Caroline Ingeborn

If you haven't started to experiment, if you haven't tried this, if you haven't joined like a education Zoom call of how to use these tools and what you can do.

I think you're missing out. And I know so many people that are like, I can't do this because all my friends will hate me. And I can't do this because I'm afraid of my job and the place where I work has a no hard, no AI policy. Your friends or your job do not own who you are or who you want to become.

Mm.

Welcome to Tastebench, I'm Bruno Marnett, my co-host is Philipp Zahn and this is a podcast about taste and the creative side of AI. Today's guest is Caroline Ingeborn, COO of Luma AI. Luma is one of the world's leading AI labs, building generative AI for video, images, and 3D, and helping redefine our creators turn ideas into content. Before joining Luma, Caroline was CEO of Toca Boca, where she helped build one of the world's most successful

Digital brands for kids with a strong focus on playfulness and creativity.

Hello Caroline. Such a pleasure to have you here today. I would love if we could start by hearing a little bit about what Luma is building.

So what we're building is see us as AI for creative work. And so that's really that's the starting point. And and it comes from us as a research lab with the mission of saying, okay, we believe that to build intelligence that can operate and generate alongside us humans, we need to build intelligence that are based across different modalities.

So they understand video, they understand language, they understand audio, they understand 3D. And this comes from when you look at where the language models are at the moment, they become really, really good in sort of like one aspect of intelligence. But anyone who's ever worked in the creative field and who's had a block of text telling a story, trying to visualize that.

If you don't understand the visual medium, it's very, very hard. So it was based on this background that we started building models, and they are inherently very, very useful if you're in the creative field. Building models is not enough because when you only have the model, then as a creative person, then you're faced with the proposition of like, here is AI.

Now all you need to do is to become a prompt engineer to get any value out of it, which is really bad. and also add add on that like your entire work is probably gonna change by this new technology. All you need to do is become a prompt engineer. And it's like I have met very few creative people that have sort of woken up in the morning and been like, I wanna become an engineer. That's sort of like n that's another different that's a different field. But that's not what

Mm.

you what you wanted to become.

and and so so then we have we have done two other things that we do at Luma, which is that we build products and we build them together with creative professionals. So 25% of the people that work at Luma today are creative professionals. I'll get soon into more what they do, but they but part of what they do is that they help inform what kind of products we should be building. and so that means that we now have a

A creative agency product where you can do zero to one work. So what does that mean? It means that you don't come to Luma and you're like, generate the video, generate the video, and I need to be a prompt engineer to get good videos. It's that you come as a creative professional and you have a long experience in your field. Say that you work in advertising.

And you know what it means to get a campaign, like what it what makes consists of a good, great campaign. Then you can come to Luma with your idea, it can be an idea, or a script. And together with the creative agent, you collaborate with it to be able to visualize. Now, to get this to full-scale production, you're still going to work with tons of humans. It's just this is a way for you to be able to bring your visualize your ideas.

to bring your ideas to life and to a largest extent when it comes to production, to produce things that possibly were out of reach from a budget perspective before. We we can dive more into in into that later.

Yeah.

I I I guess what's really unique with Luma is that you're I I guess you're you're quite famous for video as a modality. And it's it's one that not many labs will work on just because it's very expensive to train video model. I I I'm curious specifically about video, how much I guess how much control do people have when they're gonna generate a video? Because it seems that so many kind of, you know, micro decision that comes into, you know, angle, cameras, colors and so on. Like how much

How much do you feel like you you need to help kind of technically people to configure like kind of kind of videos they generate?

Less and less. If if you there's an aspect of like if you know what great looks like, if you come into Luma agents and you work with the agent, you can figure it out together with the agent. I will also say that what we are doing very successfully together with enterprises is we coin this term called forward deploy creatives, which which it and it came from

It came from primarily studios asking us like, Do you know any AI creators? Do you know any AI creators? And we know a lot of AI creators and many AI creators are very talented. But there were one thing that was like some AI creators might not be the right kind of people for those specific enterprises, but it's also like inside of these companies, there are thousands of great creative professionals.

And so what the companies need to do is that they need to not just be like, hey everybody, now you should use AI. Here's a subscription to Luma.

