26 Incredible Use Cases for the New GPT-4o

OpenAI's GPT4O model is here and one of the main questions is what can I use this for? And that's why today we'll be looking at all the different use cases that OpenAI showed us, but also what the Internet has come up with so far. If you want to learn about the details of this announcement, what it includes and how it works, I created a separate video on that that I'll link on screen right now. But with that being said, let's get into some of the first use cases, starting out with their announcement blog post. But I'll be adding in various use cases that I found across the Internet in between some of these. But this is a great place to start because they created a batch of videos showing off different things that you can do with it. And I'll just be giving you quick summaries and my take, you can watch all of these by yourself in full length if you so desire. Links below. But hold up, it doesn't even end there because look, a lot of people are viewing this channel and all of us are AI users, some of us are AI power users. And I was just wondering, wouldn't it be amazing to find out what everybody that is watching this video actually uses this for? So we're doing a brand new thing where we issue a challenge and then we have a public space where you can review all the people submission to the challenge. And the challenge is all about finding GPT4O use cases that work for you. So stick around because at the end of the video I'll give you more details and I'll tell you exactly how to participate and how to view what other people have been doing with GPT4O. So let's actually start with a clip that came out in the recent hours. It's an interview of Sam Altman and he was asked the following question, now check out his answer, which hints at something that all of us might be doing in a few weeks here are.

There use cases that you've gravitated to? One surprising is putting my phone on the table while I'm like really in the zone of working and then without having to like change windows or change what I'm doing using it as like another channel. So I'm like working on something I would normally like stop what I'm doing, switch to another tab, Google something, click around or whatever. But while I'm like still doing it to just ask and get like an instant response without changing from what I was looking at on my computer, that's been a surprisingly cool thing.

So let's talk about this one on top. I think they picked this because it has the most human characteristics, and it really shows the capability of this new version of ChatGPT to be an AI companion. Matter of fact, the voice is so hot that people across Twitter have been complaining. What is this? This is not an assistant. This is a flirty girlfriend.

From what I can see, it looks like you're in some kind of recording or production setup.

