Episode #2: Getting Creative with AI

"Yeah, if I can do this, now what else can I apply that to?"

• I'm now going to leverage this and I'm gonna become way more productive or you can go out of business.
• If you're an expert in what you do, you're in an advantage right now. And that's the bottom line. (upbeat music)
• Welcome back to the Big Cheese AI Podcast. I am Andrew Herakles, your 29th best moderator, joined by Sean Heise and Jacob Wise, two tech leaders in Indianapolis, and last but not least, Brandon Corbin, one of the smartest AI minds in the Midwest. Today, we have a super exciting topic that I've been using and actively pursuing for the past few months in getting creative with AI.
• So let's sit back and just talk about how AI is impacting the creative industry, I think, as a whole.
• Totally.
• Because there's a lot of people who are concerned that AI is going to replace a lot of creative rules, right? Like the writer's guilt.
• The writer's strike.
• Yeah, the writer's strike was a good example that they're afraid that studios are gonna end up using AI to help write scripts. You can't put limits on technology. We've seen this. Every time you try to stop a technology, let's go back to Metallica when they're like, "Well, you shouldn't be allowed to download our music." And we're gonna go and sue people because they're downloading our music, right?
• Yeah, the only way to distribute music is through a, what, phonograph? Then a tape, then a compact disc, right?
• What was the name of the, was it LimeWire?
• No, Napster.
• Napster was the one that ended.
• And AudioGalaxy, I'll date myself. I'll show you how old I am when I was pirating from Napster and AudioGalaxy. But the fact is is that the technology's going to unfold and it's gonna be here. So you either can say, "I'm now going to leverage this "and I'm gonna become way more productive," or you can go out of business.
• Creatives should be putting out more content, not less. They shouldn't be arguing about if AI is gonna help them or not 'cause it's here and it's not going anywhere. But as you'll see when this conversation evolves, it's not necessarily about AI completely replacing a human. If you're an expert in what you do, you're in an advantage right now. And that's the bottom line.
• I also kind of see the market, expectations tend to go up when capabilities go up. So I mean, I imagine expectations continue to go up. You need more people to fulfill those expectations. So I mean, new markets will emerge from this.
• We talked about this a little bit last week. And as you look at these things coming out and the big question is, are people gonna lose their jobs because of AI? And I think that no, like Sean said, people are gonna be more enabled to create better work and maybe take out some of the busy work you're working on. But I don't think that you're gonna find a situation where you're seeing mass layoffs for people for an entire job role.
• That's the argument, right? It's one of those things where it's an inevitability.
• Yeah.
• And it's, but it is creating, it is creating an economic opportunity as well.
• Yes, yeah.
• So how do you balance that?
• So my dad ran one of the largest print shops in Indianapolis. He launched it like in the whatever, the 70s, 80s, doesn't really matter. But then he got some digital presses and he bought, 'cause there was this transition in the print world where you went from analog, where people were literally like cutting out film and gluing it together and scanning or shooting it in camera and all that, to where it's now all digital. And there was this period where he had a bunch of people who had worked for him for 20 years, who was doing it all by hand. And he's like, now here's the situation, we're now all digital. And you're either gonna learn the tools and now leverage what you do, 'cause again, like you're producing film that gets turned into a press sheet that then runs through the press. You can either adapt to that or die, basically. And there were the people who said, okay, well, this is our new world, I'm gonna have to learn how to use a computer versus just sitting there with my X-Acto knife, cutting out strips of paper to now do it on the computer. And those people succeeded. And the ones who didn't, didn't. And that's really where we're at here. Now, this obviously can impact a much wider range of skill sets, it's not just the print, right? It's virtually every aspect of business now can be impacted by AI. And so it's either you're going to adapt to it, you're gonna leverage it and you're gonna exceed at it, or you will potentially die.
• So Sean, as a developer, are there risks to your job? 'Cause I know we just talked recently, so you were working on VZero.
• Yep.
• And you created the entire website for the Big Cheese podcast in how much time?
• So to answer your question, to a certain extent, yes. But it's also to a certain extent, no.
• Right, because what were we just talking about, the caching, right?
• Right.