Mm-hmm.

What they need to do is that they need to partner with companies that can provide a forward to deploy creative arm, which means we have creative professionals from your field coming in, sitting in s embedded inside of your creative teams and working to

Teach and educate everyone on your teams on how to get the best results out. And so, yes, can you come in today on self-service a fast growing part of our business? But we also see that a a lot of education to get like really, really great results. But it's it's n it's not what it was like six months ago. The creative agents have really changed the game of.

How steep or not the learning curve is?

Interesting.

Can I ask you, how do you select these creative forward whatever, forward deployed creatives? what's kind of the archetype and how do you select for them?

It's a very specific archetype. Because

They need to be very, very curious, and they need to like want to learn new technology all the time.

But they also need to be good creatives. I don't think that you can you sort of like

If you I'm a really you know, if I'm a soccer player and someone is telling me I need to learn to play soccer in a new way, I wanna learn from someone who's really good and who knows what good soccer means. So then you can't throw in Caroline who's not really good at soccer and say like I like the new way is done like this, but I can only show you like in a certain way. Like you need to have someone that is also great creatively to talk to other creative professionals.

Yes.

and then the last thing is that before we deploy creatives needs to be passionate about the opportunity of being part of the educational wave of educating all creative professionals. Like you need to be passionate about, like it's important that everyone in my guild gets to learn this.

Because it's a superpower. It's gonna help multiply your talents. And there's so there's so much fear and there's a certain extent of like you need to unlearn some things that you know, but you need to double down of other things that you know. And so so I think that that's like if you don't have that passion and if you would rather be like, no, I you know, I wanna make my own movies, you probably shouldn't be a forward deploy creative, or at least not for very

Mm.

long.

Yeah. This is I I I I imagine this is very hard to evaluate taste in specific domains that you are personally not knowledgeable or the team might not be knowledgeable. What's the what's the secret source basically that you're looking for when it comes to this feature of or this ability to be creative?

I think we were

Very lucky, honestly, to hire a good strong ones early on. And we

Yes.

did that. So the the guy that came up with this entire concept internally was a a person on our go-to-market team, Prince. And he also he has a very good eye. It's like when you when you see some with someone with taste, you're like, yeah, he he got it. And and the way that he showed that was like.

the kind of creators he would surface and the kind of like, okay, this is like this is not just cool explosions. I g this is like actually artful and tasteful. And so in the first people we hired

he was like he would show their boards to me and he was like, You need to see their boards. And

Mm.

both him and I did all of the first interviews and so I think we just spent a lot of extra care. And then when you have that first group of people that have that creative people are not gonna let that like

Flock together. Yeah.

they're so so then in the subsequent interviews arguably like

There were people with even better taste than me that were in the interview loop. And so so then that kind of snowballs. But I think it's like it's one of these like you gotta go slow in the beginning when you're setting up something new to then go fast.

Be extra careful because this is basically determining the trajectory.

can you sorry, d on the Luma had a kind of an arc on the technology that evolved. And I mean we all are stunned by how AI has evolved in general, but I guess you are I mean you are on this one domain on this frontier, the very corner, the details basically. Can you maybe from your personal experience tell us what were your kind of aha moments along this trajectory where you thought, okay, this is amazing, and then what's the next thing that was amazing?

I mean I can tell my personal like I th I mean I around LLMs it's like so long ago that I don't even that that I

Mm-hmm.

Yeah.

don't even know, but then

It was really big when Dall-E came. Like that was like a

Mm.

big moment for me personally. Then there was on the audio side, there was this app. I don't know if they're around anymore. They were called Hume.

Yeah.

And they made

Yes.

an app where like their audio could basically understand where you emotionally were. Yeah.

Emotions. Yeah. Yeah.

And I remember when that app came out and I was playing with it and I was like,

I actually felt understood emotionally by

Mm.

AI. And this so that was really interesting. then just before Luma, and that was sort of like how I ended up at Luma, I got really into like what you could do with 3D.

