Goodbye Fitting Rooms: Hello AI-Powered Virtual Try-Ons (no more returns)

Jasper Schoormans joined the Behind the Bots podcast to discuss his Artificial Intelligence powered platform, Outfit AI. Outfit AI allows users to visualize how clothing items would look on themselves by uploading a photo. The Artificial Intelligence then renders the selected clothing onto the user's image. 

Jasper originally began experimenting with AI image generation by manipulating photos of his children. He was amazed by the technology's ability to realistically integrate images. This sparked the idea for Outfit AI as a way for online shoppers to "try on" clothes prior to purchase. 

The current challenges include improving image generation speed and expanding integrations with e-commerce platforms. The long-term vision is providing retailers with a "virtual fitting room" to reduce returns. There was also discussion around AI progress in the medical field and optimistic outlooks for artificial intelligence.


OUTFITAI 

https://outfitai.co/
https://twitter.com/jasperschoormns


FRY-AI.COM

https://www.fry-ai.com/subscribe
https://www.youtube.com/@TheFryAI
https://twitter.com/lazukars
https://twitter.com/thefryai

Creators & Guests

Host

Ryan Lazuka

The lighthearted Artificial intelligence Journalist. Building the easiest to read AI Email Newsletter Daily Twitter Threads about AI

What is Behind The Bots?

Join us as we delve into the fascinating world of Artificial Intelligence (AI) by interviewing the brightest minds and exploring cutting-edge projects. From innovative ideas to groundbreaking individuals, we're here to uncover the latest developments and thought-provoking discussions in the AI space.

Hunter Kallay: Yeah, so you got into AI imaging, right? Like using images, training different models. So can you talk to me about your project Outfit AI? So it's outfitai.co. So tell me a little bit about what that is, where it came from, and what it does.

Jasper Schoormans: Yeah, yeah, sure. So Outfit AI, it's an ID, I think I've had it for over a year now since I was playing with stable diffusion and in-painting. So originally, I was just playing with pictures of my kids. And then with stable diffusion in-painting, you can give the model like a picture and you can give it like a black and white mask, and it keeps some parts of the image. So it was quite fun to add different backgrounds, put the kids somewhere, make team things.

And I was so amazed by that because it's this model for some reason. If you in-paint an image and you would cut out a person and just change the background, it's really good, keeps the lighting sort of correct. Even if you have a bad mask, it solves like strange colors or things. It makes an object out of that. So I was amazed by the possibilities. And turned out you could quite easily change clothes as well, just by masking. You know, the clothes that people wear and asking the AI to change the red dress into a blue dress or do something. But then, of course, the cool thing would be if you could do this with real clothes, because then you could sort of check how these clothes would look on yourself on a webshop and see if you want to buy this.

So for that, of course, the model needs to really know the items that are available for sale. So I've been doing a lot of experimentation with that. First with the dream boots, retraining of stable diffusion. So there you can give 10 or 20 images and it starts retraining the entire model. It takes quite a while, but it learns a representation of the object. And then a bit later, the Loras came out sort of the faster way to do these trainings. And yeah, I played a lot with it. And in the end, I did find ways that worked well.

So they were reusable enough. I could give five to 10 images of a model wearing a piece of clothing, could be a shirt, could be a dress, could be anything. And then it needs some training. And at that point, the model knows, well, this is, you know, this item. And you can add it on another person. So there's many challenges, right? So you want the results to be the same person.

So keep the face, you know, keep the body shape, keep the skin color, no strange things. It should be happening at that point. And it should be maybe fast enough as well.

So ideally, I would like to have like a Shopify app where you can just push the button. And then like within a second, you see the next outfit. So at this point, I think I'm very far with the consistency. Image quality looks good. And I can train the model easily on, like, on scraping a web shop that works well enough. The biggest challenge now is that like my image generations take 10 to 20 seconds. So that's one thing I'm working on at the moment.

Hunter Kallay: You can go on to outfitai.co. You can see a demo. I tried the demo, works pretty easily, pretty seamlessly. It's cool. But that's one of the things about shopping, right? It's like when you're shopping in the store, you can go in and try on these clothes and see what they look like in you. When you're online shopping, it's like, OK, instead of going through that painful process of I don't know if this I'm going to like this, I'm just going to order a bunch of stuff and then return a bunch of stuff that I don't like.