AI girlfriends have arrived. And look, this is one of the big new improvements. It's way more human. It doesn't just express emotion, but it also understands emotion from the phone's camera right here. And here's just guessing something based on the information you provide it with, which might not be very practical, but some of these others definitely are. So let's move on, because in this example, Greg Brockman actually uses two phones with two GPT4Os talking to each other. And this shows off a capability that we already had in GPT4, but now it's upgraded to the next level. Namely, I'm talking about the capability of setting up multiple Personas to have them talk to each other. With two phones, you can simulate various conversations, and you could apply this to your very own context. This conversation might be a debate you might be preparing for, or an argument. You could just play it out between the two phones. I mean, that's amazing, but also kind of freaky, right? And in the end, they also show the ability for the app to modulate the voice so it can sing for you. It can sound like a robot, like they showed in the demo. Have a quick listen. This essentially hints at the fact that you're going to be able to filter the voice. Just like I have a soundboard here, and I can do things like this. These capabilities are going to be very surprising to most. I can't wait to show my grandmother these updates. Now your AI assistant is going to be able to do that, too. And look, some of these use cases might seem like a nice toy to have, but if you get a little serious about it and you think about how this is applicable to some professional fields, you get something like this top comment on my latest YouTube video, which, as I mentioned, summarizes this entire update as I asked for use cases that people are excited for. Male care here says melanoma detection, retina exams, pulmonary distress analysis. This one I'll quickly have to look up. Okay, so it's a diagnosis for breathing difficulties. And yeah, comments point out this is going to be for diagnosis, not treatment. But these things are amazing. And admittedly this comment is a bit speculative, but as you can see, it clearly captured the imagination of a lot of people and talking about work related capabilities. Apparently the benchmarks do translate to use cases because as we looked at in the summary video, it performs better on almost all benchmarks when you compare it to the other best AI models in the market right now. And this includes upgrades to vision, but also things like the code interpreter so you can actually upload files and and do things like this now more effectively. Where you give it an Excel sheet and it analyzes the spreadsheet and does deep technical and statistical analysis and generates charts and visualization. This is a very simple prompt that you could reapply to your own charts and get these results immediately. And this is not the voice assistant that might not be available to you yet. This is the GPT4.0 model that all users have now and all free users will have very soon. Okay, and here's one that really caught my interest. This comes from within the AI Advantage community where remember used it to analyze the conflict between Drake and Kendrick. Not sure if you caught this recently. They're beefing publicly now and they created diss tracks about each other. And if you're not following this closely, it gets messy very quickly. There's a lot of news coming out. I personally am not following this closely, but look at this conversation with GPT 4.0. He uploaded two CSV files that included the different events that happened with dates next to them. Now here's an important detail. These data sets also include Google Trends data with how popular were both Drake and Kendrick in Google Search over the last few months. So how would you compile this into some visualization? I'm not exactly sure, but we don't have to worry about that. Let's just let GPT4O figure this out. So the whole interaction starts by Daniel uploading these files and then prompting it. And then with simple conversational prompts, he managed to make sense of all of the data and throughout the conversation. If you want to read this, you can stop the video at any point in time. I'll slowly scroll over this. He managed to create visualizations on top of which he could iterate. I'll show you the final result here in a second. Okay, because this chat history doesn't include the images, but it basically goes ahead. It maps out everything that is in the Excel sheets, it reconstructs a timeline from all the events scattered across the Excel sheets. And then using the web browsing tool that is super snappy now because we have increased speeds. It adds further context so you can look for new stories, add that into your conversation, like so look, it pulls up Wikipedia and Hollywood Life articles that recently happened and then it extends the timeline that you provide it with. Now you have all of this in your conversation, so you can prompt it with something like redo your analysis with the added context and then it finally creates a visualization that wraps all of that information into one image. And then here it is. And I think this is actually really useful if you're interested in this topic. You see a timeline of events dating back to 13th April 2024 with the first Freestyle releasing. And then it's all mapped against the Google Trends data on how popular the terms Drake and Kendrick were in Google Search. Matter of fact, it creates two different views with two different scales of the timelines. Isn't this incredible? I think it is. And sure it's not perfect. Look, it made a little mistake down here with the date, but this data does track with the original files that were provided to it. And just think about the possibilities here. You could easily be pulling Google Trends data and mapping that onto different events by yourself. Now, really simply. And it's not going to take you 20 minutes anymore because before running the code interpreter it took about a minute or two. Every single time running web search gave you one or two results and that also took about a minute or two. If you didn't like the results, you had to re prompt it and sit there for another full minute. That's not the case anymore. Look in real time I'll put in a simple prompt and it's already searching the Internet. It found five sites and before I managed to finish the sentence it gave me all the links and a summary of what happens in those articles. This is a massive change in user experience and the fact that they're making this available to everybody at this speed level is going to open a lot of people's eyes. So how about this next video where he's preparing for an interview and he puts on a hat that might be a bit inappropriate for a job interview. In this interview prep use case, the thing that fascinated me was actually the fact that, that it picked up on the fact that I had to deploy empathy to talk to him. Just listen to this answer. This is not a neutral factual answer like take off the hat. This is inappropriate. Have a listen. What do you think, Rocky?

That's quite a statement piece. I mean, you'll definitely stand out though maybe not in the way you're hoping for an interview.