Mm.

and and that's sort of how the name of Luma came up. Then when I

started spending time with Amit at Luma, I understood like, okay, there's a three then there's a video model that is going to be launched. And that was that was a huge moment when we launched our our first video model because that was nothing like it was nothing like the world had ever seen. And I honestly don't think I fully understood it until the weekend before the launch when I spoke to

A guy called Dave Clark from Promise AI. He is then like gone on and started his own studio in LA. and I I we had given him and a couple of hundreds others early access, which was like there were no video AI creators. So we were just reaching out to anyone that had done interesting things with 3D and images and super cryptic messages. Sign an NDA, get something interesting, I can't tell you what.

And for whatever reason they did, like it's that part is still inexplainable to me. because it was

Mm-hmm.

like, and I'm Caroline, by the way, it was super, it was super hacky, weird thing that happens early on. but to see his reaction or what he could do with that first

Mm.

video model, he was it was like late on a Sunday night, he was so excited, and to see the technology through his eyes was really, really eye-opening. And

But at that time it was still sort of only text to video.

Mm.

And and so that capped out and and the models were I mean, they they we thought they were great then, but they were bad. Like things were morphing, stuff was like

Mm-hmm.

it was all sorts of things were. Yeah, yeah. I I remember

Movements made no sense, right? Like that. Yeah.

we had this moment with when I was sitting with the research team and they were really excited and we were looking at a video of a man who was walking, and then all of a sudden

one of his legs disappeared and he fell. And they were like, Yeah. And I was like, there's nothing yay here. He's like losing his leg and we like no one And they were like, hey, but

Yeah.

the great thing is like the video understood that the leg disappeared and so he fell. So it was like physically accurate

Mm.

in that part. But the losing of his leg was like, Yeah, that's just that's another problem. Like we're solving that as well. So so it but it was a lot of there was a lot of like weird weird things like this that were happening. And so

who could access it was very limited. And then the kind of things that you could do with it was limited. At the end of like two to three months there was mostly explosions.

And then there was like mostly explosions. And then the second thing was when we then released image to video, something interesting started to happen. Because it and and it's the same reason why most, you know, most creatives that are selling a story, they they do storyboarding. And it's because it's it helps you explain the story that you're going to tell. And and it's the same thing when you interact with AI. Like if you have an image.

It does say more than a thousand words. And so you can start you can start to control what AI should do. And then we launched keyframes. So you could have a start image and an end image. And that increased the level of control significantly.

Mm.

But again, the kind of people that could tell stories using this and where the technology got you were still very capped. The story I always tell is the story of Culver Cup fall of 2024 and

the stories that were being told there. And there's this great person called MetaPuppet. He's he told a story about his grandparents' struggle with dementia. And he also won. But that was the first story that I saw with AI. And if we would look at it now and pull it up here, it would be like, my God, that's like weird. The story doesn't make sense. But he had like and he's a very good editor. That's his background. That's how he could make that happen.

and then the last sort of like big shift was when we got video to video, because that increases your level of control significantly. It solves for a lot of problems, but it solves for you're able to take human human performance and put it into any environment that you want. And that's huge. The second thing is though that you can also control.

Different objects with exact precision. And that's very

Mm-hmm.

hard to do if you're just working walking directly into video. And then the last thing is from a production perspective, is that you can take you can take anything really and make it into anything. And so what I mean with this is like our Dream Lab, I think, are becoming like world renowned in sort of sort of what they do, but they will build like a cardboard.

space and they will make that into a city. They will take like a I don't know, they will take a chair and make that into a motorcycle. They will take two shoes, tie the shoelaces together, pull it through a garden and make that into a car chase. Like it's it's just like and you can do that now. So that means one thing if you're a big studio and it means another thing if you're creating for your own platforms.

about big studio versus I guess more independent creators, I would imagine big studios have a lot of I guess expertise, right, in managing their assets, their I don't know, their characters, their world, the must I would imagine they're quite well organized already. is Luma also kind of trying to bring that sort of organization and workflows to people who are kind of a smaller, smaller studios, smaller independent artist?

Yeah, I think that's like one of the things that we got a lot of feedback on so far on our Luma agents is that we got feedback booked on the organization, but also like how can you connect to already existing organization platforms such as Dropbox, G Drive, etc. And so we're like every week we're shif we're shipping new connectors. and that's

that's coming both from large enterprises of like, hey, we have our existing places, we already have all of our files, but it's also coming from independent creators.

