Loading summary
Transcript1 lines
- [00:01]
A
Welcome back to another installment of Is AI Taking over the Style World? In this episode, you'll discover if ChatGPT can replace your favorite stylist. I know you're dying to know, so let's get started. Welcome to the Everyday Style School, the show that teaches you everything your mom never did about getting dressed. I'm your host, Jennifer mackey. Mary, after 25 years of dressing women with real bodies, real budgets, and real lives, I know great style isn't about following one size fits all advice. It's about learning what works for you. Hello, gorgeous. Here we are, part two of our little mini series. Can AI Make Style Easier? In the first episode, I talked about those AI style apps that you are constantly seeing on your social media feeds. And it's funny, I was talking to a group of women at an event last week and someone asked what I was working on. I told her about these episodes and she really didn't know what I was talking about. So I said, well, have you seen those ads on social media for like the ten piece Amazon capsule wardrobe? And she was like, wait, those are AI? I had no idea. And if we're being honest, I bet a lot of us see things and don't realize that they're not real or really created by humans. Anyway, I will fully admit to sending my husband some cute baby penguin videos and then having to send a follow up text that says, never mind, it's not real. Point is, AI is creeping into all corners of our lives and it's getting harder and harder to tell the difference. So it's important to be aware and educated and. And that is what these episodes are all about. If you haven't listened to part one yet, I really recommend going to check that out. The reality is you're probably more likely to use the apps for style help, which is what I talked about in that episode, so don't skip it. But in this episode, we're talking about what we usually think of when we say AI, which is large language models like ChatGPT or Claude Gemini or that kind of thing. I think these are important to talk about in the style arena because a lot of people are going straight to these tools to get information instead of using Google or another search engine. And these things work differently. AI gives you the answer and search engines show you where you can get the answer, which is a completely different way of doing it. Like most things in life, there are pros and cons to both. It is so much easier and so much more efficient to just get the answer right. And as opposed to going to each site realizing it's not quite what you're looking for and then you go to another one, it is definitely more convenient to just get the answer you're looking for. But on the other hand, a single source of truth is only as good as that single source of truth. And that's what we're going to get into today. How good are they really? Can this type of AI make style easier or make style better? We're going to find out. So if you remember to form my opinions on the AI style apps, I relied on reviews and hours of YouTube videos. For this one though, I went to the source and actually did the thing. And when I started, I posed the same two questions to three different models. I asked Claude, Gemini and ChatGPT. The questions I asked were what styles are best for apple shape women? And then I told them to create a 20 piece capsule wardrobe for an apple shape. Now, I know that these prompts were simple and kind of vague, and that was completely on purpose. I just wanted to see, with very little information what did they give me. A lot of the responses from all three models about what styles were best for an apple were kind of odd. And everything was about elongating the torso, which as an apple shape with a long torso, I'm kind of confused by because that is a vertical body shape thing, not necessarily a horizontal body shape thing. And sometimes they go together, but I'm living proof that they don't. So it was just kind of strange to me how much of the reasoning behind the advice was about torso length. Otherwise though, it was pretty standard. Here's what an apple should wear advice. I don't think there was anything I was like really, really surprised with. There were other issues with the style advice that it gave me, which we will get to later. When it came to the capsule results, there was definitely some overlap between the three models because the advice on what styles were best for an apple had some overlap. But Gemini and ChatGPT gave me 20 clothing pieces, while Claude gave me 16 clothing pieces and four accessories, which made for a much more limited capsule, but you could actually make complete outfits out of them. Gemini gave me a wardrobe for all four seasons, which again made it kind of less functional because only 20 pieces for four seasons doesn't leave you with a lot to work with. And all three models seem to believe that a capsule wardrobe simply meant all of the pieces were in the same limited neutral color palette, which, to be fair, when you have a capsule that small, you kind of have to have It a limited neutral color palette, there isn't room to put some fun colors or patterns in. But there is a lot more to capsules than color. Like do the silhouettes of the tops work with the silhouettes of the pants, that kind of thing. Those are things