Okay, I got it. What a fantastic response and a display of emotional awareness of this new ChatGPT model. Impressive. Let's move on to the next one, which is the fact that it can act as a game host. And as a parallel to this, they also show this other use case where you can use it as a meeting AI essentially where it facilitates the meeting and then in the end summarizes everything. And I think this is absolutely amazing. If you give it access to your screen and all the audio, it can effectively direct the conversation and wrap up the meeting. I don't know how well this will work in practice. The main thing that I would be looking at here is the context limitation. They only show a two minute demo. How is this going to perform on a one hour meeting? It probably doesn't have an enough context length, but we've yet to find out. This voice assistant is rolling out over next weeks. Right now we just have the GPT4 all model without the voice assistant. Either way, you're going to be able to use this as a game master, whether it's Rock, Paper, Scissors, Dungeons and Dragons or the same concept, but in a work context, which would be a meeting. Let's have a brief look.

Get ready. And three, two, one, shoot. Let's see those hands.

Cool one.

And it's another tie.

But going back to the voice assistant use cases, I even featured this one in the summarization video because I just figured that this would be so groundbreaking for education young people or anybody trying to learn a new skill. You could just open up chatgpt on, let's say your iPad on one part of the screen and on the rest of the screen you're solving some problem and then by talking to it and using the pen and highlighting, having a conversation like you would have with a tutor, it can guide you through the entire problem step by step. Kind of like a human would.

Can you find which one is the hypotenuse? I think the hypotenuse is this really long side from A to B. Would that be correct? Exactly. Well done.

Amazing. And look, I know educational is a controversial topic. We actually hosted a little post event discussion in our private AI Advantage community where several people brought up that actually the educational sector is the one that has the most resistance to this technology because the teachers just feel like this is a direct replacement to what they're doing. It lacks the human touch. Plus students can cheat on all these assignments, homeworks, etc. Right. So look, I think that makes Perfect sense. But simply put, a lot of people that struggle in school will have this technology of available to them and they will have an alternative to simply failing out of their classes. I remember moments in high school and university where, particularly on accounting and then later in university during statistics classes, I was so lost I didn't even know where to start. And at that point the tutor wasn't really an option because we simply couldn't afford it. And I very vividly remember these moments of pure desperation where I just didn't know where to go from that point. I looked at the textbooks, but I didn't understand anything. I attended the class. It didn't make any sense to me. The friends that I had were busy thinking about the weekend, so I was completely lost. And if I had something like this, I would have gave it a shot and it would have probably helped in some of those classes. So just from that experience alone, I do see the benefit of this for many people. While I don't think it replaces a human teacher, I think for now it's a fantastic addition to how our system works. And let's be real, the school system is broken in many places. Maybe rethinking it while considering tools like this could be the change we need. Just think about you learning a new piece of software and this guiding you while you do that. Amazing. But hey, at the end of the day, I'm just here to point out what's new. And I think this is one of the most amazing use cases. Now let's move on to the next one. Okay, here's an interesting one. That really was a big problem up until now. It's sarcasm. Rarely was AI able to pick up on sarcasm and now it can actually replicate it and use sarcasm. This is possible because this new model is natively multimodal or omnimodal as they name it. It's not three separate steps of transcribing the voice to text, then using ChatGPT to process the text, and and then afterwards turning the text back into voice. It all happens in one and you get things like the capability to be sarcastic.

Being sarcastic all the time isn't exhausting or anything. I'm so excited for this.

Kind of incredible. More of a capability than a use case, but interesting nevertheless. If you're enjoying this video that obviously took a lot of work to put together, hit the like button. It really helps the channel. But with that being said, let's move on to the next one. Okay, this next one is particularly interesting and that is this accessibility feature where you're going to be using GPT4O vision to help people with no eyesight. Try and tell me exactly what they're doing right now, please.

Right now, the ducks are gently gliding across the water.

This last part left me a little speechless. I'm just gonna play, too. I even know when a taxi is coming with its orange light on. I think I'll hail it to get home.

Yes, I spotted one just now. It's heading your way on the left side of the road. Get ready to wave it down. Great job hailing that taxi.

