I think is what most people's instinct is. It's just if you go back in history, people have felt the same way about radio novels when they first became a thing. Right. Parents would say kids have no attention span. They're just lying in their rooms, they're listening to this low quality radio novel content. They should be out there reading real books. Whatever it happened when I was a kid playing computer games, people thought that if you play computer games, it'd become violent. We always think that whatever is currently happening is bad for us.

Let's talk about dragons, playing about dragons and stuff. I don't understand what violence.

Yeah. There's a counter idea, which is that maybe it's just because we have so much content available to us and we're so used to only consuming the best of the best content. And you can of course argue on how you measure that. But things that keep our attention that we just don't have a lot of attention for, content that's way too long, way too wordy, and not very interesting. Right. If I want to learn about something, as someone who loves reading, I can buy a book on Amazon and that's probably going to be like 250 pages. Not necessarily because 250 pages is the right amount of pages to learn about this topic, but because you cannot write a book that's 50 pages long, you can't sell it. Right. It has to be longer for you to actually sell it. It's probably going to be outdated. It's like, why would I spend the time reading 250 pages if I can watch an amazingly well produced YouTube video that takes half an hour for most people? You learn better like that. So I think there's a lot of. Right. Critiques around social media and what floats to the top of the content parallel. But I just think that things are rarely like as simple as we make it out to be. And I think there is this narrative right now. Right. Like everything to do with watching videos online is like a bad thing.

Look, everyone's talking about AI video tools right now. Runway Cling, Veo Sora. They all claim to be the best, so we actually tested them head to head. Same prompts, same budget, real results. And honestly, some of these tools are way overhyped. If you want to know which AI video tool actually delivers the best bang for your buck, grab our AI video showdown guide. Click the link in the description. Now let's get back to the show.

I think the book community in general doesn't really like the idea of AI being merged with that, which in my mind it's like, why don't you like it? I would love to see my favorite characters on TikTok. I've been seeing them on TikTok while Doomscore and this is what I really wanted to see using AI. Like I don't want to wait for a production house to create that fit that movie because it's going to take a lot of money. I would rather have like a 19 year old create that and me seeing it in real life, so.

Exactly.

Yeah.

And I think it's still you can consume whatever you want. Right. I don't think books are going to go away anytime soon.

They're still going to be weapons of like sci fi, but.

Yeah, exactly. Right. I think it offers more creativity to people who want to tell stories, want to communicate and I think for a lot of people they prefer that. And if you don't prefer it, then that's also completely different.

Exactly. I mean you can have both. You can have the choice, as we said. So for anyone who's like probably played with avatars, knows that the demos are very pristine and then you try it yourself and suddenly your digital twin kind of looks like an NPC or you don't like it. If you had to pick one kind of thing that matters most for realism, what's the big lever?

I think a lot of them come to the body language. The earlier versions of these technologies, they had this bit strange body language and I think as humans, we're just so sensitive to even the slightest imperfections. Babies learn to recognize faces as the first thing. It's such an important way of being a human. It's like they were very sensitive to these kind of things, which is why when you go to a lot of Hollywood films, even though they spend millions of dollars in creating digital humans, it still feels off because it just doesn't look like the real thing. And AI kind of bridged that gap because instead of a human sitting down, defining what a human looks like, we can learn it from the data. But even the slight imperfections, and there are still a few left, they kind of stick out to people. And once people notice it once, it's very hard to not notice it, and then they have to increase the quality again. So I think it is a lot of these small things that gives people a sense of unease. But I think, to be honest, I think where we are right now, we're very, very close to have narrowed that gap.

Yeah, I think they canceled Lord of the Rings, the new one, because most of the actors are dead. In my opinion, they could always use AI to do that. But whatever. This is my opinion. So when someone goes like, it doesn't feel like me, what's your go to? Adjustment. When it comes to that, other than body language, what do you tell them? Kind of like a practical fix for that.

