Hybrid Realities Lab - SOOT-Talk - Week 1 Session 2

A diverse group of individuals, including artists, filmmakers, painters, sound artists, and fashion students, introduce themselves during a virtual studio program. They discuss their backgrounds, interests, and the intersection between physical and digital art. The program was planned a year ago and is led by Chris and Mary, with assistance from Mattia and Lynn. The group is excited to share their work and collaborate in the program. I'm not going to be sharing publicly the recording for the next couple of weeks, that would be nice. But, yeah, actually, I'm curious to just hear a little bit about people's backgrounds and who's in the call. I know it's a super diverse group, but, yeah, we're really interested to know who you are. First of all, we really appreciate you coming in to speak to us. I'm just, I can't wait to hear what this is all about. Obviously, I've seen some guys, the bit you've got online, but I'm interested in seeing, I think you've got some other stuff that's not released yet. So in terms of this recording, it's just for us, it's just for our purposes at the moment, so it doesn't go anywhere else, just for our reference, so people can, if they've missed anything, they can come back to it. Basically, Chris and I set this up about a year ago. We planned it on the back of some research and residency programme. We did with a group of other people and it took a lot of planning, but we got here in the end and we got, we were joined by Mattia and Lynn, who are helping us to deliver this programme, basically, a virtual studio programme. So in a nutshell, that's kind of where I am. I don't know how well you know Chris. Yeah, Chris, if you don't mind introducing yourself as well, that'd be great. Yeah, for sure. So yeah, I'm Chris. So Mary was in here, I know Mary, and I also know Noah. Oh cool. Yeah, yeah, and I think so does Basil, who's in this call. I think they met, I'm not sure. Anyways, yeah, so I'm what you call, I'm a PhD student in Buffalo. I do like new media stuff, like do a lot of media education now, but also, you know, my own personal work is sort of about spirituality and virtual worlds. Cool. And I lived in Ireland for like, you know, four years, so that's where I met Mary. Cool. Yeah. Yeah, yeah. Really? Was I a drop? Yeah. Yeah. Yeah, yeah, but I don't know if, yeah, I don't know if we met a drop. I'm trying to remember, but a lot happened. We were there, alternative years. What is it? Oh, alternative years. Yeah, yeah, yeah. Drop is a residency that Mary runs on a tiny island off the west coast of Ireland, which is really, really special and fun. Everyone should check it out. Cool, and yeah, if you guys don't mind, just like a quick minute, like from each of you, maybe just like popcorn to Basil to kick it off. I'd just love to hear kind of like what your interest is, what kind of stuff you're working on. Yeah, hi, I'm Basil. I'm usually just, usually just like a painter. I like to draw and paint acrylics. I don't have a lot, I don't have a big educational background. I've just been drawing and painting with friends and weaseling my way into networks for many, many years, and I think my interest in this residency and figuring out more of my relationship to the digital art world is influenced by an interest I have in familiarity between experiences and now like, you know, digital showrooms, digital connections artistically seem to kind of be the new frontier. So I'm trying to figure out how I can put my tangible physical practice into a digital reality with everyone else. Cool, cool, cool, cool. Nice to meet you. Maybe popcorn, Nadira. And feel free to just like pop around, like, you know. Yeah, Nadira. That sounds perfect. Hello? Can you hear me? Yeah. Okay, hi. Yeah, my name's Nadira. Well, I'm based right now in West Yorkshire, so that's the north part of England, but I'm from London. Yeah, my practice kind of just involves me, like, using my, kind of not using my family, but they're my inspiration. Like, during the pandemic at the time I was in uni, and they inspired me to make, like, a physical and virtual world that they're like, they've got these powers, and they're like, kind of navigating themselves and their stories to do with, like, Afro hair and Black beauty and stuff. And then I just like making, kind of, like, so I'm really nervous, making, like, sculptures and stuff that are 3D, like, stuff with a lot of stuff. And then I try to put it into, like, a digital, like, art space or virtual art space. So I like coexisting between the two, and, like, exploring the idea of how I can have a space in both worlds and stuff. So, yeah. Cool, cool. Do you want to add something? Dani? Hi, how are you? I'm Dani. I'm from Argentina. I started in music, improvisation, trombone, and at the same time I was working with, audio editing and recording. And later I started working with video and filming, and kind of mix all these things together with editing, with collage, started to work a little bit with painting. And now I'm trying to find a new way of putting all these things together with video and installation and all this kind of stuff. Cool. So, maybe, Alonia? So, I'm based in London. I come from the fine arts world, and I am interested in spaces that kind of traverse the physical and the virtual. So, like, with performance. I knew it was physical sculpture, but then I also worked digitally. So, I worked with video art, and then also I used Cinema 4D mainly as a program to do animations and building worlds. And then more recently I also dived into ZBrush and probably working a lot with sculpting, with digital, like sculpting digitally. Or maybe translating some of the things I did physically before, digitally now. But I'm more into an organic approach to working digitally. And then I'm interested in maybe bringing some of these things back into the physical. And then I also incorporate some of the physical things I do. I also do painting into the digital. So, I'm always interested in that sort of free space and also creating relationships between those realms, basically. Oh, wow. That's super cool. Yeah. Hi. My name is Alexa, and I'm a filmmaker and 3D animator. I mainly make experimental animated short films, visualizers, and 3D motion art. I'm currently a freelance 3D designer, and I'm based in New York City, though I grew up in Los Angeles. I'm originally from Los Angeles. I moved to New York last year. And I would say my work is really, really stylized and character-centric. I focus a lot on these very sort of beastly, femme, female characters and creatures that I like to sculpt digitally. And then I place them in the assortment of, like, vast wastelands and really stylized worlds that I like to build out. And I usually use Cinema 4D as well, and I use Unreal Engine a lot, too, in my work. Oh, cool. We'd love to see you. Yeah, come by the studio sometime. We're just down in Water Street. Oh, super cool. Definitely will. Thank you. Yeah, yeah, yeah. How about Lynn? Yes. I'm also part of the, yeah, what do you call it? The people who teach you and lead the course. I'm from Luxembourg. I'm a media artist. What do I do? I work mostly with video, also