Ep 38: Runway CEO Cristobal Valenzuela on the Next Frontiers for AI Media and The Role of Human Taste in AI Filmmaking

Cris is the co-founder and CEO of Runway, which builds breathtakingly real video AI tools, including the incredible Gen-3 Alpha foundation model. Cris sits right at the intersection of technology and creativity, and in 2023 was named to TIME’s 100 Most Influential People in AI. Runway is reported to be in talks to raise capital at a $4 Billion valuation.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Jacob Effron?

  Runway is at the forefront of AI in media, a tool used by folks in Hollywood, enterprises, and creators around the world. Today, we had Chris Valenzuela, the CEO of Runway, on the podcast to talk about this transformative new technology. Runway as a company has raised over hundreds of millions of dollars and is most recently in discussions to raise a devaluation of $4 billion. And Chris hit on a lot of really interesting topics in our conversation today, including the future of AI in media, how Runway structures its research team to allow for creative exploration and close collaboration with artists, and how Runway thinks about product, given how fast models are moving. I think folks are really going to enjoy this episode. Chris, thanks so much for coming on.

• 0:36Speaker 01

  Yeah, thank you for having me.

• 0:37Jacob Effron?

  So I was trying to think of where to start, and I feel like one of the things I always hear you talk about when you talk about Runway, and you actually just said to me before we started recording, is that we're early. And there's a lot to come in this space. I'm wondering if you kind of reflect on where we are today, how would you describe kind of where we are in like the AI for creative tools space? And what can we do? What can't we do? How has that changed in the last year, the last six months?

• 1:00Speaker 01

  I do think we're early in the sense that most of the things that you're able to do today are just like the beginning of what we'll start to see over the next couple of months. I think Gen 3 Alpha was one of those moments where you can realize realism and control and fidelity has largely been improved, but there's still a long way to go. Better control tools, more fidelity, longer generations, just a world like better consistency of models. And it feels like there's something that has been unlocked with scale that has proven to be really useful. But I think I'm excited because we've just at the early stages of unlocking what's to come.

• 1:41Jacob Effron?

  Are there specific milestones that you think about? Like, wow, it'll be amazing when I get my hands on the first model that can do X or Y?

• 1:47Speaker 01

  I mean, real time is one that I feel it's pretty interesting that will come very soon. And also, like, more customization when it comes to the type of controls that you have for when you're generating, for specific styles or art directions that you want to like go. With and then the other one that I think it's gonna be interesting that I'm excited about is just multi model controls, so creating sequences of medias, videos using like audio inputs or more than just text or images.

• 2:18Jacob Effron?

  Yeah, I think even with Gen 3 Alpha, you guys are doing kind of fine tuning with specific art styles, right?

• 2:23Speaker 01

  Yes, yeah, yeah. So it's a it's a very flexible model that allows you to use particular datas or inputs or like art directions, even like mood boards.

• 2:34Jacob Effron?

  Yeah.

• 2:34Speaker 01

  And then steer the model in a much more like stylistic, consisting way to whatever you're working on.

• 2:39Jacob Effron?

  What are the people that do that best do? Like, what have you seen? What's like the best way to steer the the model stylistically?

• 2:44Speaker 01

  Uh, taste. You need to have good taste. I mean, but that's that's that's the, to be honest, like true in any creative art form, form of expression, just having good taste and good ideas. And so knowing how to pick those ideas, evolve them, and then fit the right data to the model.

• 3:01Jacob Effron?

  I'm curious, because you obviously spent a lot of time at Runway testing new models, playing around with them. Do you have like a go to prompt you use? Or how do you, when you get your hands on a new model, how do you figure out like what it can and can't do?

• 3:11Speaker 01

  You know, they've there's um, I've learned over time that the best way to use the models is come much less with very specific, concrete, like, determined ideas and much more with a willingness to like experiment and explore and see this as a as a system that can aid you in exercising a part of your brain and uncovering new ideas. And so sometimes I come with like some concepts, like I want to explore particular characters or particular camera angles or a particular idea around something I want to tell. And then as you start iterating with the model, you start realizing perhaps new things, things you you didn't thought of before, combinations of things, a model like produces interesting results.

• 3:52Speaker 01

  I think the the part that adds to that is the speed. The model is pretty quick and fast. And so you've um, it takes a few seconds to generate, and so you start visualizing things quicker than ever before. And that allows you to explore new ideas. So the best things I've gotten and the best results I've seen around are people who come with that like exploration. Experimentation mentality and then just like set themselves to like be surprised.

• 4:19Jacob Effron?

  Is there an example of a creative journey like you went on with the models that resulted in something fun?

• 4:23Speaker 01

  Yeah, I've I've created this very interesting like what I think I label it as a B come. It's like a camera.

• 4:31Jacob Effron?

  I love that one.

• 4:32Speaker 01

  You haven't seen it's a it's a like a first person view camera attached to a bee. And you can see the bee like flying through like like grass and like different like landscapes. I didn't thought about that when I was prompting. Actually, I think the original idea was more of I wanna I wanna start visualizing different insects living in different locations. And then the model gave me a very interesting camera angle that I never thought could be like interesting to like or possible. And then I ended up with this B come that I've just iterated on like fifty times or so into different locations. And it's just fun fascinating because it's a kind of like it, I think encompasses or summarizes a lot of what these models can do, which is

• 5:16Speaker 01

  create versions and ideas or concepts that are either extremely hard to make before or have never been seen before. And that I think we're seeing a lot in the aesthetics and the kind of like creative outputs that are come from this model. So things you have never seen before.

• 5:31Jacob Effron?

  I love that idea of brainstorming with the model that like you didn't come in saying, hey, I want a first party view from a from a insect. But like, as you were iterating, you're like, oh, that's an interesting output from the model. That's something that might actually be fun to explore further.

• 5:42Speaker 01

  Yeah, yeah, if the way I've been thinking about it, because I've we do a lot of internal testing with artists and creatives. And and you just get to see them like have fun with this and spending countless hours just like in flow state, creating things. And it feels like they're exercising parts of the brain that they never thought they could exercise. They weren't used to exercising. And this comes for both new like up and coming creators and artists who are using this and like traditional and like long time and experienced folks as well. Where they're heading into runway and trying new things and iterating really quickly on and then just having fun. And it's kind of like, I think I've. I thought about it as like you're it's like going to to the gym for the mind.

• 6:27Speaker 01

  You're you're exercising something not for the sake of becoming a like an Olympic athlete or becoming like an award winning like filmmaker and like you wanna you wanna win a like awards. It's more of like, oh, it's just fun. You be end the session like invigorated and you're happy about it and you wanna do it again. And I feel like that combination of like reactions is a really interesting one that we're still exploring.