we think through, and that was clearly not a consideration for any of these models. In the end, all three were pretty similar in terms of the quality of the responses. There was not one real clear standout winner, but I thought ChatGPT was slightly better. So I chose to use that for the rest of my quote unquote research, which, by the way, I have to admit I had a ton of fun with. But for the next part of my little project, I went to our community and I pulled questions and photos from actual members. Don't worry, I cropped out your faces, any identifying information, all that good stuff, and and I put them in to see how my answers that I feel very confident in would compare to ChatGPT's answers. In the last episode, I went over some of the common app features and at the end I gave you my overall thoughts. But today I'm going to turn to the last page in the book and give you my final thoughts. And then I will tell you why I think that. And here goes. You ready? Overall, this is bad. It is not good. Please do not rely on tools like ChatGPT to reliably make your style easier or better. And I've got five reasons why. Here goes. Number one, the goal of AI is to make you happy, not necessarily to be right. This is something I heard on one of my favorite podcasts actually last summer, and it really changed the way I look at these models. And have you ever noticed if you've used them, how it agrees with everything you say and validates you no matter what? That's what this is all about. It's about making you feel good. Because what happens when we feel good, we do more of it. Like social media, it is either designed to get you addicted to using it, or it just naturally happens that you become addicted to using it. But that happens often at the expense of truth and good information, which to me is kind of counterintuitive because you know what makes me really happy? Correct information. That really lights me up. I guess I'm weird like that. But anyway, I will tell you what this looks like in the style arena in just a second. But first I got to give you a quick sidebar. I want to share how I use ChatGPT in my business, because it really does matter. Here I do not use AI to write my content for a lot of reasons, which I will. I'll share some of them later, but you know, mostly because most of it's AI slop, right? But I will use it for things like project management. Let's say I'm working on a course. I will give it just a big brain dump of all the things that I want to include, and I will let it create sections that make sense to people outside of just me, and I will let it suggest things that I might be missing. I'll also use it for research and brainstorming. Let's say I'm doing an episode on I. I don't know 10 ways to improve your style for free. I will say here are my ideas. Now give me 50 more ways to improve your style for free. I always know what I want to say ahead of time. But in case you have not noticed, I am human and I forget things sometimes. Like super obvious things, and I don't want to leave them out because my brain just didn't come up with it at that moment. So I'll always ask for a ridiculous amount of responses and normally they're just very, very, very stupid. But every once in a while I will be like, how did I miss that? How did I forget that? Or it will spark something else that I want to say under another point. And most of the time it's not useful, but now and then it is. And I tell you this because over time ChatGPT has learned my business. And when I started this little research project, I could tell it was using my ideas, my voice, my wording, all that good stuff. But I kind of got the feeling that that's not the experience that everyone is going to have. So for this I opened an incognito window and the results were vastly, vastly different. So what you get is not the same as what I get when it comes to making you happy instead of giving you correct info. Here's how it played out with this little project. I gave it pictures of outfits and the question the member asked like is this flattering or is this current? In my chat GPT, the responses were pretty decent. I would say they were maybe 75% of the way there. But in my incognito window, which mirrors the experience of someone who doesn't feed it style advice and style words on a fairly regular basis, almost every response to is this flattering or is this current? Was yes. At the end it would say, if you'd like me to suggest one small tweak to make it more flattering. I can do that. But it most certainly did not lead with that. And even when I said, yep, give me the tweak, it softened it quite a bit. It was definitely on the side of making the user feel good. So the next time you use ChatGPT and it tells you that you are the smartest, prettiest, wittiest woman in the land, take it with a little grain of salt. I mean, I am sure that you are. I'm sure you are the smartest, prettiest, wittiest woman in the land, but you get what I'm saying, right? If you are looking for brutal honesty and feedback about your outfit, this is not the place to get it. All right, issue number two, these platforms make stuff up. There are lots of ways AI gets things wrong, but this is by far my favorite. And by favorite, I mean the most hilariously frustrating. Now, if you are not aware of AI hallucinations, here's