How People Are Using The New ChatGPT Upgrades

In today's video, we'll be looking at some of the most disruptive use cases that people have come up with since the massive, massive updates that were announced at OpenAI's Dev Day. Welcome to our first ever OpenAI Dev Day. As you might have already heard, they overhauled ChatGPT and there's a bunch of new APIs. And that's why now is the time to look at the first innovations that the people have built with this new technology. Already people are already building live AI commentators. This one creates music playlists. Or what about this example where the guy built a yoga instructor that uses his webcam to guide him through the training, training and give him feedback depending on his live performance. All of these open up a world of opportunities and we are here to talk about them. And instead of just looking at the most popular ones today, I wanted to go ahead and pick out some of the ones that have the biggest potential to change the way we interact with applications, the Internet and essentially all software down the line. Let's get into it. So the first example is a live commentator that someone built on X right off the bat.

He's a one man show, ladies and gentlemen. He shoots messy, messy, messy.

Okay, so the first game changing use case here is combining two of the APIs. So one of them is Vision, where it allows ChatGPT to actually see and understand images. We talked about this extensively before. And the other one that is new is text to speech where you can take text and turn it into human speech almost. It's close enough. And what Gonzalo built here right off the bat is a football commentator. And the reason he managed to build this so quickly is because OpenAI has the so called cookbook where there's recipes for various apps and guess what the first one is processing and narrating a video with GPT's visual capabilities and text to Speech API. So for any beginning coder, you can just copy paste this and build something similar to this. But look at that, it's just live commentating the game based on frames that it's taking from the video. So it's not taking the full video, just a frame here and there and it commentates it like a magician on.

The field, dodging 1, 2, 3. Unstoppable. Look at him go. The crowd is roaring.

And when I saw this, one of the first things that came to mind is, wow, this is going to be incredible for video games because for something like FIFA, you're going to be able to play live and the AI commentator is just going to be there commentating it live.

Messy, messy, messy.

I mean, they're doing it already, but they're not really thinking independently. Right. This is going to take it to a different level. And one example that got pointed out by one of my team members was that in a game like NBA 2K, after the games you get to talk to the players sometimes, but all you get is two options, right? Junior in the house. What up, bro? You might just have a free flowing conversation with the NBA super superstars in the game. That's an actual game changer, no pun intended, I guess. And yeah, people are already doing that. Look at that. For League of Legends, they built a live commentator.

T1's positioning is crucial here as they look to establish dominance and secure the next objective.

