- 1 hour ago
AI unicorns such as Midjourney and Stable Diffusion aren’t competitors many would dare to go up against. But Victor Perez thinks he has a chance to stand out with his $83 million–backed, one-stop-shop visual tool, Krea. Launched in 2022 with cofounder Diego Rodriguez Prado, Krea makes it easy to generate and edit images and videos. Marketers, photographers, game developers and architects pay between $10 and $60 a month to turn their rough sketches into photorealistic images—think transforming a simple white triangle into a mountaintop. But in the crowded market of AI image generation, Krea’s taking it one step further with customization. Enterprise clients like Shopify and Perplexity train and fine-tune their own Krea models to fit their taste and preferences. That way, anytime they want to generate an image or render a new product, Krea knows their aesthetic and implements it into the results. “The novelty that comes out of creating with AI is not just empowering people who are already creatives,” Perez says, “but enabling people who didn’t think of themselves as creatives to express their ideas in a much easier way.”
Read the full story on Forbes:
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Read the full story on Forbes:
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Category
🛠️
LifestyleTranscript
00:00I don't think that is something meant to replace.
00:02I think that is something meant to allow creatives to have more possibilities.
00:06And I also think that requires a new kind of skill.
00:17Victor Perez, thank you so much for joining me today.
00:20Thanks for having me.
00:21So you are the co-founder and CEO of CREA, an $83 million backed AI creative suite.
00:26But for people who have not met you before or don't know much about CREA,
00:29how do you describe it to someone?
00:32CREA is a creative suite that makes sure to bring whatever the latest research is on the
00:39on the AI space to creatives through very intuitive tools.
00:41What are some of those tools that people are most using today?
00:44I would say the image generation tool, video generation tool and the enhancer
00:48are definitely the three that people use the most.
00:51And the fourth would be the real-time tool.
00:53That is one of the ones that we recently changed and people are enjoying it quite a lot.
00:59And you guys are working across many, many mediums right now between the products and
01:04offerings that you guys have and also the customers that you guys have.
01:07You are also a creative yourself.
01:09You have a background in music and many other things.
01:11Can you tell me a little bit about your experience as a creative growing up
01:14prior to launching this company?
01:16So I guess on my first, like the first creative things that I started doing were related to music.
01:21I started playing the classic guitar at a quite early age, inspired by my father.
01:26And from there I got like really interested in music in general.
01:30So I ended up like creating a band and that was kind of like all my teenage years were around that
01:35music band.
01:35And through the music band, I started getting into photography or I started getting into graphic
01:40design, mainly everything for the band.
01:42But I started like exploring many different things.
01:44And then when it was a time, I was like 17, 18 years old and it was a time to go to university.
01:50That was the moment where I like, I was like really close to go to, to study like professional,
01:57like classical guitar in the conservatory in Barcelona.
02:01But I ended up like choosing to do like an engineering that it was kind of like a middle ground
02:04between music and computer science called visual systems engineering.
02:08And from there I got into coding, I got into AI.
02:11And that's where the both walls that I, that I was like interested growing up merged together.
02:17And that's where kind of I started to mess around with creative uses of AI.
02:22And that's like how CREA ended up like coming from.
02:25For that audiovisual degree that you have in the school that you went to,
02:28what are some things that you learned there that maybe has changed over time?
02:31I feel like so much about the creative industry has because of technological advancements really changed.
02:36So being a student in that kind of intersection, what are some things that you learned there
02:40that really either inspired you to keep building or that were just like shocking to you?
02:45I think that more than the things that I learned are the things that I don't learn that inspire me.
02:49Because I was, I was like reading a lot of papers on the internet.
02:52I was like looking a lot of code bases from researchers all around the world.
02:57I was looking what they were doing and I was looking at how deep learning, that was like 2017.
03:02So deep learning had been around for a while, but there was not still like the strong belief
03:06that it could acquire some of the capabilities that we've seen, that we are seeing nowadays.
03:11So I guess that it was me looking at all of this happening and like seeing how teachers were
03:15not really teaching us those things.
