Learning about the benefits of AI image generation. But also how it can go badly wrong!
Learning about the benefits of AI image generation. But also how it can go badly wrong!
Hello everybody and welcome to the Crush Bank AI for MSPs podcast. My name is David Tan and I am your host, as always, and I appreciate you taking the time to tune in and listen to hear what we have to say. We're going to try something a little bit different tonight. It could go very well, it could go horribly wrong, I guess. We'll figure that out as we go. But what we're going to be talking about today is art, sort of. You'll understand what I mean as I go through it and, as you can imagine, when we're talking about art and creativity and visual mediums, some of what I do when I talk in this next little bit on this podcast will not resonate as clearly as it could if you could see it. Hopefully, I'll be able to make everything I'm talking about super clear and concise so you'll be able to follow along and understand what I'm saying and what I'm showing just based on the audio cues. But, that being said, we're going to release, simultaneously with this, a blog post on our website, so that'll be at crushbankcom forward slash news and I'll post the link. We'll share the link in the description of this podcast and we'll I'll mention it again as I go. But what we're going to do basically is I'm going to talk about some images and some visual representations that I am sort of creating on the fly You'll understand what I mean by that in a little bit and then we'll post the images on the blogs. You'll be able to see them visually as well. So, again, a little bit of companion piece, not necessary to understand exactly what we're talking about per se, but I think you'll find it fun and interesting. And the other thing we're doing a little bit differently on this is that I'm going to be, as I said, doing some creating on the fly as I go, so hopefully that won't make for too disjointed of a listen. Hopefully we'll keep it pretty smooth. Hopefully we'll make it through it. I'm hoping to not have to go back and do this six or seven times You'll never know, because I'll probably not admit if I had to record it over a bunch of times. But you'll get the point and hopefully you'll understand what I mean. So we're going to start off talking specifically about Photoshop, and I know what you're saying.
David Tan:This is a podcast about artificial intelligence, specifically for managed service providers or really for any businesses, particularly small to medium sized businesses. What does Photoshop have to do with that. Well, what I'm going to focus on if you're not familiar with it is the new fairly new generative fill functionality in Photoshop. So give you a little bit of background and a couple of disclaimers. First of all, I am a huge Adobe fan, so I am going to be talking about some of the ups and downs of this technology a bit as I go. That is a no way reflection on Adobe. I think their products are fantastic and I think what they've done here is incredible and I highly recommend people work with it.
David Tan:The other thing that I think will be interesting as a point of background for this conversation is that I've been using Photoshop since probably version one or two, like legitimately. I got my first Mac in 1990, 89 or 90. I got Photoshop soon thereafter. I have zero artistic ability legitimately zero but I am really good at Photoshop. Think of me as the engineer's guide to using Photoshop. In other words, I can't do anything artistic to save my life, but I just know the platform and the tools so well that I can accomplish a lot. Like example, if I needed to retouch a photo or do some photo editing really intensive, I can do that and again, it's not through my artistic ability, it's just through understanding the platform, understanding Photoshop and how the tools work. Again, that'll become important as we talk about this a little bit. So let me state the stage for you a little bit here. So I'm logged into, I have the Adobe Creative Cloud Studio and I'm logged into Photoshop 2024. And they've added, like I said, some really cool AI based functionality, specifically something called generative fill.
David Tan:So what I have up on my screen in front of me and this is one of the images I'll share is a very quaint picture of a white house that seems to be sitting amongst a bunch of trees with a mountain in the background. There's some very colorful flowers, the house is white, a lot of greenery, very cool little landscape. It's not landscape. Cool little still life of a house, or they call it a. The name of the image is a ranch, so I guess it's a ranch, it's a house, doesn't really matter, it's not important for this. But if I look at the picture, like I said, there's a white house sitting in the middle of it, some purple and red flowers in the forefront, but there's this very big, obnoxious looking tree that overhangs the photo in the front left of it, and it just for some reason, it just bothers me aesthetically, right, I'm sure it is supposed to be there. Obviously, if I saw it in person I'm going to assume that that tree makes a lot of sense, but it bothers me in the photo and I want to take it out.