• 6:52Jacob Effron?

  There's something deeply human about expressing your creativity and you know, I think to to date, there's, you know, a lot of us are told in school, only a small subset of people are the right people to do that. And

• 7:00Speaker 01

  Yeah, I think that that that it's kind of becoming more obvious. I think that's the role of any tool to like tap into people's like potentials, specifically when it comes to like artistic and creative expressions. I think there we need to detach artistic expression with like the craft and the tools. I think those for me are different things. I think creative creativity is a state of mind. It's a way of you looking at the world. You can be creative in anything you do. You can be creative in playing soccer like Messi is a highly creative player. And you can tell it's different. You have creatives working in all sorts of different disciplines. But I think more common than than than not, we tend to associate creativity to arts.

• 7:44Speaker 01

  And those two things come together. And so art is synonymous with creativity and creativity means art. And I think it's not really like a lot of what we're seeing right now is proving that point where you might be creative, you might wanna find ways of testing ideas, but not for the sake of creating a beautiful piece of art that will make you an artist, just for the sake of your own like creative expression, you know? It's like you don't go to the gym again for becoming an athlete. You just go for your own sake. No, you don't care what other people think about it. You just do it for yourself.

• 8:15Jacob Effron?

  Totally.

• 8:16Speaker 01

  And I think there's something there that these models are unlocking that feels exactly like that.

• 8:21Jacob Effron?

  How do you think about like the different user segments of runway? Like internally, as you think about the different types of folks that use it? How do you guys break that down?

• 8:27Speaker 01

  Um. It's a it's a wide spectrum of people. Uh. Again, it's it's for the most part these days, uh, professionals who work in creative industries. So think about studios, uh, production teams, and then inside those teams, of course, filmmakers, art directors, editors. Um. And then on the other spectrum, you have just like casual creators, like where people sitting on their bedrooms and creating like amazing, amazing content. And then everything in between those those two, the filmmakers with like years of experience and the new and upcoming creators, younger ones, new generations. I think we're pretty much opening the doors for all of them. I don't think we've settled like, oh, we only want to make this or that. I think the reason for that is that the models can be flexible enough to serve many of those like use cases.

• 9:12Jacob Effron?

  Do you feel a tension in building product for like the really sophisticated, you know, everything everywhere all at once, you know, uh, filmmakers and then kind of like the more novice folks that are just, I'm not much of an artist, but I'm messing around with this or or how do you think about that internally?

• 9:23Speaker 01

  Um, not really. I think I think the needs are the same, really, in in a way. It's like you you wanna tell a story. There's something in your in your that's that's pushing you to like say something. And sure, there's different requirements in terms of like media formats and uh, quality of outputs and consistency of subjects. Um, but I think there's still spaces to explore avenues where you can serve both. Or you can serve a wider spectrum of people. I do think long long term, um, you're just gonna start seeing a new type of professionals that we have never seen before. I mean, even if you think about

• 10:02Speaker 01

  the history of filmmaking, like the history of like visual effects and CGI, most of the people working in those roles um, were really like unthinkable forty years ago. Like those professions and those uh, economies and those teams and those, the things those those people are making are relatively new. And their professions just didn't exist it. And I think you're gonna start seeing similar versions of that with with with what's enabling by models like Gen three. Where professionals will start doing things that you couldn't classify in any. Category of like market or user persona before.

• 10:37Jacob Effron?

  We were talking before about the blank canvas problem and I think you were saying that basically, look, if you come to runway for the first time and all you want to do is write one prompt and be done, you're probably not going to end up being, you know, super satisfied and that ultimately the best process is really iterative. I think you guys have been really thoughtful about how you teach new users and solve that. I think you have runway academy and obviously a bunch of other things like, what have you learned over the last, you know, nine years of how you get kind of people that maybe don't get it right away to learn how to use these types of tools?

• 11:03Speaker 01

  Yeah, that's a great question. I think one of the things, key things we've learned is, ideas and taste matter more than anything else, like having great ideas and knowing how to convey those ideas is really like most of the journey and being able to experiment with those ideas and explore them and then having a sense of how to use whatever tools you have available to

• 11:30Speaker 01

  mold that idea in the way that you want. And so I think the common misconception around AI and I think part of it specifically when it comes to video and media generation comes mostly influenced by how the world has used AI these days, which is, I think for the large part, like using chatbots and like conversational interfaces. In a conversational interface, you type in something and you expect an answer that's correct or a summary that basically solves your question, right? Or like your query or whatever is. And so that's the world state and the state of mind that most people when they're approaching like video generation come with.

• 12:13Speaker 01

  And what they expect is you type in a prompt and you get exactly, exactly like word to word what you had in your head. And if it doesn't work, then the model is wrong and it's not, it's not making me like an award-winning filmmaker. And I think the point for that is like, it's probably never going to be like that, right? You're never going to type in movie enter and you get this phenomenal two hour long movie with everything you thought of. It's a process. You have to go iterate with the tool, prompt a few times, see what you come with, like if you like it or not, do it again, do it again, and spend a lot of time doing that. And I think that's the moment where you unlock real like. Potential of using this in like consistent ways? And that's, to be honest, the case for every other tool out there.

• 12:58Speaker 01

  If you think about, again, filmmaking, cinema, you're not becoming a filmmaker by having a camera and pressing record. You're becoming a filmmaker by taking that camera and pressing record in particular moments in time with an intention, with an idea, and then knowing how to edit all those things together in something else.

• 13:17Jacob Effron?

  Yeah.

• 13:17Speaker 01

  And so that's for me like a very similar analogy for this.

• 13:20Jacob Effron?

  So some people I'm sure just rock that right away and they start messing around with it. And then I'm sure other people, it takes them a little longer. And obviously you've started working with enterprises now. And I'm sure some people in the companies get it. Some people take a little longer. Have you found any good ways to kind of get people to to realize the power of the tool?

• 13:33Speaker 01

  Yeah, uh, just, uh, just start with like perhaps things you've done in the past. Um, so if you're like a traditional post editor or you're working like traditional like uh, 3D systems like Unreal or Unity, just start with the stuff you've done and try to recreate that here. And so that's, that's for you, that's kind of like a constraint. I think creative constraints do help a lot. If you start with like anything, do you have that blank page issue where like what, what should I do? You know, like I have no idea. But if you start like briefing or creating constraints, then I think people start getting it faster than than ever before.

• 14:08Jacob Effron?

