45 | Maximize the ROI of ChatGPT Vision with 7 Types of Use Cases with Isar Meitis

0:00Hello, everyone, and welcome to a live episode of Leveraging AI, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. I am really excited to do this another new live session of this podcast, and I'm really excited to have everyone join us right now. We have a lot of people on the Zoom call, and if you want to share on Zoom where you are in the world and what your interests in AI are, that would be really cool so you can share it in the chat and we can get to start knowing everybody. So we really have a big variety of crowd here all over the world.

0:44And our goal today is to dive straight in and talk about different use cases of the vision capability of ChatGPT. So this capability has been possible for a while now, but now that they've introduced the new GPT version that combines all the capabilities together, it became even more powerful. And we're going to cover multiple use cases or types of use cases that you can use in your daily life as well as you can use in your business with that capability. And I think at least some of them will really blow your mind. So for those who are live with us on Zoom and LinkedIn, we'll share the screen so you can see more stuff as I'm going through it.

1:29So let's get started. Okay, ChatGPT vision, seven types of use cases. This is the topic. And the first thing I want to mention before we dive into the details and the use cases is what does it even mean that ChatGPT can see? So ChatGPT came out as a large language model, and I'm sure everybody heard that phrase multiple times now. By the way, it's almost exactly a year ago. So November 30th was the day that ChatGPT came out and really changed the world as we know it. And it came out, as I mentioned, as a large language model, which, as the name suggests, has to do with language, meaning it's a model that could understand language.

2:12And in the beginning, it was written language, meaning it can understand what you're writing to it in a chat interface, and it can respond back in a chat interface and do magical things. But over time, more and more modalities were introduced. And if you can see the image of this AI bot in the picture, you can see she has a thinking cap for analyzing data. She has glasses for seeing things. She has earrings for listening and analyzing data, but she can also create stuff like music and movement and text and so on. So each and every one of those aspects, vision, music, sound, voice, video, text, et cetera, each and every one of those is a modality.

2:59And OpenAI, the company who created ChatGPT, added more and more capabilities. But in the beginning, until not too long ago, they were in silos. Each and every one of those was a standalone component that you could interact with for that particular option. But now what they've done is they've changed that and you can actually see each and every one of them separately where, sorry, now they're all merged together where you can use all of them in a single unified chat interface that gives you a lot more flexibility and capability. So what does it mean that ChatGPT has a vision capability? It means that you can upload an image and that could be a document, that could be a screenshot, that could be a picture that you take with your phone or a camera, any visual thing you can upload into ChatGPT, and it knows how to understand what is in the image.

3:56But what it can do with it, what it can do with the content in the image, falls into many different categories. So one of those categories is the ability to describe describe what is in the image. So what do I mean by describe what's in the image? That's very simplistic. So if I switch to the next tab, you can see I've uploaded an image of a very old Nintendo. And I literally just asked it, what is this? So Chachupiti said, that is a Nintendo Entertainment System NES. That's, by the way, for those of you who are not geeks like me, this is the first Nintendo ever. And then it gave an explanation of what it was and what the pros and cons of it.

4:43And by the way, the pros and cons of everything is something I have in my custom instructions. So it doesn't come ready like that out of the box in ChachiPT, but you can, in the custom instructions, ask it to give it pros and cons to everything it talks about, and then it will do that for you as well. I've asked you to do that. So it gave me all that description. So this is describing what is in the image. So it's just a description, but it can do a lot more than that. It can assist you with different actions that are related to what is in the image. So what do I mean by that? So if we use the same exact concept here of the Nintendo box that is very old, it came out in the 80s.