So I think the algorithm is getting a lot better, like reducing variety in how avatars come out. But you can always try and record it again. You can submit more data for the voice, for example. Like, if your voice doesn't sound right, that's usually a big thing. Actually. A lot of people have the problem where they like the visuals, but their voice would have the wrong accent. And that's actually sounds like a small thing, but it's a huge thing, right?

It is a huge thing.

If you live in the US for example, and you have a Midwestern accent in real life, but your avatar has like a New York accent, you're not going to want to use it, even though it can seem minor. So we actually trained a voice model, our own technology, which is specifically trained to preserve accents. And it works phenomenally well. It tests way better than any other providers out there. And that is one of those things that actually helps block the gap, because if it sounds like you and looks like you, it's going to be right. I don't have a British or American accent. I have a Danish accent because I grew up in Denmark. And so it's weird if I get a British accent. Right. It just sounds off. So it is a lot of those small nuances that really do the trick.

If mine doesn't know how to say hummus, I will not be happy about it. It's just FYI. Also, it supports 140 languages. It's more now.

Is it including all the accents and things? I think it's actually almost more than 200.

I speak three, but I still need to Google Translate to my mother because my mother speaks two. So accuracy is one thing, emotional stuff is a harder thing. How do you keep the feeling consistent across all languages other than the accents?

The accents are very important. One is how does it sound? Does it sound like you of. And another one is the actual translation. And I think actual translation has gotten a lot better in recent years. We don't develop that technology in house and partner with other people there. It has gotten really good. But most of the time you still want to look through to see if it's done the translation correctly. I think ours is better than Google Translate. But sometimes with Google Translate you can get sentences which sound grammatically, they're correct in whatever target language.

The nightmare.

But no one would ever say that in real life.

It's a nightmare.

Yeah, I don't think technology is there 100% solved for that yet. But actually using LLMs and like GPT for this sort of stuff to correct it is proven to be like very, very good because it just has a better sense of like what do people actually say? Not like what's grammatically correct.

Right, right. GPT. Sometimes whenever I try to speak with it in Lebanese, it defaults into Syrian. I don't know why I'm a Lebanese. I would like to hear Lebanese, you know, which is very funny to me. So let's talk about outcomes. You know, what are the top three use cases where avatars actually earn their keep? Where the team goes like this paid off.

Yeah, I mean it depends a lot from team to team. I think what we generally see is like the sort of first order effect is a lot about just reducing cost of creating video very simply. Right. If you're not using avatars and AI to create videos, it's a lot more expensive. And so what we generally see is that you can kind of begin to move budget away from external agencies or internal agencies and you can allow the actual team to create the content itself. So that is first and foremost cheaper, but it's also much, much, much faster if you always have to go through like a different department. Takes less time, doesn't it, it takes a lot of time, which means people create less content and that we see customers like Heineken, Zoom, who all have put numbers essentially like how much time do you actually save by using AI instead of using real cameras? Right, yeah, that's like one really great example. You also have customers like BSH which does a lot of training content. They've seen an increase in 30% of the engagement of that content, which when you're a company with many, many thousands of employees, that's a really, really key business metric because ultimately if you take half an hour out of someone's day to train them on something that's business critical, you want to make sure that they actually remember it. Right. They don't just sit through it and forget it because then they might as well spend that half an hour on doing something else that's productive. So it depends a lot on the different use cases, like how people measure it ultimately. But it is generally about cost, engagement and speed of creation.

I love that. So let's say we drop Synthesia in like a 100 person agency with like zero instructions. If they want fast wins, where should they start?