  On a previous podcast, I believe you said that UI doesn't matter, which I thought was like a juicy statement. And so I'd love to just unpack that a bit. Like, what do you, what do you mean by that?

• 14:15Speaker 01

  It's not that don't, it doesn't matter. I think in the state of the world we're in right now, where models are advancing at really fast speeds and it's more about leap rocks than anything else. There's a common misconception that by spending too much time over engineering the right UI, you're going to figure out things more than just scale. Um, and so we made that mistake before, we've learned from it. And, and I think you'll start to see more of that where it's an easier argument to say, just add some like UI on top of it and it will be great. The truth is that probably you will get steamrolled by just better models that can do all of it. Because it's not a perfect UI, but who cares? It works.

• 15:01Speaker 01

  Right. And I think it's not that it doesn't matter, but that's the first kind of like thing that I kind of like have to keep in mind when building products. And interfaces. And the second one is more of a, an interesting one when it comes to even thinking about how to design those interfaces in the first place. I think for most of the part, creative software across the spectrum, across different disciplines is very prescriptive. There's things that you can and cannot do, and every update on the software requires months, weeks of like planning. I think I want to come to a world where those interfaces are on their own generated as you use them.

• 15:42Speaker 01

  And so if you're working on a particular thing in runway, I hope that at some point in the near future, instead of us as designers and engineers defining what the sliders would look like and the representations and the primitives for control, the model will adjust them or generate them based on what you're trying to do. And so if you're trying to create a 2D animated film that's different from like having a 3D like short film or hyper realism, then you might need different things. And so I think there is a world where the model just prompts and creates those interfaces for you.

• 16:13Jacob Effron?

  I love that idea of like a dynamically generated UI for, for the product based on what you're doing. And I feel like what you're alluding to, I feel like this is like the question for AI product builders right now where it's like, the models have some set of capabilities today, you could build products around those capabilities, or you kind of just wait for the models to get better. And then, so do you build any, like for the, the here and now capabilities, do you build any kind of workflow around it? Or is it like, hey, that might be so irrelevant in six months or nine months based on where this is going? Like, how do you think about how much effort to invest in it?

• 16:42Speaker 01

  It's a hard problem, to be honest, because it's exactly right where you're saying, which is how do you keep up with like the velocity of change? I think we cultivate more of a philosophy and a state of mind, which is like, try to think long term, try to think about the truths that you believe are going to be there in a couple of years and build towards that. I think there is, it's less of a focus on particular points and more on the line where those points are going. Because the world will be very different in like 12 months and it will be very extremely different in like 18 months. And then figuring out the right system will make a huge difference.

• 17:18Jacob Effron?

  Yeah. What are those truths that kind of guide for you guys?

• 17:21Speaker 01

  On the one hand, I think.

• 17:24Speaker 01

  Like quality and temporal consistency of all sorts of like elements in video are gonna be, or if not, are there, in real-time generation. So inference times going extremely down in terms of like generation spits. Even like now, Gen 3 is our best public model out there, and it's significantly faster than all of our previous generation of models, while being also significantly better. And so you start to see more of that. I think technology tends to move in two waves. Expansion and then optimization. Like expansion is like you discover what's possible, you understand new opportunities, you tinker with them, and then once you figure out something works,

• 18:09Speaker 01

  I think society is really good at like when it works, just like optimize everything out of it and like just make costs like extremely like reasonable and like just make the best out of it. I think you're starting to see some version of that with language models.

• 18:22Jacob Effron?

  Totally.

• 18:22Speaker 01

  Where like now it's kind of like cheap to run more stuff. So I think that's one truth.

• 18:28Jacob Effron?

  In your world, though, it feels like we're in constant expansion mode. It's like right when we figure out the capabilities of Gen 2, Gen 3 comes and it's like.

• 18:33Speaker 01

  Yeah, yeah, it does feel a little bit that because I think the constraints towards where we're building particularly are different from just building language models. Like we're much more interested in like finding, like in in in language world, in question answering systems, in chatbot interfaces, hallucinations are bad and like getting errors are like not ideal. In art, it's something like the best thing you can get, right? You want the wordness, you want the uniqueness, you want the creative interpretations of what you're seeing. And so it opens up a completely different like approach towards building this. And so there's so many things still to explore. But going back to the question, I think another truth that I see will become more obvious over time is that you can.

• 19:20Speaker 01

  Build systems that understand the world and its dynamics in the same way that we do. And so if you ever are or have ever been in a like set of a film, you'll see directors and editors and our professionals in all sorts of like aspects of the process, speaking about what they want to do by pointing to things, by using references to previous films, by directing actors, using like gestures and words. I think I want to come to a world where you interact with these models in very similar ways. You're directing them. You might be speaking to them. You might be referencing like ideas or previous films, or maybe some concepts you had.

• 20:03Speaker 01

  And the model makes mashups that and understands what you're trying to do, the intent of it, and then suggests something.

• 20:08Jacob Effron?

  I love that idea, especially as you think about the multimodal inputs, the idea that you could just be like directing your film, not just through text prompts.

• 20:15Speaker 01

  And maybe like for me, I'm a very, sometimes I come with ideas just listening to like music. And so music, that's a sensory input, like triggers things in me, ideas and states of minds. And so I feel music is an inspiration. So maybe the model should also be able to behave in similar ways where music can become the input.

• 20:37Jacob Effron?

  It's so interesting. We had Mikey from Suno on a few episodes ago, and he was saying the same thing, basically, that like it shouldn't just be a text inspiration for music. Like, why not you see a sun or some sort of image, or you hear something in the street and like that inspires a song.

• 20:47Speaker 01

  Exactly.

• 20:48Jacob Effron?

  I love that idea because it's so how creative processes work.

• 20:50Speaker 01

  Yep.

• 20:51Jacob Effron?

  You know, I think of Runway as like maybe the quintessential, you know, an original company that kind of combines really cutting-edge research and building models with actual product and deploying it. And I think now that's become a very trendy thing to be as an AI company, where you want to both be building the models and the product. And I'm wondering if you can speak a little bit to some of the advantage you get in doing that in terms of, you know, the feedback you get from the product being deployed that then improves the underlying research and what you've learned about like running both those organizations together.

• 21:17Speaker 01

  I think over time, it all goes back to like your vision and what we're trying to do. Our vision has always been this is, I think, eventually a new form of art. And you're building for people who want to use this to tell something around the world. And therefore, you're building for humans and people who want to express themselves. You can do that. By only, I would say, by really understanding deeply and having empathy for that, but then really understanding how to do it at a core fundamental research level, at the lowest level possible. I think the thing we've learned is if you find the right people that can speak the language of art and the language of science, then that's the sweet spot.