• No AI is gonna fix you with that.
• No, so just getting into that journey, there is a company called Vercel. And they are heavily funded. They're like one of the most hot startups in Silicon Valley. They kind of want to be the Apple of software development. So they have, everything they put out is usually super polished on the marketing side. And in the back end of it, it's pretty good too. It's good tech. And so what people have done with their platform is they kind of, they went developer first. And so they started building developer tools that were transformational in terms of how we build both apps and websites. And so they went, literally, their hook was, you want to use our package to build your site. And guess what? It comes with this little deployment option, right? Where you can deploy that application to our servers. You don't have to learn AWS. You don't have to learn all these different hosting platforms. Like we'll just deploy it for you on this platform called Now. And that was what it was called. And then eventually, they eventually renamed it to Zite. And then they rebranded to Vercel. But essentially, their CEO is highly technical. And he still writes code to this day. So you think about a company that's raised almost a billion dollars and the CEO's still writing code. It's like my hero, right? Either way, so this company goes out, sets out to build an ecosystem of developer tools that are very focused on front end technology, which is what I focus on. And most of my company does as well. And so they came up with this idea to really leverage AI. And one of the things that they're doing is they built this tool and it's just kind of out of the alpha stage called v0.dev. And v0.dev seeks to do something that many have tried and everyone has failed at. I don't know if you've used a no code solution to build a website before, whether it was Wix, it was WordPress, just anything, right? These tools ultimately allow you to drag and drop and kind of have like an experience where you don't have to know anything about code to get to. But as soon as you get past the elementary level, you need to eject out of that situation, right? You need to say, hire a developer, or you need to integrate with a third party system. You need to take the control over it. Well, these code generation platforms were generating garbledy gook, right? It was all horrible, right? And there enter a couple different new technologies that have made this easier. One of which being this presentation framework called Tailwind CSS. And that is, and we can all sit there and talk about that. Forever. But in the last three years, there's a framework that came out that made building web-based applications somewhat easier, more ubiquitous, and more transferable across platforms. But anyways, they were like, you know what? I think we could build a UI tool that is AI driven using React, using the technology that we built our platform on using Tailwind. And they set out their, Jared Palmer, I think was the lead developer. He's an influencer. Look him up on Twitter. He talks about this platform all the time. But they said basically, V0 stands for the first prompt you send to this application. And you describe a UI you want it to build. And so I went and I said, build me a podcast landing page. (Jared laughs) And it took me a month to get approved for the alpha. So I literally, Jacob mentioned last week, he was like, hey, we got alpha access. I was like, oh, okay. And so I'm just experiencing this tool for the first time. Well, first things first, you are using this prompt. And it's that traditional chat GPT prompt where it's trying to maintain context of your entire conversation.
• Okay. - Right? So V0 is the first thing you tell it. Then V1 is the first thing you tell it plus the second thing you tell it. And then so on and so forth. And so the challenge is to trick this prompt into keeping track of all the things you want it to do. And I tried to build an entire website with it. I didn't just build a component. A lot of the things that people are trying to do is build individual components. I said, build me a whole thing. Now, going through that from a technology perspective, you definitely have to have, you know, kind of experience writing the right prompts.
• Like I told you. - Can I do it?
• Like I started talking about the hero section and I would then talk about the podcast cards under the hero section. And it didn't understand what the hero section was. But then when I started over and referred to it the same way every single time, it started to understand what I was building. And ultimately, this thing can change the game for people and for developers that are using these. I mean, this is not just outputting something that you can go click through. It's outputting production ready code.
• Well, I think the interesting thing, and Andre, to your question, could you use this? Yeah, I think so. I think that, you know, maybe developers would have a little bit more insight of what questions to ask. But absolutely, if you learn some of the patterns of what prompts to use, you'll get better at it. The first time I used it, it was garbage. And it wasn't because it is garbage, it's because I didn't know how to talk to it.
• Oh, after about 30 minutes of using it, I messaged you on Slack. And I was like, I don't think this is any good. And then I just kept trying 'cause I knew I was gonna have to talk about it. (laughing) And I was like, how am I gonna get this to work? And I just started thinking, okay, what does it really need? It needs me to be more descriptive. I was kind of trying to boil the ocean.