I mean, wow. You can only imagine how transformative this would be for people without eyesight or other limitations. And one interesting thought I had here was that there's situations in life where you yourself can't really see something or your attention might be divided between multiple tasks, and it would be just great to have a second pair of eyes, right? And I found a similar idea on Twitter here where Kitze, I suppose, came up with this use case for GPT4O where as a parent, you could set up your phone to watch your kids for a second and look, I can already see a comment section of like, seriously, you want your AI assistant to watch over your kid? Look, fair enough. Maybe not in this version, then you absolutely have to test it first. But just this idea of, like, hey, just let me know if my son starts crawling off to the side and you place him in the middle of the living room, and then you're gonna be able to leave the room for like 20, 20 seconds. I think that sounds pretty feasible. But again, this is for people themselves to decide. And over time, the tech is obviously going to be good enough for that to be a reasonable use case. Overall, I just want you to realize that phones are soon going to turn into a second set of eyes with a certain amount of intelligence in them, and that's just going to open up opportunities that we haven't even considered right now. Oh, and while we're kind of on the topic of childcare, there's this one use case in here where he actually lets the phone sing a lullaby, and then he adjusts the voice to. To be softer, more silent, a little louder, and then it comes up with a song on the spot. I think this one is indisputably amazing, and I guess this falls in the entertainment category, but that also is a thing that you could use this for.

Oh, majestic Potato spoons are closed.

Okay, okay. It's a little too whispery. Maybe go like a little louder.

Got it. Let's find that sweet spot oh, majestic. Potato in the moon.

Soft.

Oh, creepy.

All right, so let's switch gears and let's talk about something that will inarguably be useful for businesses, which is a customer support rep that is going to be able to handle tasks. Now, they do have this two minute video in here where they use two phones to simulate a conversation between a customer and a customer support rep.

Awesome. All right, I've just sent the email. Can you check if Joe received it?