I see. In a I I I guess in a way you're you're already kind of solving the operational aspect, like the the convenience of you bringing in material and so on. I I guess what I'm curious about is you you you seem to have presented like the Luma scope as very end-to-end almost. You you seem to be interested in the final execution, but also kind of the early, I guess, ideation stage almost. maybe focusing on this kind of

ideation stage. I I wonder how much how much you see the I guess the process changing now? Like it it is are we are we going to a world where people are gonna come with very clear ideas and they're gonna press a button and it's gonna be the final video? Or is this gonna be a more I don't know.

No, I mean I mean

can I mean can you go in to to Luma today? Ideate with a lot of images, press a button and that Luma edits it together to a video. Yes.

Is that going to be like a great trailer that you can go and sell to a studio? no. and

I think it's like this. If you are a person

with ideas and you are a person with a story.

AI is fantastic for you. you can get your entire storyboard, you can ideate around the characters, and then you can put all of those videos together, you can try out different storylines, you can

Have a collaborator that you can sort of like bounce ideas with with the agent and say, like, what if I went there? What if I went there? And you can visualize that. Be like, no, maybe that's not right. There will be things that it's like, I didn't think about this. Maybe I should go there. but to really get something like a great trailer together, you're going to want to have a really good editor involved at this point.

you can what I see is like a lot of people like stitching things together to get like I kind of want like this, but what I like here is this and this and this and this. This is something that creatives are very used to doing, like regardless of medium. Like this, like I'm using this to show directly to Shannelia what I want. But then what I what what we see on the production side is that you then take that trailer and you productionize it. And then sort of you work move into the next part of the workflow.

Which is often when it comes to to at least to filmmaking, you want real live actors, you want a hybrid production, but it means that you can make things now that was never possible before. and and that part you're still working in agents and they still help you move faster, but you include part of the quote unquote old production process and you bring those assets in.

using video to video. And and what the beauty of everything happening inside of Luma is, is that you get a nonlinear filmmaking going. So so it means that you can you can instead of like, we're starting here and then the production

Mm.

budget and everything goes but like things are locked. And now instead if you're working

on on a movie, you can go and say, like, no, what if we change this? And you can go back. And this sounds like a circus, but it's actually leav leading to better stories and better creative output. But the same thing is like if you take the advertising use case here, it's like you can now not like you can truly reach content at scale in advertising because you don't you can keep certain things constant, but pretty much everything else you can iterate on. Yeah.

Yes.

And and and that's that's from a consumer perspective, that's fantastic, because then maybe I can see more content that is specifically for me. because there's so much content we need to get through that are like actually means nothing to me. So if the content that I see can actually resonate better than me, I'm all ears.

On the arc that you described, on the evolution, an important theme was the control, basically, that you increased control over time and steering in some sense. is there also a role for the randomness still and and can it be leveraged? And do people leverage it? Because in a corporate world where I need a gr and a spreadsheet, the last thing I want is basically hallucination. But in the use cases you woke you work on, randomness seems to be an essential element. Is it can people use this? And

Do they ever leverage this in some form?

Their randomness, yes. So I do think that there's like the specifics groups where the randomness resonates more. My favorite group where like r randomness resonates more is in music videos. Not not everyone that makes a music video, but like to a certain extent it's like because they start with a song.

Mm.

So they and they are like

like I want to visualize it like this. there was another thing that turned up. Like that's interesting. Like they are much more fluid to a certain extent

Mm. I see.

Mm-hmm.

of because they they are coming with it from a song and and from a an experience more. And

Mm.

again, this is not for everyone, but there is more

Of ness, maybe.

Fluidity,

Yeah.

then like when you have, but then you you also have music videos, but you have many others. It's like this is a character that is going here, and I need that feeling, and I need that thing, and and but again, I almost everyone I meet, they say, I was trying to generate this, but then this other thing showed up, and that was interesting. So, like, there is always like going to be

situation when the randomness is appreciated. And then there's gonna be a ton of situations when the randomness is like, no, I just I need this to stay constant. I want this to stay constant. How do I keep

Mm.

it? I don't want any randomness. It needs to be exactly like this. So that's that's all for sure a you know a continuous thing that we're all working through.