• You know, one thing we were talking about this week about as you were playing around with it is just like every AI tool, there's generally introduced a concept of a vector database that has context of your problem set. Well, in UI, your problem set is your design system or the UI components that build up your application, your website, whatever tool set you have. So like we were talking, and Sean brought this up, it was like, the future of that could be very bright if you could inject an existing design system or build one and put it on there and then say, okay, now build a component that does this, that follows these tool sets or these rules. And it was like, okay, now we're talking, like that could get very powerful and you could iterate applications and very quickly.
• And one of the things that I found by using this tool is this is where the chat GPT and the LLMs come into play. If you give it context of what you want, not just what you want, but how you want it to look, it'll actually perform differently. So when I told it to build me a card for each podcast episode, it built like the most bare bones skeleton looking thing you've ever seen. But then I changed the prompt, 'cause the cool thing about it is you can time travel back to any version of the prompt you want. So if you have 12 prompts, you can be like, you know what, I really liked where I was at a V5, you can blow away all the other prompts. But that's what I did. I went back and I said, build me a beautiful episode card.
• No kidding.
• And it was just beautiful, triggered it to put some finesse into the cards.
• And you know what it did? It made the, it put object fit on the image. And so you know what I mean? It made the image, take over the card. It literally, and so it has somehow, it has context to different styles, which is insane to me.
• Well, and another, well, I'm sorry, but one other thing that I wanted to mention that I thought that you said was important was that the output is code that's using Tailwind. Now for us, that's beautiful because we use React and Tailwind.
• But that's the industry standard best practice right now.
• And you remember a couple--
• But you could say Bootstrap, and I guarantee you would probably be able to follow the Bootstrap too, because you know that all the Bootstrap code is probably within the--
• That tool specifically only outputs Tailwind.
• Oh really?
• Yes, it either outputs HTML or React.
• No, really?
• Yes.
• So is it their own, are we sure that, is this their own private LLM, kind of like Adobe with Firefly, or is this--
• So their products, their AI product integrates with any model. I'm guessing that they use chat GPT.
• Okay, I wonder if it's like a fine-tuned--
• It must be, they must have sent stuff. The concept of them being able to generate this code is honestly mind-boggling. And here's the kicker. So you're a developer, you go use this tool, you build yourself a component, you can literally, it gives you a command to install that component into your project. It gives you a UUID, you go npm install this UUID, and the component pops up in your project in the React code and you can use it immediately. You don't have to copy and paste anything.
• So they're pushing these changes to npm so you can then just install the--
• It's using, actually it's a MPX module called V0. And it takes a hash of the component and then it just copies it into your code.
• Nice.
• So it doesn't create a dependency. If you update stuff, it's a one-time eject function at the current version that you're at. And so, I mean, these are some nitty gritty details, but for me and for Jacob, and for you, I mean, we're all experts, level front-end developers. We've been doing this stuff for years. And this is something that I had to make very little edits, if any.
• Well, how did you specify the blue?
• I gave it a hex code from our brand.
• You actually just gave it the hex code.
• Yes.
• And that's--
• I straight up pasted the tailwind config into one of these as I was playing around with it. It did a beautiful job.
• Did it really?
• Oh yeah, that makes a lot of sense.
• So I think the takeaways from it was, you're talking about a generative AI product that's on the market that can augment you, even at an expert level. The caveat here is that I spent two days after that doing all the things I needed to do to get that website to work. Right? We integrated a headless content management system. We built the deployments on Vercel. We did some fine tuning of the code. We handled all the things that we would still need to handle. So that's where I get back to the argument from the writers guild, right? Is like, if you know character arcs and plots and subplots, and you're a great writer, you're still gonna have a job if you're really good, you're gonna be better. If you're not any good, you're--
• No, so here's the thing. If you're not any good, all of a sudden now, the mechanic who's never really had to deal with any technology, hypothetically, could literally do things in just English and output some pretty amazing things.
