Hey, it's Eric newcomer. Welcome to dead cat. Another great episode, I'm excited to invite two people onto the show. We have Alexis gay, who is the host of the podcast, non technical, and I went on her podcast and we've stayed in touch. And I thought maybe she would enjoy coming on and play the CO hosts role on a very sort of technical podcast. So hey, Alexis,
Eric, thank you so much for having me. I am so ready. I'm ready to be technical.
I know you. You've got metaphors galore. You've been prepping? So I'm excited. Absolutely. All right. And then the one who gets to, you know, be subject to the actual questioners. Here. We've got crystal ball Velin suela, the CEO of runway, who I actually encountered him one of my early, crazy generative AI hype cycles. And I thought, normally, you know, I DM founders, just like I'm writing about your company. And you know, they ghost me, but you're like, oh, yeah, let's get on the phone and talk about it. I thought we had a great call. So then, I thought you'd be fun guests to sort of explain what exactly is going on in generative AI? So welcome to the podcast.
First. Thank you for having me. And I'll do my best to try to explain what's going on.
It's crazy, right? What just to start off, I mean, can you just give us some backstory of how you got into generative AI and the artificial intelligence world? Like how do you come to this space in the first place?
Yeah, for sure. So it all started back in like 2016 2015. I fell in love with this algorithm and technique called Deep Dream, which is, I would consider like one of the earliest like Ayar, kind of like models that was out there. It was very nice. She was just like, a handful of parties and researchers were playing with it. And I fell too much in love with that algorithm. To the point I just left my job in Chile, I got a scholarship at NYU and spent two years researching, kind of like computational creativity, or applying for like aI models for creative kind of like purposes and workflows. And that's a rabbit hole that's hard to get out of how
does one fall in love with an algorithm
I was going to ask?
Not an experience I've ever had no,
can't say the same? That's a great question. I think for me, what's the aesthetic quality and the aesthetic kind of like possibilities that I was saying, and it's kind of like early earlier algorithms. This is again, before kind of the wave of outputs and you might have seen today, it was more about that deep, I would say two things. One, I fall in love with the idea of using neural network techniques to generate images to generate video right to generate to kind of like explore and type of aesthetic and creative possibilities just just wasn't like even imaginable before. I really like that idea of like, exploring Uncharted, unimaginable territories. And and I thought, I saw this early, kind of like, researchers and artists kind of like, as a way as a first kind of like, step towards like just experimenting with this and seeing how how we could go, I think I follow them on a loved one single algorithm would be a vo of experimenting with technology, right of trying and use it in ways that perhaps weren't meant to be used. Right, and you go in, that's the rabbit hole I feel I fall in love with. But long story short, that one thing led to the other and runway was my to where they were my co founders kind of like a research project was my thesis at art school that I went to. And the whole idea is basically I still it's basically the same that we have today, which is how do you take these algorithms, it's your the VI systems and use them in the context of art making in the context of filmmaking in the context of art or design, just in the context of creative kind of like, processor. And I think what we've seen so far right now, like last year, 2022 was like a breakthrough year where we're models got really good, right? And for me, there's first way was perhaps 2015, to 2020, to where like some techniques and algorithms, some researchers and artists were playing with ad was supposed to be seen as a niche thing, right? So now when I remember pitching like, and telling people like, Hey, here's the thing you can use to generate an image. The first reaction I got was like, Chris, this is a toy, right? This is you being an art student doing art go dive. This is just I will never use this. Right. And I think that like this missing talent. Yeah, like that happens often when it's very early. And I think the way that we see now, which is I would say 2022 onwards, and we're just starting on second wave is models have gotten really good, right? So no, my mom like asked me about like, and I saw sees and understand what I'm doing right, which is just like the outputs of them are just so much crisp and invisible that I think it changed the understandings of what we can do with it.
You said you went to art school? I did. Yeah. So did you spend a portion of your life considering yourself in artists Do you currently
I tried, I tried I did. I studied like econ, and then like design and also like dabbling like software engineering, and then spend time just in like art and research. And they quickly realized that I guess my I wasn't really a good artist, to be honest. And now I think my art, I always find excuses to make tools to make art. And then I realized that those tools were far more interesting and useful than the outcomes of those right, at least for me, I think I've come to come to the realization that perhaps my my art is just too making any real like, making something that then a very talented artist can take and use in ways that I can't even like imagine, right. And I think that just that's a very kind of, like, interesting position to be in.
And in terms of our runways, specifically, like a video tool, is that right? Or am I thinking about it in a too limited way? I mean, it's helping introduce sort of these generative AI tools into video making, is that right?
Yeah, definitely. I mean, runway you can think about as a company. First of all, we are an applied AI research company. So we develop and build kind of like low level, the basic fundamental algorithms for content generation. And I mean content, because and we work with images with video with just multimedia formats, right. So we have a research team that sits at the core of the company of the product, that researching drives, innovates, and develops novel techniques for kind of like driving generation of content. And then we'll we'll products that leverage that sort of like algorithms and techniques. And there's things you can do on the video side. So we have films and post production companies and broadcasting companies and studios and just teams in general working on the video side of things, leveraging and using our kind of algorithms and products. And we also have a set of like image base solutions and products as well.
Do you have a dolly or like a stable? Do you have sort of an image creator this public? Yeah,
I think you mentioned stable, efficient, we,
I know you're a part of that, I'm going to talk about it. But do you have your does runway, the company have a separate sort of image product?