Yeah, one good way to start is looking at all different communications that you have as a company, right. And there's going to be a bunch of really important like marketing and brand moments. Like Maybe you're running TV ads, you're running YouTube ads, you're doing tuple whatever, right. That kind of content is probably not what you want to use these years there. It probably makes sense to invest in, you know, create a really, really awesome video with camera and actors, et cetera, where it becomes much more interesting. It's this kind of middle layer of communication, right. So this usually where a lot of like PowerPoint content would live. You know, you take a deck to a customer to explain about a new product feature. You have a deck probably when you onboard new people in your sales team for them to learn about like how to sell the product, how to send a pricing policy, whatever. And in this kind of layer where there's a lot of PowerPoints, that's where video can be a really, really powerful replacement. Right. Because videos can be consumed asynchronously, they have much higher information retention. People are generally much more engaged with video. And so what we see successful companies do is they actually focus more on replacing documents, wiki pages, PowerPoints with video than necessarily going all the way to your most important brand moments, which I think will get there eventually. But I think the technology needs a year or two more before it really is able to produce that level of quality. But for us as a company, we also think that this middle layer of the content pyramid is actually more interesting because there's so much more content there. Right. If you think about how many marketing videos or company creates every single year versus how many PowerPoints they create, the differential is absolutely gigantic. Right. And so generally it's like, where are we communicating with text or slides today where we want to be more efficient? That's where Synthesia comes in. It's like a really good.

If Synthesia was there in 2020, like 2021, I would have done my master's with it. I would have presented it without having to go with, like, people on and. Yeah, so what's the rookie mistake? Because there's silicon mistake that people make when it comes to that.

I think it's actually like trying to be overly ambitious of, like, let's create like a big advertising campaign with Synthesia. You know, that's just not what the technology really is like, developed for. And it's not really where it's the most impactful yet. So I think one of the things we always have internally is like a mantra called utility over novelty, which that. I think there's a lot of AI products, AI companies right now that promise a lot of things that don't actually materialize. And we try and be the opposite with our customers. We always try and be like, hey, the technology will get very, very good. It will be able to do all these things that you're dreaming about. But for where this industry and this technology is today, these are some of the use cases that make sense right now. And that is very much about communication, product marketing, customer support, videos, training videos. That's where the technology really works. And if you try and push it too much, it can become a bit too much of an innovation project or something that isn't really going to deliver like a real ROI back. And sometimes that's okay, you want to experiment, but we always try and help our customers towards, like, where do you actually get like real tangible ROI today?

Yeah, that makes sense. Yeah. So we're going to talk about the guardrails because people need to talk about that. So let's say political sensitivity and like sensitive content. It's very gated to enterprise and like custom avatars. Where do you draw the line? How do you draw the line? And what does safeguard actually mean here in practice?

So this has always been very important. Since we found the company in 2017, we have this ethical framework called the Free Cs Consent, Control and collaboration. Consent is just about not creating advertising people without their consent, which you would think is baseline for every platform offering this kind of technology. Unfortunately isn't. Control is about content moderation. So we take a strong stance. What you can and cannot create, I'm not going to lie, it's a really, really hairy thing to get right because there's a lot of opinions. What is one person's fake news, another person's truth? What is one person's scam, another person's risky investment and opportunity and so on. The way we've done it is we have three categories of content. We have the green content which is self explanatory. We have the red content. So that'd be like hate speech, violence, things like I think almost everyone would agree we don't want that. Or it gets much harder in this messy middle where it's like you're talking about politics, maybe you're talking about cryptocurrency investments or all these things which are highly contentious and everybody will probably have different opinions on where to draw the line. What we've done up until now is we've been kind of probably a little bit overly restrictive, you could say. So we've just been like erring on the side of let's not allow things that are potentially controversial because these technologies are still very new. I definitely think that's going to change over the years as these technologies become more part of our daily life. But right now, for example, we don't allow things like news content unless you're like an enterprise plan. So current events, politics. We're also pretty restrictive around even some legal categories like gambling investments. We also take that.

And you have someone that kind of oversees in terms of.

Yeah, it's kind of like a product internal integration. So we have both engineers working on it, of course, but we also have a team that essentially owns the policies. And when we detect content that looks like it's breaking our content policies, they'll be the ones who review, look at the customer, talk to the customer and figure out if it is within the content policy or not within the content policy.

Yeah, that makes sense. So I imagine you have to protect against misuse without strangling creativity. How do you hold both of those without going great already?