Now you can actually use your own body and own picture of you and see what this particular clothing item might look on you, not on some model, but actually on your person. Is that the idea, something like that?

Jasper Schoormans: Exactly. Yeah. And because, of course, usually the models don't really look the way you look. Unfortunately, maybe colors can can be quite different with your own skin tone. Hair color.

It could be. Yeah, like trying on clothes is really different than looking at the picture. And that's even before deciding which size you should even have, because there are definitely some. I think there's there's a lot of smart machine learning on that side that some shops employ, they sort of know your measurements after some orders, like you can send back the ones you send back.

But even to step before that, the colors, the general fit of things. Yeah, ideally I would like to have some sort of TikTok style thing where you could swipe and then don't like this, don't like this. Yeah, I like that. And then it's fully personalized. You don't need models at all.

Hunter Kallay: Jasper, can you just walk me through the long term vision of this project? Because right now it looks like you have like a demo mode. You're working on things. What's your long term vision for this? Do you see this on like online retailer stores? Do you see people taking clothing? You know, do you have your own store? What does it look like?

Jasper Schoormans: Ideally, I would have Shopify integrations. Like where people that own a store or brands can just install this, it would automatically scan the inventory, find pictures, train the model, and then have this little like virtual fitting boot app they can add anywhere on their site. And if people then decides to use it, people can upload their own picture on this virtual fitting room and then see any piece of clothing they want on their own image. And at that point, I would love it if they decide to, if they like it, if they decide to add their email address and then they can get personalized, like maybe weekly pictures, newsletters of their own, you know, the clothes that would look good on them, maybe picked by the retailer or a stylist. And it would be like a very personalized form of marketing.

Ryan Lazuka: For sure. That'd be awesome. So you would get like an email newsletter or something from, I don't know, like a gap or what are the hot clothing companies? Not like any kind of clothing company. Now they'll send you an email newsletter and it might show you with a new outfit on yourself. And the outfit you like based on what you bought in the past, maybe something like that.

Jasper Schoormans: Yeah, yeah, indeed. Yeah. And then you wouldn't need any models anymore. You can only look at yourself, which may be a bit strange. I can imagine it being personalized to that level.

Ryan Lazuka: Well, I mean, that's one thing is, I mean, this is more tail. I think people going to shopping malls and not in the real world, it's becoming less and less. So it's not, it doesn't sound like it's geared toward those type of people and sort of going away in a way. But one thing I do want to say is I think every guy out there hates trying on things, you know, going to shopping and trying things out on if they go out to a real store, you know, maybe their wife's with them or girlfriends with them. And it's just a big pain in the butt. So yeah, just back alone. This is this tool is awesome.

Jasper Schoormans: It might even save some, you know, like people ordering way too much and then sending sending 90% of it back, which is a waste, of course. And then part of that clothing that sent back is destroyed, you know, so if you would reduce that even a little, it would be very, very useful as well.

Ryan Lazuka: You know, when I grew up, people would go to the mall and shop and try stuff on the mall. And then, you know, if they didn't like it, they would they wouldn't buy it. But now, like all all these people just buy whatever they want on Amazon or any online store and they just they just return it.

It's like a big waste of time, you know, like my wife might buy 10 outfits and only have one only buy one of them and ship the rest of the nine back. So that's awesome. And then it's cool can help.

Hunter Kallay: And then it's like backing up all the shipping for these companies, you know, the shipping 10 packages when really they only needed to ship to that. There's another supply chain issue there on that other end as well. So in the cost to do that, the labor to do that could be a lot more efficient.

Jasper Schoormans: If I mean, I mean, it could be at the point where it's if it looks exactly like it looks on your person, like your own person. Yeah, you maybe like it wouldn't feel maybe the touch could be different. But if it looks at least how it would look in real life, I think you could reduce a lot of this.

Hunter Kallay: Now, are you looking to implement anything about sizing? As I know, a lot of times when you shop online, it's like, I'm not sure what's what size to get. Sometimes they run small, sometimes they run big. I mean, they have the measurements there. So I was wondering if you're implementing something like with people can measure their weights or measure their their height or something and incorporate that somehow. Do you see that in the future?