Yeah, we have around 30 different tools, right. And every tool uses a different algorithm or a combination of different algorithms to either generate a video, edit a video, or generate an image or translate from an image to a video, right? And so what's interesting for us is that we're not for our customers, we don't you get access to all these models behind the scenes, but you are not necessarily kind of like aware of those models, right? So from a filmmakers perspective, if you're an art director, if you're a designer, if you're editing a video, you don't have any care have to know about like the complicated like the ins and outs of these models, and the techniques and the names and all this, in the same way that if you use Photoshop, you don't really care about the blurring function, or the algorithm that allows you to blur an image or just use the image, right? You just use a tool. And so for us, it's that primitive, it's like it's about human intention. It's about you coming to the tool. And you're working with these algorithms in very intuitive ways. To the point you forget you're working with this kind of like, quote AI, Rajesh, you work with a tool that helps you achieve something in particular. But behind the scenes, yes, we have like those models that basically just power the whole set of features and products that we have
my understanding correct me if I'm wrong, is that sort of generative AI really blew up in the public consciousness when stable diffusion, an open source project with ties to stability AI, a company and your co founder, I believe at runway was one of the head researchers on that project. And then that project became open to the public right before that, you know, these projects had all been sort of behind the scenes, open AI, despite the name hadn't released theirs. And then stable diffusion comes out and sort of the world goes crazy open AI puts out, Dolly, am I getting that? Right? And where do you sort of fit in to that project? And how much are you involved with him? Yeah,
I think directionally you're right, I think so we it might look like there's an overnight success, or a specific model that let's say the space has gotten like so much attention in the last couple of months, right? And might look like, Oh, something happened in a specific like day or month. But the truth is that the research has been brewing for like years, right? We've been working on runway for the last four years. And it takes a lot of like iterations and learning and training and learning a lot from like, what works, what doesn't work, what datasets you can use, what models can you train, what are the possibilities of that, right? So the actual code and model and paper that powers and the fundamental architecture behind civil efficient was was released actually, like almost two years ago, right. And that's open source. It's been open source since we publish it. That work is the collaboration of runway and LMU. Munich is a research organization in Germany, and we do this a lot. We collaborate with other researchers, that Patrick I sort of was our research Principal Research Scientist led those efforts, right. And we published that code on open sources for a conference that was like released in 20 late 2021 And then since then that was actually called late and deficient, I was late in deficient work, right? It's deficient is a technique and late and it's like a word to describe like, the space of like possibilities in like a in a in an AI. So we basically call it later diffusion. There's a technical term that's actually longer than that it's high resolution image synthesis would lead in deficient models.
Sure. Good consumer faces right
off the tongue.
So that yeah, you can see. And then we iterate on the model, we first read the first model, and they need to rate it on a more training on more data sets for longer. And then a company called stability, kind of like offer, compute, basically came in as an infrastructure and just say, Hey, guys, I saw your open source project, I'll just throw in some computer and you guys can train it on a larger data set. And that's that basically, was we're stable division was born.
And then there was, you know, I love drama, there was a little drama when you updated. You guys updated that software. Right? And there was a little back and forth about whether you were allowed or not, or what was that whole thing?
Yeah. I think there was some confusion to Venice of again, we published that work almost two years ago, and it's been open servers, we've been improving aid and somebody donated compute, and I think they had an internal like, confusion or conflict about like ownership that they kind of like apologize about both publicly and privately. And, and I think we're just like steer an unnecessary conversation. For us. It's like, we've been working on this for life, a lot of time for a lot of years. And we publish new papers, we will continue to upgrade our related deficient work. And we have already like new models that are gonna like, change the landscape, if they will deficient was like, inflection point like, q1, if necklift this year is going to be like, probably ask you guys that one. But I think I'm less concerned about like, the drama to be honest. I know people like like the drama, but I think it's just like a entrench confusion. I'm like, I'm glad that some Yeah, I'm glad. I'm glad it's figured out, like I know, I think it's just like, got
the stakes here are just and this is classic, you know, we experienced this in open source software projects before, but it's who owns what, and the stakes feel bigger with AI, because it's been presented as this sort of, you know, everybody's going to lose their job or Well, you know, like, it's going to change the world. And so I guess my question is just like, yeah, like, all these new, you know, when, when open AI comes out with GPT? Four, when you update your software, new versions of like Sable, defusion, how much of those projects will be in the public domain? Where any company can use them? And how much are they going to be owned by particular people? Because my understanding with runway, was that like, part of how you're trying to differentiate was sort of like, the tooling and making sure you can apply the publicly available artificial intelligence technology to specific use cases, or are you also trying to own proprietary algorithms?
I think every company might have a different strategy and a different like, focus. I think, for us, something we've we've defined very early on, that has taken us some time to like, really nail down has been this idea of owning your stack, right? So we don't build our products on third party like models, or software or datasets, like we tried to build the whole thing end to end. And the truth is that that's, that's hard, because it requires you to really nail or understand how to build research, how to build infrastructure, how to build products, right? The advantage of that, if you've built that muscle and you train enough is that you can change any part of the stack, right? So if I want to change, you want to improve a model, because specific customer wants a better version of whatever we have, I don't have to ask someone else, or I don't have to wait for someone else to do it. I know how to do it. And they can go there like do it. Right. They can go to the lowest level possible have a model is trained on all this optimize how the model gets deployed, and then go back to the surface level and change that for them. Right. I think that's
your proprietary algorithms that you've built.
Yeah, of course, we have we have we have custom research, and we've open sourced some research. So if you go to our Research website, you will see that deleted deficient, the civil efficient work was something we open source. There's other work that we haven't open source yet. Well, perhaps open source at some point, it's like, but it's a combination of both, I think, our open source strategy. So we started to community building, like the the wave of creativity that emerge after releasing stable, efficient was just like, really, really incredible to see. And that just helps drive the field forward, which is exactly what I think happened. There was like an inflection point. It's a weight grade of hiring as well. You attract talent and you attract researchers who want to work with you. And like, people read the papers and people read the authors and you go back and kind of like, we get a lot of like, really talented applications from that. But we're not over kind of like approach perhaps different from other companies is that we're not an open source company like our product is not open sourcing products, right, or products to help creatives to use these techniques in comprehensive ways in their day to day lives.