David Tan:So, traditionally, what I would have had to do in this type of scenario is use the selection tool and the rubber stamp tools and the lassoes and the Matting and masking and all the other things you do in Photoshop. That I'm not gonna get into and it would probably take me again on Artistically because I couldn't just paint over it hours to do, but I'm gonna try Adobe's generative AI. So what I'm gonna do is I'm gonna grab the lasso tool and for those of you that don't know, the lasso tool is sort of a free mode selection tool, what lets you draw Free lines and then it closes the loop of the line you draw and it serves as a selection. So I've drawn a lasso Very roughly around the tree that I described in the front left, and when you do that in Photoshop it pops up the generative fill Box, that the box where you can kind of have those controls, those tools on demand and there's a field there called, literally, generative fill. So what I'm going to do in that fill it says when you click on it says what would you like to generate.
David Tan:I am gonna say remove Tree and press enter, and and when you do that it goes and it Pills up a progress bar. It does a bunch of things in the background. I am talking because it's actively generating. It Doesn't take very long, but it does. It's not instant, it does take some time and now it's done and literally that Tree is gone. And again, I'm gonna post some before and after pictures so you can see exactly what I'm talking about. But the other thing it does it's kind of cool you can give some feedback to help train the model. So you can click on it right here and I can give it a thumbs up, I can give it a thumbs down. It also gives you three variations, so it lets you click through them and choose the one you want. So in this particular case I really like number two, so that's what I'm gonna choose. So that's the one I'll post. So you'll see. It's the same picture again, but the but the tree is gone.
David Tan:The next thing I want to do to edit this picture a little bit is I want to Change the front bottom of it. Right now. There are some purple flowers there, again very pretty. But there's something about this that looks like it should be sitting on a lake. It just feels like a lake house to me. I don't know why. Maybe it's like the Sort of the optical illusion of a reflection short of in the bushes. I don't know, it's again, it's hard to describe, baby, you'll see it when I post it.
David Tan:But in this case I'm gonna do the same thing. I'm gonna drag, grab the lasso tool and I'm gonna lasso about the bottom, I don't know 10 or 15 percent of the screen. And then I'm going to do the same thing. The generative fill box is gonna pop up and here I'm gonna type Lake. So again, you could probably just hear me typing sorry for the noise of my keyboard. So I type add a lake and it's doing the same thing. It is generating it. We're gonna wait a minute or so and that should be done In just a second. Now You'll see it comes up and again same thing. It puts a lake in front of there, gives me three options to choose from. One of them's just a lake, one of them actually seems to have what looks like a boat ramp, and one of them has a floating dock in it. So that's pretty cool. I'm gonna choose the one with the boat ramp because if I live on a lake house I'm obviously gonna want a boat right. So that's kind of cool. And the other thing again, this is not an advertisement and or it's not meant to talk about Photoshop or say, but it does create these in layers, so you'll be able to kind of go back and add and remove them.
David Tan:One last thing I'm gonna do. This photo is I live in a very house now, or this is a picture of my cool little lake house now. But the house just doesn't fit anymore with that that vibe. It just needs to be, in my mind, a little bit more rustic. I'm in the mountains now. I'm on a lake I.
David Tan:So there's a new tool in Photoshop called the Object Selection Tool, and what's really cool about it is it's super smart. So in the past, if you wanted to select an object, you would have to sort of draw a fine line around that object and figure out what to highlight and how to highlight it. But now the tools are so context sensitive again all by AI that if I hover over, if I go to the Object Selection Tool, I hover over it, it selects the house for me. So I now just have the house selected and same thing generative fill, and I'm going to say turn this house to a log cabin. So I want something like I said, much more thick, a little bit more quaint, and we'll see what that looks like. Again, it's going through the generative fill, but you get the idea and again, this will be kind of cool, hopefully, if you look at the images of it. But super easy, super intuitive, really simple and beautiful Came back, gave me three to come in this case, so I'm going to leave it. So now I have turned this ranch and the woods to a log cabin on the lake in the 10 minutes or so that I have been rambling and talking here. Very cool technology, very powerful. Like I said, I love what Photoshop is doing. I love the idea of using the generative AI and making it accessible and available, so you don't need to know anything about the technology in this case in order to take advantage of it and to leverage it and to do something really cool with it. That being said, though, that creates some dangers too. So when I was playing around with this I had some ideas. So let me give you a little bit of a round.