Jasper Schoormans: I'm not interested in that at the moment, but I think it could it would work together very well. So the visual part that I'm now focusing on, like, you know, do I actually like this? And of course, the images that I generate, they should be close enough to the real thing. So if shirts like run wide, it should be like a little bit baggy or some things are tight, that should look tight on the image.

But then the next choice, like what actual size do you do you choose? I think that's a whole different whole different field, actually. So I would rather work like partner with with other companies that that have the knowledge.

Hunter Kallay: So you mentioned when you were first starting out this project, you kind of came up with the idea because you were trying different images. You were trying to manage tools with your kids and like putting them in different locations and backgrounds and playing around with that, which is pretty cool.

What kind of tools were you using then and how have you seen those kind of tools evolve over maybe the past year, even past couple months or even weeks at this point?

Jasper Schoormans: Well, I think since the last 10 years I've been working with like programming images with with Biden usually. So I've been doing that like forever. And even one and a half years ago, I was just like making simple Python scripts that did some in painting. So making masks, running these models. But I don't really have easy access to good GPUs.

So I've been using replicates a lot the model hosting website and to do the actual like AI model running and hosting my own models there as well. And yeah, things have changed a lot. So I think this fall stable diffusion XL became available with very good training of Laura's. I've been using that and it's it's it's a lot of trial and error, right? So I love going on replicate and there you can explore all these new models and just like trying them all online.

Ryan Lazuka: Is it sort of like a high interface? Replicate is a replicate that AI. What's what's the actual URL? You know, yeah, replicate.com.

Jasper Schoormans: It's it's it is like hugging face spaces. But so the ID, I think is they host open source models that you know, you can find in papers and where if you want to implement these things yourself, you're busy like for a day, you know, to get all the installs and everything. So they make all of these open source models available with an easy API to and then you you can run these on their GPUs for and you pay per second. So especially to, you know, to have small scale projects or, you know, fast prototypes, it's really easy and. You can add your own models there as well and a lot of people do and.

It's a great resource to just explore things. And even they have like these like I'm hugging face. They have these spaces where you can just try and upload your own images and your own prompts and everything and see the results. That's awesome.

Hunter Kallay: I've never heard of that. That's sweet. So are you using replicate for your platform or what are you using behind the scenes to make this magic happen on your site?

Jasper Schoormans: Yeah, so I'm using replicate for hosting the models. I have a like Python Django back and you said what is that?

Ryan Lazuka: Jasper, what you said, but

Jasper Schoormans: no, but Python Django, like a back end. But most of the sort of the AI stuff is on this replicate hosted on replicate. And it's stable diffusion and a few other models to do image segmentation, like, you know, like finding the face, finding the background, handling all these images. And then a big part of it is training the model on outfits. So it needs like 10 good pictures of an object that could be a person wearing it.

It could be just the object with a white background. As long as you have a few angles covered, you have a bit of variety. And so for that, I use Laura training on the stable diffusion model, but that's also on replicate.

Ryan Lazuka: So you just like how does the process work? If someone wants to, you know, they've got a small store and they want to upload all their clothing to your site so that you guys can fit, use it to fit somebody, you know, the end customer that comes to your site to get fitted.

How does that work? Do they just upload? Is there a process on your website right now where I upload all the retailer, for example, uploads all their inventory? And then you call out the API to replicate and then it does its magic at that point?

Jasper Schoormans: Yeah, yeah, yeah. So I have my own sort of web app where you can upload your pictures, like you said. And, yeah, click the button, start training. And then it's, you know, it starts training in the back and like 10, 15 minutes later, you can try this with your own pictures or with like any of a few example pictures as well.

Ryan Lazuka: Do you have to like lay out every shirt or dress or pair of pants or whatever it is, like perfectly? Or does like, how does it, is it better to have it laid out in, you know, perfect detail? Or does someone just have to take a quick picture of it on somebody else wearing the shirt? How good do the pictures have to be that are uploaded for it to work well?