From your perspective, taste plays an important role. You already alluded to that. How do you define taste in the context of, let's say, someone creating a video?

I mean, I think to certain extent taste is

something very personal and it's also something that is

I wouldn't say judged, but recognized. And so what I mean with that is like for

For a certain set of people that are dressing in a specific subculture.

I might look at that and be like, that's just weird. But if you're in that subculture, you know who the tastemaker in that subculture is. And how do we make taste makers? Like, we under we appreciate their craft, right? And

Mm-hmm.

and so there's often this like, that person has taste, that person doesn't have taste.

You can have that conversation, but then you need to be like, what kind of taste are we talking about here? do I think that there's a lot of slop going on with AI on social channels? Yes, they there are for sure. but that is not that is not so much that AI doesn't have taste. Like

AI is just it's just technology. And if you have taste, you can make tasteful things with AI today.

You describe so yeah. Bruno, you

W when you s sorry, w when you said just

tech j when you said just technology, I I think sometimes people say something is missing in the machine, right? That that that just prevents it from having taste. For some people they they think of it as, you know, the lack of care, like the machine doesn't care. This is the way like a human would care about like a specific vision. For other people it's more embodiment problem or emotion problem. Do do you already have a I guess a mental model of why your machine is just

different from a human in in its capacity to have taste or not.

I mean the the one way of looking at this is like I've I've seen so many beautiful things made with AI. But that was made by humans that put a lot of care and detail in what was made. They had a vision, I would argue they have taste, and they put a lot of care into bringing that into reality.

So I do think that anything that we do that has a

that brings things to the next level has a a lot of care and effort put into it. And I think that's true here as well.

It is also true that AI won't come with

It's learned from everything that we've ever done.

I mean not everything we've ever done, but like we Yeah it's like

What's public at least on Reddit. Yes.

Yeah.

it's based on what has been done by humans. So what this means

Mm. Mm.

is like it's not gonna come up with something that we haven't seen unless there is a human directing it to go in that direction.

And so with AI you can make things that are completely new, but AI won't come up with that. Because you need a human to come up with their own original creative ideas.

Think

that's a fantastic segue for the other big topic we wanted to cover today, which is creativity. because before Luma you also had a a great career in another company which also cared about creativity. I think Toca Boca. I I don't know if you want to say a few words about this as before we dig into creativity, but

to so Toca Boca was a company that was started in Sweden and it was based on this premise that A, kids were going to use screens, B play is very important and everything that is being built or was being built in the early twenty tens was building for gaming and getting to the next level.

So the open-ended, the role-playing, the creative play was not gonna have a place in this new platform. And that was problematic if children were going to spend a lot more time there. Because that meant that unless there were products where they could role-play, construction play, creatively play, they would not be doing that because they would just be gaming. And so

We looked at the App Store from the point of view, like if it was a toy store, and we said, okay, which toys are we gonna put there? And it cannot be gaming because other people will do that. And it cannot be education because other people will do that. Which is like we we took away the two biggest groups, right? And we said, okay, but in this toy shelf, what are all the toys? And then we bought the physical toys and we invited kids.

And we looked at when them when they were playing with the toys, and we said, what are the underlying play patterns here? And which which are play patterns that could be enhanced by technology? And so in the early days, like one of the big successes was something called toca hair salon. It was then copied by Disney and everyone else. But that experience is actually better with technology because when you cut someone's hair.

you can then regrow it and you can start over. But where like

Ha ha

I my youngest daughter, just like me as a child, she has this she likes to cut her own hair a lot. I mean I I now I'm in the situation I can't be angry because my cousins tell me I did the same. So but she does that and it's it's like and I I look at her and I'm like, my God, she's so cute, but she looks like a crazy person 'cause there's just hair all over. It's much better if they do that in a digital product than in real life.

Yeah.

and and so that was a

Something that worked very well. The the last sort of ingoing hypothesis in this is that we we said it should be gender neutral. because and this is like a tension I feel more when I'm in the US, but play patterns are not divided into boys and girls. They're just not. And so again, coming back to Her Salon as an example, those characters they looked weird.