03:17So I guess that it doesn't really answer your question, but I think that I got inspired
03:21more by the things that they didn't taught us than the things that I learned in there.
03:24Yeah.
03:25So coming from a creative background and seeing maybe what you know can happen in the industry,
03:29what you hope to happen in the industry, but that not matching up with what you are learning.
03:33Walk me through then that decision to start building a company around it.
03:36How did you come up with this idea? What was kind of the, you know,
03:39what, what pushed you in the direction of wanting to solve those things on your own?
03:42There was not a plan. Like it was not something that I,
03:46like I saw a problem and I started a company to fix a problem.
03:50Everything was kind of chaotic and kind of serendipitous.
03:53So I, I got like really interested into, into AI and into coding coming from a creative background.
03:59And I think that that was something that there was like a small movement around people that came
04:03from similar places from where I was coming, that it was growing, especially on Twitter.
04:08Like there was a lot of people that either learned how to go to use AI to generate visuals,
04:12to generate videos or even like do interesting stuff with, with text.
04:17So at the beginning, I, I got like really interested in this like small community that it was growing
04:21around people, um, kind of like researching and kind of like coding and doing like creative stuff
04:27with AI. And as I got deeper and deeper, what I realized is that most of the people that they were
04:33in the space, they were more programmers than artists, if that makes sense. Like there was definitely
04:39a mix and I can see that mix in myself, but I know that naturally myself, my brain works better
04:45for computers and works better for the kind of challenges that computer science has,
04:49even though I love art and I love creative tools. And as I got like deeper and like I saw like really
04:55cool things happening, like I really saw like a, like a monotony around like the things that people
05:01were trying. And I was like, I, I guess at that moment I also saw that there were not a lot of tools
05:06for artists who didn't know how to code, uh, to use this technology. So as a side project, not thinking
05:12about a company, I started building interfaces, like up until then my background was first creative
05:18as a kid, then research, I get research growing up in university. Then I switched towards interface
05:24and kind of like figuring out how can we build like an interface for using this technology that
05:28is completely new and that we don't really know what it's like, how is it going to be used apart
05:33from experimental visuals, but it's so cool that I really want to want to see what artists can do with
05:37it. So I started like building side projects and one of these side projects, we got, um, investment
05:44from, uh, from a venture capital called H of zero, uh, from the U S uh, also in a very chaotic story
05:50around my co-founder stealing my idea and us flying to Miami. So, um, I was, I was in, I was in Barcelona
06:00and I, I mean, I guess that the, the whole story, it's, um, Diego, who's my co-founder was in New York,
06:07we both got like one of the most prestigious fellowships to be studying in New York. He was
06:11already in Cornell and I was supposed to go next year. And while he was at Cornell,
06:14I think that he was more bored than what he expected to be. And, um, he set himself up to
06:20like improve his communication skills. Uh, and the way how he approached, like the way how he tried to
06:26improve his communication skills was through trying to convince people to give him a job, which was
06:32kind of, kind of, uh, funny. But so he applied to Stripe and he got a job at Stripe. He applied
06:37to Jane Street, uh, without really knowing much about like what they were asking. And I don't know
06:42if he got that job, but he was applying to many things. And one of these things that he applied to,
06:46he was H of zero that he was a startup accelerator and he applied, he got into the interview. First
06:52thing that, that they asked on the interview is, what are you building? He blanks out. Uh, and, but it
06:59turns out that, uh, one or two weeks before I was teaching him everything about this side project
07:03that I was building. I recorded myself for six hours, like going through all the code and like teaching
07:07him everything about like this project that I was working on. So he knew everything about, about it.
07:12So he blanked out and he pitched my whole, like my whole side project, like my whole little
07:17interface that I was building at that point. Investors liked it. Um, right after the meeting,
07:22he calls me freaking out. Yo, yo, yo, like I just did this. Do you want funding for your idea? I didn't
07:26want to sell it. I just like blanked out and I like sell it out for you. If you want me to stop,
07:30tell me I'm going to stop. Uh, but if you want, like I'm getting to the, to the next, uh, stage of
07:36the interview. And I was like, no, no, no, I go for it. Like sell the out of it. Uh, so he did. Uh,
07:41and on the second, on the second interview, people like they, they played with the interface and,
07:46and the investors liked it a lot. So they invited us to Miami and then he called me again. He called me
07:51again. He's like, yo, can you get a flight in 24 hours to go to Miami? They want to spend a week with us.