Cristobal. I have some questions about that. So one thing I'm curious about, you just use the term wave of creativity so that after these tools came out, there was a wave of creativity. I largely saw on Twitter, people typing, like, bird flying over a truck that's on fire. And also Elon is driving the truck. You know what I mean? So my question is, when you say wave of creativity using these tools, like, can you help us understand a little bit about what kinds of things were happening using the tools?
Yeah, for sure. I think every new technology enables us to think about are creative in different ways, right? Like the ability for us to type something and generate an image just like it's novel. It's new. It's we've never seen it before. Like, some people were experimenting with it 10 years ago, but now it's just you open a browser and just type something, you get something out, right. And I feel like there's a lot of times where this like novelty component of it is just trying to think you just tried the weirdest impossible? Sure, when I like see, you wait a second, what you're saying you want to see how far you can go. Right? And it really like a curiosity component, like everyone is trying to break the system is like, Okay, I'm gonna like, really try really hard, right? And I think
that's like when you realize you're dreaming, and you're like, Oh, my God, I'm dreaming right now. What should I try?
I think that's the that's the i 100%. That's the experimentation of condensing curiosity phase, you're like just just experimenting with it, right? You're like, this is new. This is different. I want to understand that. What do you mean? But I think true creativity comes after that right? After you've like, okay, that has settled down a little bit. And you start experimenting more seriously, right? You start to think about it, how you might incorporate it in like a filmmaking process, right. And if you talk to a filmmaker, or for an art director, it's not about the crazy Slyke prompt. And you can generate some things a lot about like, control, like, I want to generate a variation of an image and I want to have I want to change a quarter or the left and a quarter or the right, how do I do it at scale? Can I do it like in 1000, like iterations per second, like, those are different kind of like components? And for me, something I hear a lot about is this idea that AI or like, the systems are automating creativity, right? I think for me, so curious
about that. I'm a comedian, by the way, for context. I'm a comedian, and I have a podcast, but prior to that, I worked in tech for seven years. It just like a classic career path, as you know. But anyway, so I have a lot of thoughts. I have a lot of thoughts and questions about this idea of AI supported creativity, or AI supported, dare I say, content or AI supported art, and all kinds of implications of that. And so I would love to hear a little bit more how
much writing Alexis Have you tried to write jokes with?
Okay, so here, I tried to what did I try? I saw the I saw AI generated versions of my own tweets. That's what I saw using some software that someone sent me which I'll have to look up the name. And it was terrifying. Because they sounded like me.
Mind would be like weird Twitter fights. And like, I wasn't shot article, I need to look
at air generated version of Eric newcomer sweets, you just
be a bunch of fake scoops. I need to look.
Good. Yeah. So I love this idea, though, that it's a phase you're saying right now? Or maybe perhaps when it first came out? We were in the curiosity phase. And that implies you said true creativity comes after that. So what do you think that means? I think,
for me, creativity is a state of mind. It's a way of looking at the world, right? It's not a tool, it's under a process, right? I think what you can truly automated is the processes of executing a creative idea, right? I think you can, you can go back to like, like previous moment in time, or like, there was a major technology like change, and understand the changes that transfer that brought to creativity, right. So I always go back to like, one of the earliest examples of technology changing art and creativity and how we look at the world, which was before, like the 1700s painting was like mostly the realm and the space and the possibilities of like, folks who have the money, the resources and the time and the knowledge to create these very complex pigments, right. And creating a pigment was a very sophisticated thing. It has like this history and tradition, you have to like hire these masters that were very expensive. And you have to like mix these very obscure colors, right? And then storing the pigments was like, you had to use a piece of leather. And like silly with a string, right? It was very, like, secure, right? And then 1800s Someone basically John brands inventor, he came up with the collapsible paint tube, right? That's the piece of technology, right? You can take colors and put them in, like a paint tube. And the most important thing is that you can take that paint tube out into the world, right. And so impressionism was born, right. You had people going outside and having like canvases like small canvases. You didn't require like a whole studio to just paint. Right. And so what that meant is that you started looking at the world in a birthing way people started painting just like portraits of like The royalty but like, mountains and trees and like social like events, right and like, you have all this creativity to like emerge. And that's what I mean by that give you the opportunity that's like your, that's a piece of technology that enables you to look at the world and express your view of the world in a very different way that everything you've used before, right? And fast forward is like that been able a huge revolution in art and like the trends forever transform how we think about painting artists, right? And you can think about very similar moments in time where technologies like, like that have changed, right? art and creativity and photography is another one, right? But they're gonna be change, we also look at the world. And a lot of the times early on the photography, kind of like wave people said that, like, photography was kind of like the end of hard, right? It's gonna be the end of like, paintings, right? Like, you're never just like, the Trinity, that it's not, it's just a different medium, right, just need to explore that medium in a way, there's new artists that will emerge from the new ideas will come up for that, right, and you open the door for something that's to goodness is very hard to predict, right? Because you try to predict something that you haven't even yet experimented with, right? And it's very hard to like make these sort of assumptions over time about everything that needs to happen. I think I come to embrace that like uncertainty as in like, we're early on something that's going to massively transform a lot of things. And we should embrace this, like, just experimentation phase, right? That curiosity phase, right. And that kind of things that you build after that, or you get steel to be like, uncovered and discovered. And that, for me is really, really exciting.