installation, some sound. I am co-founder of the Collective Mimousine. Matthew's also part of that. And I studied in Dublin, first fine art, then media studies. And, yeah, that's pretty much it. Now I'll go over to Matthew. Thanks, Lynn. Just very brief. I'm also from Luxembourg. My background is in theory. I studied philosophy, an MA in philosophy in Berlin. I also studied in Dublin before that, but I'm now moving into art as well. Mostly kind of like video installations, those types of things. I do lots of photography. And I also co-founded this art collective, Mimousine, which is co-running this program here. Not sure who we're missing. Who hasn't spoken yet? Is it Doone? Is that the correct official name? Yeah. Thank you so much, Matthew. You're welcome. You seem very philosophical. Oh, thank you. He's the father of the group. The sense of father. I'm pretty sure, Dan, you are the father. I'm just pretending. I'm not really. The child is the father. I think it's Berto or Johan who's not spoken. And Tammy as well. I think Doone, you were about to introduce yourself as well. Sure. So I'm Berto. I'm from the U.S., specifically Texas, but I live here in Germany, Nuremberg. I come from a background of graphic design, art direction, creative direction. Working for Adidas. And my main topic of focus is topics around surveillance, military-industrial complex, and environments of trauma. So, yeah. And my background in a past life is military. Oh, cool. Where in Texas are you from? Murillo. Okay, cool. Cool, cool, cool. Awesome. I think that just leaves Zora and still Doone. Hi. Who else do you want to? Yeah, go first. So I'm Zora. I'm Tunisian Luxembourgish. I'm an interaction designer and visual artist. And I mostly work on immersive experiences, interactive installations, or performances. And I think that, yeah, I actually work a lot with stop-motion or, I don't know, printing and all of that. And I like working with these textures in more generative and digital playgrounds. Cool. So maybe a follow-up. So I'm Skander, and I'm a sound artist. I'm half French, half Tunisian, based in Paris. My academic background is in innovation and technology management. And on the art side, my main focus is on a more generative way to create music and still keeping it into the emotions, memories, telling stories. I mean, a lot about the storytelling part of music, let's say. And now I'm also a music live performer, music producer. And now I'm focusing more on the immersive installation part of sound. Cool. Amazing. We have a lot, actually, to share more. Last but not least, Doon, and I apologize if I'm mispronouncing your name. Hi, my name is Tammy. I'm from South Korea. And I'm currently studying fashion communication and promotion in London. And my work is mostly very, like, MIM, like, internet culture-based. And I use different digital tools, like Blender, Stable Division. And because I didn't go to high school, I spent a lot of time just watching, like, Instagram and TikTok. So that's why I'm into, like, internet culture and digital stuff. Yeah. Amazing. That's super cool. Well, thank you, everyone, for introducing yourselves. I feel very at home in this group, as will become evident very shortly. But it's really interesting, just the range of experiences from physical to digital practices, creative technology. And, yeah, just to share a little bit about myself. First of all, thank you all for being here. My background is as a sound artist. So I'll just share a little bit about myself, about some of the inspiration for Soot, which is largely influenced by the design principles of nature and then how that translates to a world powered by machine learning and distributed computing. And then we'll share a little, like, demo. And we can just kind of, like, play around. And I'll show you, just given kind of the background that you all have, I'll also show some of the incremental steps, just in case that's interesting at all. But, yeah, so my background is as a sound artist. I was actually, for a while, with the Paris, formerly Paris-based sound art collective called Sound Walk Collective, if you're familiar with them. But I was interested in mainly, like, mass notification events. So doing kind of, like, public exhibitions with, like, big military sirens and, you know, really, really loud kind of alerting sounds that I'm sure drove Mary crazy 10 years ago when we were working a lot together. But transitioned from there to working on what I kind of think of as, like, applied art and technology. So I ran the social robotics lab for a self-driving car company called Zoox. I can show you a picture of what that looks like. They're now the Amazon self-driving cars. This one. You guys can see there's a little vehicle here. As you can see, it's got a light bar at the top. Underneath that, between the two headlights, I don't know if you can see, there's, like, a little strip. And that's actually a sound bar. So it has 32 channels of, 32 speakers on either side of the vehicle, which uses this very badass kind of beam-forming, acoustic beam-forming technology to shoot beams of sound at pedestrians rather than, you know, kind of, like, honking and causing all this noise pollution more broadly. So my lab invented and developed that technology and kind of the vehicle behaviors as well, kinematics, all around this idea of, like, you know, how does a self-driving robot with no human behind the wheel communicate and interact with pedestrians? And this is kind of a segue into Soot because, you know, it's one of the challenges of developing a language, an auditory language for a vehicle that is, you know, powered by computer vision and software and advanced hardware engineering to be able to collaborate in these very complex and dynamic noise environments, like a city or like a highway, is simply around, like, what it takes to be audible. And I felt like it was a very important kind of, like, moral stance to take that we should not be reinventing mobility while also caring for this centuries-old baggage of noise pollution in the form of a horn. So we really wanted to redesign the kind of auditory language for how autonomous mobility should interact in cities. But it's, like, it's not super obvious, right, how to do that where you have, for example, really quiet pedestrian street and then suddenly a bus drives by, right? So suddenly there's this very unexpected spike in the kind of noise profile. And so, you know, how does one go about building a language that's not only intuitive and efficient and safe but also reliably audible? And so I started looking at a lot of kind of, like, high-performance soundscapes in the world, you know, particularly in nature. For example, if you walk into a rainforest, what's really amazing is, you know, just kind of, like, if we can all close our eyes and imagine the lush soundscape of a rainforest, you can hear thousands of pieces of information all at the same time, right? You can hear tens of thousands of insects, apex predators, birds, primates. And it's actually really intuitive while you listen to peel apart the layers and understand, you know, each of the different parts. And the reason for that is because sound in nature is