And I think this one is particularly interesting to talk about because it hints at the future of these products. Right? Because for this you absolutely need integrations with other tools. And that is something that ChatGPT doesn't really offer. You have access within the GPTs, but doing a full customer support agent in there, it's just not feasible as of now. You would need multiple actions, you would need reliability, you would need longer context length. But just by them uploading this video, this clearly signals to me that this is a direction that they'll take it in. Not as if that was any news to me, as already when GPTs came out back in November, I remember saying that this is the clearest sign by them that this is the future of the product. It's this independent AIs that have a set of tools that they can access and then use those tools to act on their own behalf. By the way, this was even pointed out by Sam Altman in a recent interview. He said that there's really two directions this can go in. One of them is kind of this assistant that will help you do your work better. And the second one is a senior employee where it will not just act by itself, but also have a certain level of autonomy to override your decision making and your prioritizing, because it's a senior employee. So that's what you can expect from this over time. I think for now, this isn't really there yet. They even pointed out that this is just a proof of concept. Nevertheless, I wanted to talk about it as it might show you a glimpse of the future and what we'll be getting soon here. Okay, so here's another use case and this one is more development focused and this is all about integrating GPT4O into this AI powered IDE. If you're not familiar, that's a software that developers use to write and test code. And basically they were extremely fast in integrating GPT4O. I mean, they took less than 24 hours to integrate it. And every single ChatGPT wrapper, as people call them, will have this upgrade very Quickly, because all you need to do is exchange one line of code and you have this better model. And all the use cases we talked about in this video will very quickly arrive in the applications that you might be using already. Plus, for the developers, it's a 50% saving in cost as it's 50% cheaper to use all of this overnight. That's pretty amazing anyway. Why is it significant? Because people across the Internet report that the coding abilities are actually improved. But this did cause a little discussion because it seems to be improved in certain areas and worse in others. And here's just a quick example of what Sawyerhood managed to do with this in the first 24 hours, and that is rebuild Facebook messenger with one prompt. And he's reporting that GPT4O did this in six seconds. I mean, look at that. That's wild. This is one HTML file. We're getting closer and closer to people being able to create games of their own in just a single sentence. We might not be there yet, but this is going to get wild soon. And talking about technical capabilities, there's also this brand new ability to generate consistent text. Matter of fact, it's so good that you can do something like text to font, where you prompt it to create the basic letters and then you can keep prompting it to give you various styles. Look at that. Futuristic here. It created a Victorian style font. And what that means is that they solved consistency of transferring text to other objects. So as you can see here, image of a coaster, their logo, it perfectly combines the two. Like a week ago you needed to know Photoshop to do this stuff. Pretty wild. And there's all these other use cases here, like taking a logo and mapping a word on top of it, or writing a poem and then visualizing it as if it was handwritten. Look at that. Want it in a different color? No problem. And while we're on the topic, character consistency is one of the main things that has been missing in many tools. Midjourney added it recently, but now also with GPT4O, you're able to generate a robot and then create multiple images of it, like so, while maintaining the same character. So you can tell stories now instead of just generating one image. And then if you regenerate the robot, it looks like a whole different character. Oh, and here's one more interesting capability, and that is creating images that represent original with a single reference image. This is not the easiest problem. It has been solved by some other tools. But now you will have all of this packaged into one tool so you can upload an image of yourself and then recreate it as a caricature. Now, just a quick note before you jump into ChatGPT, a lot of these things will be shipping over the course of the next week, so you might not have it yet. The voice assistant or the availability for free to all users. All of these things are coming over the next weeks, but that's not going to stop us from exploring what's possible because I have a few more here and one of them is this 3D object synthesis. Okay, this is something nobody really expected from them in this announcement and they didn't even mention it in the live stream. It's just hidden in their blog post under this one option here. So from a simple prompt and giving it view 0, view 1, 2, 3, 4, and then view 5, it generates five images of the same same thing, and then you can reconstruct that object from the six generated images. So this is a combo of the consistency that we talked about a second ago with the robot and then this new capability to construct multiple images into a simple 3D model. Here you have the same thing happening with the sea lion. And if it wasn't clear yet, they really did solve how to represent letters inside of these images. All of the AI image generators could not solve text yet, maybe with the exception of Ideogram, but their quality was not up to par with some of the leading models like Firefly Midjourney or Stable Diffusion Excel. All of these top models couldn't do text, but now you can prompt it with the exact text you want there and create mock ups like this, where it gets every single letter right. And sure, if you look at the buttons here, you get some funkiness like what kind of spacebar is this? What's up with these buttons? But that is being nitpicky. If you're creating social media content, the main thing you want to get right is the fact that the text is correct and that is a capability. Now, there's actually one more thing that I found in reference to the 3D object generation, and that is the fact that you can actually use the code interpreter to create 3D objects in around 20 seconds. As Min Choi reports here, this table is the result of the generation and he simply created it with a simple prompt that says make a STL file of a table with four likes and random attributes. 20 seconds later, finished working and here you have the result. And just mind you, this is the beginning of the 3D capabilities. So I just want to point towards one of these top comments, which is like, hey, Just remember, this was Midjourney version 1 in January of 2022. Okay? These things move fast and a simple table like this is version one. So I had an idea on how to really superpower this video and how to show you what not just you can do, but what also other people are already doing with this. What I did is I set up a community space where you can simply sign up for free and you can share what you're using GPT4O for. And at the very least, you can review what other people are using it for. Matter of fact, I packaged this in a challenge that is now running for one week. If you want to check out the details and participate and win prizes, you can do so. It's going to be the first link in the description. But basically we created a step by step guide and clear judging criteria and what tools you will be using and how we recommend you present the challenge. It's all in this brief guide if you want to participate. But what you really need to know is this GPT4 to GPT4.0 is a major leap and they made GPT4.0 available to everyone for free now. Fair enough. A lot of the capabilities pointed out today, like the voice assistant, are not going to be available to everyone. But as you can clearly see, this opens up a whole new world of.