5:28I then went ahead and said, you are a home automation expert integrator who specializes in entertainment systems integrations. I have an NES system. I am in the process of upgrading my house and the entire entertainment system. I'm using an OST MX-1280 amplifier. It's a very high-end amplifier for all my sound. And I have a Samsung

5:55QN900C8K QLED TV. It's a very high-end TV. What do I need in order to connect my NES to my new set for best results? So now it's assisting me, going back to the assistance related to the image, it's assisting me in doing something that relates to the image. And it says, first of all, you will need video conversion because the NES is old RCA output, which is analog, and my TV is obviously digital. So I will need a composite to HDMI converter. It says your NES composite output, yellow, red, and white RCA cables get composite HDMI converter. I need audio integration into the amplifier for the same reason.

6:37I need RCA to amplifier, and I will need to configure my TV. So it now tells me how to configure my TV for the best results, use game mode because it's going to be very slow refresh rate compared to the old system and so on. Fine tuning the TV, set it to four to three screen instead of a 16 by nine and so on and so forth, additional tips and other stuff. So it's assisting me in doing something related to the image that I have uploaded into it. And you can use this capability for multiple things in businesses as well. But the other thing that it knows how to do, it knows how to interpret information, meaning a much deeper level of analysis from an image that you upload.

7:25So in this example, I've asked it to look at an image of an icon that I have on my website. And I said, this icon appears on my website. Here's the URL, so it has reference of how the webpage looked like. You are an expert designer specializing in creating icons and custom website graphics. I would like you to create several icons that would mimic the style and the exact colors of the attached icon. So this icon was created by a graphic designer for me that I'm no longer hiring or in to be more specific, I don't need to hire anymore. So I'm asking you to create a family. Your goal through this conversation is to create a family of icons that uses the same design language, but are representing different concepts that I would share with you.

8:22The first icon from this family that I would like you to create for is for an AI course. So I'm teaching an AI course and I want to have on the AI course website, I want to have other icons, but I want them to mimic and look like this one. And the first thing that I asked it to do is before we start with the actual graphic design, please give me three ideas for what the icons may show. So the cool thing about this, now it understands my website because it has access to it. It understands how the image of the icon looks like, so it can understand the language of what I'm trying to create. And based on these two things, it's giving me ideas of what the graphics could be.

9:09So the first option it gave me is book and circuitry, an open book with circuitry pattern, blah, blah, blah. The second one was graduation cap and chip, a graduation cap resting on top of an AI chip, et cetera, et cetera. And the third one was laptop and an AI brain, a laptop with an AI brain graphic on the screen. So it gave me three different options. I told it I like the second one better. And it gave me a graphics that somewhat resembles the first one, but it's not very good yet of a cap on top of a chip. But then I gave it some feedback. And one of the biggest mistakes that people make, and it doesn't matter how you use Chachapiti, whether you're using the vision in order to go back and forth with it or not, is that they give up when they get the first not great solution or great answer or great graphics in this particular case.

10:01You should not give up. This is a conversation. You need to go back and forth. And I always tell people when I do my consulting or when I work with people or when I give presentations is you have to keep on digging to hit gold. You're not going to get a great solution on the first time around. And you have to think about ChatGPT as an amazing intern, like a really smart, capable intern that still doesn't know you very well. They still don't know your company very well. They still don't understand your preference very well. But if you give them feedback, they will accelerate and they will provide you amazing capabilities much faster.

10:41So this first graphics is really using the same color. It somewhat resembles the idea of the first one, but it doesn't have that 3D view that again, if you're listening on the podcast, you don't see, but the first icon is very much has a 3D feel to it and the graphics that it created didn't. So I gave it feedback. I said, this is a good start, but I would like to provide some feedback and let you try to create another version. The original image is more 3D, meaning there's clear, intricate shadows on each of the components, making them look 3D. Also, there are less things happening in image, meaning it's a cleaner, more airy design.

11:21Please recreate the image. And the second one is perfect. It's spot on. It literally takes the concepts and the look and feel off the first icon that I've uploaded that, again, I paid a lot of money for to get created. And it created a variation that relates to the course with the idea that it has provided. So you're seeing here how you can take an existing something you have in your business, attach your instructions to it that needs to be detailed. And it's a process that goes back and forth and create something new. It's an incredible assistance that can do multiple things for you using this capability.