self-organizing. In other words, each species, and even within a species, different subsets of a species will communicate in distinct acoustic channels. And through those distinct acoustic channels, instead of being a cacophony of noise, what you get is something that's more akin to an orchestra. And so the bandwidth of information that you're able to perceive in a single frame, just in the auditory domain, is, like, immense. It's actually really, really surprising. And so this was kind of the main inspiration for what we were doing at Soot, but this was kind of the bridge. So I was, you know, researching this for self-driving cars. We wanted to develop a sound that could cut through the noise, and we used soundscape ecology as kind of, like, the core inspiration for how we went about developing that sound. The result was, I'll just share, and then we can kind of move on. But, you know, like, you have a metal bar, and you hit it with a metal cable, and you hit it with a metal bar. It's called a phaser, right? And it's kind of like the sound that you hear in Star Wars, like when people are, like, shooting at each other. There's also another phaser. But the thing that's unique about phasers is they basically span the entire audible spectrum. And so what we found was a way to create a sound that's unmaskable is by making it basically like this phaser kind of quality, so that if there was a big chunk of noise that surprisingly popped in some area of the acoustic bandwidth or audible spectrum, then it wouldn't be masked in other areas. So the result was a sound that could actually be quieter than a honk, but equally perceivable. And, yeah, as I was doing all that research, I was also Googling around and doing all these things, and it kind of struck me as being really strange that, you know, in the soundscape of a rainforest, you can hear tens of thousands of pieces of information. But when you look on Google, there's a tag on the top left corner of the search that I almost kind of, like, overlooked almost, like, for decades. I was like, once I noticed this, I couldn't believe that this was kind of, it almost felt like a false promise that was being made every time I did a Google search, which was, like, let's say you search rainforest. What you'll see is, like, 3.9 million search results. It's, like, tagged in the corner, right? But what do you actually see on your screen? What you see is, like, 10 to 14 things, maps, right? Like, there's a few links, there's a few tabs, a few ads, but the bandwidth of these interfaces, at least with Google, is extremely limited. And I found it insane that, you know, like, just in the auditory domain in nature, we can hear tens of thousands, we have access to tens of thousands pieces of information, but in the best, most high-performing human interfaces, we can see maybe, like, 9 to 15 pieces of information. So that was kind of, like, the inciting incident for me to kind of move away from working on self-driving cars and go deeper into kind of first principles, look at digital interfaces. And so with that, I'm going to share my screen and share a couple other things kind of along this journey, and then let's see here. Where is the right place to jump in? All right. My computer's running out of memory, so let me actually just delete something real quick. I'm always at, like, 300 megabytes of free space, and it ends up fucking up with the slide deck, so give me a quick sec. But yeah, so that was kind of, like, the inciting incident, and as I started to look deeper into this, I saw something else that I thought was really, really interesting. All right, can you all see this? Is this full screen? Yep. All right, cool. If you tap on the box, it comes up. It's full screen now. Yeah, okay, cool. Thanks, you all. In spite of working on this crazy AI tech company, I'm still somehow incapable of using basic technology. Okay, so here we have this kind of foil between Google and the Rainforest, where you have all this information, and it's so hyper-accessible, because the information is self-organizing. And it's kind of interesting to imagine, right? Like, what would happen if our computers were designed like this? Instead of having browser tabs and application windows competing for surface area on our screens, what if all of the information that we wanted to interact with was self-organizing and visible all in a single view, so that you could rapidly navigate from any point to any other point, the way that it kind of works in natural-forming information ecosystems? But of course, our computers don't work like that, right? Instead, they look a lot like what's on my screen right now. There's something kind of strange about this picture, if you look at it for long enough, right? On the left is the Mac one, basically the first commercially available graphical user interface, based a large part on the research that was being done earlier at DRX Park. And we talked about Tanglebar before that. On the right is a screenshot of my desktop recently. And what's crazy is, it's kind of a 40-year gap between these two interfaces, but they look basically the same, right? And in that 40-year span, we've had so much happen in the world at large, right? Satellite TV, Google Maps, Viagra, it's all happening. But in the same span of time, if we look at the evolution of the graphical user interface, we see a very different picture. You can kind of imagine, this is like a very generic timeline of different user interfaces across the last 40 years. And what you'll notice is it kind of seems like we've been pushing rectangles around on a screen. And actually, all of these different interface patterns you can reduce down to one very simple geometric shape, which is a line segment. In other words, almost all graphical user interfaces in the modern world are based on this kind of core geometry, where you have something at the top spot, and then it cascades down into infinity. So whether you're scrolling through social media, you're reading the news, you're shopping online on an e-commerce site, or you're sorting files in Finder by alphabetical order, these are all different manifestations of a file system whose kind of core architectural geometry is the line segment, which is kind of interesting to kind of realize. And in some ways, the line segment made a lot of sense for the world of 1984, for the world of the Mac 1, where most of your interactions with your computer were with some kind of text editor. You had on average two megabytes of storage, and also wasn't connected to the internet. But the world's changed a lot since then, right? And so what happens when you match this design pattern intended for the world of 1984, what happens when you match that against the amount and complexity of information that we're interacting with today? Was there a question? I hear like a little sometimes Discord notes, and feel free to just like interrupt me if you have a question or comment. Jake, I think it's people coming in and out of the room. Okay, cool. Let's see. All right, well this is in the middle part of this isn't loading for some reason, but... So what you get is, what is meant to be in the middle here is a picture of Instagram, but the background