• The thing that really hit me when I was going through this experience of building the podcast website with AI and using vZero is the feedback loop is completely different than anything you're used to. It's like, well, maybe it's a little bit like back in the old days when you didn't have a local development environment, and you make a change to your thing, and you're like, let's SFTP this up, and then you got a white screen, and you're like, holy shit. But that takes time, right? You know, you have to upload the file, and it has to register, and then you might have a cat, you know, clear your cache. But the thing about this is that it doesn't necessarily, for me, I'm not sure, it didn't, it took time. So I'm 14 prompts in. Guess how long it took to get, for me to do 14 revisions on vZero.
• I don't know. - How long?
• How long?
• It was about an hour and a half.
• That's incredible. - Wow.
• You're right. So because, and I was deleting, so it was actually more than revisions than that, 14 acceptable revisions. You know, it's like when you send a crappy ticket to your developer in Jira, and they come back, and they're like, instead of asking questions, they just push up something and pull a request, and it's terrible, right? That was some of my prompts, and I had to throw those away. But each prompt is like 10 seconds to respond, and about 10 seconds to paint it. I mean, it takes time.
• Well, I think what we were talking about, too, is the final product, you could have probably built that just as fast as it took you to do it in vO, or vZero, sorry. But the value is the cycles you can do. Like, we all know as creators that the cycles are what eats up your time and your energy. So if you can do something 50 times that would have taken you two weeks, now it only takes you two hours, doesn't matter if the output is something you could have done if you knew that's exactly where you were headed, and that amount of time is like, that's the value, is like the iterations and the creative process, and also, too, your value, or people's value is walking through that process.
• This is meant to pipeline current developers who are in web development doing that much faster, but obviously leveraging all the knowledge you guys already have when it comes to this and the experience.
• And then that's why I just continue to say, fundamental knowledge, and it's gonna continue to be important.
• Yeah, yeah, and it just reduces the cost of risk, right, or of iterations, right? So, you know, you can make a mistake, like Sean was saying, he blew away entire versions of that thing, and that would have been pretty painful. It'd be you've been coding for an hour, and you're like, all right, that was shit, now you gotta start back from zero, you know? So that's kind of--
• From V zero.
• Yeah, from V zero, oh God.
• Well, and you would be able, like if you know what you're doing, again, so what I found is that the generative AI stuff was really good at coming up with like logo concepts, right? Not full deliverable logos, but it came up with a bunch of unique ideas, and I'm like, oh, that's cool, I'm gonna take this, I'm gonna take this, I'm gonna now combine it into my own original thing. And I think that that's kind of where we are. You could absolutely see in five, 10 years that low-code, no-code platforms are literally just like, enter your prompt, and brrrr, but it all just gets generated.
• You know, like in Minority Report, when Tom Cruise has that visual computing layer that he's going through, and he's just, he's kind of, that's the way I kind of think about some of this stuff. He's taking these sources, and he throws them up into this thing, but then he wipes it away, right? He wipes it away, and he's like, nah, I don't like that, let me start over, right? There's that concept where you're using this assistant to, there's another movie that does that exact same thing, I can't remember what it was, but you basically have this assistant guiding you the entire time. You're throwing stuff at the wall, and you're seeing what's six.
• The only problem with Minority Report, though, is that everybody underestimated how tired your hands are gonna be. If your hands are up like this the whole time, and you're moving shit around, you're just like this the whole time, by like an hour in, you're just gonna be like, oh, everyone is--
• I thought you were gonna diss the whole concept of it, like etching something into a sphere, when it could have just printed it off on the screen. It's like every time someone's gonna die, it's like this Plinko game, and you're like, what the hell was that?
• Well, Jacob, Brandon, you guys both, all week this week on the Big Cheese Slack channel, were talking about DALI, and creating these images, and things like that. Tell us more about that, what were you guys doing?
• Yeah, well, it's fun to play with new tools, and I kinda just, to Brandon's point, once you get a starting point, so I didn't know what to do. I've used DALI 2 before, it was crap. I started playing with prompts, and the stuff that comes out of it is pretty amazing. And then you get your creative juices really flowing. It's like, okay, if I can do this, now what else can I apply that to? So I just went down this very, very long path this week of creating blog post images, content for the podcast, you know, even personal stuff. It was just little cute cartoons of me and my girlfriend, you know, on a trip to Glacier National Park, kinda thing. So like, some really, really cool stuff.