12:06Another thing that it can do is recommend and convert. And I want to extract. So I will give you a few examples now that touch on all of these together. So let me go through the seven things that it can do. And now we'll go to the more sophisticated examples that incorporate a lot of these capabilities. So describe is just the ability to describe what's in the image. And it could be a physical something that you take a picture of. It could be a image of a document that you have or a graphics that you have or a flow chart, et cetera, that you can load from somewhere in your company or from anywhere on the internet.

12:47So it can just describe what's in there. Assist is really helping you in a process that is using this. Interpret helps you understand something from within the image. Recommend allows it to recommend you things, let's say a comparison of two different things. Convert allows you to convert one format to another. So let's look at interpret, recommend, and convert for a second, and even extract. So extract is extracting information that otherwise is not easy to extract from an image of something. And again, you'll see examples in a minute that will make it clearer. And evaluate is, again, allows you to evaluate the level of quality of different things within the image, which the next examples will help you understand.

13:35So extracting information. I did something that is very simplistic, but I think will help you understand the concept. I literally took a piece of paper and in a pretty lousy handwriting, I wrote notes as if I'm watching a presentation about AI. And I literally wrote in there, I took some notes during today's presentation, and there is like a really bad hand-drawn flowchart with even worse handwriting within the boxes of the flowchart because the boxes are not big enough and my handwriting is tilted and going out of the boxes and so on, and probably some of the things that are misspelled as well.

14:10And I asked it, can you help me extract information from this note? It is critical that we do this accurately. And Chachipiti said, certainly. And then it extracted information. It said the context is the author of the note mentioned taking some notes during a presentation. The main topic, the note revolves around the relationship between an image and AI. The process involves image to AI, describe, recommend, extract, and interpret. And then I can go ahead and ask you to actually recreate that in an actual graphics that I can put in presentation. So a handwritten, bad handwritten note can turn into a graphics I can use in a document that I can share with my company, that I can share with my colleagues, that I can put in a presentation in literally seconds.

14:57seconds. So the ability to extract information, or if you're a person that likes to take notes, handwritten, but then you want to put them in somewhere that is searchable, notion of your company or a Google doc or anything that exists that other people have access to and that you can search afterwards, you can still enjoy the best of both worlds by taking handwritten notes, taking a picture of it, loading it to ChatGPT, asking to extract information, and even interpret the information, put it in whatever format you want, and then save it as you want to save it. So that's just one example. Another example is of interpretation and so on is something that almost every company struggles with and every individual struggles with when we go to trade shows is following up.

15:48You go to a trade show, you get 500 business cards, and it's very hard to remember which were more important, which were less important. What was the follow-up? What were the topics that you wanted to talk to with each and every person? That ceased to exist when this technology was released. And that's something I've been perfecting in the last several trade shows that I was speaking on. What I'm doing right now is I'm actually using the ChatGPT app on my phone. So ChatGPT has an app. What I'm doing is I'm going to the ChatGPT app and I'm taking an image, a picture of the business card from within ChatGPT.

16:26And I'm asking it to extract the information from the business card. And you can see in this particular example, there's an image of a business card. And then the extraction of the information is, it gives me the name, the email, the phone number, and even the tagline of that particular person. Now I asked it to turn it into a VCF file, like a card format that I can then import into any tool that I use, whether it's Google Contacts or Microsoft version of it. It doesn't really matter. I can now take this vCard and put it into whatever system that I want. But it goes beyond that because since it's on my app, my full process right now is I'm taking an image off the card.

17:13I am then clicking the microphone button on the app, and I'm literally saying what I need to remember about this particular individual. so i shared the image with chachipiti and then saying i met this person at this and that conference we were discussing this particular thing and we wanted to follow up with this person on this and that topic the highlights that we need to discuss are this and that and i here are the key points that are important from our conversation. And what happens is at the end of each day, I have a hundred of these individual snippets. But what I do in the end is I ask it to create those cards for each and every one of them.