I mean, I have such conflicting thoughts on this, like, on the one hand, so many, like amateur visual artists, to me, like what they enjoy is like the physical process over the output, right. And even if they can get better outputs, from a sort of an AI system, is that cutting into some of the enjoyment I get the digital art and also photography, and everything is cut into that and people get their choices. On the other hand, I'm like, so excited by the idea of like, pairing the human mind with like, another sort of intelligence or, I mean, there's such there, I feel like you know, Alexa is wearing a chess sweater, it's like, if you get good enough at chess, you feel like a certain Connect, you know, you're you can use these bots to like, figure out how you should play better. They're, they're like these, I feel like they're already experiences today where you can sort of feel in sync with like a robot system where I play a lot of bridge, and my partner in Bridge online is, is a robot, I sort of understand what they do. And so having sort of this partnership with some sort of intelligence is, you know, a very charming experience. But yeah, it's definitely a terrifying future. Are you worried about it at all? Like, do you ever worry, it's just inevitable to you?
Yeah, I'm not worried a lot. I think actually, it's a very exciting time to be. And I'm very optimistic about it. I think, like, radical technologies like this are like one of the things they did do is to reduce the costs and enable things to be more convenient or easy to use, right? So think about like filmmaking, the people who had access to like professional filmmaking tools, even nowadays, there's like a handful of people that know how to, like edit, and professional software, right. And I will dislike complicated VFX workflows, and you need to understand computer graphics and a bunch of things. Picture having everyone with like ilm, or like be effects professional like skills, right? It's a wave of like storytelling in ways that we haven't even thought of before. That just like right around the corner, what kind of tools we need for that? Well, let's just build them. I don't know yet. I'm more excited. I feel like more excited about the outcomes of those and enabling more people to express themselves in creative ways. And I think that's ultimately the role of technology, like at large. I think I'm less interested in like, this idea, again, that you will automate it. And that will become like some sort of like, one click off solution, and like, everything will become the same. I think that's just boring. And like, it doesn't really, like, interest me, to be honest. And it's also like a partial view of how the world actually works. I think I think there's always this human component, and that that's what gives us like, meaning. Alexis, you were saying, like, someone just generates tweets that are like, just actually there's a I've heard there's a journey to podcasts, right? So you can generate the script, and you can generate the voice. And you have this like conversation between like Steve Jobs and like, Lex Friedman. Right? This everything is generated, right. And it's an interesting, it's like, yeah, I mean, works really well. You're convinced halfway through that this is actually happening. But I feel like there's still a lot of potential of using that in way more creative ways than just generating the entire thing like Antoine right.
I hear what you're saying. And I agree with a lot of it. And I especially the point about the way access to pieces of technology can really draw out someone's creativity. I think my personal experience is a good example of that where I was able to teach myself to edit video because iMovie came on my Mac, right? And then I know how to use YouTube and then I graduated to other software, but it was all accessible. It was all just downloading things. I'm looking at how to do this. I'm 20 year old dude, his basement told me on the internet, and then I was like, great. I know how to make videos now. But here's what's different to me about generative AI and some of the other tools that have preceded it and assisting creativity, which is that generative AI models, and please correct me if I'm using the wrong terminology, they have to be trained on something, right? And so when we're talking about visual art, or God forbid, comedy, or anything else, don't we have to feed it existing artists work in order to teach it how to do it? And doesn't that raise some implications around ownership, credit and profit participation of the output of the AI?
Yeah, totally. I think those are very valid questions and things that are really worth kind of asking ourselves as we develop these new technology, and, and again, I'll go back to like, Premier moments in time, there's this video and this recording of 35 years ago, same questions being asked when Photoshop was first released, right? Yeah, it's the same conversation or share with you guys. The same purpose, right.
It was original. I was like, this is an interesting question.
But it's but it's a valid question. It's part of the it's, yeah, it is. Right. And the interesting part is, like they were discussing at a time, it's like, are you allowed to edit a photograph? Right? A photograph is like, the truth, right? You can edit it, right? You can go into National Geographic on to like TV Magazine and edit an image, right? And the half of the panel was like, it's just illegal, this technology should be banned, right. And he can imagine now like not having like a tool like Photoshop, in our disclosure, so obvious, and like Photoshop became a bird, right? And so for sure, there's a lot of things that you need to figure out, like, Photoshop had to figure out all the things and kind of like, concerns about how to use the technology, people were using it to like for, like, fake bills, and like create fake money, right? So if you try to open up Photoshop or building Photoshop, Photoshop will like prevent you from doing that. Right. You can't just like it's your back, right? Yeah. So you build, you build over time, ways of like securing the system, right, but the 99% of the outcomes that are emerging from this are going to be net positive, right? I'm going to make society do you
think 99%? I'm sorry, I just want to go back. So you, I'm just you on the record just said that 99% of the outcomes?
That's I think that's just a believer, but I think yeah, most of the I would say that,
I put that on a slide somewhere.
Oh, the screws, I mean, it's a it's a very powerful technology, it's, it's gonna be as massive as cell phones were like, I feel like mobile was like, really net positive for everyone, right? Like, you're able to hold like, a lot of knowledge in your pocket and able to connect with anyone in the world. Right? That's, I think that's net positive. I think AIS and Jedi are similar technologies, right? You're, there's of course, like people that are going to their bad people in the world, and then to try to use everything they can to do bad things, for sure. And like, will be controlled to prevent those. But overall, the outcomes of the technologies net positive, like, Will we're allowed to democratization of content? For example, as you were saying, of anyone having access to these things, just like that, to them article change? Yeah. But I think to your point, there's, there's a lot of conversation that needs to be had, and people are having around like data set and training the models and when which data's and like how you do it. But again, like Google had the same kind of like, discussions early on, right? I don't want my data to be constrained by by Google, right? Kind of have and kind of have like a waist opening out if you don't want to have my website being like, searchable? And the answer is yes, you can, like just put this file on your website, and Google will not scrape it and they will not index you in the search file and the search kind of like options. So there's, I think there's still a lot of things yet to be uncovered. And models are not 100%. Like ready yet to be like using all sort of like professional environments, we're still very, very early.