Jasper Schoormans: So I've been trying a few dozen objects that I scraped from, from web shops. And so these are usually very high quality pictures, of course, studio pictures. But there's usually five to 10, one or two models, a few without any, without most flat.

And I've seen it usually works for all of them. Sometimes there's a little bit of confusion if it has pictures of the back as well. So imagine that the front of the shirt might have text or something in the back doesn't. So the model in the end could, you know, turn it around and actually show the back of the shirt at the front. So there might be, it might be useful to have a little bit of manual checks on those kinds of things.

I think as long as it's high quality and there's a bit of variation in terms of angles, it's fine. Because I do also automatically crop the object itself. So next to the images, you have to input what it is. So like a shirt or a dress or whatever. And then I use like a blip vision model that sort of finds this thing in the image. And it will train only on the actual image itself.

Hunter Kallay: It is very realistic. If you go to outfit AI dot co and you look on their domain site, if you just scroll down just a little bit, I'm looking at like the image comparisons.

And I'm looking at it's like it's the same picture of the person and you can just drag drag the screen across and you can see just different outfits on the person. And it's just crazy realistic. It looks like they're actually wearing it. It really does. In the background is very cool. I mean, it's very nice, very well done. I am just in awe of like how this image generation and how these manipulations are coming.

Coming across. I mean, I just remember even five years ago. I mean, when you would try to take an image and try to put a different background on it, it looked like the person was literally just cut out. They'd have like a little glow around them or little like edgy. And sometimes it would get the background. And, you know, this is just so seamless and it's it's crazy realistic.

It's very cool. One thing I wanted to ask along those lines is like, obviously we see like deep fake images all the time coming out. Celebrities, whatever it might be. People are worried about its ability to deceive people. Obviously you're in the image generation space. So what are your thoughts on all of this and how to best deal with it?

Jasper Schoormans: Yeah, it's an interesting thing. It's not something you can actually stop, right? All this misuse of faces of voices. And I think it's also a thing I'm a little bit worried about with this project. Like what do people think of uploading their own face to such a to such a service, right? But then again, so for myself, I thought it was really important that this worked with just one picture. So I don't know if you've been looking at these like these headshot apps or these AI avatar apps where you need to upload 10 or 20 pictures of your face and then you can make anything, which is really cool, of course, but it's it's inconvenient and it's also I think that's a bit scary that you have a model that can make like these deep fake type things.

And I think there's there could be some hesitancy for people to use that. This is just the picture you upload yourself. At least the face you get exactly your face back. Yeah, I think it's the least scary thing there is for people. But I'm not sure about that.

Ryan Lazuka: Do you think deep fate like outside about the AI, do you feel like deep fakes and these image people being manipulated in images for nefarious reasons is going to be a problem? Like for politicians or celebrities, they might be put in, you know, pictures that don't put them in a good in a good way. And, you know, it looks bad on their image for a politician or for a celebrity or something like that. Do you think that's going to be a big problem down the road?

Jasper Schoormans: Yeah, I wonder I wonder how how influential that that would be because like with Photoshop, we've already had the capability to make very good looking fakes for maybe 20 years. Of course, it's easier now.

For sure. I think these so these voice scams scams for for instance, where they can clone your voice with with like a minute or something. And they call people like parents with the cloned voice, you know, it's bad. But yeah, there's nothing we can do about this, right?

Ryan Lazuka: We're screwed having our voices on this podcast.

Jasper Schoormans: Yeah, this is ready way too long. Yeah.

Hunter Kallay: In my personal opinion, I see these as more of like comforting the public and being like, oh, you know, this is our way of dealing with it. And saying that they did something about it. But at the end of the day, I'm like, these big companies must know that like the train is is is rolling and there's not much you can do to stop it at this point.

Jasper Schoormans: Yeah, I don't think there's any anyone who can do anything about this.

Ryan Lazuka: It's just gonna it's just gonna like sort of play out and it depends how how you ride the train, right? Like you can watch it roll by, you can jump on board. So I don't think it's gonna stop anytime soon.

Jasper Schoormans: Yeah, and here in Europe, they try a lot of regulation, but of course it's never gonna work.