• 22:02Speaker 01

  That's when it really, something special happens. And that's what you get a little bit in Runway, where you get in the same table, some of the brightest minds in like research, working alongside someone whose work on visual effects or editing or has a traditional art practice for the last 10, 20 years. But they're both speaking the same language. They're both thinking about what needs to be done almost in the same way. And then the connections and the influences between the two are made obvious in the results of what we're shipping and doing.

• 22:34Jacob Effron?

  What have you learned about making that work? Like having artists work really closely with researchers? I'm sure you've gotten better at it over the years. What have you learned there?

• 22:41Speaker 01

  Well, the first thing is you have to have them have a process on their own, like figure things on their own. Again, it's a conversation, it's a language that both need to have. The second one is like removing all preconceptions or how things should work. I think from the research perspective, what you want to do is limit your variables and your constraints to a set of predetermined things you can measure. But the thing is like in art, you don't want to measure too many things in the way you're being perhaps as prescriptive in like science. And so there's a world that I've seen in research where what happens is people are obsessing around generating videos, right? Like the problem was like some sort of like you just generate videos

• 23:25Speaker 01

  and if you generate videos, then art is solved, right? But like a video, it's not the same as like a film. It's not the same as like art. It's just a piece of media. And the way the media itself can be combined with many other things, that's how you make something much more interesting. But that in between, the process of like generating a sequence of pixels and making that thing something way interesting, it's it's hard. And it's not as obvious as like, oh, just put some variables and constraints and we're done, right? And so I think what we've learned is like, you want to make sure you leave it open for tho- that those teams and those people to find the ways to like unlock that.

• 24:07Jacob Effron?

  I mean, how do you do evaluation? Like, how do you know if you're getting better, uh, given the kind of, yeah, that it is a little

• 24:12Speaker 01

  Well, there's, there's two like, uh, answers to that. The first one is just use benchmarks and human level also like make sure like, uh, preferences are like, uh, the best or and whatsoever. This is like the, the typical approach to just every model out there. Um, I think in a way that's a little bit flawed. Um, and what we tend to rely on is just taste, like literally like go through the stuff, like, do you like it? Like, it's hard to tell, but go and find the best people with the best taste, with the best, uh, know-how of, um, like styles or ideas and have them judge stuff. Do you like them?

• 24:53Speaker 01

  Maybe it's not, uh, performing the best in like any particular benchmark, but fuck, it's interesting. You know, it's like, there's something in the quality of it that I've never seen before. I like that. And I think we tend to rely a little bit more on like the tasteful part of

• 25:07Jacob Effron?

  And then they ask me to give that back to the research to figure out how to push the research frontiers.

• 25:10Speaker 01

  Yes.

• 25:11Jacob Effron?

  Yeah.

• 25:11Speaker 01

  Exactly.

• 25:12Jacob Effron?

  It's super interesting. I mean, one thing I feel like that also must come up is, I feel like with the state of these tools today and models, there's a lot of like individual features you can build, right? Like there's a bunch of these companies that have gone and built avatars or, you know, dubbing or, or, and I imagine at some point there feels like a bit of a tension between, you know, shipping individual features that are really cool of, of things you can do with AI versus kind of like an end-to-end product. Like, how do you guys think about that internally?

• 25:36Speaker 01

  Yeah, um, we've been, we've been in, um, In stages of the company where we've, um, we tried to optimize for the wrong product at the wrong time. And I think, again, we've been building this for like six years. And so we've learned a lot of things that don't work, you know? And so we'll give you an example, um, like a couple of years ago, like three, four years ago, we, we trained this video segmentation model for rotoscoping. Rotoscoping is basically removing objects from foregrounds.

• 26:04Jacob Effron?

  Right.

• 26:04Speaker 01

  Or backgrounds.

• 26:05Jacob Effron?

  My understanding was that was a lot of the original, uh, like where people, where the product got started in many ways, right?

• 26:10Speaker 01

  There was one of the research, like areas that we started with this idea of like, uh, vision task that can be simplified with like, uh, training models specifically for that. A bit of historical context, most of their rotoscoping and traditional editing out there is still very manual. You still do it like frame by frame. You go pixel by pixel and it's extremely expensive, extremely hard. You have armies of people doing it by like the second and every future film has probably a version of some rotoscoping work done. And it tends to be very, very expensive. And so we trained the system that does it in like a few minutes or seconds. And so you're feeding like 4K footage, you select with the UI, uh, subject. So you want a rotoscope and then the system does it for you.

• 26:56Speaker 01

  And it doesn't, it's not like a hundred percent, but it's like 95% most of the times, which is still like huge saving.

• 27:03Speaker 01

  And so you've, uh, you can build specific products for that. You can build APIs, you can build interfaces, primitives, you can optimize for it. Um, and that's kind of what we did. And then we decided to also start investing more time in driving models and more of the, the beginnings of gen one and gen two and gen three. Gen three alpha right now, um, can do out of the box with zero shot training, like rotoscoping and green screen. And so it's becoming a much more effective tool for generating that, that any previous version of our particular narrow task model trained on one particular thing. And so it's also cheaper and like overall just better. And you can also generate as many as you want.

• 27:48Jacob Effron?

  Yeah.

• 27:48Speaker 01

  And so it's one of those things where, again, if you start to focus and obsess on particular points, you're gonna miss like the line and it's more, much more about that line over time. And again, it's a reminder for our team of like, hey, okay, this is like the green screen or this is like Gen 3 and Gen 4 and Gen 5. And then coming back to the principles of like, let's do Gen 3, Gen 4, Gen 5, more than like any specific thing. Even if it's, even if for the short term is good, because you can justify it for the short term. Like, green screen was great at the time. But I was, we're very glad, I'm glad that we didn't spend too much time on that.

• 28:22Jacob Effron?

  Right. I mean, looking back, do you regret doing it, period? Or is it like, look, at the time, that's, that's what the most cutting edge capabilities were and it got you some of the brand of Runway and got people starting to use it. Like, did it still, in retrospect, did it make sense to spend time on that?

• 28:34Speaker 01

  Yeah, of course. I think it made a lot of sense. I think we learned a lot from it. We built a lot of the like streaming capacity that we had, we built. And again, it's less about, like, that was right or wrong. It was more like, we always saw that as a milestone into many things. We never thought of Runway as a rotoscoping or company, as our inpainting company. We had a bunch of different tools. We just thought of Runway as like, here's algorithmic improvements on models would unlock a whole new set of like creative toolings. We need to build towards that.