• So with a platform like DALI, what worked really well, and what wasn't working really well as you were creating those images?
• Yeah, I would say stay away from trying to use infographics or anything heavy in text, didn't work really well. Photo realistic things, I would say, better serve with a Mid Journey, which is a Discord app, which is a little bit less approachable, I would say. But the stuff that worked really well was you give it a direction of, first of all, it uses chat GPT, so you can just say, here's what I'm trying to accomplish. You don't have to know exactly how it's gonna accomplish it. Then it'll give you a prompt that says, oh, a illustration using a cute cartoon of a mouse with this, this, and this. Copy that prompt, adjust it however you see fit, and boom, you've got another iteration of closer to what you actually have in your mind. So super cool, super cool stuff.
• Mid Journey, I've used it a couple of times. Actually, Aaron, one of my colleagues, uses it all the time. But Brandon, you were mentioning something about earlier when we were talking that there's a certain way you can prompt these things. There's certain ways that you would prompt a Mid Journey that might be a little bit different than a chat GPT. Can you talk a little bit about that?
• Well, it's always hard to tell because you don't necessarily know, so you just have to go see what other people are doing. So things like Lexica, L-E-X-I-C-A dot art, maybe, Lexica dot art is a great resource where you can actually just see a bunch of photos and then you can actually see the exact prompts that were used to generate it.
• Wow, didn't know that, that's awesome.
• Yeah, no, so those are super helpful. But with Mid Journey and DALI and Stable Diffusion, a lot of times what you find is if you know the artist, so people who are art historians are actually very primed for being able to be excellent prompt engineers for generative art. So they can go in and they can be like, find me something that's Pablo Picasso meets DALI or whatever it is. And they can basically describe it because so much of these are trained on actual artists' input. You know, Kirstanne and all these like modern visual artists, you know, their names, you can just add their names to it and all of a sudden you're starting to get more and more of their style of prompts. So I think that's really cool. But yeah, in terms of like, how do you actually perfect it? You just end up, it's just complete trial and error. And then trying to find other resources like Lexica and different things to go and see like, oh, here's this photo, I can go copy this prompt and paste it in.
• Yeah, I've got to check that out because that is the thing that helped me the most was like getting started, right? Like, all right, how do I even get this engine going? Oh, that's great.
• Yeah, so you get all your prompts. So this is lexica.art.
• That's super good.
• You can basically then just copy the prompts here. They actually have their own version that you can actually do generative art and then you run out after like five images and you got to pay for it.
• Well, what you see in social media, like as just a person that's not really in, this is so much of the like the hyper realistic stuff. Right? And I think that the takeaway for me this week with what you guys were working on is that there's this whole other arena of art and stuff that creatives will get into and love. And like, we created some mascots for Big Cheese that are like, I want that guy. I don't want him to go away. I like him.
• Super awesome, yeah.
• I'm creating this AI character. And I think that's gonna be an interesting thing that's gonna happen. Is people are gonna have these ongoing, like worlds that they almost create that they're from these prompts and they keep building on those prompts. I think brands will be created with a something that was just an AI took a base V zero prompt and then built off from.
• The key will be, can you reproduce them too?
• Yeah.
• That mouse, the--
• I wonder if we can reproduce the mouse.
• Right, so stable diffusion, you can actually give stable diffusion an image and it's gonna generate a prompt. So it does the reverse, right? It's image to text. And you can then take that prompt. And then there's a thing called LORA, L-O-R-A. I'm not sure what the acronym stands for, but it's basically another way that you can basically add learning on top of these different models to generate consistently. And so that's one thing that I think we're still waiting for to be just kind of built into all these systems. But once it's there, then video generation becomes that much easier. 'Cause right now, if you look at any video generation tools, you see them kind of shift. The characters, they kind of look similar, but then they kind of are shifting and it looks like this weird dream space that you're in.
• We have to do a whole nother episode about that 'cause those videos are beyond funny.
• It is, it's trippy as hell.
• Has Adobe got into this space also?