Ryan Lazuka: Yeah, but Europe's at least doing something here. All of our politicians are like, we're gonna come down tough on AI, we're gonna regulate it, and we're gonna do this, and we're gonna do that, and then nothing, they don't do anything. So it's just like all, just talk about it, nothing happens.

Jasper Schoormans: Yeah, but things go too fast, right? It's for this whole process as well.

Ryan Lazuka: Yeah, yeah, it's a fine dance. It's a dance, you know?

Hunter Kallay: It's tough, because if you do anything, the question is like, okay, well, whatever you do, it's gotta be right, because if you don't do it right, then you're gonna get criticized because you're really screwed it up big time.

But then if you don't do anything, it's like, well, now you didn't do anything. So they're really stuck in this middle ground of like, okay, well, we're looking into it. We're making another announcement, we're having another meeting, and we put these in our newsletter occasionally, but we can put them every day if we really want it to. Another act is signed, another meeting has been done to discuss AI threats and stuff like that. But it seems like this stuff's going on all the time, but at the end of the day, it's like, only time will tell what's going to happen.

Jasper Schoormans: It's very difficult because it's a global thing, right? You can't really regulate things. Well, like, I think in China, they do a lot of, they have a very good grip on things because they're very tough on everything.

Hunter Kallay: Right, right. It's tough because countries are also feeling like they might fall behind if they regulate too much because other companies or other countries are full steam ahead. I feel like, well, if we pull back, we might be way behind the curb in two years on the innovations that the other countries have, maybe in military warfare or technology or just really their economy, whatever it might be. So I think that's another factor as well.

Ryan Lazuka: One thing too is like, say if 75% of the world's countries come down hard on AI and regulate it heavily, well, all it takes is for a couple of countries to say, we have no regulation in AI and AI runs wild because everybody, all the big companies work out of those countries without any regulations and what are they gonna do? I mean, I could see it's starting a potential war one day between countries, depending on what the regulation is on AI, but we'll see. It's gonna be interesting.

Jasper Schoormans: Yeah, it's gonna be very interesting, but even then, it's easy to hide some GPU from somewhere, right? And you're talking in a cave.

Ryan Lazuka: Yeah, all it takes is one company to work around the regulations and who knows what will happen, good or bad. I think one of the biggest plays, and you mentioned it a little bit very early on, is the marketing aspect of it, right?

Like on your site right now, Hunter alluded to that you guys have, or he said, he went on and saw that you have awesome demos, which you guys do. And one of the coolest things about it is like, say if you're a small or medium-sized retailer and you live in the middle of the country and United States or New York or wherever, and you want your model to have the clothes on your website in New York City, or you want the model to be on a tropical island, well, you can do that on outfit AI, right? Like you can take a picture of your model and put the clothes on it and put them anywhere you can imagine in the world.

Jasper Schoormans: Yeah, yeah, and that's very easy to do, of course. So right now, I always use like the fashion cities as backgrounds like Paris, Milan, New York, but of course everything works, beaches, runways, whatever you would like. So I think that would be a very cool thing for stores to sort of adapt these things to their own brand feeling, right? So it would be like a personalized way to market this, but yeah, you might like next to the piece of clothing and how it looks, you might also want to sell a feeling. So you can put the customer with this piece of clothing, like say you might be selling outdoor wear, put the customer in the mountains or in the forest, and it looks like a proper fashion photo shoot.

Ryan Lazuka: Right, and I can imagine that there's probably stores on Shopify that your product will be a great fit for, and it sounds like there'll be a plugin or something on Shopify that the store can download and then use your tool, but I can imagine there's a lot of stores on Shopify, small, medium, and large that just have generic pictures of their models right now, and if they put their models on a different light, like out of beach or in a mountain, if it's a mountain climbing gear store or something like that, it could increase their sales by like 10 or 20 or 30% or something, like that probably can happen for them right now if they have their pictures in a better light than what they are right now.

Jasper Schoormans: Yes, it's technically definitely possible, and it's of course a fraction of the cost of actually going there and finding a model, a photographer, and the whole crew.