• 29:03Jacob Effron?

  Yeah.

• 29:04Speaker 01

  But you need to do stepping stones every time. Because six years ago, it was a completely different world where we're in now.

• 29:10Jacob Effron?

  What are like the next frontiers for video models? The things that you think about as still unsolved problems?

• 29:14Speaker 01

  I think better control needs to be like critical. That's the trade off that I think you're having right now with traditional computer graphics and video rendering pipelines, where in traditional like simulation software, you have pixel level control about almost everything. But it's, it comes at a cost, which is extremely hard, expensive, time consuming. In Runway, you can do it in a few seconds at the expense of like pixel level control. I think what needs to happen and towards, kind of what we're working towards is that pixel level control that would unlock perhaps the similar level of like fidelity and fine grain aspects of creating stuff that traditional like computer graphics tools have you or have right now.

• 29:58Jacob Effron?

  How far away from that do you think we are?

• 30:00Speaker 01

  Couple of months, I think, one, two years, yeah.

• 30:04Jacob Effron?

  And then, you know, obviously you release a model like Gen 3 Alpha and it's, it's the capabilities are kind of feel like a total step change better. What's like happening under the hood to make a model like that so much better?

• 30:14Speaker 01

  Well, there's a different part. I think Gen 3 Alpha, for context, Gen 3 Alpha, it's a smaller version of our larger effort of Gen 3. That's why it's called Alpha. And so it's a early-stage model that is magnitudes of time smaller than Gen 3. Now, some of the things that we did to get there were, first of all, we built an entire new infrastructure. I think it's not, it's common to underestimate how much it takes to get to the point where you can train the models at this level because you have enough time spent in both developer tools, infrastructure, and finding the right, right research environments.

• 30:56Speaker 01

  And so what we've done is just build a very cohesive, like solid, like foundations. And that allows us to do like quick fine tunings or quick improvements and quick iterations on models over time. It took us like a long time to build that infrastructure. I think that's been one of the most, I would say, interesting aspects of the results of the model.

• 31:16Jacob Effron?

  What were some of the main challenges you had to solve on the infrastructure side?

• 31:19Speaker 01

  Just like training, like at the scale of what we're doing is, it's hard.

• 31:25Speaker 01

  Running the, the model is like accessible for everyone. So the unit economics need to make sense. If generating 10 seconds, it's too expensive, then you're like not incentivizing people to, again, experiment. And so one of the challenging part for us was like, how do we make it so it's fast, as inexpensive as we can? And it's also is good. And that a combination of both research and infrastructure done the right way.

• 31:49Jacob Effron?

  And where do you think the future improvements come from? Is it data, kind of scaling compute further, architecture, the feedback you're getting?

• 31:55Speaker 01

  Scale matters a lot. It's the bitter lesson. Just better scale, more scale. We're still far from like reaching a tipping point or like the, the, the max capacity. I think we're still like, there's a lot of growth to do. Data matters a lot. And data, again, taste matters a lot, like the type of data you choose, the way you create the data, the way you caption the data, the way you work with data.

• 32:19Speaker 01

  And then finding the right way of scaling both. I think it's it's less about one particular thing and more about the real knowledge that a team can have around what works and what doesn't. And then use that to to build, yeah, the best models.

• 32:32Jacob Effron?

  I'm curious, when you think about, you know, the research team, obviously a really large research team internally, and I feel like there's probably some of these like longer term bets you're making. Like, I think you guys are working on general world models and things that clearly are going to matter in the future. Unclear when exactly that comes into the product. Versus like, I'm sure you're working on whatever comes after Gen three alpha. I don't know if it's just Gen three period or whatever. How do you think about allocating the resources of that research team amidst some of the the near term and long term bets?

• 32:59Speaker 01

  So it's a it's a team that's focused on different things. There's a pre-training team on the first hand. Pre-training is just training the baseline models. There's a team focused on controllability. How do you make the models steerable and aligned to what you want to do as a creative? We have a team focused on quality and like safety as well. And then a team focused on fine tuning, which is like, we work a lot with studios and they might want to customize the models. So we work with them. You need to work with the data that they have. So there's different aspects of the research that I think we've we've iterated on that I think have provided us with the best structure. We were still like constantly changing, perhaps the and most of the times like the structure of how the team works.

• 33:44Speaker 01

  And then in all of them, you embed some part of creatives or artists working alongside.

• 33:49Jacob Effron?

  How have you changed that over the last like year or two?

• 33:51Speaker 01

  I think the companies have changed a lot. I mean, again, we've we've learned from iterating on things that didn't work. There's, I mean, we're putting Gen three alpha out. But behind Gen three, there was like hundreds of things that we did that just didn't work. And by doing that, you figure out the best like structure for every team and the best way of doing that. We tend to be less short term goal oriented, like we don't have very like concrete, specific, hyper determined goals at the beginning of the year. There's like a master ambition and vision and goal. And then we mostly have the teams like figure out the right organization mechanism that works for them rather than being too prescriptive.

• 34:32Jacob Effron?

  I mean, from the outside, it feels like it's one series of things that are working, but I'm sure internally, like what are some of the things that haven't worked?

• 34:38Speaker 01

  Uh, I mean, there's many, I mean, most of the things that you try don't work. That's kind of like science, like you're iterating. If, if it was perfect, it was a one-off like thing you always knew how to do. I think we'll be in completely different like worlds. But like the aspect of, of what we do is again, you need to be comfortable, understanding that most of the things that you might do might not work. But there's something around it that might still be applicable to something else. And then if you have the right mindset of like, it's science, you're training stuff, you're iterating, you're finding things that work. And then it's gonna work. Um, and I think that's more, again, it's a state of mind, uh, more than anything else.

• 35:15Jacob Effron?

  I've heard you on the work structures hat. I think you've talked about how you organize your team into like ensembles and, and kind of a different way of structuring things. Like what motivated that and how does that work?

• 35:23Speaker 01

  Again, based on that idea of, of stepping stones and milestones, sometimes you don't know exactly what will be possible in like the next iteration of a thing. And if you're too prescriptive about what would happen, I think then there's not enough like ability for the team to wander around.

• 35:41Jacob Effron?

  Yeah.

• 35:41Speaker 01

  Uh, and so you want to have the ability for the team to wander. I think that's where real, true innovation and invention comes from. Um, I think there's a distinction between like marginal improvements. If you want to do marginal improvements to things, then you can be very prescriptive because you kind of like know where you're going and what needs to be done. And like, again, it's more of the optimization part. For invention, for newness, for wondering, for doing things that have never been done before. You don't need a lot of structure. You need to find the right people with the right mindset to be able to question everything they've done in the past. Just feel, think first principles, think about what's true in the world and then go and experiment.