How important do you think the humans are to generative AI? Or will be in the next couple of years? What we're talking about is humans sort of as creators experimenting using this tool. But on another conversation on this podcast, we've talked about tick tock and you know, how sort of the content sorting algorithm is so good. And you know, there was this question asked of like, How far away are we from tick tock just saying, oh, you know, not only do I can guess which videos that exist, do people want but I can just create the video that you are likely to want like, how far away do you think that world is where somebody like tick tock can just deliver the video that I want without a human involved at all?
Oh, we're very, very close. I think we're not far away from having laughing
but I'm not like oh, yay, you saying that makes me scared. That does not sound fun.
Are we?
I don't want to commit to a specific date or moment but I think like we're heading towards a world in general where a lot of the content that you consume online will be generated picture like a YouTube like generated like stream right? You can do Netflix, you aerated content like tick tock generated content, right? I think that's somehow festival like today if you combine a bunch of different things, but there's still a lot of things to be developed to get to a point where like, you'll do it real time. And I think it's, it's gonna happen if it works if it's possible. And he felt immense and Hala was to explore new avenues of like creativity, I think that's, that's going to be great. I think a lot has to do yet with developing these models in more safe manners and more, kind of like align ways to our human intention. And there's a lot of work to be done still yet there to prevent possible misuses, for sure. But overall, I think we're going to be in a time and moment where like, you're gonna be in every single movie you ever wanted to be, right? You're gonna be, you're gonna be amazing about
it is not only that, it can create something, but it can create something specific to me that I will find compelling.
Yep. 100% 100%. And,
but and we're using this is like three layout, like, yeah,
what's soon? I'm like, which? How do I need to prepare my job security? Like, what does my timeline need to be to learn to code? I just need a sense for when I'm going to be replaced by GPT. Three,
I think very soon, a couple of years, I would say. Yeah, that's your your, I mean, if there's definitely like an exponential like Greg progress fade that you can see and perceive more more period now that like, here's the progress, it would take years what took years of progress. He's now taking like months, right? Yeah. And will what will start to take months was gonna charge starts to take like weeks, right? And you go by that, like, all the time, like, large language models, I'm now going to be writing like papers, I'm going to be writing code and you start like accumulating the amount of progress that you can make. Right. And I think that's yet it's happening, and it will continue to happen. But I think that's going to happen, like sooner than I think I mean,
I I respect you for doing your best not to give us a date, because, you know, I covered the self driving car industry. And your that was an artificial intelligence powered industry that always said, you know, full autonomy is just over the horizon, like, what did we take from that experience? Why is this different than that, I mean, things people have gotten excited about chatbots, before things can appear. You know, if you solve 90% of the problem, but you need to solve the whole 100%, it can be easy to convince yourself that you're like almost there, but the last 10% can be can be extremely difficult to impossible.
I think it depends on what your goal is and what you're trying to accomplish right in, I guess, in the case of self driving cars, like full autonomy, right? Can you drive a car, in any street in the world? And like with no driver? Right? That's, that's the end goal, right? For for asset runway is not like, can you make a piece of art? Or can you make a video with no human intention or no control at all? Right? I think that that's not the goal. The goal is, the goal is like, can you take an idea and execute that idea in the fastest way possible, right? And right now, if you want to create a video, like Alexis, you're gonna make a video 2030 years ago, you will have taken your like, months, right? You have to like rent equipment, and then like, and have somewhat of been cost prohibitive as well, it was yeah, it was too expensive, right. And so your idea is like, I want to make comedy, I want to make this you just, you can't, I mean, usually it's gonna take you months, and you need to find a producer. And like all this. Yeah. And the fact is like, technology and like, smartphones on the internet, God, you like 90%, there, there's still the first work you have to do like coordinating on recording and editing, etc. But like, it's made like 90% of the work, right, you're like, so it's so much better, so much visible even to even consider doing something like that. So if the goal for us is like to help create, it's just like, get over that, like 80 90% of the work that you don't want to do as a creative. Maybe just like no one wants to like spend time, for example, searching through hundreds of videos, and then copying those videos and then placing them to express the idea that you want to have in your head. It would be great if you can just like, automate that right. And that's like, that's 90% of the process. Right. And I think that's the goal for us. If you measure by that, I think we're very close to going into that point, right. And that's the ultimate goal that would serve God is like, taking the cost of content creation down to zero, not the cost of like ideas, right? IVs are still like ideas and like, the best ideas will still win, right? I think this is the time where everyone will become more of an editor and curator and like ideas, the best ideas will be executed not because they have the funding or the resources just because they're the best ideas.
I feel that we have to we're playing sort of the cynics even though I'm extremely excited about this technology, but you're you're doing a good job of making the bookcase Do you worry that we're gonna have even more of like a bullshit problem? In the world of social media, there's already been the problem if you have a big account, you have people writing to you and just the work of like sorting through among humans. What's like a reasonable critique to spend time intellectually and engage with and what sort of a waste of your time is a taxing exercise even if you're like, sort of a with IT person. And then if you add to the mix, I literally on Twitter had someone I'm 90% sure use GPT as like an auto response to a bunch of tweets to like juice engagement, I responded like GPT. And they sent me a smiley face. I mean, it was obvious to me, but it's still like, oh my god, if I'm gonna run into that all the time, where I'm gonna have to engage with the cognitive task of is this GPT. And, of course, the problem with it being GPT is GPT, or whatever text text tool. There's sort of bullshitters they're very, they're 100% confidence, and maybe 90% accurate. And that's very challenging for human beings. Yeah. So what do you say to the bullshitter? Problem of AI?