I don't think you can hold both. I think we are definitely giving up some level of creativity and also probably you could argue like some level of growth. We could potentially grow faster. Like on our self service business we allow more content in, but we Think it's the right decision. I also think long term it is also good business because when we speak to the enterprise, people want to work with companies that they trust and companies that kind of do well. Right. And I think that's something that is as immediate as like just opening up the floodgates and getting revenue in from people who maybe have more creative use case. But I think the long term, it's one of those things as a founder you choose to invest in and I feel good about it, both from a business perspective, but also from a moral perspective.

So I want to go through data and license data in general because everyone talks about it and everyone has concerns. So let's say because you license data from partners like Shutterstock, which is extremely rare for Genai, Honestly, in terms of the Genai world, from a product perspective, how does that show up better training signals, like more safety, performance gains?

I think it's just you want to make sure that in the training data set you have high quality data because higher quality input means higher quality output. And so it's great when partner with companies like Shutterstock, for example, get data in.

So with lawsuits like that storm in around OpenAI anthropic perplexity, do you think licensing becomes the standard path or is this more of like we're ready, everyone else will catch up later kind of situation?

It's hard to say. I think the world will convene on not regulating the inputs to the models, assuming it's public available data, but regulating the outputs of the models. Which means that if you're like one of the big LLM providers, for example, it's going to be less about what data you put into it and more the fact that it cannot spit out content which could be copyright. Right. So you can't make it create an image of like Mickey Mouse, for example, because that's copyright. You should not be able to create images or videos of like celebrities or other things that are in the dataset. We'll see how all of this nets out. But I think you look at the lawsuits and where they are currently, my kind of sense is that obviously it's not going to be okay to use data that's not available on the public Internet. But I think there was a ruling recently which at least put it in the direction of it's going to be deemed for the greater common good that you can use content that's available online, but you have to make sure that you regulate the outputs.

Right. I think you can't really right now also generate pictures of let's say political figures on the journey till this day. I tried to do it for the newsletter and it just banned me for a day, which hurt my feelings, but that's okay. So in terms of talent and future, with AI, you can just show what you're capable of instead of telling a recruiter about it. Do you think portfolios are quietly replacing resumes right now? Should I send an avatar of me in terms of that?

I think as an engineer, like an AI researcher is obviously different because you'll have a publishing history of papers and so on that you've published before, which is going to be a very important part of getting a job anywhere. But I don't think actually that much change. I still think your resume matters. I still think that you probably don't want to send an avatar to your job interview just yet. Maybe at some point. But as to my comments earlier on, you have to use the technology where it actually makes sense and not where it doesn't make sense. I think in an interview, right, you want to show up, you want to show the real authentic view that you're excited about the job. Maybe in the future it'll be two avatars interviewing each other and you'll just get a notification when your avatar, I've got a job for you. But I think we're still.

That's so helpful, honestly, rather than go through the interview process over and over. So where do avatars fit into in terms of. I've seen my fair share of people that go through online and they have to go through the video thing and they stop sending their resumes. Now it's just strictly videos. Do you think people are going to use that specifically?

Yeah. So with our own agent product that's going to run a capability you're going to be able to do, right? Like give someone a case study, live on a call, for example, which instead of, as you said, submitting a portfolio is actually also a much more fair way of interviewing people. Because I think what happens a lot today is that companies will open a job, you'll get 10,000 applications in, like, I don't know, two days. And you cannot realistically sift through all of those different people. Right. And so the way that the TA teams have to do it is they have to be very quick. And unfortunately, the best way of being very quick is like looking at purely the CV of that person.

Right? Yeah.

And so I think what this will enable is actually make it more fair because it's going to be more like, let's give everyone the Chance if they want to to talk to an agent for 10 minutes and go through a case study and then you can look at that as a signal in relation to the cv. Right. So I think it's going to be great for candidates. I think it's going to be great for companies who will also get a bigger talent pool of more diverse talent that doesn't necessarily have the top five top logos and they just purely select based off that.

Yeah.

So basically talent based but through the videos. So I want to move into the future like three years from now. What's the moment you're building towards that most people don't see it coming right now.