And I think this is so amazing because he takes in all the data, he looks at the minimap, he looks at the itemization, the gold count, all the little intricacies of the game, the commentator considers them. And I don't see this really being better than human commentators. But in a video game setting where you don't have a live commentator at all times, this might be really good right now. Okay, so let's switch gears and let's move on from sports and video games to actually using some of the innovations that were announced inside of ChatGPT. Because one of the biggest things we got is the addition of GPTs, which are essentially single purpose assistants. So instead of just having GPT4 that can do it all, you have GPT4 that does one thing like create stickers, as does the sticker wiz in here that was built by OpenAI. Many people have access to this already. And here's all the presets that are available for from OpenAI. But while that's a great start, most of these are not going to be supremely useful to you because the real potential hides up here. Create a GPT, right? And we'll be talking about this extensively on this channel. I think this will be one of our main focuses, creating these single purpose assistants. Because this really actually non clickbaity completely changed the way we use ChatGPT. Because up until now you had to craft prompts for specific instructions. If you followed some of the more advanced tutorials on this channel, we, we went into the custom instructions to infuse it with all this custom context. But in the future you're going to be able to preset a single assistant for a single use case. You're not going to have to switch out custom instructions, you're not going to be have to switching conversations and feeding it with all the context again and again and again. You can do it one time, set it up as the preset for your GPT and then return to that. You'll just have them in the sidebar here. That's absolutely fantastic. So while the custom ones are not available yet, they're supposed to come later in November. Some people have access already. So we'll look at that. And we can use some of these OpenAI preset GPTs right here. So we'll look at that too. Okay, so while some of these early examples are just very primitive and maybe not super useful, that will come with the custom GPTs. If you look at a fun little app like StickerWiz right here, I actually have to say I really enjoy using these because I can stop thinking about all the possibilities. All I'm doing here is creating stickers with the stickerwiz, right? And there's preset prompts right here. Just use one of these. Can you make cool stickers? For my laptop, we'll just say include AI Advantage and it's creating the sticker. I don't have to prompt this in a way that it creates stickers because this has been preset to use Dall E free, as you can see. So it uses browsing, advanced data analysis, Dall E free, all at the same time. And the instructions are set up for this to create stickers only. And then I can switch over to another one, which is for negotiating. So these are some of the prompts we talked about on the channel months back. Can you role play a salary negotiation with me? You can do it right here. And no need for a prompt that is this long because it's set into the GPT by default. So there you go. But as mentioned, the real power here lies within the custom GPT. So let's talk about those for a second, because some people on Twitter already have access to this and they shared what they found. For example, Rowan, right here, was testing the GPT builder, and if we have a close look at this, you will find that this is very similar to what I've been teaching on this channel since months now. You give it a set of instructions, right? You give it custom instructions, and then optionally you can give it conversation starters. You don't really need to. And down here, you can now upload knowledge and pick the capabilities. So do you want to use the code interpreter? Do you want to be using Dall? E to generate images, or do you want web browsing enabled? All of this can be preselected here and you can add actions which are additional functionalities that to me look just like functions inside of the Assistants API. But I would have to play with this to confirm that and then look at Rohan using all of this. So what he did is he downloaded his Twitter post data from Twitter analytics and then what he did is he provided custom instructions here inside of the instructions field and he uploaded all the analytics data to his custom GPT. And now when he asks it to optimize his tweets, it's not going to be classic GPT, it's going to be classic GPT4 with knowledge of all of his analytics and his preferences that he installed here in the custom instructions. And it has access to the web, can generate images for his posts, and can even use the code interpreter to create visuals, mind maps, whatever you might need for the posts. So what he essentially did here is he turned a general purpose GPT4 model into a single purpose, one that is there to optimize his Twitter xPosts. And that's really the bigger story here. These GPTs are going to be very niche. They're going to be there for one use case, but most people have a few use cases that they keep returning to. So what I expect here is that beginner users are going to use some of the preset GPTs and intermediate to advanced Users are going to be building their very own GPTs that have their own custom instructions, have their own custom data that have their own prompt presets. And all you need to do is navigate to them here on the side, just like here I navigated to the Tech Support Advisor. And what you'll do with the GPTs is you'll just add them to the sidebar and all their favorite use cases will be up here. A greater user experience. Really what you are going to be able to do is share these externally because there are some advanced examples in here that I don't want to go too deep into. But for example here, Nick Dobos on Twitter, he built something that he dubs AGI. It's basically one of the more advanced versions of these GPTs that I've seen so far. And what he does is he uses a SQL database with the GPT in here to store data in. And then he set up like 20 keyboard shortcuts where you can press different buttons and interact with the database. So he really expanded this. And just keep in mind this is all day one builds. Okay? So people are already pushing the possibilities of this. And if you Care to learn more about examples here on 350 he does a little demo of how this works. First he creates the database and then he uses the various keyboard shortcuts to navigate around this. And yep, there it is. He has a database on his machine that interacts with ChatGPT, enabling all sorts of fancy use cases like daily task management with the support of AI and then long term memory for all the tasks that have been previously completed. Again, this is all very early, but basically if he shares a link, we're going to be able to go in here and try this ourselves and experiment without any coding knowledge, without any usage of Playground or anything. We're just going to open this up inside of ChatGPT and then here's the prompt presets, he said. And then here we are. I don't even have