• Oh yeah, yeah, yeah. So that's where, so Adobe, the Illustrator was the vector, the vector ones that we showed. But in Photoshop, you can now select an area and be like, add an arm here, right? And then you can cycle through multiple iterations of it. And so you can be like, all right, no, no, no, yes, that's the one. And then you go and you select another area and you'd be like, fill this with a unicorn head, I don't know. And it just goes through and it does it. So Photoshop, absolutely. And that's what their flyer fly model.
• And that's another thing where if you're a Photoshop person, that's like in depth, right? Like I see that the Photoshop tutorial, I'm on the Photoshop tutorial Instagram algorithm. It's insane. But like, that's just another, oh yeah, I'm on that. I'm on the PowerPoint one too. And I'm like, I didn't know PowerPoint had this many capabilities. This is insane. But the point is, is that you're not gonna open up Photoshop and become a badass because of this tool.
• Right, yeah.
• And that's what I keep going back to is like, the winners are gonna be the big guys that are augmenting their existing platforms by adding these types of features. And the winners on the user side are the people that have good expertise that are smart.
• Yeah, if you have a good eye, you're gonna be able to generate good stuff. If you have a bad eye, you're gonna generate bad stuff.
• Well, Brennan, this AI music creation has been crazy. Was it two months ago that Drake's song came out and the guy with the little cape over his head came out with a better Drake song than Drake's came out with? So with platforms like Google Research Music LM, what potential lies in AI generated music samples?
• So music makes all the sense in the world that AI is gonna go and attack it, right? Like again, eventually you see that I should be able just to like stream, so I'm in this mood, I want this type of music, and it's just gonna constantly generate a radio that has just those songs. It's all gonna be highly personalized. But so what I wanted to do is really what I was trying to find was a way to come up with some intro music for the podcast, right? And so I wanted some like ska, reggae, like dubstep, backbeats thing going on. And so I went through and I tried like as many that I could find, and so soundfull.com was one of them. The other one was vid.io.
• 'Cause that's a whole video platform too, right?
• It is, but it's got its own, Vid's got also a music generator. So they're kind of going under that whole guise of we're just gonna do the whole production suite.
• That is a great platform.
• They got a super cool story too, by the way. The founder like completely failed for the first two years. They pivoted, yeah. It was super cool.
• Oh, see, yeah, I had no idea. I just happened to find them just kind of doing it. So I haven't done anything with their videos.
• That platform is super solid.
• The thing that really took them off, the pivot that they did was they really focused on a niche, believe it or not. And the thing that they started doing was one, the founder started creating a bunch of YouTube videos and putting them out there. And two, the main thing they focused on with their video editing was creating the transcripts, a lot like Opus Clip is doing.
• Oh, and it does it quickly.
• They were the ones that originally started that. So then that's why you started seeing all these videos with creators, now they had all these things and they realized that it got a lot more people's attention. So once they did that pivot, they started promoting that content that, hey, this is how you do it, they blew up.
• I literally Googled, I had a video. It was a podcast, video podcast I did, but I did it terribly. I didn't do it legit like Dre. And I Googled, I tried to find a way to auto, 'cause I don't know anything about TikTok and I didn't realize that TikTok does that shit for you. And Instagram, 'cause I literally can't use either of those apps, now I'm learning. But anyways, I was like, how do you generate a transcript? You know, blah, blah, blah. There goes Veed, and then you pop that thing in there, it uploads super quick. It transcribes it super quick and you can completely edit every single character and adjust the design. And you're doing that within minutes.
• And that's V-E-E-D.io.
• That's right.
• If you wanna go check it out.
• It's super cool platform.
• Check out that platform for sure.
• Again, so that was what I realized, it was super fast to generate the music from there. It is prompt driven, so you can just kind of go in. But the reality is, is a lot of times the music just, it all sounds muddy, right? Like it's the only way to really describe it. The best one, hands down, was MusicGen. And that's a model that Meta has released. And the only way that you can really play with it right now, I mean, you can go in and you can install it on, I think Bass 10 has it as a model that you can go and install and then interact with it through their API. But if you go to, if you just search Hugging Face MusicGen, G-E-N, you'll then go to where it's on Hugging Face and you can just go put in a prompt and it's gonna generate you a 12 second clip of whatever music you want.