David Tan:Over the years, in all the companies we've owned, we've generally had someone young in the marketing department. For a number of reasons. My partner's got a background in marketing so he can sort of lead from a senior level there. So we didn't necessarily always need someone senior in marketing. We have some other people we work with that have really strong marketing background. So, though it just seemed like logical, we love giving young people an opportunity, especially on the marketing side. They can be kind of energetic, fun to work with when you're trying to be creative and get a business started. That's no comment about anyone else, it's just something we had done traditionally. But that has its ups and downs as well. So one of the things that we've always found all of them.
David Tan:Over the years we had six or eight of them and all of them were great, but they all had their limitations. What I mean by that is someone would be really good at writing but have no artistic skills at all, or someone would be really good at photoshop but couldn't write to save their lives. So we never had, especially when you're dealing with someone sort of young and starting out in that position. We never had that, that expert that could wear many different coats and and and carry on many different roles. So I had the. I was thinking that this would be kind of a cool Way to supplement that skill, right? So let's take one of the marketing associates we had that could write really well but had no artistic skills. Maybe we could equip them with this Photoshop, with generative AI, and then we say, hey, here's this image, I need you to touch it up for me. Photoshop could do that for them without having any artistic skills. So that got me thinking, because I have a bunch of, in the next six weeks or so, crush Banks gonna be.
David Tan:It's at six or eight different trade shows and I'm doing some speaking engagements at most of them, quite frankly, and a lot of times when you do that, they like to use a headshot, both in the you know and the Agenda leading up to it and some of the marketing publicity. And also sometimes they insist in putting it on the screen when I start talking, and the bigger the room and the bigger the audience, the bigger the headshot. So, for example, a few weeks ago I was at the IBM think exchange conference out in Las Vegas and I was one of the speakers During a keynote on the main stage. So you can imagine that's a big room with a big screen and I had my big face there staring out at the crowd and I was horrified. I'm gonna show my Vulnerability and vanity in the next few minutes, so please bear with me, but anyway it got me thinking like, hey, what if I wanted to use Generative AI to touch up my headshot a little bit?
David Tan:So I pulled up my headshot and I went and used the lasso tool again and I lassoed the top Quadrant of my head that's got hair on it or what's left of it. My headshot definitely shows my ball, my balding and my age. It is what it is. I've come to grips with it. But hey, if it's gonna be full-size and an auditorium for 5,000 people, if I could add a little bit More hair on there, I'm gonna do it. So I'm doing this again as we go.
David Tan:I just generated I sorry, I just pulled up my headshot and I lassoed the top of it and in the generative film prompt I'm gonna type add more hair, and the results of this were absolutely Horrifying. So, unlike the previous examples where it pulled out the tree or added a lake or turned a simple house to a log cabin, it had no idea what to do with my hair. And when I show you these, when you look at these, you will laugh hysterically, believe me. So obviously this was a no-go. I couldn't add more hair, I couldn't clean up the Some acne I had on my chin for this picture I couldn't, you know, I couldn't take out a couple of grays that I had. I couldn't do anything to touch up this picture. I just had to live with it, which is fine. Again, like I said, it was sort of a joke and experiment and I'm okay with that. And again, I'll post this and I will show my vulnerability because you will laugh at it hysterically.
David Tan:But it got me thinking a little bit and what it got me thinking was that, in this particular scenario, I have no idea what the model that this generative fill AI is trained on. I don't know if Photoshop and this is true, I have no idea Photoshop did this themselves. I don't know if they partnered with anyone. I don't know what training data went into it. It's completely Obscured for me. It's what we call the black box problem with AI. Right, something goes in on one side, something comes out on the other side, we don't know what's in the middle of it, and that becomes the real danger Of this. So I said, like the technology is really cool and I would love to give it someone. Give it to someone to sort of ride side saddle with their job, function and their Responsibilities, but I have to be super Aware and I have to be super careful.