07:56And we did that. We went to Miami during like the whole web three boom. We were like the only
08:01AI startup in there. Those were like some crazy times, but that's how actually like Korea got,
08:05got started. And what year was that that happened? 2022. I guess at end 2021, that's when this
08:11happened and early 2022, that's when we started with Korea. Okay. Gotcha. And so from just 2022 to
08:16now, or like you're talking about in these kind of like Twitter ecosystems, people just starting to
08:20experiment with it. So much has changed in AI from then to now and like its capabilities and what people
08:24are using it for. What have you noticed from your perspective in terms of people's use cases or
08:28why they're coming to an AI platform in general, and then how Korea kind of taps into that?
08:33Sure. I mean, this technology got orders of magnitude better than 2022, 2023, and it keeps growing. Like,
08:41I feel like we're still at the, at the very beginnings of, of the capabilities that this technology is going
08:46to have. But I feel like in 2022, 2023, after Dali 2 was released, I think that it was the first time
08:53that it switched from a toy to something that it can actually serve for certain creative use cases.
08:59And I think that the, the main way that people are using it right now, or that creatives are using
09:04it right now, it's to, to get ideas. Like, I think that is a great replacement for something like
09:09Pinterest or like they use it a lot for creating mood boards. That's on the one side. And on the other side,
09:14I think that is also like all this novelty that comes out of creating with AI, that is not just
09:19empowering people who are already creatives, but also enabling people that they didn't think of
09:24themselves as creatives. And that now through this technology, they can express their ideas
09:29in a much easier way. Yeah, that I totally understand that. So for Korea specifically,
09:35what are some of those specific ways that people are using the platform today? How, and how is that
09:39different from when you first launched in 2022, like on a very practical level? Sure. So in terms of users,
09:44I would divide them into one is a professional creative. One is the non-professional creatives,
09:49kind of like the hobbies. The professional creatives is paying for the subscription because we are
09:54making their work more efficient. Examples of these are like, again, like people like doing,
09:59doing mood boards, like getting inspiration because, um, you type like a very simple prompt and the AI can
10:05give you four different ideas or eight different ideas. You can pick some of these ideas, rephrase a
10:10problem or add like some other conditioning into the model and get something new. And there's this
10:15conversation between creative and AI that is very powerful to get ideas. And we see a lot of creatives, um,
10:21getting inspired by like having a conversation with AI. And then on the other side, like depending on the tool
10:27inside Korea, we're like making people efficient in different ways. Like the enhancer is used by a lot of architecture studios,
10:34uh, in processes of, uh, uh, enhancing or upscaling. Like they come in with very basic 3D renders.
10:41They pass them through our AI upscaler and they get like 4k images with an insane amount of detail, like textures
10:48being extremely realistic. And this is like a process that they used to take them. It was extremely expensive
10:53and it used to take them hours. Now they get it in like a minute or two. Uh, or we also have industrial, uh,
10:59designers coming to the real time tool and being able to connect like their 3D platform, like their 3D
11:06tools that they use like blender or, or cinema 4d. And again, like out of like a very simple,
11:11like 3d shape and out of like very like, uh, basic doodles, they can, they can like imagine how
11:18a product could look like. Like we see people like, um, designing chairs or designing like helmets or
11:23designing like, like different kinds of things in real time with like extremely realistic renders happening
11:28out of like a very, uh, basic shape. Uh, so we, we see a lot, a lot of efficiency happening for
11:33professional creatives and on the other side for the hobbyists, I think that this is more of an
11:38enabling, like an, like they're paying for the subscription because they are able to do something
11:43that they were not able to do before without AI. And I think that some of like, most of them are just
11:48having fun. Like, um, having an idea and being able to see a, uh, like representation of it in an image
11:54or on a video, it's, it's very fun. Like it's kind of like in the same way that people get into
11:59photography because they have fun, like going on the weekend industry and taking pictures.