• 36:21Speaker 01

  Uh, and so the teams that we try to assemble over time and the way we structure the teams is basically emphasizing that. Being not, if someone comes to runway and has a 60-page memo on all the things that they know are gonna work for the next like 12 months. It's not gonna work. Like, we've done that a few times where people come and it's just my proposal for like six months of work. It's like, great, now train a model. And just start with that and see what happens and prepare yourself to be surprised, in a way.

• 36:50Jacob Effron?

  Yeah. Is there an example of wandering that comes to mind of like going down some some weird place and coming back with something that was really valuable for?

• 36:57Speaker 01

  Sure. We've, um, we have a tool called Motion Brush. And Motion Brush is a way of like, taking an input image or frame, brushing over a specific subject, and then like defining the motion of that. That was not something we thought like prescriptively 10 years ago, five years ago, two years ago, and say like, that's the thing we're gonna build. It was mostly like two people in the team, like chatting with each other. One being a researcher, another one being like, um, an editor. And they're just tinkering around, uh, for a few ma- hours, days, minutes, working with some prototypes and then seeing this idea. And then it evolved from something that was completely different to like a final thing that now it's become like very useful for our users. And again, that was one of the things where like, if you just allow the right environment

• 37:42Speaker 01

  for people to be first of all, not constrained on like exploring things, even if they don't know exactly where they're going. And second, just the right environment to like do the research, capacity wise, compute wise, then like great things will emerge.

• 37:56Jacob Effron?

  Have you gotten better at creating that environment over time? Or imagine actually a new problem has just been the team growing a lot. Like, how has that changed, you know, from the, the early days?

• 38:05Speaker 01

  It's become, I think I've become more conscious of how important there is, at least for our team. Um, and I've become more obsessive around like cultivating that and maintaining that. I think a few things that I've tried to keep in mind, a scale and size of the team really matters. I think there are some thresholds where if the team gets too large, then you start losing that just because it's harder to coordinate people. And so you need more like- Meetings and more like discussions. The second one is really emphasizing to people and everyone who joins that everything will change. Like if you come to Runway and you feel that what we have is what will forever be, then you're wrong. Like you have now a responsibility to actually change it.

• 38:49Speaker 01

  And you want to find people who feel very comfortable, like extremely comfortable with uncertainty. People who might not know what they're going to be working on in two months. And that's great. And so it's more about, I think, organizations, companies, everything you see out of Runway is not, it's the result of people working together. And so if you obsess around people and the way they work, then that's the only thing that matters long term.

• 39:14Jacob Effron?

  Yeah, I love that. One other thing I'm curious about is how you've just thought about pricing the product over time. I mean, I feel like there's this one world where you could have just, you know, get it as many people's hands as possible and, you know, you certainly have access to capital. So you can, you know, do that just to get ubiquitous. And then there's, you've obviously charged for it over time on the consumer and enterprise side. Like, how have you thought about that?

• 39:33Speaker 01

  Again, build towards like the truths that you think are going to be obvious in a couple of years and the line instead of the point. With that in state of mind, I think pricing right now is a function of the world being explored and extended and not the world being optimized for. And so I'm not too, I know prices will go down. I know the cost of running and creating media will be the cost of inference. That's the world that will become a reality very soon. If you optimize for price right now, you might miss the discovery phase. So right now for us is more about unit economics that would allow us to just wander

• 40:15Speaker 01

  and have our users wander to then when it really works, like start optimizing for it.

• 40:20Jacob Effron?

  It's kind of like the lowest price point that's not, you know, obviously if you make it completely free, yeah, that makes a bunch of sense. And then I think, you know, obviously there's been some interesting developments in the broader video space. When OpenAI released Sora, I believe you tweeted like game on, right? You know, curious how you thought about the competitive landscape, other folks building video models and kind of where this all goes.

• 40:40Speaker 01

  Yeah.

• 40:42Speaker 01

  I mean, we started the company with like the same vision that we have today. And I think we've been able to create a market, a business, and instead of like realities that are becoming more obvious to more people. And I think that's that's true when like something works, you attract more interest from others that want to do part of that. I think Sora was was was one of those models where you start seeing more people like interesting and doing interesting things in the same way that you might be doing them. And the scale of of those companies are just very different. Like, we are a small fraction in all measurements to the size of a company like OpenAI.

• 41:24Speaker 01

  But yet, we've managed to build better models, better like products, and have them even before it was obvious for many people. And I think I'm happy that that competition is starting to like happen, where we're going to start seeing more media models and video models out there. But I think long term, it's less about the models and more about the vision of what to do with them and how to use them in really interesting, compelling ways. And so for us, it's still making sure they were at the frontier. And being at the frontier will invite more people to try to like compete with you. And I think competition is great. It incentivizes innovation. It makes our team and most teams much more eager to like win and continue to like change. And so game on is basically an invitation

• 42:08Speaker 01

  to get more people to think about the things that we care about.

• 42:10Jacob Effron?

  Do you think the future is like one kind of dominant video model? Or could you imagine? I mean, I think even in the image world, people have kind of been like, oh, well, if I want this kind of image, I go here. And maybe this style is better actually with this other vendor. Like, do you think we end up in that kind of distributed world? Or is there going to be kind of one one home that is that is clearly the right place to go for most use cases?

• 42:30Speaker 01

  I think most most markets tend to be condensing to like a few winners. I think this is probably going to be true for media models. And I tend to say media models instead of video models because I think video is a transitory stage for at least for us when it comes to like building the type of things that we want to build. And it's less about any video modality or Put control for videos and more about media pixels and the the combination of making those pixels together with audio as well. And I think at least for us, there's advantages in both being first and being good at building the right interfaces and products and systems that allows users to use these models.

• 43:11Speaker 01

  And I do think you're going to get to a world where you're going to have a handful, a very small handful of people able to build that large scale of models with that large scale of like offerings. And I think it's still early, but I'm confident it will be just as small handful of people.

• 43:27Jacob Effron?

  Yeah, it's interesting. Do you see a lot of people today, they'll like maybe generate an image in Mid Journey, bring it into runway, put some like Eleven Labs or, you know, Suno Audio behind it. It seems that the world's definitely going multimodal. Like over time, do you think is audio something that you eventually build yourself or something you partner for? Or how do you think about that?