Hunter Kallay: So how do you see this changing the model industry? Because you're looking at it, you talked about, I mean, everybody wants to talk about jobs, but less concerned about that, more concerned about, when people see models, they're like, oh, well, this model is unrealistic in their body type, and then some companies go the whole other way, and people are like, well, now you overboarded it, and that's like, this is unrealistic for the average person. So it's like, okay, let's do away with all the models, we'll just put you in that place, and there you go, personalized. How do you think that that sort of thing changes the modeling industry in general, and how people view models maybe, what happens to models and things like that?

Jasper Schoormans: I don't know, it's gonna be very interesting. I think some image generation models, so mid-journey, for instance, this looks so realistic, and you can make these models look so good that I feel like we're not far from the point where it's really not possible to distinguish between real and fake anymore, and at that point, why bother with models, why bother with real photo shoots? Then you're also not limited to any physical constraints anymore, like gravity, you can be anywhere, do anything you want, maybe famous people, that could be like a selling point, right, but unknown models that just look good, I don't feel like any, would have any advantage of over fake images.

Hunter Kallay: The interesting thing is that for years, people have been upset because they've been editing the images of the models, companies have been doing this forever, ever since Photoshop came out, as they're taking these human pictures and editing them a ton anyways, what's the difference if you just start with the generated image, rather than an actual person, what's the difference at the end product anyways, I guess on these company sides, right? Yeah, exactly, yeah. You have a background in the medical fields, can you just talk to us a little bit more about medical imaging, because that seems to be something that's really important, that's coming to fruition, I mean, you're talking about like, analyzing like screens for breast cancer, and all this different research that's coming out, seeing like people diagnosing autism based on like, eye scams and different stuff like that, given your background in that sort of field that combined with the imaging, where do you think the potential of this has to go, is it as great as people say it is right now, is it better than people think it is, where's the imaging at right now, and where do you see it going in the future in the medical field?

Jasper Schoormans: Yeah, great question. So I must say I've been out of research for a couple of years now, but yeah, we've seen like great improvements in the last years, we've been one limiting factor has been lack of data. So it's very difficult to find good medical imaging data sets because of privacy laws, and you can just use images as you want, right? So collecting data is a very long process that academic institutions, they might take years and years, and then they come up with like a few hundred or a few thousand patient data points or images, which might vary a lot of like in quality and how they've been made. So compared to other fields, the data just isn't as great as it is. But I think that just means that there's like a lot of room for improvement still in that field.

So it lacks behind a couple of years, but that's good. But then of course, you know, there are like constraints in physics, right? In measurements you do. So whether it's an ultrasound or an MRI or CT, you only have so much data to work with.

And it's different from like regular data, gen AI where you can just imagine things up or try to fill in the blanks because you need to be working with actual measured data and things need to be very reliable as well. Right. That's way more difficult and way slower than other fields of course. Right.

Hunter Kallay: I mean, this is one of the things that I've seen as a trend in the medical applications of AI is that a lot of it is trained on like these closed cases or really old cases, not on predicting actual real time cases. We had this issue also in some other industries that we've looked into. Ryan, if you remember when we talked to the Cleveland, the lady from Cleveland who did the, she looked at the different cases, like some criminal cases and rape cases. And it was able to do really well with the data that it had, but it was all trained on old data, closed cases from long ago, because that's all you could get access to. I wonder if there's going to be a way to utilize AI for real time cases in the coming years. Maybe I'm thinking when people get medical imaging done, there's like a waiver that they sign that's like, oh, this could be used for AI training data or something like that.

I could see that sort of thing happening and it could get really interesting. But yeah, there are various industries, like you say, that I think struggle because they don't have that real time data. It's all these closed cases, these past cases that's like, okay, great, it predicted with 80% accuracy things that were done 10 years ago. It's like, okay, what are we going to do now if we don't have access to data?

Jasper Schoormans: Yeah, but then one thing that's very promising, I think, is synthetic data. So the better like generative AI becomes, the more useful the synthetic data you can make to train like other types of models in that as well.

Ryan Lazuka: So in Jasper, when you say synthetic data, for the people that don't know what that actually means, it means that the AI creates its own data to train itself off of.

Jasper Schoormans: Yeah, exactly. Which of course has many pitfalls and things you might get worried about. Like, is this synthetic data, how representable is it gonna be of the real world? And so that's a whole like field and discussion in itself, but it's one of the possibilities, I think, to solve this.