David Tan:So in this scenario, if I had given my marketing associate my headshot and said, hey, can you just run this through Photoshop generative AI I know you don't have any artistic skills, but just run it through Photoshop generative AI, I'm sure it'll come out really well and then just send that image to the, to the conference, to them you know, the sponsors or the people running the conference and have them post it. I would have been horrified when I walked in. Now, when you see the picture, I can assure you nobody would have sent this to someone. That's not the point. The point is that if we don't know what are in these models, we don't pay attention to the inputs and outputs and we don't chaperone it. Quite frankly, as the expression we use, there is a real danger of unintended consequences, unintended results, unintended output.
David Tan:Super easy if you're looking at a picture, you can tell if it's a horrible result. If you are listening to music that was generated by, if you're even reading something that was written by generative AI, you can generally tell, with the caveat that if it's something that you don't understand so if I ask you to write a paper about thermodynamics and you know nothing about it, then you're reading the output of it is not going to do you any good. You still need to be able to chaperone it. You still need to be an expert to write side saddle with it. But this also comes right.
David Tan:A lot of people use it for writing code and again, I think that that is a really cool, really powerful use case for generative AI. I've been playing around with a model called StarCoder, which supports something like 80 programming languages, and, yeah, I can take my JavaScript code, or I should say I could take my NET code and convert it to Java. Ibm you know our strong partnership with IBM. They've built some functionality where you could take NET and turn that. Sorry, you could take COBOL, a really old programming language that's got no expertise left anymore and turn that into Java. So, again, there's real value there. But if you don't know how to test that or how to understand that, then that becomes really dangerous and something you need to be super careful about.
David Tan:So, really quickly, I want to change topics here for a minute and talk about something else artistic, same type of vein. Right, talking about the dangers or the dangers is too strong of a word. The concerns the things you need to be on the lookout when we're talking about trusting generative AI and trusting these platforms blindly and exclusively. And this comes from an article that was in the Times a few weeks back now, towards the beginning of September, and what the article is about is there was a museum the Duke University Art Museum, decided to put on a art exhibit that was completely curated by ChatGPT. This is not meant to make fun of or pick on ChatGPT. They are just the biggest target when it comes to talking about some of this stuff. A lot of people just sort of throw content and info at ChatGPT and they think it will solve all their problems for them and be an expert. This is sort of a cautionary tale. This isn't as bad as the story about the lawyer who created a brief with ChatGPT that quoted case law that didn't exist or some other examples I can tell you, but this is a kind of an interesting one.
David Tan:So these Duke University Art Museum decided to use ChatGPT to curate an art exhibit and basically what they did was they actually did this the right way. They spent a bunch of time fine-tuning a model. What that means is you're taking you're essentially taking the focus of ChatGPT 3.5 or 4 whichever large language model you're using in this case and narrowing it down. In this case, they're describing the 14,000 or so pieces of artwork that they have to help it create a curated collection that they could show as an exhibit. The results were completely. They weren't bad as in dangerous or you know any other negative consequences. They were very unexpected, really.
David Tan:Again, you talk about that black box problem really impossible to understand, and what I love about it is there's a quote right at the beginning of the article from one of the curators that basically said I'm going to read it to you. We naively thought it would be as easy as plugging in a couple of prompts and if I could crystallize the issue with generative AI in one sentence, that would be it. It's not about feeding a few prompts into a system and getting outputs. There is a lot of work that goes into this, and if you don't understand the models that you're dealing with and the technology and how it works, you run the very real risk of it not meeting your expectations or, quite frankly, making your job worse or making it harder. Right Again, if you're building output that doesn't work, how do you troubleshoot that?
David Tan:We're a very big believer in what we call trustworthy AI, and what that means to me is that you have transparency. It's essentially, I like to say, breaking open the black box right. Monitoring transparency, making sure things like bias don't sneak into models, making sure hallucinations, making sure models don't drift, things like that. All of this stuff is going to happen to these large language models over time. The more they're used, the more they get trained, the more they learn. It's only natural that bias is going to sneak into the systems, right? That's how these things work. If you're a tech company, for example, and you're training your large language model about which resumes are the best fit and you're using current employees, but you happen to have an employee base that's 85 or 90% male, like a bunch of tech companies did back in the 90s and early 2000s. Well then you're naturally going to inject inherent bias into that, because all these resumes are men, so it's going to automatically assume the women's resumes in this case are not any good. That's just a very simple example of what bias is. But again, that stuff all creeps into the model if we don't do it carefully.