12:03There's like a similar factor, uh, here going on with non-creatives and non-professional creatives.
12:08So you guys are across photography, videography, rendering for architecture projects or, you know,
12:14products that people are going to build. What is the technology though, that makes all of this
12:18possible from the back end for someone who doesn't know much about AI or the coding that you guys have had to,
12:24you know, kind of go through to build this. How would you explain how the technology actually works today?
12:29So there's not one technology that is doing all of this. Uh, at Creo, we serve like probably like 20,
12:3430, 30, like different AI models. And each AI model is specific to a particular use case or to a particular
12:41workflow. But the way that I would explain how AI's work is that they look to a lot of images,
12:49they look to a lot of videos, they learn about them. Like they learn like the patterns that form them.
12:56And after they have this learning, they can, uh, what like, like they can generalize, which means
13:03that even though like, like what would be a good example, like I've seen a picture of
13:09a chair and I've seen a picture of an avocado.
13:13And now I can create a
13:16chair with a picture, with the shape of an avocado, even though I've never seen a chair like with the shape of an avocado.
13:22So there's like this factor that happens with design models. That is what brings creativity.
13:26And it's what brings like the interesting emerging capabilities that these models have.
13:30That is what makes them, in my opinion, very interesting for creative use cases.
13:34Have there been any use cases that you guys have been really shocked by that early users
13:39were testing out or things that they told you, we really want this. Like, can you build this on
13:43the platform that kind of changed what you guys were offering?
13:46I think that one, one of the, of the best examples for these, it was like the AI trainer,
13:51if that makes sense. I feel like, um, I personally, and some other, and some other people in the team,
13:57we thought that this idea of training your own AI model would get obsolete relatively fast
14:04because of how complex it seemed. But we saw the, the community, like the hardcore AI community
14:12doing this by themselves. And they curated like their own data sets and they like train like their
14:17own little models with, or they fine tune their, their own little models with these, uh, data sets.
14:24And they were able to get like really good results and a lot of controllability, uh, for, for, uh, these models.
14:29And we thought that this was not the way that creatives would use this technology. It didn't feel
14:34intuitive, but people were asking for it. So we ended up like building it and, uh, we ended up like
14:40building the, uh, what we call it, like create train. You can like upload a bunch of images on a
14:44specific style of a specific person or of, um, of an object. And, uh, the AI can learn that style,
14:50person, object, or pretty much like whatever you put into the data set. And it turned out to be one of the
14:54most sticky and successful, uh, launches that we've made in the platform. And it turns out that
15:00creatives are interested in learning about the technology. Like, I feel like there was like a,
15:05also a learning from our side that, oh, wow. So like for creatives, it's so important to have this
15:11controllability in AI models. It's so important to have that specific style, that specific coloring in
15:17the images, or like a specific composition on the images that they create, that they are willing to
15:22learn all this process around like getting a data set, tuning a lot of like complex parameters that
15:28are like almost coming from research and do like 10 different runs until they have this model. And
15:33once they have this model, they create all sorts of images with this. And I assume that that like
15:38branding capability really helps your enterprise enterprise clients as well. Cause you do have,
15:43you know, those individuals or consumers that are using it, but also major enterprises like
15:47perplexity or Pixar or Shopify that are using the platform as well. So is that how they
15:52are most often using it or how else are these like larger companies using the platform?
15:56That is the number one feature that these kind of people love. And that is also like what it's
16:02starting to show like, like a new kind of creative workflow with AI that is like more complex than
16:07just like typing a problem and getting an image. Like we see people, like we see design teams at
16:11perplexity, for example, or Shopify training their own Loras for different campaigns. I mean,
16:17and Loras is like, um, like, um, fine tuning of, of one, one of our AI models. Um, so they,
16:25they fine tune these models for a specific campaign and then they start combining them. And that's how
16:29they get to the level, uh, that, that they have right now when they like put a post on Twitter or when
16:36they make like an ad that is like fully AI generated, but it doesn't really look AI generated because of
16:40how good it is. So yeah, like, and we were, by the way, like very surprised. Like when we started,
16:44like, we just like recently started to talk with enterprise customers and we thought that they,
16:50that they of course, like wanted to stay ahead of the curve in, in, in the sense that they want
16:55to learn about this technology, but we didn't know that they were already so knowledgeable.