That's a good question. I think, overall, I'm optimistic that that will get solved in the same way that like spam and like scammers got solving like, mail and in the internet is like, if you can receive an email from like a prince in Nigeria, you're like, Yeah, this is this is. Right. And this is like a very common like thinning the internet and like, you build filters, and you've been mechanisms to detect bad and like, people that have figured out and governance, like, there's ways of like improving those systems to like, avoid that right now, what's happening is that it's still very early, right? Like, if you consider like Chad GPD, tragedy was like a research release, right? It wasn't like a product, right? And it was just like, right,
they were taken by surprise, how popular was Oh,
yeah. In the bottom, it says like, this is a research release, right? There's a lot of things that you haven't yet consider, but you're just trying to learn how people use it the possible, like things that need to be fixed. And I think that a big misconception is to look at this and be like, Okay, this, this is it. This is the final thing we'll use ever. And it doesn't work, right. I'm like, No, it's just like a point in progress and time, and I will continue to improve it until a point where like, all that bullshit that like, will you will find and these things will be like, either reduce or remove or like prevented, and there's still a lot to be to be done. For sure.
I think that there is cool stuff. I agree with a lot of what you've said around taking the burden of some of the more like task task pieces of the creative development process. I think that there's a lot a lot that could be good there. I'm just scared. I'm scared. So we've talked a little bit about like, oh, eventually people will figure this out. And I'm sure that like things will happen down the line. I'm kinda like, this seems to be going awfully fast. And I don't know who is ultimately responsible, like, who's the adult in the room here that's supposed to put a hand up and be like, we should probably have some rules.
fatalism to I think so much of what happens is just if it's possible, it will happen. And that's how it Yeah, I think, sort of assumption underlying some of this conversation is that like, the learn to code crowd is going to win. And the word sells writers among us are the ones who are gonna be screwed. Do you actually agree with that premise? I mean, we're seeing these, like programs, do some coding or like, is it clear to you like, which human skill sets come out on top? If if this all happens is you see it? And which ones are the losers? Yeah,
that's a good question. I think I don't have a full answer two minutes, because I think a lot has to yet be like discover and like, understood for, like, where models can actually go. I think with every kind of, like, previous moment in time in the past, or something like that happened. Like there's, like jobs and disciplines and like, things that just disappear, right? Like, you don't need the the market for, like people who do analog film editing has, like, vastly been reduced, right? No one is just cutting films like with scissors anymore, right? You can find these people, right. And the reason is that those folks have to adjust their jobs, and like they're tasked to, for a digital age, and they have to learn a new task and understand the limits and the kind of like, all directions of using that. I think overall, my hunch is that we'll start to see that like professions, and jobs will radically change, right? Software engineer, I think will become more about, again, having an idea and helping a system, like execute the idea in a secure way, right? So if you want to write right now, you can just like something like copilot and just write a function and have like copilot, like, complete the whole function for you. Right? So you're not going into documentation to try to understand API's and then going back and just like it works really, really well. Right. But you still need to give guidance, and you need to the human need to understand your intention, right. And you're like having the system just help you along the way on that. Right. So I think that not sure, like software engineers are like, the job itself is like not, it's still going to be here, right? Just reading you're going to be different, it's going to be it's going to function differently. It's going to skills that might require gonna be different in might not need to know this very obscure, like levels of like API documentation, because now you have something that can do it for you. And for writers, I think it's the same. It's like pretty like, think about like how, like computers. So even typewriters change the process of writing, you're able to like raise things easier. They're able to like compensate us more quickly, you're able to like remix things because you have them saved in different ways. Right? And I think that that just changes the nature of like writing and who else who can work and right if I'm not an English like native speaker like these algorithms can help me like, significantly improve my writing skills. And now I can have a conversation with someone else in another language I don't have any can be exposed to because I don't have and that can kickstart something else. So I think it's more interesting to have a like a very open mind and very kind of like learning mentality of like, okay, how is my job or I think I do be augmented and how we need to be transform and change. And I think there's like the hesitation on the word conflict component of it's very natural, like, silent film was like, the thing for 20 years, like in the early 1900s, right. And when audio came to be a feasible technology to happen in cinemas, the first reaction from like, everyone in the Hollywood industry is like, we need to bind audio, right? This is going to destroy, and this is true, like, trolls, do not want to be on that. Are we still recording? Charles sharpening was like binding audio, like, I will show you like the ads just like the Association for like film and audio in like the musicians from like Hollywood or bonding audio from movies, right? Because like, who's gonna pay for like the orchestras that are in the theaters, right. And, of course, like, something happened. And for the for the,
a couple of things you're saying that stick out to me are one, there's potential here that the proliferation of these tools that let's just focus on writing, like help you, right could actually raise the bar for what type of writing we're willing to and interested in consuming. Because if everybody can, let's say, in a year, quote, right, like to be at three, the bar is going to be much higher for your average, I think article or newsletter or whatever, because you're going to have to be bringing something really special to the table in order for people to spend their time reading it. Yeah, 100%, maybe that's just me trying to be
I just want to know, I mean, I've sort of been in Silicon Valley long enough and watch the government be slow that I'm, I'm like, whatever is possible will happen and sort of nothing will put the cap back in the bag. So worried about it too much is, you know, it's good to think through these things. But, but yeah, the technology is going to progress. But I do think it's an open question. You know, what this does to literacy, like all of a sudden, people are just like outsourcing. You know, their professional writing? You know, most writing is bad. Most writing is formulaic. Yeah. But people learning to think, is important. I mean, I feel like there's been a lot of chatter about this sort of, I don't know, highschool cheating problem, I literally see it on teacher subreddits, you know, worrying about the cheating problem. I don't know. Are you worried about it? Like, one solution I was saying is, I assume the AI systems will just get better and better over time. So it'll become easier and easier to identify old cheaters. So it seems like a big risk to cheat when we're gonna be able to ferret out cheaters more, I don't know, do you worry about that?