They looked funny. Some of them looked exactly like us, but some of them were animals. One was a Christmas tree. They were all over the place. These products wildly outperformed Disney, Nickelodeon, anyone that ever tried to compete with us. And you could say it was the craft. It was also that it was speaking to both boys and girls. And then I would meet hysterical parents that said,

I have a boy and he's playing with hair. Is that a problem? Like. No, of course it's not a problem. Like, it's not a problem. It's great, but would he, if he was sitting with like a princess alone and playing with hair, like A, would he ever want to do that? Probably not. B, would his parents freak out even more if he did that? Yes, they probably would in certain parts of the world. And so, the the last thing that was, it was like it.

We put a lot of effort on arts and design.

And we did that because why should we not? Just because we're designing things for children, should we make them not have taste from our perspective? Not have a signal of the arts? And so I think like the the first real

Mm.

wave of users for Toca Boca was parents that had creative professionals because they were like, my god, this looks good.

Interesting.

so anyway, that's what that's what I did for almost a decade.

And

I I'm gonna ask a question that's gonna sound too trivial for you because it's very obvious, but how does play help people be more creative or like coming up with new ideas? How how how is this formative?

So I think

My definition of

Creativity is quite wide. I believe that people that are best in their profession, whatever it is, are extremely creative. Because if you're if you're not able to think about things from new angles or be open to like rewriting the path of where things can go,

I think that's sort of like so the I think the best doctors and the best lawyers are very creative. Because they th they look at what is and they they think about it in new ways. I think play is very, very important for creativity. I mean, I can extend play to be important for other things as well, but if we stay on creativity, I think play is very important because what what research shows us is like there's five types of play.

There's playing outdoors, there is role-playing, there is creative play, there is construction play, and there is like learning and gaming. And what happens is like when we are kids, we are all we play across all of these. And then what influences us is a little bit is like in which culture and which household we grow up. But then

Mm-hmm.

there's also just what we are naturally inclined to play with. And anyone that has a sibling or has more than one child.

knows what I mean because they just see it play out in their own lives. I was very into role-playing, my brother was very into construction play. I also built a ton of Lego, but then I continued to play with it afterwards. Whereas he would build and collect. And and and so there's there's this thing that happens then when we grow up, which is that we

Are all pushed into like the learning and the gaming part, getting to the next level. We're predisposed to this from parental pressure, school pressure. School is all about getting to the

Mm.

next level. So that's sort of what happens. I think creative people have managed to, for whatever reasons, stay open in more of these. They were probably like naturally, if I think about the more visually creative people, they were naturally into one or more of these play patterns.

But then they were able to, either because of where they grew up or because they were just so stubborn they stuck with it. They found a path where they continued to have a creative outlet. Either because they made that into their profession full-time, which is like, you know, that's a a bunch of the users that we have at Luma today. They like, I'm making this my career. or they found another place where they where they got

to here have that creative outlet. But I I think if they had been pushed to not play as they continue

Mm-hmm.

to grow up, they probably wouldn't be doing what they were doing today. And they we probably wouldn't have as much great creative professionals around us as we have. Last thing I will say to my like wider definition of play. I think play is fantastic for just like general life skills.

Mm-hmm.

I mean, I could extend it and be like a little bit what we're doing here today is that we're role-playing.

Mm-hmm.

Like we're all some some version of ourselves when we go home tonight, and now

Mm-hmm.

we're here in a professional setting having a professional conversation. anyone who's building anything has been is doing some one version or the other of construction play. And anyone that is trying to

Visualize something that don't exist today in whatever profession you're in is doing some version of creative play.

If you have a positive outlook onto how AI will evolve, how can it maybe bring more people back to certain patterns of creativity that they are neglecting get driven out of?

I think I mean if I take very

concrete I'll I'll make this super concrete and this is like if I look at where Hollywood has been the problem has never been that there's not enough creativity in Hollywood. The the problem has been the business model and the current trajectory of like all production costs a lot and that means that safe bets needs to be greenlighted and that means that

We're watching more and more sequels.

Mm-hmm.