16:59And that when they come to the platform, one of the first things that they ask is, okay,
17:02but can I train my own model? It's like, all my creatives are already using this and using that.
17:06And they are already like having like so many models for, for us. Like, can they do that in Korea?
17:10And we're like, yeah, like actually, we can make it happen.
17:13Yeah. No, the AI community is growing so much, like throughout so many different industries.
17:17I'm curious to the, the concept of like, especially from coming from a creative background,
17:23having taste and like bringing that to the forefront of what you are creating is really,
17:28really critical for the success of the outcome of the creative project. Does AI understand that?
17:34Or how do you implement taste or like human creativity into these things that people are then,
17:40you know, typing up or the prompts that they are creating?
17:42Right. Like that's why Korea exists. Like Korea exists because taste, it's a new form of IP
17:48and you need tools in order to like put this taste in an AI and have an AI that is like effectively able
17:56to reproduce that taste or that specific idea that you have in mind. And again, like through this whole
18:02process of like training a model, like training your own model, that's exactly what you're doing. No,
18:06like you're by curating a data set, you're putting your taste into it. And once you have like that
18:13taste, like that visual, like that, that specific thing that you're looking for in, in this model,
18:17you train it, it's not going to work well on the first shot. So you will need to do like a few
18:23iterations until you get it. But after you get it, you can produce like that, that, uh, like that visual
18:29that you were looking for. And I think that these kinds of workflows show how creating with AI,
18:34it's not just about typing a prompt and magically getting an image. There's a whole workflow and
18:39there's like a whole process for people to really get, uh, like production level, uh, results out of it.
18:46Just like a creative has drafts or many iterations that you go through when you're painting a picture
18:50or writing something as a human as well. Exactly. And for these workflows, like you need tools.
18:56Yeah. Like you need to, to have like very, like you need to have thoughtful people thinking about
19:02what research, like what things need to be abstracted and in what way and what things need
19:08to be exposed to the creative and how this communication between an AI model and a creative
19:12needs to happen. And how does the, like how does the tool allow you to effectively communicate with,
19:19with the AI. So what are your thoughts on this conversation around AI replacing artists or replacing creatives?
19:25How do you see this conversation evolving as we get more and more in, you know, involved in what AI can
19:30and will do? I think that the conversation is like definitely softer than a year ago. I feel like people
19:36are more openly adopting this technology, even on like famous creative agencies, famous companies like
19:42Shopify. Um, I see like, like a trend about like people being more open about them using it because I
19:48think that there has been like a, I don't know if it was a majority, but definitely like a silent group of
19:51people that they have been using this technology and not telling anybody. And I think that the,
19:56that what is going to happen, it's a shift rather than a replacement. And it's a shift similar to
20:01what happened when we switch from drawing logos to making logos with illustration, like through like
20:06computers or like similar to happen with CGI and like, uh, I don't know, like being able to, uh,
20:13like CGI or, or, or, or photography, for example, instead of like throwing a portrait now, you can
20:20like take a snapshot of it and that creates like the job of a photographer. I think that in this case,
20:25like what we're going to see with AI is a shift and we're going to see how there's new kinds of formats
20:30that appear out of this, uh, technology that they become extremely powerful and extremely popular
20:36that needs people to have like a lot of understanding of AI, like creative people to have a lot of
20:40understanding of AI in order to be able to succeed in this new kind of form, this new kind of formats
20:46that are going to exist that underwants it. And then like working with like AI at the end of the
20:50day is like a medium. It's not like a magic thing that gives you everything that you need for your
20:54marketing campaign. You always need the creative behind and you need a process for the AI to give
21:01you something that you like or to be able to have something tasteful, something that really looks good.