Ryan Lazuka: Yeah, it's hard to even comprehend that, right? It's like once the AI becomes so good, it doesn't need any kind of real world data anymore, it just trains itself. That's kind of hard to wrap your head around, but it sounds like it's possible. Just there's gonna be some glitches along the way. Hopefully they're not glitches that make things terrible for humans, but we'll see.

Jasper Schoormans: Yeah, we'll see, we'll see, maybe shortly.

Hunter Kallay: That is interesting. So what do you think about these doomsday scenarios, Jasper? Are you a doomsday AI guy, or are you more hopeful about AI in the future?

Jasper Schoormans: No, I'm hopeful, but I do understand where they're coming from because it's amazing to see how good, for example, GPT-4 is in reasoning, like in writing codes, which just works now, like long scripts work, you don't have to do anything anymore. So I think it's easy to see how things could go wrong. But then, so the one issue I have with all these doomsday scenarios is they always sketch the scene where everything's going fine in the world, and then this AI decides to do whatever it's gotta do, and as a consequence, the entire world dies because, I don't know, it didn't take that into account. But then I feel like there could be issues and dangerous things, but why, we'll see it when we get there.

And I don't think the first issues are gonna be that big, right? Like the first dangerous, there might be dangerous things, but it could be that AI does something that kills 10 or 20 people. Well, if we're there, we can think about it, but I don't think we're gonna be at the stage that it's safe, nothing happens, and then a million people die at first. There's gotta be steps in between.

Ryan Lazuka: Well, same thing with any new technology, like driving cars is dangerous, but we still do it, right? And we get better at it. Like, you know, car safety has gotten exponentially better over the years, so it's probably gonna be the same type of thing with AI, hopefully. One of the interesting things, like Elon Musk, he's making these AI robots, you know, the Optimus 2, I think they're called. And one of the interesting things that he did with the robots is he made them so they're lightweight, they're only like, I think maybe 50 pounds around there.

And he said the reason why they made them lightweight is because if they do go rogue, then a human can overtake them if they need to, you know, they can overpower them.

Jasper Schoormans: So, right. But they could work together, right? With like 10 of them.

Ryan Lazuka: Yeah, the great point. Right, but maybe you didn't think about that, yeah. You should only have one robot in your house cleaning for you then, don't get more. Oh, sure, yeah, yeah, yeah.

Hunter Kallay: Yeah, I mean, Ryan makes a great point as far as like, I think that there's gonna be things that come up along the way that we don't see right now. I mean, when they're in, people are inventing cars, it's like, they didn't know anything about, you know, the stoplight technology or automatic braking and stuff like that. It's like, those are things that weren't even thought of when cars came out. And now that cars have been implemented, we've seen some incidents, we've learned from them, kind of like you said, Jasper, there might be some things that come along. We learned from those and then we're like, okay, well, maybe we need to implement this or maybe we need to put this parameter in.

And we kind of explore along the way. Same thing with planes, like when they came out, it's like the first planes being flew or like, okay, I'm not getting on a plane. But now it's like the safest way to travel, right? You got like autopilot, you got all these like safety features. So I think AI could be something like that, that there's a lot more that's going to surface as time goes on. Yeah, yeah, agreed, yeah.

Jasper Schoormans: We'll see it when we get there.

Ryan Lazuka: It's gonna be fun, a fun ride. I mean, even right now, it's like every day you jump on and see what new news is out there for AI in particular. And it's like, it's, our world is entertainment in and of itself. Like you don't need to go out and watch TV, just see what's happening in our real world in terms of how AI is sort of becoming a new and better thing every day. That's entertainment right there. Just jumping on the news every day and seeing what new stories are out there.

Jasper Schoormans: Yeah, yeah, yeah. It's overwhelming as well, of course. And I feel like there's this fear of missing out. And so many people do so many cool things like you wanna be like keep on top of it. And try it out.

Like I wanna do my own stuff. And you know, but it's really very, it's amazing. This like these last like one and a half, two years. Like every, I think every month there's been a sort of like a thing there, like how is this possible?

