20VC: Why AI Models are not a Moat, Where Does the Value in AI Accrue; Startups or Incumbents, What the World Has Got Wrong About AI, Why AI Needs a New Story and Who is the Right People to Tell it with Cris Valenzuela, Co-Founder & CEO @ Runway

0:00AI is not just chatbots. I think models are not a mode, models eventually they'll murder. I have so many investor emails so people telling me, love your team, etc. but journey via is not a thing. Stop saying it's a thing it's not a thing. Good investors are the ones that are not looking for patterns. This is 20vc with me Harry Stebnings and this we were doing an incredible feature week on the two hottest AI companies, Runway Today and character .ai on Thursday. So to our show today and I'm thrilled to welcome Chris Valenzuela, CEO and co -founder of Runway, the company that trains and builds generative AI models for content creation.

0:35To date, Chris has raised over $285 million for the company from the likes of Lux Capital, Felicis, Cotoo, and Plafian in video to name a few. Runway's customers include Academy -nominated movies, TV shows, media companies, and creatives across industries. But before we dive into the show's Today this episode is brought to you by TIGAS, the go -to research destination for bold investing. TIGAS curates, expert in -size analysis and financial data to give you powerful perspective for your investment decisions. With lightning fast access to over 60 ,000 transcripts across 20 ,000 companies, you'll discover a wealth of unique insights to fuel your fundamental research, gain perspectives, synthesize information, and even model outcomes all on TIGAS.

1:20As a 20 minute BC listener, try for free at TIGAS .com, forward slash growth, and spare you tools we can't live without like TIGAS there, Mayfaer helps companies protect and grow their cash automatically. I access up to 50 million dollars of FDIC insurance and earn up to 4 .7 % interest on your deposits from a single account. After the failure of SVB and First Republic is clearly the standard FDIC limited $250 ,000 is simply not enough for VC -Bass startups. So for founders, that means finding a safe home for cash is now a priority and a major burden. Mayfair can take care of that few in under 10 minutes, through his partner banks and sweet providers, Mayfair automatically spreads customer deposits across a network of over 200 banks.

2:05Therefore, multiplying the level of FDIC insurance coverage and limiting the risk of any single point of failure. Think for it like a high yield savings account, it actually pays a high yield and is actually safe. Go to getmayfair .com for what's slash 20 VC to get started today. And finally, Secure Frame is the leading all -in -one platform for automated security and privacy compliance. Secure Frame simplifies and streamlines the process of getting and staying compliant to the most rigorous global privacy and security standards, like SOC 2, KIPER, GDPR and many more. Secure Frame's industry leading compliance automation platform paired with their in -house compliance experts and former auditors helps you get audit ready in weeks, not months, so you can close more deals faster and secure from users over 150 integrations.

2:52So you have built -in security training, vendor and risk management and more to make compliance uncomplicated. Secure from MACE it fast and easy to achieve and maintain compliance so you can focus on serving your customers. to make your security and privacy compliance with Secure Frame and you can schedule a demo today at SecureFrame .com. You are now arrived at your destination. Chris, I am so excited for this. I had so many great things from so many different people on your cap table. So thank you so much for putting up with my terrible British tones today. Thank you so much for inviting me, a huge fan of the podcast.

3:26Happy to be here. That is very, very kind of you. Now listen, I spoke to Sunil Lamplify before we started and he said to me that you and runway is the perfect embodiment of founder product fit. Tame me to the founding moment of the company and that aha moment for you. Why does he say that? So LillistuKind, runway, journey dates back to 2015 -2016 where I met my school founders Alejandro Anastasis in actually art school. The best way to think about it is art school for engineers and engineering school for artists. And while we were there, We were all coming from old schools with your backgrounds and experiences from programming, engineering, business and we started just tinkering and playing around with state -of -the -art AI at that time.

4:07I know a lot of things that happened now, but tried to imagine where models were and where research was 10 years ago or 80 years ago, and we just built a lot of different things. And so we started tinkering with this idea of taking AI models or research at the time and building creative tools. And so we were just building experiments and experiments over experiments and realizing that there was something here something special And then we started delving into Imagine video generation and I think we've realized at some point that after building so many things There was something special around both where things are we're heading and how much passion we had for building this thing And so we didn't found the company.