And so a positive outlook on AI here is that because we can now in the ideation phase visualize more things, because we in the production phase can reduce budgets for really bold ideas, more things can be greenlighted. And

That means that more people are gonna get a shot at getting things greenlighted from large studios. I'm convinced about this. It also means because we already have all of the distribution platforms that anyone needs, it means that.

If people that maybe before were trying to get someone like they needed to get something green-lighted from a big studio to make it happen, they can find other ways forward because the budgets look different and they already have the distribution. And so I would say this production itself will be solved. It's about the ideas, it's about having the stories.

And it's about getting access to distribution. Anyone who's willing to do the work on distribution today can get it. And it does it's not like locked behind the studios anymore. So that's sort of like my very positive of like what it can mean. What I've also seen is if I look at our partnership with Innovative Dreams, they that team were doing productions all across Europe for a long time.

For budget reasons, and they are now able to move jobs back into Hollywood. Now, I don't think all creative jobs or filmmaking jobs should be in Hollywood, but that is a signal of like they created hundreds of jobs in LA. And so jobs are gonna change a lot, but I think as an end consumer, in terms of the content that we're gonna see and in terms of job creation, there is a very positive outlook as well.

You mentioned there was a keyword, jobs will change. How do you if you have advi if you have give if you have to give advice to young people, how do they prepare? And maybe also to people in their career already? I think you're pretty strong in the future that it's not gonna AI take a take over everything. It's just that humans will do different things. What would you recommend them?

I would say that

I understand that there's a lot of fear because a lot of things are changing and so it's very natural for us to fear that. Mm. I don't think anyone can jump on a podcast and say, You don't need to be scared because again it's a per it's a personal thing and it's about how all of us handles change. And whether that

Mm-hmm.

changes AI or moving to a different place or whatever it is, like it's change and it's a personal journey.

What I will say is

If you haven't started to experiment, if you haven't tried this, if you haven't joined like a education Zoom call of how to use these tools and what you can do.

I think you're missing out. And I know so many people that are like, I can't do this because all my friends will hate me. And I can't do this because I'm afraid of my job and the place where I work has a no hard, no AI policy. Guess what? Your friends or your job do not own who you are or who you want to become.

Mm.

And you should just like try this out. Talk to people, find friends, be curious.

This is true in creative field, this is cru true outside. Like how I was doing my job last week is not how I'm doing it this week.

Mm-hmm. That's

And that's because

now I found a new a a new little thing that works for me. And that's exciting. So I was really excited. and and yeah, I think that's sort of like how I would view it. It's like just stay like put like don't tell this the hat that is scared or angry or whatever it's feeling that it's not allowed to be here. Just say like I'm today I'm putting on my curiosity hat and that's what I'm gonna do.

I wanna one one question I always have around this is the the the so sometime when people start using AI, they have this feeling of being a bit deceived by the machine. Like one I remember talking to quite a few sub stack writers, for instance, who were starting to use more and more AI in the writing process. And w one thing, for instance, they would a few people told me was they would write an article one day and they would on the moment they would feel like it's

It

looks great. Like the AI has helped them find good sentences. And then when they read again the next day what they had published, they realized actually that was very sterile or that was very bland and it's actually not me. And

Yeah. No, you need to be very careful.

And I I guess yeah, I guess what is the words of caution? Like how do you help people avoid this sort of bad experience?

And you need I mean, I think you need to audit

yeah, you need to audit yourself as well. And you need to audit others. Like I have this sense now where like I'm flooded with AI research documents.

Mm-hmm.

And what happens to me is like first time I read it, I'm always like, this was great. And so then I will tell the person, this was really good, good thinking. Then I read it again. And I'm like, What the f

Yeah, yeah. Yeah.

freaking fuck is this?

and I probably also create documents like this, but I'm like, this makes no sense. And I go in like if it's like a competitive like comparison and I'm like, these like this just sounds like AI travel. Like there's no this is no original thought. Like who's we're gonna take action on this? Like this is not a and so now I'm doing hard push of like, hey, actually like if you're doing research on competition.

I expect you to go in and use all of their products. so it's not that AI can't help, but I think humans also need to think and needs to question and needs to audit themselves and needs to audit whatever tool they're working with.