image, I think, also tells the same story. What you get when you match these two things together, which is like the amount and complexity of information that we interact with today with the line segment, is you get the feed. A feed and a feed lot, kind of similar concepts, actually, when you think about it, right? And when the kind of like, highest ideal of an interface that you have is something in which information starts at the beginning, starts at like the top spot and disappears into infinity. When that's kind of like the interface pattern that you're working with, the maximum value that a company can extract from that pattern, if they're trying to release a product or service, is one in which you have to basically commodify the user's attention, right? You have to keep them hooked to that feed. So that's kind of like where we are today. So we were trying to think about, like, well, what if instead of a feed, what if we use a different geometric shape, right? What if we use something like a well? You know, a place where information, instead of disappearing into infinity, is able to grow and accumulate and increase in value over time. And borrowing from the design principles of nature, what if that information were self-organizing, right? So that, you know, you could see tons and tons of information and it was all extremely accessible and you could navigate it rapidly. So that was kind of like the question that we posed and our goal was to try and redesign the graphical user interface based around those principles. And there's a couple of superpowers that we're now able to use that I don't think have been really as available at any other time in the history of the computer. One of them is, obviously, machine learning. And so in order to kind of enable this self-organizing functionality, we relied heavily on a branch of machine learning called clustering algorithms. Clustering algorithms basically work by extracting all the features of a data set and then trying to run this optimization where you put near things near each other and dissimilar things far from each other and from that you kind of create this map of all your data. And I'll show you what this looks like in a minute. But the second one was also around computer graphics and graphic streaming over the browser, which a large part came from video game technology. And why that matters is because if you want to show 40,000, say, images in a single view on your computer, you have to be able to render that. And that wasn't really something that was possible until literally within the last year. So, you know, using some of these kind of technologies to redesign the graphical user interface was the kind of springboard for where we started to go. And with that, I will jump into kind of the first step along that journey. Before I go any further and kind of show you how this all came together, is there any questions or thoughts based on what I've shared so far? Cool. So let me jump into this. So about a year and a half ago, we started to release what we call Playgrounds, which were our first kind of prototypes of what a graphical user interface based on self-organization and super-fast graphics streaming could look like. And I'll just take you in. So Dr. Me is a design studio based in Manchester, and they came to us, they had this hard drive called Black Magic 2, and it was like 10 years of projects, project files, rejected proposals, school work, all these kind of things that were kind of like just like fucked into this hard drive and kind of like forgot where everything was, and you know, hard drives are extremely inaccessible. And so we just mapped all of it in foot, and this was the result. So I'll just start by kind of highlighting that, you know, this is thousands of images, it's all super high-res, loaded on my browser just now, and if you start to scan through it, and one thing that's kind of fun is it's very playful, and just to kind of take you on a tour of this hard drive, I'm going to put it into a little bit of a more easy-to-understand shape. But something that's really interesting is our brains are so kind of engineered to discover and see gradients everywhere we look that when you take something as big and chaotic as a hard drive and map it by visual similarity, suddenly it becomes really easy to at a glance just get a sense of the vibe, right? Like you can just kind of understand roughly the different kinds of projects that are in here, which is a really kind of important first step for opening up the accessibility of something like a huge data space of, you know, over 6,000 points. But you can also do really fun things like, you know, it doesn't have to just be organized by visual similarity, you can also organize it by color, and suddenly a totally different view of relevant gradients to the data become available. Another thing that's kind of unique that we wanted to draw out was this idea of similarity and relation, right? The idea that the things that link any two pieces of content together live not in the metadata, not in the file name, not in the folder that it's stored in, like not in any of these extrinsic properties, but everything you need to know is actually intrinsic to the content itself, right? So let's give an example of what's cool in here. So like if we were to grab this orange jumper and say show me everything similar to this. We wanted the ability to completely reorganize instantly the entire hard drive like so if I zoom in you can see the entire hard drive here but as a heat map of similarity to any individual item. And it becomes a really kind of fun and interesting way to just like leap around and get a feel for some of these space or some of these data by traversing these gradients of similarity. And from there we were like okay cool, but also what happens if you want to like search for something? So one of the really nice things about AI is when you bring everything together into this model, you're able to do multimodal it's called co-embedding, text-image co-embedding. What that means is like I can describe something in text let's say like black and white with a spiral of text and the AI doesn't really have a super strong distinction between the text and the visual image and so I'm able to use text to basically power search features by saying like here's a text string using the AI sees the text string as being the same as the image embedding let's just like find everything that's really really close to this text string. So it's kind of a clever way of like backing into getting like text search. So yeah, this was kind of like the first step along this journey was like first of all, is it even a good experience to be able to see tens of thousands of things in a single view? And if it answers yes, then like what kind of things does this unlock? And one of the key ones was like it makes huge archives of data just like extremely accessible in the same way that hearing all the parts of a rainforest are accessible. You know, the other thing that we were curious about is like what other things can this power? I kind of hinted at this earlier when I described, you know, shopping, social media, reading the news as all being manifestations of a