I mean, I don't worry about it too much. I think like, it might help us, like really understand, like, from, perhaps from a more first principle, like thinking perspective, like what the goal of like schools are, like, it's not to memorize like words, it's like, you need to have critical thinking and like, I'll learn you, I'll teach you how to think and like, view the world and process data and come up with ideas, right. And that's ultimate goal of like, going to school to be like anything, right? If you want to learn something, it's not about memorizing stuff. It's not about managing understanding how to use a complex process and our workflow or whatever it is, it's about, like, looking at the world in a critical way, having like, a way of understanding things, right. And I think if like, if I learn how to code with something like a language model that can teach me constantly that can fit me they can give me examples and, and they can do a good software, ADR or I can execute something by doing that. That's fantastic, right? Didn't have to go through like a formal training of reading this, like, books and this traditional process. Maybe it didn't, then that didn't matter, right? I think that might happen. And of course, it's normal. There's like pushback at the beginning, right? Because it's just, it's different. Right? It's, it's changed, and we're humans. Yeah, we're very reluctant to change, like, we're very reluctant to change. And then the thing is, like, we get used to change a lot, I was actually trying one of the self driving cars in SF a couple of weeks ago. And the first like, 10 seconds is just very scary. You're like, I want to get out just like, where's the driver? Like? And then after a minute, you're like, Okay, I'll just take my phone and just relax, right? For a minute. We're, like very reluctant to change and then you you tried it and then you kind of like assume okay, that works. And it like doesn't like crashing or dying, it can move on and like it works really well.
And the human psychological journey on embracing new technology is a little sad, right? When something's exciting, we're very fearful about it. And then like you're saying, we just sort of embrace it and then now you know when I'm 10 minutes late because of the subway I'm like furious even though prior worlds were getting across New York City could have taken me like days right and required a horse Right, right. There weren't the bridges to do it all you know, like, so it's sad how much we take existing Technology for granted and then spend all our time obsessing over what's right over the horizon. But right that's, that's humanity.
Great. Your optimism for this technology is truly contagious because I really am I am thinking about it in less scary, negative cynical ways. I'm thinking about it in more positive ways as what could it do to like you use the word augment? Or Eric, maybe you did? Like, how can we look at these as tools that will help augment creativity and not replace it?
Yeah. In terms of particular companies, obviously, you're excited about your own. But what are the projects sort of using generative AI? That you're most excited about right now? Or what do you think people should be watching for?
That? A lot of things I've seen there some, I think there's some movie in particular that I think for me exemplifies a lot of other will continue to series over the next couple of years. I don't know you guys have watched everything everywhere all at once.
Oh, I love love. All Time. Yeah. Okay.
That was fully written using generative AI.
Kind of kind of I know, that's a it's a beautiful movie, right? If you haven't watched and you're listening to the World Watch it. It's like phenomenal movie, right? Yeah. And I think are quicker I see the thing moving guys, I don't know if you guys remember is it has a lot of visual effects, right? It's very intense visual moments, right? The most interesting fact for me when I learned this, and it was just like, astonished about it was like the majority of those effects were made by fight people, right? Fight people, right? Just fight people that didn't, that's your thing that didn't have previous experience building like feature film complexities and feature film effects like that. Were able to put together that kind of level of like quality and like they were like, very talented people, for sure. Right. But they just fight people that used to take hundreds if not doses of like cheese. And when I did, it was like, Wow, just like this is just insane. Who are these people? These are like super humans, right? What's going on? I search for those five folks a search for the like IMDB profile profile, then we got the five kind of VFX people behind the movie. And of course, like wanting to runway and see if they had like accounts. Right? And all of them had accounts. Right? That's it. That's phenomenal. There's so reach out to all of them. It's like, yeah, we'll chat and we chatted and they use the US parts of runway to edit the movie right to this small part.
Yes. That's yeah, very validating.
Yeah, and, and the main central is like, well, if we wanted to, like, I guess what I had the insight of like, I will search for them in our database is because if you think about creating a movie like that the complexities of the movie are immense, right? But you to be able to execute that you had to automate it, you had to simplify it in a way, right. And my assumption is like, maybe they came across runway, because basically what we tried to offer, right. And so they did, and we wrote a case study about it. Now, some of them are already using for other phones, I'll send you the case study. And I think like, what I'm really excited about is those kinds of things, right? They're very highly creative teams, with a lot of motivation, a lot of like greed to just understand how to get their ideas out without having to have this budget restrictions or like that, or no other set of limitations and just execute it. And it worked. It worked really well. It's like the bullies gonna probably gonna win, like those dozens of awards. And so Germany would say, I'm really excited about those teams that are embracing technology like this, that are like, experimenting with it that are using that are trying to push it in ways that I haven't even thought of like no one else, I thought, it's just like, you need to like expose it to more people. And I'm also excited about if you think about that same approach taken to product building, right. So a lot of the research in this space has been mostly led by like domain experts, right? The researchers who are like, very deep into the way it's like PhDs who are like incentives to like make benchmarks and X, Y, or Z metric better. But I think the outcomes and the results of using these models are going to impact way more than the researchers, right. And the insights are going to come from like multicellularity. And you need hackers in RDS need, like more people jumping in here. So the companies that are more excited with the products and the products I'm excited about are products that can blend those things, do you have technical folks at the same time you have RDS like speaking and sit in the same table and be like, Okay, here's what's possible, right? Would that be useful? Would that be not useful? How I'm gonna, how can I help you augment something? Right? I think that's the second way that I think I was telling you guys early on, like 21st wave of AI was probably the new wave of IPOs. They've been, they've been a few historically, but 2015 to 2021 was the first one and 2022 onwards. The next one is going to be about figuring out how to collaborate with humans, right? And how to take these algorithms from research domains and silos into real world examples.