what they are. This is the definition straight from Google. AI hallucination is a phenomenon where artificial intelligence, particularly large language models, LLMs, that's what ChatGPT is, generates confident, plausible sounding, but entirely false, nonsensical or ungrounded information. Now, we're not talking about referencing slightly dated information or misunderstanding a prompt. This is straight up just making things up out of thin air. A while ago, maybe last year, something I don't remember, I was working on an episode and I wanted to include some research studies. You know, I love those. So I asked ChatGPT to find me peer reviewed scholarly articles on this topic and pull relevant quotes, which is way, way, way faster than doing it myself. And I think this episode was something about clothes and emotions. In seconds I had a ton of studies, tons of quotes, including one that said something like, our emotions are like clothes hanging in our closets. By choosing the correct ones, we could become the most beautiful versions of our inner selves and change the world for the better. And this was attributed to someone with a PhD after her name. I don't remember who or what the name was, but we'll call her Sarah Smith. Now, if you have ever read a research article, you know that doesn't sound like something you'd find, right? There's no flowery, poetic language. They're just like boring and dry. So I questioned it and I said, can you show the full article from Dr. Sarah Smith? And the response was, no, I made it up. Sarah Smith doesn't exist. And I was like, yeah, no, no, we don't, we don't, we don't do that. And in all fairness, I did not come across any hallucinations while doing this little project. But for me, it's enough to know that there's a possibility. Like you could ask ChatGPT what's trending this season and it could tell you that because of the Artemis mission, there's a nod towards space with metallic jumpsuits and moon boots being hot this season. Which that's definitely not a thing, right? This doesn't mean that all the answers are totally made up, or that they're even wrong at all, which ironically, is my third issue with them. If the information you get from things like ChatGPT was right all the time, we wouldn't have a problem because you could rely on it. And if the information was wrong all the time, we wouldn't have a problem either because we just would ignore it. We wouldn't use it. The problem is, is that it's right some of the time, but not all of the time. Unless you have the skills to know which is which, you can't trust any of it. Imagine I invited you over for a party and I've got this big table of food laid out and I told you 70% of these dishes are great. The other 30% will make you sick. Now guess which one? You wouldn't trust any of it, right? You wouldn't. You wouldn't eat anything. You would be driving through McDonald's on the way home because you're hungry. At that point, you're not taking a chance. It's the same thing here. If you don't have enough foundational knowledge to tell which advice is good and which is bad, you can't trust any of it. Last episode I briefly touched on an app that promises body shape and color analysis and a big criticism all across the board of this app was the accuracy. Some of the social media comments on this app said this was spot on, but enough of them said this was way off to plant that seed of doubt. If you don't know what your body shape is, how do you know if you got good info? A while ago I got an email from someone. She's like an Instagram expert or something. This had nothing to do with her business, but it said I found the best life hack of all time that you have to try something low key like that. The hack was using ChatGPT for color analysis and she said just take a couple of pictures, upload it and ask ChatGPT to tell you your season. Easy peasy free. Which again, like I said last week, this could be an absolute game changer. Later that day, I was meeting with our capsule creative director and we were talking about this, and she said, yeah, I did that. It was great. So I tried it. Wrong, wrong, wrong, wrong. How do I know it was wrong? Well, a couple of ways. Number one, I have enough color knowledge to question the results. But also I have been analyzed by every single color system out there. All whatever their version of the lightest spring is. That's what I am call it. By any name, I am light, bright and warm. The colors on those color cards, those are my best colors. But if I didn't already have that knowledge, and I just took Chat GPT at its word, I would be walking around in some dusty blues that make me look paler, older, and more tired than I really am. Which, no, thank you, I'm doing fine in all of those departments. But what I realized is that both the business owner and our creative director have very easy colors to determine. There's no question even I could do their color analysis, which is really saying something. I can tell you how to find your colors, but I lack the eye to really see it for myself. My coloring is a little bit trickier. My eyes are warm, my skin is neutral, and my hair is naturally very cool. So I decided to give ChatGPT another crack at it with another difficult to type test subject, who is my younger daughter. She has dark, dark, dark brown eyes, very