21:06So is this going to change the jobs that people have or how creatives come
21:09into the workplace or what might that look like in the future? Like I think that in the future,
21:13we, we, we will see like job postings for AI creative or job posting or, or maybe like the
21:19requirements of a, of like a creative job, like include AI. Yeah. And do you think that artists
21:25are excited about that? Has there been any like skepticism or for, from the people that you're
21:29talking with, is there fear around that shift happening? I think that most of the people that
21:33I personally talk to, they are all quite excited. They see this as something that is empowering them
21:38rather than replacing them. But I know that there's like a lot of discussion around people feeling
21:43like they are getting displaced. I think that what is important is to like adapt. I think that this is
21:49something that it's better than the previous generation of creative tools that we, that we had
21:55and that people will need to adapt. And I think that at the end of the day, it's like, I, I don't think
22:01that it's something meant to replace. I think that it's something meant to allow creatives to have more
22:06possibilities on, on, on creating like with formats that maybe they could have been, they could have
22:11never been able to create. And I also think that requires a new kind of skill. Like I think that
22:16as a creative, something that is going to be very important now is that you have a taste, that you
22:21have an idea. And there's like a lot of parallels to be drawn also around with, with coding. Like AI is
22:27like much better at coding than generating images. Yet there's not this like crazy pushback from coders,
22:33as there is from, from artists. I think that this is also a lot about like the, the interfaces that
22:38coders have and the work that coders do versus like a work that, that an artist do. And they're
22:44like completely different industries. So I understand that the, the artist community is like more
22:48affected by this, by, by the coding. But I think that there's like, you can do like, you can prototype
22:55an app as a coder and you can like by code it and get something create out of it, but it's not going to go to
23:00production. And I think that is like very similar to what's happening right now with artists. Like
23:05you can create a video with AI, but I doubt that it's like going to go to production right now.
23:11But I'm sure that some of the new films that we're seeing right now have a lot of AI processes
23:16and a lot of AI spark new ideas and a lot of AI, um, allowed them to move faster on, on a lot of
23:23different processes. And what are investors thinking about this? You guys have 83 million in funding,
23:28correct? What have those conversations looked like? What are they most excited about you guys
23:31developing or growing throughout all these different industries?
23:34So at the beginning, I think that our, our strategy of focusing first in product
23:38and then going down, down the stock to train AIs have turned out to be very successful. And I think
23:45that investors like that, like investors like that, the first thing that we care when building, when
23:51creating a strategy is the user. And that first, like we became a sort of like horizontal platform and we
23:57made sure that anything that is out there is available to the user through a great product.
24:02And only after we have that in place, we go and build our own infrastructure to serve these models.
24:07And only after this infrastructure is stable and is in place, we go and train our own models.
24:12And I think that what they like is that we're like so horizontal, but vertically integrated at the
24:17same time. And I think that they see a lot of potential around us becoming one of the first,
24:21platforms that is like fully, like vertically integrated with a laser focus on creatives,
24:28because the things that creatives like, or the things that are needed for creative workflows are
24:33very specific. And we are really, really excited to make new AI models for, for, for those. And,
24:39and most, most of these, like most, most of the needs that I think that the creative community needs
24:44from AI right now, it's, it's around controllability and aesthetics, like really being able to control
24:50these AI models and how can we make these models even more controllable. That's one of the main things
24:55that we are like researching right now. And secondly, like, how can we have models that truly understand
25:00like certain aesthetics or that truly, that truly understand aesthetics and that truly understand you
25:05as a creative. I think that this personalization layer, um, that, that, that we can build right now
25:10through our platform is something really, really, really interesting. Uh, and I think that that's
25:15what we're excited about and our investors were also excited about, and that's why, uh, why they,
25:20uh, what they decided to, to invest. So a lot of this conversation around AI has also surrounded
25:25sustainability and how things like data centers or the, the amounts of water and just like energy that
25:31has to go behind AI is potentially causing problems in the future. How are you guys thinking about that?