• My God.
• And it was great. Like, I mean, MusicGen from a audible, like how it sounds standpoint, to me felt like the highest quality. And you can go and generate these things. The other thing that's really fascinating about a lot of these, and just in general, the generative space is, so Adobe, Microsoft, and Google now will, depending on which models you're using, if you get sued for copyright, they'll go and they'll protect you. So they'll indemnify you. And basically, so Adobe specifically with Firefly, they're saying that our model is built on no copyrighted material. So if you get sued for something that you generate, we're willing to back you. Same with Google, except for BARD. BARD's not included in there for whatever reason. I'm not sure why, but BARD is not part of this. But this, it's an interesting space that we're getting into, 'cause like the courts basically said that AI generated content can't be copywritten. And so now we're gonna get into this weird space where people are gonna be generating things. Someone's gonna be like, well, that's mine. And now these big companies who are producing this stuff are willing to back you legally from getting sued.
• Is it, it's hard to attribute the owner of the original content that the model used.
• Sure.
• Right? It's not like Spotify where everybody, you know, oh, I got a million streams. You get two million, you know, I can compensate everyone evenly.
• But with OpenAI and DALI, right, those are trained on the open internet, right? Like everything, like, you know, all of the different sources that are out there. So it all is like suspect. Where Adobe though, with Firefly, was trained on all of the things that they own. So they own Shutterstock. They own a bunch of different sources for images.
• So that's why Bard can't do that.
• I would assume so.
• 'Cause it's using, you know.
• And I haven't heard about OpenAI in terms of if they're willing to go to court with you.
• Well, and that kind of just points out to kind of how the natural way that the legal system and intellectual property ownership works in this country is if you created it and you own, it's like owning real estate. If you own the music, you know, if you own the painting, if you own the patent or the idea, like that shit lasts. And so like Adobe doing that on all their own data, so glad we had that acquisition of Shutterstock and you know, and those types of things.
• Yeah, back in like 2000, you know, it just happened to pan out.
• And then the thing that's kind of gets sticky is that other things you use content from, they have those little agreements at the bottom, like you can't use this for commercial purposes. And then you get through, go ahead.
• Yeah, I was just gonna say, I was doing a little bit of research for the pod, Dolly 3, you generate it through chat and I believe that is commercial license. You can use that for commercial purposes, but if you do it through Bing, I believe because it's probably going out to the internet as I don't know why, but you cannot use that for commercial use. So Bing has--
• Bing you can't?
• Bing's image generation tool.
• Do they have an image generation tool, Bing?
• I think it's using Dolly 3 under the hood, but it's through Bing search engine. I didn't actually play with that one.
• Well, you know, going back to what I talked about earlier with Tailwind, CSS is a company that owns their framework and you are not allowed to build products off of that, that do certain aspects. You can build a website, you can build an app that uses it, but you can't, there's restrictions on what you can do with their framework in terms of repurposing assets that were created. And I can guarantee you that Vercell's throwing them money for this.
• Yeah.
• Oh yeah.
• So like there's another situation where they created this model, if you will, and they're monetizing it because somebody built an AI tool that layers on top of it. So go create things, own those things, and you're probably going to win long-term in this situation.
• Right. Well, with the adoption for sure, like with Tailwind, I mean, who knew that like five years ago when Tailwind launched or however long it's been, that it would become kind of like the de facto.
• It is.
• And again, you can go to chat GPT right now and be like, I need you to write me a function or a component in Svelte that does X, Y, and Z, and I'm using Tailwind, right? And then it's just like, here you go. And it's BG, blue 500 or whatever it is.
• Yeah.
• It's incredible.
• I could go on for two hours about how that platform enabled all this and how it wasn't previously possible. That's definitely too nerdy a conversation for a different day.
• Yeah.
• Well, everybody that was the Big Cheese AI podcast. Again, I'm Andre Herakles, joined by Sean Hise, Jacob Wise, and Brandon Corbin. Today, we got created with AI. See you guys next week.