Hunter Kallay: I remember just about a year ago, Ryan calls me on the phone and tells me about how there's this thing online. I didn't understand what he was saying. He's like, there's this thing online. You can plug in and say, you know, cause he was talking about his, one of his, you know, his daughter and he's like, yeah, I can type in and say, hey, write a bedtime story. And it literally writes a bedtime story. And then it's like, and you press regenerate.

And then it just regenerates something completely different in like a second. I'm like, no, this can't be real. I'm thinking like Ryan's like got some sketchy thing going on. I'm like, what is this? And then it turns out like there's just, that was just the start.

I mean, right? Like some people are just amazed with the chatbots in general. And that's just the start of what AI can do. So it's really incredible just to have seen what's going on in the past. Like you said, just the past year and a half. Yeah.

Jasper Schoormans: And I wonder how many people are still skeptical or don't even realize all of this. Like what's possible. I think many people that are not like always online and just do, I guess like normal people stuff, they don't realize this at all. Or there might be skeptical and don't really see the possibilities yet. Yeah.

Ryan Lazuka: There's, I think there's a huge crowd of those seven people out there. I do like short YouTube informational news videos. And some of the comments they get, you can just tell the people, they don't know about the technology yet. And most of them are just haters, you know? They're just hatred. Cause they're afraid of it deep down inside. That's the problem. Yeah.

Jasper Schoormans: Do you think that's just because they're afraid or?

Hunter Kallay: I think so, Ryan. What do you think? I mean, I remember the first time I used to chat to you, I was like, okay, well, it is scary, right? I mean, you get on there and you start putting stuff in and you're like, oh wait, maybe this can do better than me at a lot of things. You're like, okay, well, now I'm trying to test it and trying to find ways. And I've been in a couple of trainings for like, how to use AI and education and things like that. And one of the things that they love to do, the educators love to do is just say how limited AI is at this or that or they love dog piling about, oh, well, they can't do this well or it can't do this well. There is a lot of hostility and negativity towards it for those who are new to it for sure. What do you think, Ryan?

Ryan Lazuka: Yeah, I think it's, I think there's a turning point. I think most people when they're confronted with a new technology, unless you're a technologist or a programmer or something like that, your first initial reaction is to hate it. And then once someone shows you a tool that can make your life way, way easier, you're like, holy crap, this is the coolest thing ever.

And the email, I know this example gets brought up over and over again, but like the internet and email, like everybody uses email now, whether you hated it at the beginning or not, even your 80 year old grandparents probably use it in some form or the other. So it's just a matter of finding something that improves someone else, improves your life or someone else's life. And that's sort of just like the aha moment that gets most people to jump on board.

Hunter Kallay: I wanna say one thing to that, Ryan. That is so true because I was talking to this one professor and he goes, I don't want any of my students getting anywhere close to AI. It's gonna stifle their learning, blah, blah, blah. And I'm like, okay, well, what if AI could help you grade, you know, all of their assignments? He goes, well, well, well, I guess that would be good. And you know, I'm thinking, well, well, well, well, well, well, well, it's the difference. You know, you can use it, but they can't. Now hold on a second. But you're right, you're totally right until they can use it.

Ryan Lazuka: Yeah, mind-numbing tasks, like grading papers, I mean, that's gotta take hours and hours and hours. And if you just have AI do it for you, it's like, holy crap, that guy's gotta be like, this is amazing, you know?

Hunter Kallay: Like, I could just like visualize his mind, just like come out of his head and flip and then go back.

Ryan Lazuka: Is there anything else that you wanna promote other than the outfit company that we have you on here for in the first place?

Jasper Schoormans: No, no, no, no, no, it's outfitai.co and I'm very busy building that. I'm on Twitter as well. And yeah, thank you so much for the invite. It's been great talking to you too.

Hunter Kallay: Absolutely, so that's outfitai.co. And then be sure to check out Ryan and I's weekday AI newsletter. You can get the hottest updates, the coolest tools, including tools like this, outfitai.co. And then we also got a mystery link too. Every single day we send a mystery link, it'll take you to something very interesting in the world of AI. And then on Sundays, we get our deep dive articles into cool developers and developments like this one. And then be sure to subscribe to this YouTube channel behind the bots so you can see all our interesting and intriguing AI interviews.