• 43:46Speaker 01

  Yeah, we're building models on the audio domain. Again, it's multimedia and our approach on world models is you want to build models that understand as kind of we were speaking before, the world in the same way that any one of us understands the world. And so that means building models that can translate between different modalities quickly and easily and understand both the objects being represented in in a video or in a piece of content and media and then combine them with other ones. And so for that, you need really good data and really good like just overall research in like multimodal. So it's it's less about text to video or image to video, more just anything to anything or sequence to sequence. It's just sequence to sequence.

• 44:24Jacob Effron?

  It seems almost inevitable at some point that folks with really good IP are going to come in and partner, right? And you'll you'll find some way to to generate cool videos with Spider-Man or I don't know SpongeBob or whoever like. Where are we on on that journey today? You talked to enterprises, I'm sure, a bunch and like, why haven't we seen more of that to date?

• 44:40Speaker 01

  Well, that's happening now. I mean, we're working with studios and IP holders and media companies on creating custom models.

• 44:52Speaker 01

  The reason you might never seen them is because they might never be made public. You might use them for internal purposes. It's your own thing you've created together with us. We've trained the model, we give you the model. Once you have the model, it's up to you what you want to make with it. And most teams might decide to use it for internal purposes. So if you want to watch the new movie about the new show that you've liked, maybe a big percentage of the scenes being there were made with the model that you created. But the model is theirs and you might never know. And that's kind of the point, I would say, with, I would say, where we see these models being applicable,

• 45:30Speaker 01

  where it's less, much less about how it was made and much more about why, how it's been used. You know? And I want to get to a world where we stop distinguishing between AI-driven content and non-AI-driven content. It's just content, it's just media, it's entertainment, it's a movie. Like, you never go and watch a movie because how it was made. You never go and watch a movie and the first thing the director says is like, "Here's the bunch of tools we use." Like, who cares? Like, you care if the thing is good. If the thing is good, then like, great, you're gonna start seeing more of those. I think over time you start realizing more of that, where these models will embed in every media format that you see. And what you will care is how good the story is, not how it was made.

• 46:13Jacob Effron?

  You know, you talk about stories and that's what Runway is about. And as you think about zooming into the future, the stories we tell, you know, as these tools get easier and easier to use, how much of that do you think is like, you know, stories we tell with like IP or shared cultural experiences that everyone knows about? Everyone knows this character, everyone knows Messi, everyone knows whoever. Versus people in our own lives versus like just entirely net new, you know, net new characters?

• 46:35Speaker 01

  Well, I think first of all, the history of art and the history of like storytelling is somehow the history of technology. Those two things come hand by hand. I mean, we've always used technology to tell stories, even to like the most prehistoric like versions of ourselves. Like, we always use a rock, a fire, like some sort of pigment, and then evolve from that into like what we have right now. And so tools- Are in some way intertwined with our ability to express ourselves. And during some historical moments in time, every couple hundreds of years or thousands of years, you get something that just feels different than anything before.

• 47:17Speaker 01

  And I think the last time we felt that was with cameras, with the invention of filmmaking at the beginning of the nineteen hundreds. That was on itself a new art form. People started to realize that it wasn't anymore about depictions of the world in pigments and paintbrushes and paint tubes. And it was more about this technological innovation, this radical revolution of technology that was able to capture light in ways that no one ever before saw it possible. And to be, to be fair, like, most of the people working on like lenses and cameras at the beginning of the nineteen hundreds didn't realize what was happening.

• 48:00Speaker 01

  They thought it was some sort of like technical, like, I don't know, innovation, but it was nothing more than like a fun thing, right? Or you use the camera to just recreate paintings or something, or something that you would. Not even that, I think, like, even, even like some of the pioneers of like what we now consider early cinema, thought about it as a gimmick. It's like, this is like, you know, it's fun, but there's no future here. That's actually the word they used, like the Lumiere brothers thought of like cinema as something with no future. And people were very aggressive around it. It's like, this, this is not the same as painting. People are wasting their times. Photographic is for the lazy. Those are all things that people said at the time. And if you're given enough time, if you have the right people experimenting with it, a new art form emerged.

• 48:41Speaker 01

  And now you have industries, people, world, professions, shared experiences we had because of a new medium that was created because of a technological revolution. Yeah. And I think this is for me somehow very similar. We're, we're at the beginning of those like early camera explorations, where now you can synthesize and render things in ways you never thought was possible. And the first thing we do is like we compare it to the previous form of art. And I think that's the right thing to do because that's the only thing we know. Eventually, I'm much more interested in like what the new form will be rather than like how this resembles the previous form.

• 49:17Jacob Effron?

  I love that analogy to like the the jump from painting to, you know, camera and film. Is there something you've seen in the early glimmers of of how people use these tools that you're like, that's the the early inklings of of what feels like a jump, not just recreating what we've done on on camera?

• 49:30Speaker 01

  I think like perspectives on camera angles or like visions or aspects of telling stories that are very, very, very hard to do with traditional filmmaking is for me beginnings of that. The second part are like customization where you're able to render things in ways that are particularly unique to you and not just generic to everyone. So you pair those things together like customization, unique new like angles or cameras or perspectives with real time. And that's where I think you get like this new media format and art form.

• 50:11Jacob Effron?

  That's awesome. I mean, runway obviously is this viral consumer product and then, you know, you've also leaned into enterprise as well. I'm curious, like, you know, how you've thought about that. And then over time, like when, you know, runway as the business in five, 10 years, is it mostly enterprise, consumer?

• 50:25Speaker 01

  It's hard to predict, to be honest. I think, again, if you go back to like, like the art forms that were like so deeply trying to like

• 50:33Speaker 01

  provoke, I think that storytelling and art form like idea will be

• 50:42Speaker 01

  you can find that in like every kind of like user persona or industry or market, professionals, non-professionals. Even if you think about what we consider professionals today, like filmmakers that are using tools that are 20 years old are going to be able to do perhaps what used to take teams of hundreds of people like in a few minutes with one or two people. They're still going to be professionals. They're still going to be working in enterprise, just in different ways, you know? So I'm much more inclined in finding those.

• 51:09Jacob Effron?

  And then, you know, obviously you've had, I feel like there's no shortage of people that want to give you a lot of money, just given the excitement around what you're doing and the vision. Like, how have you thought about the appropriate amount of money to take in and also like, you know, how much you'll need to raise in the future?

• 51:21Speaker 01

  I mean, it depends on scale, really. We've been extremely efficient and effective with managing like our both burn and overall spending. And I think the reason for that is that we want to keep the team like small, very focused, very concentrated in what really matters. But there's a level of like investment that needs to be done to like scale the models for the next generation of what needs to be done. And so we just, every time we've raised, we've just might make sure that we're training at least for the next 24 months.