fair skin that is clearly cool, very pinky, and medium brown hair with golden highlights. A couple of years ago, she wanted her colors done, so we did it. And even the professional said, yeah, she's tough. Eventually, though, we figured it out. ChatGPT was so far off the mark, it was ridiculous. It recommended her worst colors. And what I realized is that the harder you are to categorize, the greater the chance the answer will be incorrect. And that matters because the easier you are to figure out, the less help you need to figure it out. It's the people who are hard to type who turn to professionals, or in this case, turn to the gnome who lives in their computer for help. But if you don't know how to determine if those results are right or wrong, what are you going to do? And if you can't trust it, what's the point of doing it? So back to my little research project. The first photo and question I put into ChatGPT for this project was a member who wanted feedback on the flattery and currentness of an outfit. ChatGPT's response was perfect, perfect. It said everything I would have Said, actually this is the one that made me go incognito. It was that much like me, but even the incognito version got it right. And I had another moment of, oh, that's it, I'm being replaced. Yes, I discourage easily. It is what it is. Good thing I recover quickly. But anyway, I did another one. Another member asking about flattery and currentness. In this one, there was one thing, one thing that was throwing everything in the outfit off. There was one obvious issue. Yeah, there were some minor tweaks or well, you could do this, but. But to fix the outfit, I would have fixed one thing that was honestly, like, glaringly obvious to me. My chat and the incognito chat both missed it. Yeah, it gave a list of other things to fix a lot of those little like minor tweaks, but not the one thing that would have actually made a difference. Again, if you're someone who doesn't know what's wrong with the outfit just by looking at it, which is the reason you're asking in the first place, you can't know if the results are right or not. And if you have no idea how often it's wrong, you might be tempted to trust it. But getting bad advice is not always the AI's fault. Sometimes it's yours, which is issue number four. The advice you get is only as good as the prompt you give. And most people do not know how to give the right prompt to get what they need. Going back to that try this Life Hack email I just told you about. The prompt was evaluate the colors of my eyes, skin and hair and tell me which of the 12 color seasons I fall into. That was it. That was the prompt. Last fall, I created a bonus course for our annual Style Circle members who joined us at our annual holiday sale. And this course was all about finding your best colors. Again, I can't tell you what your best colors are, but I can teach you how to find them. I am the literal definition of those who can do and those who can't teach, and I am completely fine with that. So in this course, I think I gave four ways to find your color season, and I encouraged members to do all of them. Because chances are, one way will click or there'll be some breakthrough that you go, oh, now I finally see it. Or there's enough similarity in the results that you can kind of piece together the right answer, which is the best way to DIY something like this. One of those ways is using a platform like ChatGPT. So I included a prompt My prompt for finding your color season was six pages long. Yes, you heard that right, six pages. There's the initial three page prompt which includes a questionnaire for ChatGPT to ask you so it's not just relying on one or two photos. And then it includes a detailed format to deliver the results so you can check for accuracy step by step. I then followed it with three more pages of follow up prompts to verify results, explain the why behind the answers, compare secondary seasons and lots more. Now, which do you think is likely to get better results? One sentence or six pages of instructions? Probably six pages, but most people aren't going to give instructions that detailed. And why would you? Who has time for that? I read somewhere a long time ago that a good prompt takes at least eight hours to create and I definitely spent that. But the average user isn't going to do that for themselves. At that point you would just hire someone to tell you your colors, like what's your time worth? And to be honest, most people don't know how to create a prompt like that because if you knew everything you needed to ask or tell it and you knew where it was going wrong during testing, you would probably be far enough along that you don't need someone to give you the answer. And I am not saying that every question needs a crazy long perfect prompt, but most of the time you can't just give it one sentence and expect good results. That's what we have been trained to do through search engines. And this is another case. You just can't use them the same way, they don't work the same way. And it's not always about the prompt either or all of these AI platforms depend on your photos. Whether we're talking about an outfit idea from an app or a body shape analysis on ChatGPT. I saw one review video when I was working on the apps where the woman did notice