4:44I think the company founded us The interesting thing I spoke to many of your arrests about was also your background you're not being in Silicon Valley and you meant an artistry being at the center of both your education and also how you think today. They called you an outsider. I'm sure in the loveliest way, do you feel like an outsider and how do you think being an outsider helps you be better? I always feel a little bit like an outsider to be honest. I have a background in the economy business but what I was studying that I was doing film on this side. So that was strange and I felt like an outsider in both of those worlds.

5:17Then I was doing art but also learning programming so also fell a bit of outsider in both of those and then it came to the US and started our company while being in art school that also felt like a bit of an outsider. It's been a common thread of my professional life. I think being an outsider is both a blessing and a curse in the sense that it allows you to really reason for first principles, you discover everything and ask yourself why a lot. At the same time you take longer than other people because you're just experiencing for the first time. I think the key aspect of being an outsider is that creativity really sparts from being able to merge different domains and different languages and different professions in a coherent thing.

5:54And so, runway, I think it's the manifestation of that. It's a company that merges our, then, science by being able to be in between. And so, instead of being an outsider, you create your own world that's much more powerful, I would say. In terms of creating your own world, you do that when you kind of grow up in the world as an adult. I always think that we're all running from something. And as I'll pass shape us in many ways. When you think about what you're running from, what would you say you're running from Chris? That's a good question. I grab you Chile. Culturally, in a country, it's a very traditional structure, strict tradition -based country.

6:27So things are very rigid and it's very hard to do anything that goes beyond the norm. And really, I don't feel particularly passionate about people telling me what to do and how should I follow or what are the norms. I'm running away a little bit from that tradition and trying to figure out more of the things I'm passionate about, the things I'm curious about. And that was for me like really going to art school. It was it was a curiosity driven approach of just I want to understand how much AI I can take and put in the art world. And that's what I did. That's what I'm really passionate about. Would you say you're fair for the mediocrity?

6:59I think I'm I'm fearful of people not pushing themselves both intellectually and creatively and maintaining a status quo. That's where companies go to die and where creatives go to not really progress is when you feel you're too comfortable. I get a little bit anxious when I feel like that. David Goggins says something brilliant which is there's nothing more dangerous than the civilized person. But I do want to, on that, you know, it takes me to performance a little bit. As a leader today, how do you define high performance then? The first one for me is being a vision and a motto and a value that we've embedded in the company since the very beginning.

7:35This idea of getting things done. Just figure it out. It's become so much of a thing that run with that and make a t -shirt that's just figure it out. But if you learn how to learn, you can figure out anything. And that's like a superpower. Unlocking the ability to do things without complaint, without understanding or without necessarily contextualizing things in the sense that you haven't done it before is really a superpower. And for us, at Runway it was. I've never started a company. I have to figure out how to stay in the US, to start the company. There were so many hurdles and challenges at the very beginning that we just had to figure out stuff.

8:06If you learn how to learn, there's nothing that you can't do. I love that. Can you help me understand how do you approach learning? Do you have a process? Do you have a preference for the types of ways to learn? How do you approach it? Very hands -on. There's something I want to do. For example, when I was diving into neural networks and really understanding the field, I was coming it from a beginner's mindset. And so really what I did before trying anything was I built from the ground up on neural network. Every single layer, every single function, activation function in between, I've curated the data set, I train the model, and it built everything that I could get me a sense of how they're in your parts of a system work, in this case, in your notebooks.

8:49And so for a long time, I didn't know what I was doing. To just try things, you try, you try your heatwalls, but you keep pushing yourself over and over and over and over, go and read as many things as you can, and then find a project that you like, and then go as deep as you can on that. Being hands -on and being active on learning by doing something really helps make sense of the world and that's something I've been trying to do in everything I've tried to learn since then. How do you not get dejected when the learning is slow or not happening? Fine, brilliant people. Fine people that inspire you and can help you along the way.