file system. But when you kind of solve this core interaction pattern of seeing everything in a single view, it turns out to apply to a lot of different ways that we use computers today. So another one of them is shopping. We kind of hacked this website called Essence, which is an e-commerce platform just to see like what would it look like if, you know, you had every shoe on this e-commerce platform also interactive and able to be navigated visually. And so that's what you have here. It's about 12,000 shoes. And what's really fun is you can kind of just like zoom around if there's things of interest. You can say again, see similar. And it creates a heat map of everything, you know, everything related to that shoe from the entire collection. And we can also do things like I really like pig pumps. It's like save search. And yeah, so it just becomes like a really fluid and fast way to navigate these kinds of things. And, you know, it's kind of fun because then you can kind of start to layer back on top of this some of the metadata and extrinsic features that are what make file systems valuable today, right? So, for example, you can have links. You can start to create collections. So in this case, it's a collection by a designer. And start to layer back on top of this core experience the building blocks of, you know, all the common patterns that have been so well-crafted over how the interface works. Yeah, so that's kind of what we've been working on. And I'll just share a little sneak peek of what happened next. You know, so after validating these different experiences and how users interact with them, the next phase was to really build a tool. Something that people could use themselves. And so what you'll see here is kind of like an updated design language. So this is the platform that we're now working on. And yeah, so like what I'm kind of sharing with you is actually my team's internal hard drive that's broken up into a bunch of different projects. I have this kind of like recent page just at the beginning. But yeah, it's basically the same principle. You know, you can see we have like a few shots from a video, a commercial that we're working on, user interviews, screenshots, branding work. And we're all just kind of living together in this one big mosh pit. And it's been really fun for me because it means that I have basically like instant accessibility to all of my team's work. And something that I find really interesting about that is I don't know how much when you're working with collaborators or with a client or you have designers that you're working with, we kind of get in this weird space sometimes of like you know that thing in the Heisenberg Uncertainty Principle, right? Like you can't observe and measure complex phenomena in the same frame. That's kind of a butchering of what that principle is. But roughly speaking, that's kind of like my takeaway from it. And often with design projects what that means is like you can have work that's in this creative, ambiguous space where ideas are flowing, where they're evolving. But then there comes time to share it with your collaborators. And in that instant when you have to fix something that is fluid and pointing a lot of directions and fix it into a linear narrative by sending a link or putting it on a deck or something, something gets kind of corrupted about that work, right? It ceases to be in motion and now something becomes fixed and somehow some of the magic disappears. And so something that I really, really like about being able to just like ambiently be connected to all the work of my team is that it means that the work always live in this kind of in-between ambiguous space. This is early technology, so sometimes things don't go as planned. So as an example, here's a project of icons that we're working on at the moment. And a lot of these things are still in flight and no one explicitly shared it with me, but I'm still able to navigate their ideas and see how they interconnect with everything else that's happening across the team and be inspired. But yeah, my dream, I would love if this is kind of how icons start to look in the future. But anyways, that's a story for a different time. But yeah, so that's kind of the story. To just share kind of what's ahead, we kind of see like... I kind of see like the personal computer as, in a lot of ways, kind of like the foundation of everything that we do and touch on a personal computer is access through some manifestation of a file system. So to me, that's like the gateway of everything that comes after, right? Like all the tools, all the digital experiences, all the functionality that we use computers for are all built on layers of file systems. And the world today is one in which all of our different applications that we interact with are kind of like walled gardens. You have to enter into this very opinionated file system that is fighting for attention on your screen, is fighting for your data to become this walled garden of value with a lot of overlapping features and then you might have another application and it kind of becomes this mess. And to me, it seems like pretty obvious that how a computer would look and feel 10 years from now is this kind of spatial, self-organizing file system that connects private experiences with public experiences and that has all of your tools and applications just layered directly into the file system itself. And so for us, kind of the next phase after we launch this file system is doing just that, the information tools, one of the things that we're thinking about that would be really funny to build is a dating app on top of Soot, a karaoke app, ways of editing images. I didn't kind of point this out, but we have, for example, Dolly 3 built directly into the search bar. So if you don't find an image you're looking for, you can generate it. Expanding outward into different content types, so text and audio and video and all the things. And kind of bringing the personal computer back up in reverse, starting with the file system. So yes, that's basically the story. And thank you all for listening. And yeah, I'm curious like, the thoughts or questions or comments. Thank you, Jake. That was fascinating. It looks like an amazing system. I have peeked at it before because obviously Chris and I have talked about you coming on. That was, yeah, I'm blown away by that and the potential for it. It's just amazing. Can I kick off with a question? In terms of, have you left the judgment of criteria in choosing similarity between images entirely to the AI, or have you had to train it in some way? Yeah, you know, it's funny. In some ways, we're like an AI company, but at the foundation, proprietary AI is not something that we wanted to do explicitly. We wanted to build a data pipeline where you could basically plug in an off-the-shelf AI model that suited your purposes pretty easily. So we used really generic open-source AI models to drive that. And the reason is that once the kind of first MVP, as they like to call it, is released, we think it's really important for people to very quickly be able to say, like, this is great, but similarity, for my context, requires more specificity. It requires a model that's