I think it is very much about evolving video into something that's different and it is going to be this idea that you can kind of talk to a video. Right. And it's going to be able to take on so many different roles. It's going to be the thing that could be your tutor if you're trying to learn something. If it's a kid, maybe it's like a teacher.

Parents will have less anger issues. That would be so cool.

They probably will. The kids will learn faster because they'll be able to ask questions. And if your kid loves soccer then they can explain math problems. In terms of soccer based personalization. It's going to be all this stuff that's going to be such a different experience to what it is today. I think a lot of people right now we're all having that experience with ChatGPT where it becomes a very, very different way of learning things. And I mean searching the Internet in some sense where most people today, I think if you have something that's not trying to find that rest of the restaurant or something, a lot of people today, including myself, will ask ChatGPT and then you will ask follow up questions and you'll ask another follow up question. You're having almost more like a conversation with your computer. Right. It's very different than in the past, which is not really that long ago where you would like Google something, you would go to a website, you would try and read what is on the website. If it wasn't what you're looking for, you go back and click another website. That's kind of how we watch videos today, right. Like you search for something, you watch a video, you hope it's the right thing and if it's not, you go back and find another one. But this is going to be much more like multiple to you. Right. So you start watching your video like you start a conversation with ChatGPT and then it's going to be an experience that goes like that.

I'm a forwarder here, so I've never really used the tube before. That was my first time using the tube. When I got to London, I had to watch a walkthrough of the how to use the tube. And that's how I learned honestly, I would never use these apps ever again. So you're building Synthesia for eight years. If you wanted to start today with the same brains, with new blank page, what other AI frontier would you chase?

I like weird fringe ideas, which Synthesia certainly was back when we started the company. I think on a personal level I find it very interesting. So I think I would look more at what's not consensus right now. I don't think you'll find a single person in the world right now more or less who doesn't think AI is going to be the future and AI is going to change everything that we do. It doesn't mean there's not lots of opportunities there. I just think for me personally, I would probably look more towards. I think there's a lot of interesting happens in bio, for example, and things like. I think nuclear is a very interesting area right now. The whole energy, not crisis, but the energy needs of the future is going to be insane. And I think there we're going to need a lot of new solutions, both hardware and software. So I think a lot of the opportunities is more things that are less consensus today.

I want an automated household, for example. I don't have that yet and I need to have someone make me.

I'm sure you'll have an AI housekeeper.

Soon, that would be so cool. I want to make the dishes anymore. So if you could teach one skill that genuinely future proofs people into the AI world, what would it be?

I think critical thinking, being able to read something and you know, not just accept it for what it is, but try and critique it, try and find other sources that think something different or that confirms it. I think increasingly that's going to be like the most, one of the most valuable human skills. That and creativity. I think what's clearly happening right now is that the kind of, a lot of the hard skills, like creating an app, for example, right, like software development in five years time is going to be something that most people are going to be able to do, kind of like by just chatting things. You don't have to actually be able to write code. But the problem is you still have to figure out what to build And I think people often forget that a really hard part like building a successful app or company or product is not just actually making it. It's coming up with the right idea for what is worth building. I think that comes back to being able to think creatively and figure out what is a good solution to some problems out there in the world.

So looking ahead, what's out there for Synthesia?

Well, there's so much to go still. The big thing for us in this short term future is putting out agents. It's also going to be a lot about helping people, the creative process. So today, making a Synthesia video is kind of like making a PowerPoint presentation. We have some agents that can help you write the script, put the visuals together. But what we eventually want Synthesia to be is that it's kind of like working with a really, really good video editor that can just do whatever you want so that you can create really high quality, great, informative content in just like a few minutes. And there's still a lot of things to do there in terms of us helping you with agents on doing that. And then there's like just the better models, you know, like what we're seeing now is Express 2 is a huge step up, but there's going to be levels after that as well, to the point where creating something that looks like a Spielberg film, even for like a simple training video, it's going to be very possible for most people. Right. I think we're getting very close to that. It's very exciting.

Yeah. Thank you so much. It was awesome. So thanks for having us.

Thank you.