9:20I think that that for me has been critical. So if you feel on stock, there's things that you can really come across. It's too difficult. Just ask someone that it's going to help you. What has been the most recent subject that you chose to really engage in a learning process begin as mindset on? I mean, everything to be honest, from fundraising to building a company to hiring top talent, everything has been a journey of learning. And I think we've really not been afraid to be in Paris or mistaken because that's usually what you're going to do it. More recently, we've learned a lot with what we're doing with Gen 1 and Gen 2, which are our two latest video generation models.

9:55And there's so many assumptions that you might think about how these models actually work. But the moment you put them out, most of the assumptions are going to be changed. And so we've learned a lot and learned a lot about how people are using and how creative are using some of the research we're putting out. What is Samsung has changed the most? From Gem1's Gen2, we assumed X and it was totally wrong. I think there's a lot. One that I've been thinking a lot about is, you eyes don't matter as much. A first approach to creative software comes from this idea that you have to be extremely thoughtful and extremely diligent on how everything is going to look and how everything is going to work.

10:31But true is since the field is moving so fast the assumptions that you have are run how models are going to be used in a creative Context might change rapidly And so if you build a lot of assumptions and preconceptions are unreally helped to get the best out of these models you might be wrong really soon And you just need to throw it out and sorry again And so keeping the value at the center unless of what you think people might get out of those models is really really critical How do you do that if you don't project out what they can get? It's kind of like a free -frual. It's almost like product marketing to a horizontal audience So if you don't quite the use that how do you give a great UX?

11:08You toss a lot like if you think about it Think people do when they come across this idea of video generation is they assume that everything they know about film and cinema and film We replicate itself to this new medium and the thing is this is a new medium. It's a new tool It's a new set of narrative possibilities And so it's very different from everything we're using the past. The most challenging thing is uncovering those things. Because we haven't had the chance of doing it in the past. It's like you find yourself in front of a camera, the first camera ever. And everything you'll be using in the past are paintbrushes and paint tubes.

11:40A camera doesn't work in the same way as a paintbrush. And the things that you can do with it are only going to be uncover and tap the moment where people get their hands on it. That's how cinema was born. You have so many people trying this device and coming up with new forms that are beyond your comprehension or beyond what the creators imagine. I think that's impartial, though, what we do. We need to build this camera, we need to put this device into the hands of more people so we can collectively as an artistic community figure out all the primitives and UIs and others that will emerge from it.

12:10How is Ui? I think straight away still in field at Figma with this. And what I mean by that is that they're really heavy buildouts technologically and you chose to kind of build in public, where bluntly you know 4 seconds is short, but is better to be out in the wild and how people use it. Dylan almost took the other side which really didn't release for 2 -3 years, until it was much more perfect. Why did you decide to release knowing that maybe it was challenging for people to use in full? In our case, this is something new. The easier that you can generate video, create video, would nothing more than work for example, as it ever been tested out before.

12:48So you can for years build assumptions on how people are going to use it and the value and opportunities on the market. But the truth is that you might be wrong and really wrong. And if you build with that assumption for too long, you're going to miss really getting feedback and really understanding and learning from use cases and from people who are going to be using it. In other cases where it domains and perhaps the values that be more defined, you can spend a bit more time and building in private and then release. It was pretty much the realization that the only way to move forward the field and the intersection of creativity and artificial intelligence was we build by as close as we can with the community that we're building towards which is artists and filmmakers and we're going to gain a lot more from being there than being too silent or too private about what we're doing.

13:34You mentioned this barimentation being key to the work of those of artists and It's so many a scar that AI will replace. You were seeing the strikes. You always say that it's an enabler. Has the media got AI wrong in terms of replacement, not enabler? And is enabler not just an intermediary step to replacement? You know, I think a lot about this. I think one that won in language models have dominated the discourse, the public discourse. And the whole field of AI has been reduced to this idea that AI is just language models. part of it has been due to the success of course of Chad Gbt, which has been on the minds of everyone.

14:11And AI is not just chatbots. I think the reductive view of thinking about what AI is bringing to the table and how revolutionary and how much change is going to do is just not an interface where you can chat with a system. And so a lot of the discourse I feel sometimes comes from the place of extrapolating the current state of language models and the challenges they have, which are true to other domains, which divert and are very different from just building chatbots. And so in the one end, I think the core realization for me and really what I mean trying to emphasize more is that we need better stories.