But part of what I'm taking you to say is that, you know, there's a lot of emphasis on the generalized intelligence and part of what you're saying is that, actually, you need people focusing on specific use cases and how they play out or is that the right contrast to be zeroing in on?
Yeah, I care more about humans. I think technology should have been the purpose of humans, right? So we're helping you there's something I think when we might lose that, like narrative or the direction when you think too much about technology for the sake of technology, and that's technology as a way of serving humans in a particular way.
Wow, I know a lot of people who could hear that not gonna say who
does a safe space? I guess you can do it.
Super safe space?
Do you? I mean, like open AI. I mean, they're obviously the elephant in the room, Microsoft is reportedly like can invest $10 billion dollars? I mean, are they a rival of yours? Is their success good or bad? Or neutral for you? Or like, what's your view on open AI? And how do they relate to what you're doing?
I think it would be nice, a research driven organization, right? They're mostly about like pushing the limits of like research, right. And they've done an incredible job at that they've released multiple, like breakthroughs over the last 10 years, right? This started much more focused on reinforcement learning and then move because they discover new things. And that's just how research works. And I think that there are pretty interesting, unlike organization that's done really breakthroughs in research. But I think research is not enough for products. If you think about products product is about, again, helping serve human needs and helping people to achieve or solve a problem in a specific way. And part of it in this case, is research because you can build the fundamental technologies, but part of it is also how do people interact with it? Right. And there's a lot more that has to be built around it. I think, I don't think so pay is a competitor to DNS, I think they're just research organization that's been do you use their API's or anything? We don't know known their product, we build our own. We might at some point,
I'm sure you know more about this than I do. But like Jasper or whatever super hot company in the tech space was building on top of open AI. And then, you know, open AI comes out with this chat, GBT that ends up competing with it. I don't I right. Am I crazy? I mean, there are lots of questions about like, in terms of companies who's actually going to be able to win sort of the war here.
I mean, our our bet is you need to own your stock, you need to own your technology, right? If you're building something, you need to be able to, like, understand every piece of it, right? Because if not, someone else can just take it off you right. And for our kind of like focus as a company has been just building that full stack component, right? That's very
interesting. In a way, it's sort of like asking the question, is generative AI more comparable to the technology that is the smartphone? Or is it more similar to an operating system? In that? will it matter? You know what I mean by that, like, is it? Is it the technology that things will be built on top of like in the case of runway? Or are we talking, which is more like, I guess you could consider it like more commodified? And less like owned? Or are we talking about it as like, oh, it's equivalent to like a fully owned and operated operating system or something like that?
I think it's pretty vague. My hunch is more close to a smartphone, to be honest, like what you have, I mean, Apple has radically made that like the case, right? You've you own your hardware, you build everything from the hardware itself, to the software to the platforms, where people build on top right. And that's a way stronger business case, or a business like argument than having like, slight integration. It's mostly because, again, technology moves really fast, right? So you can't consider anything or you can't give anything for granted. If things are moving so fast, and we're becoming obsolete so fast.
We can let you go or Alexis, you have any final thoughts?
No, I think I've really, I've really grown on this podcast. Glad, what do you think, I think because I think I started a little more fearful. And now I think I'm more open minded. And also, I am terrified of the fact that 20 years from now we're gonna listen back to this. And I don't want to sound like the person that got interviewed about the internet in the 90s. And was like, You guys heard of this internet thing? So I'm really trying not to,
I've been clear in my I think it's huge. I mean, I was super skeptical of self driving cars, because they need to be complete. Whereas this, you know, it's about the human interaction, I'm already seeing, you know, like, professional type people, generating questions and drafting emails. And I feel like even if you're plenty smart to write a good piece of text, it's just like, solving it. And then the point that you've made sort of very clear on this podcast that nothing is more valuable than when a technology saves human beings, time. And so even if you don't view it as like, okay, it's going to be more creative than us in certain ways. It's like, if it's saving us time to do creative tasks. That's sort of the core value of technology, and people will always want to save time. So I'm super bullish on that. I, I still think I mean, we saw like lens there, you know, there's still a hype cycle going on. No, I mean, people get so excited something, it's cool. And then it falls off. So it's about sustaining that enthusiasm. Yeah, I
agree with that. It's staining the work is the staining due to some under development rather like specific spikes of hype. And so here for the for the long run to make that happen.
I think sometimes get things get better after the hype cycle dies down a little bit actually because there's less like fervor and like hand waving in the field, you know, and the people who are genuinely able to and passionate about those tools, like I think of the crater economy is a great example. Remember the five minutes when like everybody pretended to care about the creative economy. And now 90% of those people are not focused on that. But the ones that are the folks that are actually passionate actually competent in that space. And I think that ultimately the tools as a result will be much better for it. And I think something like this and self driving cars are probably also parallel examples.
Totally. Cool. Well, thank you both for coming on. Listen to Alexis on non technical Chris runway check it out. All right, thank you very much for coming.
Of course. Thank you for inviting me. Thanks, Chris. Silicon Valley