David Tan:So basically what happens is sort of close the loop on this art exhibit and I recommend checking out the article in the Times. Like I said, super interesting. Basically, what ChatGBT did was it shows 21 works of art. So that is not nearly enough for an art exhibit. So it didn't really understand inherently what an art exhibit was. So that was sort of the first problem and it was very eclectic. They said Again, that's a line right out of the article. I would say it was an eclectic show, very disjointive, even though it matched the theme. And what was interesting about it was they couldn't understand again to sort of touch the black box problem. They didn't understand why it shows the pieces that it shows.
David Tan:And the prompt the simplified version of the prompt was act as if you're a curator and, using your data set, select works of art related to the theme of dystopia, utopia, dreams and subconscious right Great prompts for AI to build, especially when you talk about dystopia and utopia. We can spend another three hours talking about that, but I'm not going to do that. So it did some things that were kind of cool, right? So it created, it insisted on using a bunch of dolly works. I'm going to be a huge Salvador dolly fan. He's one of my favorite artists, so not surprising when you talk about utopia, dystopia, things like that, that a bunch of Salvador dolly works and makes it way into the selection. But it also shows some really weird ones, right. It shows some ancient sculptures that have nothing to do with it. It shows a Mayan vase called consciousness again, makes no sense why that was in there.
David Tan:It just didn't understand the assignment, quite frankly, if I can be trite for a moment. And the other thing it did poorly was it did not do a good job of describing the works of art. So they asked it to write the descriptions and again I'm quoting from the article here it created a bunch of bromadic taglines, so, for example, taglines like experience the art and immerse yourself. So, rather than you know, still life with flowers or you know whatever the tagline would have been for a dolly painting. It went for this much more kind of rudimentary, obvious, less descriptive titles, which I thought was kind of interesting. Again, that all has to do with how the prompt was fed in, how the model worked. Now, again, I don't know the level of data science expertise that these people had. I don't know if they worked with anyone with data science. They could do a better job by a bunch like multiple shot prompting examples and probably more fine tuning Really prompt tuning would probably been ideal for them. So there are certainly ways they could have done it better, but that actually, in my mind, demonstrates the problem.
David Tan:Most people believe this technology is very approachable, which it is. It gets embedded into systems we use, it gets added onto platforms we use and that's really cool. But if the vendors don't understand it which I found to be the case in a lot of cases if the users don't understand it, there is a very real risk. And again, I think the thing we all need to focus on and worry about is this sort of AI governance and trustworthiness and transparency. So, yeah, I thought that might be kind of an interesting exercise to take this what we talk about here again specifically around AI models and generative AI and turn that into something in a different realm, but that you could see how it relates right. The thought of models that you don't understand and using them the wrong way and trying to get something created that you're not qualified for to either create or review dangerous in any walk of your business, whether you're in managed services or whether you are a managed service provider listening to this and dealing with your clients. So I thought that'd be kind of fun.
David Tan:Like I said, there will be a visual companion rather to this podcast that we'll post on our Crush Bank blog. So wwwcrushbankcom forward slash news and we'll put the link in the description of this podcast as well. So check that out. Laugh at the pictures of me. I am not going to leave them up there for very long because they will be used against me in a multitude of ways. Anyway, thank you so much for taking some time to listen to this episode of the Crush Bank AI for MSPs podcast. Love to hear some feedback. If you have any thoughts, questions, comments. You want to challenge anything I say? You want to ask me to clarify something? You have ideas for future episodes? Hey, you want to come on and chat? I welcome it. I'd love to have it. I set up an email address just for this podcast. It's just podcast at crushbankcom. Shoot your feedback there Again, like I said, absolutely love to hear from you and thanks again. So much for tuning in. We'll talk to you next time, music.