14:42We need better narratives. We only hear one story these days, which is horror stories. And I think part of it has been heavily influenced by fiction, by movies we watch, by books we read. But that's one story, that's one way of seeing the world. There's other stories and other narratives that can help us really on tab another set of possibilities. And so we'd run away, we're entapping those possibilities of, hey, not everything is a language model. It's more and more complex than that. Why does screenwriters like strike if it's an enabler? What do they not see? A lot of the discours around our replacements comes from this rather unsophistic view of the world that you're going to type in something, just write me a script or write me a movie, click enter, and you get entirely the exact same script that you did a month ago.

15:29And for me, that's come from a place of me, maybe we would never use these tools. Like writing a book and writing a script is way more than just pressing enter. It's a process. I go back to experimentation. It's about feedback and it's about people and it's about using these tools in the context of a macro project. And so I think I would invite more of those more people who are very skeptical of these tools about to use them and to experiment with them and see where they are and see how they provide value to your creative process and understand that this is not a zero sum game, which I think a lot of people think about it, in that way, which is you're going to type in, get me a movie, click, enter, and you're going to get a movie out of it, that perfect movie exactly the way we've been making movies for the last 100 years.

16:06And of course, there's nothing weird like that. It's much more, it's much more complicated than you've also a lot of feedback processes within. Can I have a more stuck here? What's more important? Data size or model size? And how do you think about the importance of size of model? I think size of model matters in the sense that we've seen that larger models, parameter -wise are going to get better at doing more things in multimodalities. But at the same time, it depends. It depends where you're trying to do. And there's opportunities to get models to be smaller, but has more specific to the type of thing that are trying to accomplish.

16:41And we're already seeing approaches like this in the language domain. We'll start to see this as well in the image and video domain. You think we'll see the verticalization of models, or like Richard Sotcher with you .com very much pronounces the importance of a single model to rule them all, which is much more horizontal. Yeah, I don't think there's going to be a single model to rule them all. That's like saying that the internet would only have one e -commerce site. There's many. You have something like Shopify where everyone can build their own website and their own e -commerce site and you can sell anything.

17:09If you think there's just a tool and it's a general purpose tool, and I say tool AI in general, there's so many opportunities to build different type of models and different ways of working with those models. and it's still very early to be so specific to say, oh, we're only gonna use that thing or that other thing. How do you think about model lifespan? Do you have to continue to see update models? How do you think about that? I've found myself hearing a lot about models as the mode and mode has been something that Silicon Valley has been discussing for some months now. I think models are not a mode.

17:42Models eventually don't matter. What matters most is the people building those models and how fast can you change your learn from those models. So I don't think that's why I go back. There's no one similar model that's going to rule them all. What impacts how fast you can learn and change? Getting to a hands of people and understanding how it works, whether their values that they're getting out of it, pushing the boundaries of both training, research, and figuring out improvements. Four or five years ago, state of the art and language was totally different from where we're right now. Same with images, the basic assumptions of how image duration worked, Just a couple of years ago was completely different from where we are today.

18:19So if you find yourself trying to optimize an engineer's love tool, optimize things, that was a wrong thing to optimize for. And so Vivaire humble, be very clear that whatever you have right now, today might change and just chip it as fast as you can, learn as much as you can and continue to read it from there. Can I ask you, you know, when we think about creativity, someone wants to describe hallucinations to me as students who kind of are brilliant, but go off their meds, which is a form of creativity in some respects. How do you think about hallucinations as a feature or a bug, especially given the artistry and creativity that's inherent in what runway is?

18:55This is a great example of being a bit renewing because hallucinations in language models are different from perhaps what we might understand as hallucinations in video models. If you're asking for a fact, if you're asking a language model to give you the capitol of Chile, and the model gets it wrong and starts rumbling and hallucinating other things. I wanted a fact and you gave me a creative interpretation of the answer, but in a video domain, in an image domain, in a created context, maybe you want to have the model rumble a little bit or go off the chart and maybe you can control the temperature to see where you want to go, because again, it's a creative process.