access to the custom GPTs yet, but I can already use this one through the link that he provided on his Twitter. And then you just add it to the sidebar here. And you can use this at any time. Another one here would be this playlist AI by Brett Bauman on Twitter here. And what it does is it connects to your Spotify and then just creates playlists for you. And again, these are just great early examples. But what about create a drum based work playlist and then I list to artists and there it is, it's interacting with Spotify and here it asks for permission. How nice of it. And we just created a playlist. Look how fast that was. I just click the link in there and I have a drum based work hustle playlist inside of my personal Spotify just like that. So what I did here in advance is I connected it. So here under the privacy settings and connected accounts I needed to log into Spotify. But this just shows the opportunity, right? People are going to be able to link all these external apps right into the ChatGPT web interface. And then you're going to get to use all these GPTs and you're going to get to build your own just like this. And this left bar is not just going to have your history of conversations, but it's also going to have a list of all the GPTs, all your favorite use cases, essentially. And then if you take that a step further, very soon we're going to have multiple GPTs interacting with each other. So this is just the first step. OpenAI did this to make you comfortable with the idea of a specific GPT working for you, but the next step really is taking multiple of these and maybe taking the Spotify playlist creator and taking the day planner with the database and maybe adding a YouTube coach. And if you add your to do list, it's going to create a playlist for you to listen to while going through the to do list, it's going to use the YouTube coach's knowledge to at least start drafting outlines for some of the items on my to do list. And it all gets organized and they all work together. So this is not possible yet. This is something that is clearly coming in the future. But through the Assistants API, which is essentially a way to use these GPTs in code, it's possible already. And Yue here, which by the way is the author of Baby AGI that we talked about on the channel multiple times, built this right away, used the assistance API and did a GPT versus GPT where he had a mean pirate and a bubbly mermaid who speaks like a valley girl. Talk to each other. And yeah, here it is. And yes, you could totally use the new text to speech API to very simply vocalize this conversation. And then he went a step further and started rebuilding the Hustle GPT. If you remember that that was a thing back in spring where they used the API to help you make money, help you build a business. It was an interesting thought experiment, not one that had particularly significant real world results. It was just interesting. And he's already rebuilding that with the systems API. So people are going to be doing these things. I would conclude that these GPTs are a real step forward in terms of mass adoption for these things, because up until now you had to have a deep understanding of ChatGPT and its capabilities to really get a lot of the benefits that came from it. First you needed to know the prompts that were relevant to you. Then you needed to be able to craft the prompts in a way that GPT had all the relevant context to you, so it could even perform them appropriately. And then ideally you had to feed it some extra context in the form of examples or custom instructions, so it can do an even better job at assisting you. And that's a lot of steps to take. If you're just looking for a few quick ideas, let's say now you're just going to be setting up these things in advance. So when I need stickers, I'm just going to open the stick quiz and have my personal custom instructions in here and it's going to be aware of me, my goals, my backstory, and any other relevant information and I can just get working right away. This hugely reduces friction. And if you've been following the email newsletter we have closely. You might have picked up on the fact that that we have a product with preset custom instructions for hundreds of workers like this, including also prompt generator that gives you all the tasks that are needed for the workers. And quick side note, yes, we're updating this for free to fit with GPTs and assistants in here. You can find a link in the description. So the last thing that I want to talk about here is a use case that shows the potential of all of this more than anything else I've seen so far. Because the commentary is a fascinating new thing that we'll certainly see many experiments with. Right? Just think about it. Every single one of us has a high quality camera in their pocket, right? So with the Vision API, there's going to be apps out there that are going to be utilizing this and developers are going to do their best to enhance our life with the use of the camera smartphones. Right? But what are people going to build? Well, we don't know that yet, but we can start speculating because right here, Robert Lukoshka on Twitter created a simple app that is a yoga coach that uses the Vision API. And I want you to really check this out and think about this for a second because he just sets up his laptop and think about this. This is day one type of stuff, right? These are going to be advanced. And here's an AI yoga instructor that doesn't just give him instructions on what to do, he also gives feedback. So right here you can see the AI picking up on the fact that he's overly bent and he's giving him advice that he should straighten them slightly and align your head atop your spine as a string is pulling you up from the ceiling. So this is exactly the type of stuff that a yoga instructor would tell you, right? And this is just one example. I'm just saying these personal assistants just got eyes. And now you can start thinking about how this will change the industry that you work in, because these things are going to get implemented faster than you think. Like at the time of the recording of this video, it's been less than 48 hours since the event happened and people are already creating all these use cases. Just think about what's going to happen once this has time to mature. Give this a year, give this three years. The question here is not if this will replace people's responsibilities. The question is which ones? And I hope that some of the examples in this video inspired some ideas of your own. And I would love for you to share them in the comments below. Let's talk about how some of these innovations might be changing the game in your profession or field. For example, for me, I can totally see a program going through YouTube, looking at different pages, looking at different data, and then making recommendations on what content to create. Instead of me having to manually prompt it manually, have to provide it with all the images, all the context. People are going to build these coaches in essentially any industry. Just something to think about. Okay? And if you want to learn about even more visual use cases, this video right here goes for over a hundred of the best ones. So there's some extra inspiration for you.