I think the the example of research I find really interesting because there's there's been quite a bit on research on this about why we think the AI research looks smart. And and I think in many ways it's kind of just hacking our perceptors, right? Like we we used to be in a world where if a document is very dense in information, we would assume okay, someone has, you know, done a lot of work. Or if something is, you know, really well written and, you know, perfect English again, we would assume okay, this is I don't know, a well educated person. And

And in a way the models are kind of, you know, passing for smarter than they are. Do do you do you think there is a similar aspect in the more, I guess, aesthetic use of use of AI? Like a g do you see it in image generation, in video generation where an image also has this first impression wow, this is very realistic, this is very realistic picture or something. It's very impressive. And then actually the next day you would look again and say, actually, this is slop.

I mean, y I I think it also has to do with like how much you have used AI for a specific thing. Because like

When I was doing research, I don't know, a couple of months ago, I think I was more optimistic. It's almost you know what it is? And I think it's the same thing as like if you're the first time you come in and you start generating images with AI, you're like, Whoa, I can do all of these things. We see this with hybrid a a lot. When people discover hybrid and like the ability to upload a video and move that person into a new environment, they're like, Whoa, this is so exciting! And it it means I'm still gonna film and I'm gonna have my crew and like

This is fantastic. And then like they do that. And then it's like, but now I actually want to like be able to do this specific thing. So I think part of that is just we get used to technology. Same thing with like first time you took an Uber, it was like, whoa, I just went to Mars. And now it's like, why is this card taking so long? Like this is unreasonable. so I think that's always true with all technology, but I do think that the other thing is like audit yourself.

If you're really good and you've been a

Director, an art director, and editor, like all of your career, you can get these tools to do magical things because you can push it into a place where most other people cannot. And so you need to like stay true to what like if you're a Substack writer and you're good at that, like you can audit yourself. Because otherwise, like you're just subject to AI writing Substack.

When I when I hear you and I I think this is a theme that came up now two or three times, it's you what you're suggesting is that everyone has to become more of an editor compared to the roles they took up before, right? If you are they have to be more on this I'm evaluating and I need to judge it more holistically than from the more narrow perspective that I had. Do you agree with

that?

Yeah.

Yeah. I think yeah. I think you need to n to audit it like from a very specific angle. And

And when you're okay with like I think it's like the relationship a bunch of us are having with Claude. Where it's like, I just need to move things forward in this direction and it's like directionally need to be a correct. But like if you're like I need specifically this thing, like you need to push it and you need to know where you're what you're pushing it for.

one of our previous guests, Eleanor, also made this comparison that you're creating something w which superficially looks like nice class, but if you're touching one point basically, suddenly it starts breaking and you're seeing that it's unraveling, essentially. That seems to be one of the the themes

Yeah.

we have also seen with we talked also to researcher with an episode there. Very similar thing that this thing is it is hard to basically do the specific thing without breaking the vault, the full thing.

Yeah. And and it's sort of like

That's kind of part of the game, like a little bit. Like how

Yeah. Yes.

l how long can you push it? How long can you rest it?

Caroline, we have already taken up a lot of your time. thanks for this. I have if or if okay, one last question. What is Luma developing towards? What's the next thing? And if maybe if you can over reveal something what you're working towards.

So we're continuing to build on the research trajectory of our creative models. Our creative agents are really like they're so powerful in helping in some of the workflows and like creating repeatable workflows. That's like a huge thing of just like

Okay, like now I've done this storyboard. Like I want to be able to implement that's a huge thing else, like the repeatable workflows. And then we also recently announced that we have an open research lab where we're focusing a lot of efforts on robotics and how this helps the creative flow, one might ask, is for robotics to work, you really need to have a good understanding of the physical world.

Mm-hmm.

And a lot of the creative problems we're seeing today in the f on the foundational layer is like the understanding of the physical world. So like these two things actually do converge.

Perfect. We

Interesting.

are looking forward for more. Thanks a lot for being part of this.

Thank so much.

Thank you very much.

Thank you. Thank you very much.

Thank you so much, Caroline.

Have a great evening, guys.

Thank you.

Thank you. Bye.

Thanks for watching. We've got more episodes in the oven, so please like and subscribe to get notified when they're out. And please check out taste-bench.com to read about the taste benchmark and the creative engine we're building towards.