18:06So now I have the cards and I ask it to write the follow-up email based on the comments that I left with voice that took me seconds to actually record. So at the end of each day, I have all the business cards ready to be uploaded into my email platform, whatever it is. And I have the follow-up emails ready to be sent with zero time. And all I have to do is literally go through the emails, make sure that they make sense and they're not bullshit. Add whatever info that I want to add or change whatever I want to change. And I can send a hundred follow-up emails at the end of the first day, 30 minutes after I'm back in my hotel room.

18:43and not to mention not losing any information that all I had to do is take a picture and record the information. It's from my perspective as running large companies and doing trade shows for the last 20 years, it's a complete revolution. That was always one of the hardest parts to get right in a trade show. And this system solves it for 20 bucks a month in addition to all the other things that it does. So I want to give you another example, another great business example on how you can use this tool for more or less anything in business. But in this particular case, I'm comparing two different landing pages.

19:19So what I did is I took a screenshot of a full landing page. You can do that with different tools. Within Mac, you can do this within the Mac itself. Within Chrome, you could do it within the Chrome itself, but not just the top of the landing page, but literally the entire page. So then I loaded two different landing pages. One is of my own course that I teach about AI, and the other is of another course creator that is a large course creator that teaches about AI. And I told ChatGPT, you're an expert landing page copywriter and creator who specializes in creating and evaluating high converting landing pages.

19:59I uploaded two landing pages for AI courses. I would like you to start by analyzing each of the landing pages separately and telling me everything you can about each one. please include an analysis of what sections are included, what pain points it addresses, what are the good things and the bad things about each of the pages from a conversion optimization perspective. And then it gives me a very detailed analysis of each and every one of the pages. So it starts with analyzing page one, Generative AI Business Transformation course, which is my course. Sections included, course title and value proposition, enrollment button and course details participants participant companies knowledge and skills etc etc what are the different things then he talks about the pain points that it addresses complexity of generative ai for business leaders need of strategic business process and optimization using ai demand for practical executable ai strategies then he talks about the pros and cons of the page and the good and what needs improvement and then he does the same thing for the other page.

21:05So I get a very detailed overview of this page. Now I could have used an A-B test of two pages that I've designed to help it evaluate each and every one of them and give me the pros and cons of each one. Then I went ahead and asked it, if you had to pick one of those pages, which is more likely to convert, which one would it be? Please provide a clear breakdown of why. So Chachupiti said, if I were to choose one page that is more likely to convert, I would pick the first landing page, Generative AI Business Transformation Course. I was very happy that I won, but I did not know that going into this experiment.

21:46And then it says why. It provides social proof, and then it explains why social proof is important. Instructor credibility, it explains what I did there that is good. Content richness, testimonials, visual and textual balance, clear value proposition, multiple calls to action that appears in several different places around the page and bonus content. So the podcast promotion at the bottom acts as additional engagement, blah, blah, blah. Awesome. So I was very excited to hear that it liked my landing page, but then I went further than that. I said, okay, great analysis so far. Thank you. By the way, if you don't know that, if you're nice to ChatGPT, you get better results.

22:23I know it sounds weird. I know that you're talking to a machine and you think it doesn't matter, but this machine was trained on gazillions of human interactions. And it learned that when people are nice, they get nicer responses. So that's the way it works. That was proven in multiple researchers. So first of all, I'm just used to writing this way, but also it actually helps getting better results. Going back to what I wrote, great analysis so far. Thank you. What optimization suggestions would you make in order to make this page more likely to convert? Please include all aspects such as structure, visuals, copy, etc.

23:02ChatGPT came back with simplify the flow. Organize content into more digestible sections with clear headers. So mine doesn't have headers, even though there's clear sections in different colors, it doesn't have headers for each section. Highlight key information, bringing attention to the start date, duration, and price, etc. etc. Sticky navigation, implement a fixed menu bar that includes the CTA that I actually have, but since it's a screenshot, it doesn't know because it can't see it scrolling through the page. Visuals, imagery, incorporate dynamic images or graphics related to the business and AI to break up the text, etc., etc., etc.