that the items the app suggested didn't really go together. And she was wearing the outfit in the video and she showed the screen grab of the app created outfit on her phone. Right? And, and they looked totally different because of the lighting in the app. They went together in real life, not so much. And that's not the app's fault. The same thing happened during my ChatGPT color analysis. When I was putting together and testing this six page prompt, I tested the photos that I had used before. Remember when I got the wrong results I was trying to figure out if a better prompt got me to what I know is the correct answer. It didn't but this time I questioned it and I said, I have been typed as a light spring multiple times. Why are you giving me a cool summer result? It analyzed the photos again and asked me, were these photos taken in winter? And sure enough they were. And I had followed all the instructions that you are supposed to do when you do color analysis. No makeup facing a window on a sunny day. But the winter sky added a gray coolness that I guess kind of overshadowed everything. And that's what it picked up on. I mean, that's not really my fault. I hadn't even considered time of year to do this. But it's also not ChatGPT's fault because it just used what I gave it. And I also found this with the photos from the community that I tested with. One thing my style circle members hear me say all the time is that I need good light in the photos they submit for style coaching because it's really hard to give feedback to someone in an all black outfit in a dark room. Like I can't see what I'm giving advice on, I can't see any details. So I gave it one of those photos just the way it was submitted and then I compared it to the advice that I would give. But for me, I lightened the photo in canva enough that I could see the details and ChatGPT's advice was way, way, way off. It was valid advice for the dark photo, but not the real life photo where this person actually did have like two legs. It wasn't just, I mean literally, I think it said that this woman was wearing a skirt, which was. She was not. She just had her legs together in dark clothes. So anyway, it was off my last issue though number five is that they often give accurate but dated advice. When I asked my original question about the best styles for apples, all of them gave Apple advice straight out of 2014 suggesting leggings and tunics. And to understand why this happens, you have to understand how these large language models get their answers. Unlike a search engine where you ask a question and it gives you a list of resources, AI looks for patterns in what it's been given in the training phase of the model and what users have fed it since then. I told you at the beginning of this episode, I do not use AI to write content and this is one reason why I am not going to give it my ideas. I have turned off the setting in all of the models that I've ever tried. That allows it to use my inputs for training because I'm not going to help it give better style answers to the world. Sorry, those are my ideas. Go do your own work. Right? And on this one I had a very meta conversation about this with ChatGPT, trying to understand exactly where these answers come from and the last thing I asked it was how do you stay up to date? Do you scour the Internet looking for current information? And here is the answer still in incognito mode. It said that's a great question and you're wise to zero in on important distinction. There's that making me feel good thing again, right? And then it said usually no. By default I don't browse the Internet in real time. When answering, I generate responses based on what I learned during training and in general patterns for something like apple shaped body styling. The core ideas haven't changed in years, so older info is still relevant. However, if you ask something that is time based, like trends for the current season, I can look up current info and incorporate it. So basically it is using all of those blog articles written a decade ago as the basis for its answers. That's why the capsule wardrobes they created for me look like I got in my time machine and went back to my closet When I started this business in 2012, all of those things on this capsule they were in my closet back then. And it is true that a lot of the principles or goals of dressing an apple shaped body are the same. Those haven't changed, but current styles have changed drastically and we've had to change the style we use to keep those principles and goals the same, if that makes sense. So it's really not its premise that's off, it's just the execution and that causes the end result to miss the mark. And isn't that why people use platforms like this in the first place? To get that quick end result done for you? Answer again, if the information you get isn't usable, what is the point? So here's just a quick recap of my cons of using AI models for style help. They are more focused on making you happy than they are on giving good answers. The accuracy cannot be trusted. And that can be because they straight up fabricate answers, they don't get good inputs, or because they reference accurate but dated information. All of that leads to that variable accuracy, which is my overall issue. When you don't know if the information