19:29And so it depends. I think in some cases, it's actually a future and it will actually help you uncover more. and in some other cases you can afford to have a hallucination if you're getting a unique facts. On the artistry and the artisan element, I had Jan McEun on the show and he spoke about how unwaveringly this will be an open ecosystem and the benefits that will come from that. With artisan, developers, creators, do you agree with him on the open -beating clothes model and how do you think about that? I think the open is always better in the sense that allows you to get to a broader audience, faster and you'll also get different perspectives because people can build on top of models and we've seen this I think again, it's more than once once you start defining where you want to build products for people and so open -serves versus Closers is only one part of that stack.

20:17You're not gonna build an entire company just by being open -serves and closer. There's way more than that and so it really depends. Some companies I think are gonna go into open -serves more and more because it helps their value. Prop some others are gonna go closer. If you want to build other types of products I think we're always striving to be in between, or really open, or some of the most relevant models in this space that have been game changers, and at the same time we have more proprietary models as well than we might never open. Chris, what is biggest rate -limiting factors to the runway product today?

20:45If I had a magic wand for your product team, for your tech team, and could take away a problem, what would it be? I think speed, I mean, and I say this because we're moving very fast, but speed really matters. The rate limit I would say sometimes for us is being able to move even faster. And those things come in place with sometimes compute. There's a limitation of compute that the industry has at the moment. And so sometimes you're constrained because of compute, sometimes you're constrained about use cases or deploying those models and making them efficient. I think it depends. But speed is a constant thing that you're thinking about when deploying models and training models is for us really critical.

21:21So many on your cap table told me about the performance element of your small team. What have been your biggest lessons on hiring a performance first, speed first, but small team? Focus on again and we're discussing before people that can figure things out and get things done. I think a lot of the mistakes that have done in the past on hiring is a we focus to watch on credentials or past experiences or people who came from all these lofty universities or companies and not on the folks were hungry and who just wanted to do things. How do you test for that in the interviews? We've been learning a lot, we've learned a lot.

21:56We have an interview process now that I think is one of the best, of course, on biases, but it measures your productivity and your level of, of course, there's a baseline level for every position and domain, but at the same time, it measures you to get things done. And there's a bunch of experiments and exercise we run and interviews, and then we'll get together the whole folks, the team that was interviewing, and try to make sure that the values of what makes RunWrg especially are there in that person. When you look at your interview process today, what would you like to know as a founder listening that you could share with them?

22:29What can someone learn from your interview process? Doing is more than talking. There are a lot of great people who are great at interviewing and then great at saying what you wanna hear. I've been cautious about that now and really understanding, okay, I get it, but can you do this? Can you show me how to do it? can do it right now. And so focus on actions more than words. What would you say your biggest mistakes in hiring other than credentials? Credentials and people that are not, I think, humbleness. It's a key component of a company. The rate of re -earning as a company, building a company is about just pushing every single day.

23:04You want to quit. You really want to quit every single day. You need to push it again and again and again. And it's a mad process. You need to have a lot of tenacity to be able to do it every time. But at the same time recognize that some things you're gonna do are gonna be wrong And it's fine you might spend months working on something and the next day you're gonna throw it to the trash And it's fine. Don't worry about it. Have a very stoic approach to it and continue moving on So people who have very definitive worlds views of their world perhaps What I was running away from before you asked me don't really match and work in runway people that feel like that's the way That's the only way I'm not gonna be able to learn as much on a field that's ever changing and so humbleness for us has been a core value, look for people who are asking questions, who are ignoring and have always sort of embracing change, which I feel is the only constant mistakes.

23:52I find people who are boring and capable. That's my two things, and you know why? Because people are never as boring as they come across in an interview process. They're always more fun. You would obviously almost boring self, but boring people are rarely arrogant. Dude, what do you think are others who might like complete start up BSs? You're outside of the ecosystem. What do you think is another like crap advice that you see given about building companies now? I'm very cautious about recipes and rules and algorithms in the sense of company building. I think there's no rules. There's no algorithm.