trained towards a specific use case. And to us, that's kind of like somebody should be free to be able to make that decision. So we wanted it to be really generic and simple, and it turns out that AI models was good enough to tell the story, to inspire somebody to come and say, like, that's great, but similarity to me, actually all I care about is similarity of face, you know, for what I'm doing, or pose, or similarity of a particular background texture, and you can kind of imagine that kind of spiraling out to an infinite range of use cases. I've got more questions, but I'll let others chime in first. I just wanted to jump in and ask if you'd seen the note-taking software Obsidian before? I haven't used it, but I've got a lot of friends who have, and when you were talking about, like, being all these, like, files or data points, like, more like a rainforest as opposed to, like, the constant feed cycle, that's just what came to mind, and I wanted to know if you had any thoughts on Obsidian, or if it inspired you. Yeah, you know, I was also like a Roam user early on for, like, 10 minutes, and I was interested in Obsidian, and I imagine it's probably evolved a lot since I last checked it out, because, you know, with the release of, like, large language models, all these kind of text note-taking things have changed, except for Roam. I think that guy's, like, a global leader now or something, I don't know what happened to that project, but you know, I think it's kind of interesting, because going back to this idea of, like, that all you need to connect one piece of information to another is already, like, intrinsically living in the data, I think that's really the power of doing AI, right? And I found it very interesting how a lot of these tools for thought are kind of built around the practice of organizing your work is like manual labor that you need to do, right? Some people are very meticulous, and for them, like, it's really nice to tag words and connect it to words in other articles and do these things, but I feel like, in some ways, you know, these are leaps that should be easy for us to make without actually having to do the work of manually making those connections. I'm curious if Obsidian now supports some of this, but we are looking at, actually, the first prototype that we built was text-based, and ended up moving to images as our first use case, mainly from a technical perspective. Like, it's really, really hard to get 10,000 images on a screen. It's really easy to get, like, 10,000 text notes on a screen, and so, like, if we solve the, if we really put product and solve the image problem first, we kind of get text for free. So, after kind of building that early text prototype, we moved on, but one of the things that's really interesting about text is so much of our lived experience is like speech, and so we've been looking at a tool that we call Teleprompt, top secret, where it basically records what you're saying, and then it builds an index, kind of like, an index. Imagine if, like, everything that you said to an Alexa device or a Siri, rather than, like, all of your private information for the very small convenience of, like, having a hands-free light switch, imagine if that was, like, a searchable index of your lived experiences, and so I think it's something that connects to what you've read and your text notes and things like that, so I think there's so much depth that can be pulled out of just connecting a large language model, a microphone, and a visual file system, and it's one of the things that we're really excited to jump into. Thank you. Anyone else got a question? I've got a list, so I could probably talk for the day, but anyone else want to ask anything? When you were showing quick examples of the design work that your team has been working on, and you mentioned how instead of outputting specific files or sorting them in any hierarchical way, how you could kind of just see the big web of stuff your team's working on, it kind of seemed to me almost like a way of stream of consciousness. It seemed like a stream of consciousness to me, a way of working and collaborating with each other, and I'm curious as to how your design teams and the designers that specifically use the tool in their design work and in their iteration work. How does that affect their workflow and how they collaborate as a design team? It's funny, there's kind of like two edges to that. The one edge is we mainly use it, or the reason why most people will open to find something, that idea of digging around a Figma file to try and find something that there's actually no search feature in Figma for a visual asset. It's just a lot faster for us to just pop up in Soot and roughly describe it or go to something that looks like it, hit see similar, and then it's a really, really quick way to get to an asset. That was kind of like the core utility that I think was habit forming for our team. But then the second edge of that is the discovery, right? Because when you see something when you hit see similar, you might find something that you were looking for, and we kind of have this internal metric that we call time to find Waldo, which is that any search should be served in under 20 seconds. The average for finders, or the upper lower core type of finders, is between 20 minutes to two hours, and we were like we want to be two minutes to two hours and we want to be under 20 seconds. But then there's that magic of all the other things that are related that you didn't think you were looking for that you stumble upon that leads to new ideas. I think that's kind of both edges of that loop, which is that really directed, I want to find something specific and then also the magic of more often than not, finding something unexpected along the way that answers the question you didn't realize you were asking. If you think of Pinterest Yeah, exactly. Pinterest does have some of this as well, right, where you scroll down and it's got all the other things in the world for things like mood boards. I have a question if I may. First of all, thank you for your presentation. It was really super interesting. We were amazed during all the presentation and I was wondering if you used the data cubes as an inspiration for your work. Maybe you're familiar with data cubes? With data cubes? No, I'm not. Let me just quickly look up what this is. So it's a format used, I mean, I knew that when I worked in the Earth observation sector, so it's a way to sort information on a three-dimensional space, let's say, where you will have, for instance, we can have a project on an axis, we can sort dimensions on another axis, and time to have newer to older projects, for instance. And it's a way also to sort things in space. My, just from a quick glance, my understanding of data cubes is, so, caveat, I'm not the technical founder. That would be other more brilliant people than myself. I think the core principle is like a n-dimensional vector, right? Or that you have many different vectors which you can map features of a particular digital object. And you listed a few, like it could be timeline, it could be a few others. And it might not be answering your question directly, but I was