• 51:47Jacob Effron?

  Yeah.

• 51:48Speaker 01

  And that scale is larger as we go.

• 51:50Jacob Effron?

  Do you think, I mean, obviously, companies like OpenAI and Anthropic talk about raising, you know, they'll need to raise billions and then tens of billions to like, does video go the same route where to train state of the art, you'll need that?

• 51:58Speaker 01

  Well, it depends on what, again, it depends on what you're doing. Like, if you're building God, if you're building some sort of like intelligence, like religion, that like you, you think will like save everyone's problems and humanities like forever, then probably you need trillions of dollars. Like, I mean, religion is a pretty like capital intensive like business, right? We're much more focused on like, okay, there's, these are tools, these are media formats, how and what's the best, most effective way of scaling them, scale really matters, how much spending you need to do, and just focus like on that. Like, our goal is not to create AGI or some sort of like God system. Our goal is to create tools for people to express themselves.

• 52:41Jacob Effron?

  Yeah.

• 52:42Speaker 01

  And that, I feel, is a much more interesting long-term goal than anything else.

• 52:45Jacob Effron?

  Well, we always like to end our interviews with a quick fire round where we get your thoughts on a few quick questions. And so maybe to start, would love like one thing that's overhyped and one thing that's underhyped in the AI world today?

• 52:54Speaker 01

  One thing that's overhyped, I think it's like text to video systems. I think it's less about prompting. I think prompting is like the wrong interface, to be honest. More about just whatever interface and input or control mechanism works. One thing that's underhyped, I think, is what you can do with models like Gen 3 Alpha in terms of like simulation systems and engines and how the model understands like fluid and dynamics. And there's a lot of it that we just haven't explored yet, and we're still like figuring out that I think is going to be really surprising.

• 53:30Jacob Effron?

  What's been the biggest surprise in like building AI features and product at Runway? Maybe like an example of one thing you didn't think would work that worked really well, or one thing you thought would work and didn't work?

• 53:39Speaker 01

  You don't need tens of billions of dollars to make state of the art models. You just need a very focused team with very clear goals and be very diligent with it.

• 53:47Jacob Effron?

  What's something you've like changed your mind on since the start of this like, well, since the start of Runway, but also the start of this kind of Gen AI craze?

• 53:53Speaker 01

  Oh, I've changed my mind every week. I like learning stuff and changing my mind.

• 54:00Speaker 01

  I think I've, I think for a long time I've, I thought it was obvious what I was seeing. And I still believe there's a long way to go to make this more obvious to more people. I think for a lot of people it's not that obvious. And getting people to understand the trajectory and more than any point in particular. It's something that I thought was going to be easier, but I think it's still, we still have a lot of work to do.

• 54:21Jacob Effron?

  Do you think we've had our like chat GPT like moment where it just dawns on a lot of people how powerful?

• 54:25Speaker 01

  No, I think it's coming where you're going to have just hundreds of millions of people making stuff.

• 54:29Jacob Effron?

  What might that look like?

• 54:32Speaker 01

  Surprising, amazing, you're going to start seeing things you've never seen before, hearing from people you never heard from before. Like the people who are able to do this stuff that you might consider like Hollywood great stuff. It's really, really, really a small percentage of the world. Most of the world has never been able to access all of those tools. And that's actually one of the reasons I started Runway is like I'm from Chile and there's not a so big a self industry in the media world as it is in Chile and as it is here. And I think there's a loss. There's so many good stories out there that are constrained, not because of the ideas, but because of capital and resources and tools. I'm really excited to like lower that so anyone can do it.

• 55:12Jacob Effron?

  You were obviously were in art school a decade ago. Like any recommendation you have for, for people who are studying to be designers, artists now, on what they should be doing maybe differently?

• 55:21Speaker 01

  Explore weird stuff, like find things that are never, that you've never perhaps thought of before. Focus on ideas more than like the tools, like the tools are an extension of yourself. So just be original, like authentic and weird, I think, yeah.

• 55:38Jacob Effron?

  Which AI startup outside of Runway are you most excited about right now?

• 55:41Speaker 01

  I think there are a few sort of driving companies that I'm like studying about. I think there's a magical experience of.

• 55:45Jacob Effron?

  Being in a Waymo, you're like, this is nuts.

• 55:47Speaker 01

  It's nuts. I think the interesting part of it is like, I think it somehow resembles a lot of what we're seeing right now with like AI fears, where people freaked out about it and like they have all these like arguments and like ideas around why not? And then you get into a Waymo. And you're freaking out for five minutes and then you're like, okay, cool. I'm gonna take my phone and you like, forget about it. The same thing is gonna happen with most of the tools that you're starting right now. People are like, oh, this constraint, you use it for five minutes and like, oh, cool. I got it.

• 56:14Jacob Effron?

  Yeah.

• 56:14Speaker 01

  And so, yeah, Waymo is one of the ones that I like a lot.

• 56:17Jacob Effron?

  Awesome. I mean, you often say Runway is about stories. I'm curious, what's one of your favorite stories?

• 56:22Speaker 01

  One of my favorite stories? Uh, I think it's the story of how we built Runway. Uh, and the challenges and hurdles and, uh, what we thought was like impossible. Um, the beginnings of our journey, uh, not having like any experience in like Silicon Valley or, uh, ever raise money or ever build research in the way we've done it. I think that story, which has still been told, like, we haven't finished. It's something that, um, it sometimes feels like fictional because it's just like every time I tell it, it feels like, oh, that's not gonna work. It's not gonna work and it works. Uh, so, yeah, I like, I like how we're still telling that story.

• 56:57Jacob Effron?

  Yeah, well, I'm excited to see it reimagined through Runway 2 and, uh, over time. Um, well, I'll leave the last word to you. This has been a fascinating conversation. Where can folks go to learn more about you, the exciting work you're doing at Runway? The floor is yours. Anywhere you wanna point people?

• 57:08Speaker 01

  Um, just like runwaymail.com. Um, I think we have a few like resources and we're putting together more content to make sure people understand this extent of what we're doing. Uh, and then basically that website has everything. Uh, and in Twitter, I post a bunch of like stuff as well if you wanna learn more about what's going on.

• 57:24Jacob Effron?

  Yeah, so we'll link to the B videos, I recommend.

• 57:25Speaker 01

  Yeah, the B videos are good.

• 57:27Jacob Effron?

  Uh, well, Chris, thanks so much for coming on.

• 57:28Speaker 01

  Thank you for inviting me.