is good or not, you simply can't trust it or use it. Now I said in the beginning of the series that I was going to give the pros and cons to both and y' All, I am trying so hard to find the pros here. I really, truly am. And I guess the pro is that it's easy and it's free and accessible and sometimes very, very good. But again, if you don't know if you've gotten good advice, what's the point? And I am sorry. I'm really, really, really trying to be fair and balanced. But I keep coming back to this idea that the people who rely on tools like this the most are those with the least amount of knowledge to tell which advice is valid and which isn't. To me, that is just a recipe for more style confusion. Like, I can kind of laugh at the advice that it gives because I know it's bad. I know it's bad. But if I didn't know, I might be tempted to take it. And before we wrap up, I want to share one final thing about using AI to improve your style. And it's something I touched on in the last episode when I said that giving you outfit ideas each day makes it easier to get dressed in the moment, but it cannot replace the skills you learn from picking a top out of your closet and challenging yourself to make five outfits with it. You remember that. That right there is the biggest problem that I have with using AI for style help. It doesn't teach you anything in the long term. But really, the truth is, it's worse than that. The more you rely on AI for something, the worse you get at doing that thing on your own. It's not just that it doesn't make you better, it actually makes you worse. And here is how. First, through something called automation bias, studies have found that humans are more likely to accept incorrect answers when they come from an automated system than when they come from a human. Humans, we question, machines, we don't. Which does not sound problematic at all, Right? Over time, you stop using your own judgment. And the answer to so many style questions is, well, what do you think of it? How do you feel in it? Your own judgment is vital when it comes to style, so don't give it away. The second way it makes you worse is through lowered creativity. A study at MIT Sloan School of Management found that AI helps people generate ideas faster, but the ideas tend to be more similar to each other and less original overall. And the more people used AI, the less able they were to come up with new ideas on their own, even when they weren't using it. And I would argue that creativity is pretty important when it comes to style. Again, don't give yours away. Don't lose that muscle that you've built. Finally, the last way it makes you worse is through something called cognitive offloading, which is also known as the GPS effect. Have you ever noticed that you used to be able to figure out how to get places you could find your way around, but after years of relying on Google Maps or gps, you don't know how to find anything anymore? I certainly have. If you have, you are not getting old. You are experiencing cognitive offloading, which is that when people rely on tools to do thinking for them, they practice those skills less and their ability in that area weakens over time. To put all this plainly, the effects of relying on AI are that you trust it more, you think less creatively, and over time, you get worse at doing it yourself. And now I feel like this entire show could have been that one sentence. But, you know, I like to talk and look, let's be real for a second. I don't think any of you are currently addicted to using ChatGPT or a style app to get dressed. You probably use them occasionally, which isn't a big deal, but like with all good drugs, that's where it starts. And even if you're not feeding an AI addiction, and you still have the issue with reliability, which in the end just gives AI a big thumbs down for me. Okay, so your homework for this episode is to do nothing. And by that I mean if you've been listening to these episodes thinking, you know, maybe I shouldn't use AI so much. It doesn't even have to be style related. Just stop, break up with it, challenge yourself to do the work on your own and build that muscle back up. Okay, AI lovers and AI haters, grab your pitchforks and light the torches. I am ready for you. I know this is a hot button issue, and I know that this was a very simplistic look at AI and all the things it can do. There are absolutely great uses for AI and there's amazing innovation happening. But I am a wardrobe stylist dedicated to helping women build lifelong style skills. And. And today, in this moment, I don't think AI is the way to get there. That is it for this episode of the Everyday Style School. I want to thank you for spending time with me today. Whether you love AI, whether you hate AI, whether you're just curious to know what I think. I appreciate you hanging out. If you are ready for easier, better style, come take my free class, Style Made simple, where you'll discover how to stop relying on tools that don't teach you anything and actually make your skills worse and start relying on your own knowledge and your own opinions and your own creativity so that you can make style easy for real for life. You can sign up@freestyleclass.com or through the link in the show notes. I'll see you next time. And until then, stay stylish. Ra.