actually, I found it very, kind of funny corollary to, do you all know the parable of Plato's cave? You know, you have like shadows, projections, and usually this parable is like, it's like a negative thing, right? It's like, you know, this is like the world of the person who's enslaved inside the cave, and they're not able to perceive reality in the full dimensions because they can only see things projected onto the cave surface. The thing is like, actually in the world of AI, like there's some benefits to having the shadow on the wall, right? Because basically AI sees the world in tens of thousands of dimensions, right? Like we live in a four dimensional universe, but AI can map features in, you know, many orders of magnitude more dimensions than humans can actually even reason about. And basically how we're able to create those visualizations where you have things organized by visual similarity is actually kind of wild because it's basically like Plato's cave applied to a graphical user interface. And what I mean is, you can imagine this crazy cloud and super high dimensional space of all these different objects, and it turns out that with a pretty good amount of accuracy, if you were to like cast a flashlight so that that high dimensional space could be capped onto a two dimensional surface and like kind of roughly see the near things, you can actually reduce the dimensions down from this like super high dimensional impossible for humans to reason about space down to something that we can see and interact with, which is in two dimensions. And so it's essentially like how the technology works. It's like mapping the thing that you see on the wall of Plato's cave, the shadows, and allowing people to interact with that, because that's something that humans can reason about, whereas the thing that's happening in the reality of the AI is like way too complex for us to reason about. And it was a very interesting discovery that it turns out that the precision of this dimensionality reduction, the precision of seeing the shadow, is a good enough approximation that it's still useful as a tool. And I'm not answering your question now at all, I'm now just going off on one, but it triggered another thought related to that which I found very interesting about this project was in data science you always have this trade-off between efficiency and precision, right? Like you can be incredibly precise, almost infinitely precise, but the more precise you are, the more computation needed, the longer it takes to process, the more expensive it is, right? Or you can be really, really efficient. You can do something super fast and quick and dirty, but it's not precise. And something that I think we found that was really interesting, and I think this kind of came from also just having a background in being a performance artist, was like, it turns out that we don't need that much precision for something to be extremely useful. And so, like of all the AI companies out there, we use probably like the worst models. We use like, you know, and we even brutalize the models, like by showing them in that squash, you know, in that elliptical shape. We reduce the efficiency, we reduce the precision of the models a lot. You know, but the result is something that's much more useful. Like it turns out that that extra precision that people are spending billions of dollars training models to be better and better and better, a lot of that actually didn't turn out to lead to more value. And so, anyway, that's just like a funny thing that we realized was like one of the keys of making this experience work was realizing like, from a technical perspective, the AI can be really weak and really poor, and that's actually a benefit. It makes it better. So, yeah. Well, that's quite an interesting lead into what, you know, I was wondering, you know, without that specificity, you know, if you're trying to locate something, you have to go in a different way to find it. If you knew what, if you'd wanted to get there really quickly to that specific item, is that you've got to go in a different way? Have you got to know how to reference it? Is it findable in some other way? Does that make sense? Yeah. You know, I think we could just like, I could just show you. Let's see. You can see me now researching what a data cube is, so I can understand the reference. We'll just go back to Dr. Me, right, and I really like this view, this kind of flat view. So, because I've been in this space a lot, I didn't know that the, like, nature-y stuff is over here, and, like, the quality stuff. And so there is an aspect of, like, spatial memory, right, where you can just kind of, like, oh, it's on this area of the ellipse, and it's just, like, you kind of navigate there, and kind of traverse the gradient visually, and using spatial memory, and you can kind of get back to something. But it's also just so fast to type the search, like, you know, that just with, like, a word or two, you can kind of find it. And it's actually kind of interesting how rough kind of you can be, right? And here's another example of a, this is a graphic, an illustrator called Charlotte Ager, who's also based in London, and, like, you know, all of her sketches are up here. The more, like, finished stuff is down here, and then she's got a bunch of drawings of pools, like, over in this corner, right? But also, I could just be, like, if I didn't feel like doing all that, I could just say pool. And you know, it's kind of interesting, but it recognizes drawings of pools as readily as, like, actual photographs of pools. So, the reason I kind of bring that up is it's always good to have, like, multiple points of entry. So, something that is, like, fixed in a landscape, but also something that you can just kind of, like, hack your way to, or improvise your way to. Yes, it's brilliant. It's so fast, as well. I mean, I'm sure there's some technological secrets behind how this functions, but I'm just amazed at how fast it is, considering the amount of images and data that must be behind it. Where does that data sit? Does it sit on a hard drive? It's in Virginia. There's an Amazon, like, data center in Virginia, and that's where we stream everything from at the moment, you know, once there's actual users on the platform. We'll share that out. But, yeah, it's in Virginia, and, yeah, it's kind of interesting. I think about this being very similar to, like, you know, like, dark matter, dark energy, where, you know, there's this, like, humming, ambient energy everywhere in the universe. I read somewhere that, like, if the volume of the Earth were converted to, like, what is the mass of the same volume of dark energy? Something like five milligrams, right? So it's, like, essentially not very dense, but because the vast majority of space is, you know, nothing but dark matter, this accounts for, like, the vast majority of the mass of the universe. But you can think of a website like that as well, right, where you kind of have this, like, negative blank space on the screen, but the thing that you're connected to is this kind of ambient, super-powerful computer, kind of distributed computer. And it turns out to be really powerful to centralize a lot