23:42Give me all these things and including copy, like it provides me better copy than I have on how to improve the landing page. Now, let's take that into any two visual things you have. These could be physical brochures that you're designing or that you've already printed or of a different company that you want to evaluate. And instead of reading the full brochure, you can actually take pictures and have ChatGPT evaluate it and give you a summary and tells you what it's addressing, what it's not addressing, what could be improved. It could be comparing, like I said, A-B testing of different capabilities of different things that you want to post.

24:21This could be two graphics. Remember in the beginning, in the very first example or one of the examples, I showed you how I'm actually creating icons for the website. I could actually take two or three different variations that ChatGPT creates, load it back and say, okay, here's two variations that should go on this page and you can actually give it access to the page. I want it to represent this and that. The goal of the page is to achieve this. The goal of the graphics is to achieve that. Which of these icons do you think is more suitable for this purpose and why? And it will give you this reasoning.

24:58So the ability to have ChatGPT work with you on anything visual, help you analyze it, help you compare it to other things is extremely powerful. But it does not stop there. In a business, it's not just about evaluating graphics and creating graphics. It's in many cases about writing proposals and delivering them to clients based on information that the clients are providing us. And you can do that as well using this tool. So what do I mean by that? Let's look at this example. So what I've done in this example is I've taken a home renovation project, but you can, again, relate that to any other graphical information that a customer needs to provide to you.

25:52This could be an electrical panel. This could be a design of a circuit. This could be a workflow that they currently have that you want to help them improve as a consultant, like anything that is visual that relates to a proposal you want to write, you can upload into ChatGPT. But in this particular case, what you can see is a floor plan. And what you can see on the floor plan is red lines that go from one centralized spot in the house to multiple other places around the house. And the first thing I did, just out of curiosity to see if ChatGPT will know what this is without getting any feedback.

26:33I said, can you tell me what is in the attached image? And he said, certainly, the attached image is of a floor plan, which is a scale drawing of a room or building viewed from above. The plan typically illustrates blah, blah, blah, blah. And it told me what a floor plan is, which I obviously knew. It also gave me the size of the house based on the measurements that are written there. But then I went ahead and it told me which rooms it found. So I identified the following rooms, master bedroom, living room, dining room, kitchen, laundry, medium room, and garage. It actually missed a few rooms, which is actually pretty funny because this room had no title on them.

27:10So it missed a few of the bedrooms. It missed general areas that don't have any title on them. It did not know what they are, but then I helped it understand it and understood all the rooms. But then these red lines that are drawn on the floor plan are actually speaker wires. So this is a plan for installations of a full house surround system that goes back to a centralized control unit. And the goal of this particular example is to show you that you can go from a drawing of lines or a floor plan to a proposal. So what I did is I told it, okay, these lines are actually speaker wires. I need you to analyze the length of each of those lines based on the scale you have on the floor plan.

27:58And it took me a few cycles to explain to ChatGPT how to do this. But the way we ended up doing this after it got it wrong a couple of times is I basically gave it a yardstick. I said one of those lines, the line that goes from here to here, that goes to the right side of this in that room that it can identify on the map on the floor plan is in real life, whatever, 50 feet long. And now it knew how to use that as a scale to measure all the other lines. And what I did then, I said, okay, a speaker wire costs X, whatever, a dollar a foot. Every installation point costs$250. The speakers that we're going to install cost 300 hours a speaker.

28:45At the central point, I need eight feet of slack to go down from the ceiling to where it connects to the amplifier. And each wire needs a 20 % slack based on its length. In addition, I need these connectors that cost this and this. That's why I give it all the additional information in addition to what it has in the image. And then I asked it to create a proposal for me. The first thing I actually asked it to do is to create a table. So I will scroll down now and we'll show you what is created. And you can see that I've done several different iterations. But in the final iteration, you can see that it has each of the rooms.