25:36Is that something that you guys have that on the top of your minds? Right. Um,
25:40I don't think that we are at the level ourselves where we are using an insane amount of compute
25:45at the moment, but I understand that for, especially on the LLM space, I understand how this is more of
25:51a concern because we're talking about like hundreds of thousands of H100s rather than a few hundreds or,
25:57or a few thousands that we're using for, for our models. So I, myself, like, I haven't really like
26:03thought deeply about it. And I guess that the, the first thing that comes to mind, it's thinking about,
26:09okay, like we're investing like all these resources and this clearly has like a sustainability impact
26:14to our society. What are we getting out of it? Like, is this going to pay off? Like,
26:18is this going to be good enough for all the expense that, that we are, that we are having now,
26:25nowadays to train this HCI or, or, or whatever it is that, that we're trying to build.
26:30And the answer is that, I don't know, I think, but I think that there's a potential of, of it having
26:35like a massive impact in many other industries that maybe that impact makes us orders of magnitude
26:40more sustainable. Like on the one side, like on the, on the medical side, just like the fact that
26:46through AI, we can advance, we can advance medical research and do things like cure cancer. I think that
26:54if we're able to go through AI, fuck yes, like this was, this was, uh, um, worth it. And I don't know,
27:00like if through AI, we can find like more sustainable materials to build cars or to build
27:04spaceships. If these, uh, technology can help us make, I don't know, like, like better, um, better
27:12infrastructure systems to delivering things on Amazon or delivering, uh, or, or like anything that is
27:18like right now polluting the planet. Can we optimize it through an extremely powerful AI system in order to be
27:24more sustainable? If the answer to that is yes, I think that it's going to pay off. Uh, it all depends
27:29on where, where we're able to, to take this technology. So maybe it's a means to an end and
27:33that might be worth it then as long as we, you know, use it in the right ways. Yes. So moving into the next
27:38like one year to five years down the road, what are you most, most excited about with CREO? What are you
27:43guys looking to continue developing? What's next for you guys? So I think that the space is extremely,
27:49extremely early. Like I think that it's like, if sometimes I hit, I think about the, the hit rate
27:55that we have with media models versus the hit rate that we have with LLMs. And I feel like LLMs seem
28:01that they are giving, they are like 80%, like they're giving you the right answer 80% of the time for 80%
28:08of the tasks, which I think that is like a massive hit rate. And we are not there with on, on the media
28:13space, like video image, audio, 3D, like all these models are still like very, very, uh, behind
28:19if you compare them with LLMs. So I think that, I think that a lot of progress in the technology
28:26is going to happen. And, uh, things like reasoning need to appear in the, in the, in the AI space,
28:34uh, better multimodal architectures need to like, I think that, that they're also going to appear and
28:40they're going to make it. So the communication between the artists and, and the AI model is going
28:44to be so much better. And the capabilities of these models is, are, are also going to, uh, improve
28:49orders of magnitude. And, and I, I'm like thinking about the future with like, uh, like an open mind,
28:55because I think that every time that there's like a new breakthrough in the industry,
29:00there are things that I didn't expect to happen. Like, and, and I think that this like still going to
29:05happen, uh, as these models get better, get better. But for example, like I, with video models, like
29:11something that I didn't think about before seeing some of the, of the technology that, that we have
29:17like nowadays, or that we like started having like a year ago, is that a really good video model
29:23can also be like a, like can give you like a 3D experience and essentially like something that it's,
29:28that can render a video. And that you, if you add like a layer of controllability, if you can move
29:33around that video, even though everything is generated by a video model, it gives you like
29:36that 3D understanding or that 3D experience. And I think that these kinds of emerging capabilities
29:43of these models are going to keep happening. And I think that they are going to bring new ways
29:47of interacting with this technology. And I'm excited to put CRIA in a place where we can create,
29:53like as a research lab, like explore what capabilities are possible and what can we,
29:58like what can we create, like what capabilities we're excited about and how can we train these
30:05models to have them. And, and on the other side, like how can we bring them to like very intuitive
30:10products to the creative community. Well, that's a great place to end. Thank you so much for joining
30:14me today. I'm so excited to see all you guys create yourself and then enable other people to create as
30:18well. Yeah, thank you.
Be the first to comment