of these, like, very computationally expensive tasks, like running machine learning algorithms or processing all the different image compression layers that allow us to quickly show lots and lots of data and make it feel really seamless and fast. Because actually there's some, like, there's some, like, David Copperfield, like, sleight of hand happening. You know, like, when you zoom out, obviously we're not showing full-res images. We're showing very, very tiny res images. And the trick is, like, we have the high-res images all ready to go, and as you zoom in, we quickly swap them in. So it's a way of managing. And as you zoom in also, you're filtering out a lot of the images. So it turns out that you only need to load a few, maybe 10 or 15 high-res images as you zoom in. So it's a really nice way of delicately managing, actually, the amount of data that you actually have to load at any one time. So the result is an experience that feels like, wow, all these high-res images loaded immediately, and it's instantly accessible. But the reality is that we're able to be really careful about never loading more than the person actually needs at any given time. That sounds really smart and really clever. Should we allow people to have a break, Chris, do you think? Yeah, I just wanted to make one last point, and then just to really relate what you were talking about, the spatial understanding of what's going on there. I find that super interesting because what I've heard from people was a lot of connecting between the physical space, the digital space, and what we've gathered from that is that there's a discrepancy between tangible and non-tangible, like spatial and kind of digital. What is that floating just in space? And when you said that we have the spatial memory of this database, you're kind of like, okay, I remember that this picture is here. That's so interesting because humans kind of remember, when I read a text, for example, I know what quote was standing where, but when I read it digitally, I think that there's a lot to explore in that matter, how we perceive digital data when we give it a sort of spatial landscape sort of to explore in a more, yeah, just like walking, like imagine you walk around and just have a map of it. I think that's really interesting and we can get a lot from that information. Great talk. Thanks for sharing. Yeah. Thank you, everyone. Does anybody have a final question before we let Jake go? Oh, no, don't go, Jake. Can I just ask one quick question? When do we get access to it? When can we play around with it? So, Mary, as I believe, is still on the call. But you can just send an email to Mary, I'll let you kind of give the info. You're muted. You're muted, Mary. Hello. Hi. Basically, if anyone is interested in having a user test account, I'm trying to find a chat in this thing. If you're on the desktop, it's top right, there's a little bubble. Ah, yes, thank you. So, if you send an email to meher.soot.com, we will hook whoever would like an account up and maybe do a little one-on-one onboarding call. So, yeah. So, feel free to drop us a line and, yeah, we'll hook you up with some test accounts. Wow, amazing. I thought you were going to say six months down the line, come back. No, I mean, it's amazing. We'd love to have some people on it and get people's feedback. You know, we have quite a few people on it at the moment. Yeah, and it's kind of fun to come on a journey with us and see how it develops, you know. Wow, amazing. It feels like a privilege. Yeah, thank you so much. Yeah, it was a pleasure to have you here. Fascinating. Brilliant. I'd love to carry on talking, but we will let you go. Thank you. And also, when you send an email to meher, please send your portfolios, your work. We'd really love for this to be two-way and to learn about all the incredible things that you're doing. Thank you so much for letting us in to say hi and to meet all of you. It's been really fun to just get around the channels here and to see all the really interesting ideas that are flowing around. Yeah, don't be strangers. Let's be friends. Thank you. Thanks again. Thanks, guys. That's so cool. That kind of software is something I feel like me and my friends have been talking about, wishing someone would start to make, and now it's just kind of happening. It's pretty cool. Well, you know, for some reason I had you in mind when I was asking that question, thinking you're doing all these illustrations. I'm sure you've got masses of material you could just... I think it's a great opportunity. I'm actually surprised they said yes, we can have access. I know, I already sent an email. I'm super stoked on it, too, with the amount of folders of just phenomenal reference images I have that are just dialed into one folder and not delineated. Yeah. Super helpful. I think Chris isn't feeling too good. We're going to let Chris go, I think, but let's have a five or ten-minute break and come back. If anybody wants to have a chat with me about Art City, I'd love to hear what you're doing, if you put anything in there that's bad or indifferent. I'll come back in. Should we have ten minutes and I'll come back and we can have a chat? Yeah, for sure. Chris, if you want to hang around, it'd be nice, but I know you're not feeling too good. Yeah, I'm going to go lay down. I'll see you all next week. Will we use the new Art City? Should I bring my PC or...? No. Oh, well, we can just have a chat about... If you want to show us anything, it's up to you. I'll just put a nojo file on it and change some settings on the visual side, but it's nothing very... If you want to show us, that'd be great. Show us anything, ask any questions. Do you need to jump on another machine? I have my computer, but I mean... You can open your file online. What? The new Art City is online, so you can maybe... Oh, yeah, yeah, yeah. No issues. That's a good part of it. You two know each other. Did we know that you knew each other? Sorry? Did we know that you knew each other? Did you know that we knew each other? Did you know that we knew... Did we know each other since before? Sorry? What did you say? Did you know beforehand, before this? Yeah, yeah. And we're currently working on a project together now, so that's why. That's cool. That's really good. Let's have ten minutes. I'll see you in ten minutes. Perfect. Thank you. See you. You're pointing at me, Lynn. Yes, I am. That looks funny. Oh, great. Are you good? Yeah, yeah, yeah. I wanted to tag Jake in a post. I have to ask Chris what his Instagram is. Do you know if Suit has an Instagram page, maybe? I have to check. You can tag a person. You can tag a handle. I don't know. I don't know. I haven't found them on Instagram. Yeah, I just keep saving videos of the thing, and then I'll make a little post. I'm just wondering what happens to the chat. It seems to preserve the chat, doesn't it? Oh. That's brilliant. There's stuff we put in on Tuesday. Is that right? Yeah, look. Yeah, it is. That's good. Maybe Mayha would give us some more information that we need if we want to connect. Let's have a quick break, shall we? Yes, yes, of course. I'll see you in a few. Bye-bye. ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...