These AI Use Cases Will Affect Everyone You Know
OpenAI's GPT-4o is rolling out now with faster responses, cheaper pricing, and better vision features. Google's Gemini Advanced handles massive documents, and smaller tools like Stable Artisan and 11Labs are adding video and music generation. The host walks through workflows and waitlists creators can access today.
Chapters
Okay, this is one of those weeks in the AI space where as a content creator, you don't get much sleep because there's just new stuff coming out every single day. There are so many new releases. But as this is AI news you can use, we will only focus on the ones that you can actually put to work today. Which, as a matter of fact, is not that many. Most of the OpenAI and Google I O announcements are things that are coming in the future. Nevertheless, me and my team cop out all the different releases from this week from both Google, OpenAI and other companies. And it turns out that there are some wait lists that you should know about and sign up to, so you can use some of these tools as early as possible. And with that being said, let's just dive into all of this AI madness that came out this week, ranging from GPT4O to new video generators by Google and some nifty new hugging face spaces that were overshadowed by the big announcements. But nevertheless, you should know about them. All right, so first things first, this was by far the biggest one this week. GPT4O, OpenAI's brand new model that beats GPT4 on pretty much everything. It's way faster, it's way cheaper, it has new capabilities. There's a voice assistant coming that is human, like a lullaby about majestic potatoes. Now that's what I call a mashup. Ant can detect and express emotions. This model is multimodal. There is so much to talk about here, and I absolutely did. If you didn't see them. There's two separate videos that I created this week about this announcement. The first one focusing on a summary of all the most important points. I'll link it on top right now. And then a second one which compiling that one was absolute madness. Me and my team collected all the use cases that we considered really groundbreaking and and put all of them into video. That one you can also check out on the channel. But beyond those two videos, there have actually been developments with the model because the most confusing part of the release was what is available today, what is coming up. And as this is news you can use, I see it as my responsibility to keep you up to date on which parts of this massive release you could be putting to work today and which parts you will be using in the future. And in that spirit, I actually created a tweet here summarizing where we're at with the rollout of this release as of today, May 16, 2020 24. By the way, I will keep posting on Twitter when things Change. I will keep updating this, so if you want the fresh updates, just follow the Twitter AI advantage. But as of now, if you go to chat.OpenAI.com if you're a paying user, you will find that you have access to GPT4. Oh, this is the first point. Virtually all paid users that I know have access to this model as of today. But they also announced that it's coming to all free users and this rollout seemingly started a few hours ago. I caught the first few comments on my videos and the first few people on Twitter are reporting that they can actually use GPT4O as a completely free user. This still blows my mind. I haven't wrapped my head around the fact that every single person on this planet is going to be able to access GPT4. Heck, better than GPT4. A multimodal GPT4 that has access to GPTs and so on for free might take a while till everybody gets it, but this is underway. Next up, if you're using it, the image generation in it is still Dall E free. Now in the use case video we look at all the image generation capabilities and they're massively improved. Incredible what they added in there. It can now do text, meaning you can generate a full font for yourself or write text on various images like so. And they also showed off capabilities where it does one shot fine tuning, meaning you give it one picture of yourself and you can recreate that in any style. This is absolutely mind blowing when you pair it with the fact that it can also do character consistency, meaning you can upload one image of yourself, recreate that in a different style and then create a whole comic strip. You can create a whole series, you can tell a story, create a storyboard. And two hours before this recording, Greg Brockman actually started tweeting about this feature, meaning this is probably on its way. But as of now, it's still the old model Dall E3. So if you're going to be testing this in the ChatGPT interface, you're going to find that still the old model, that is kind of meh meh. Okay, but what does work as of today is this upload feature. So if you upload images to GPT4O, you will be engaging with the new multimodal GPT 4.0 and you will get all the improved capabilities. If you check out the benchmarks on Vision Understanding, this is the new best in class vision model. It beats Opus, it beats Gemini Ultra, it beats GPT4 on pretty much all of these benchmarks plus as a power user I can confirm it is the best vision model that we have today and it is available in the web interface today. And for the last ones I'll just speed run this. We already have improved web browsing and code interpreter available today. They made under the hood improvements to these, but the main thing is that they're super fast now, so it's easy to iterate and create multiple generations, whereas before it took forever. One thing that is not here yet is GPT404GPTS and this is surprising to me to be honest. Seems like an easy thing to implement. But if you have GPTs that you use for specific tasks, these still run on GPT4 as of today. Okay, wait a minute. 12 hours passed since I recorded that segment and actually some screenshots surfaced on Twitter or X from Jeremy here that shares that there's a new module when you build GPT. So this wasn't announced. We didn't know of this and I don't have it, neither does anybody else. But it is an interesting preview look. Basically when you create these specialized versions of ChatGPT called GPTs, you have this interface where you create them and there's a new button at the bottom where you can add blocks and states. And it's so interesting that they added this to me because when I teach building GPTs. Matter of fact, when the GPT Store came out, I created a video outlining how to build a GPT with just one prompt. And the entire prompt was based on building blocks. These are different blocks with modalities that the GPT can do for you. They integrated this building block approach into the GPT builder itself. Seriously, I don't want to brag here, but it's so cool to see that the channel is months ahead of these feature rollouts and I teach you techniques that they later on implement. I mean, this has been the case with the prompt templates that are released in December 2022 for ChatGPT. It took over a year, but now they have these buttons, Anthropic has their prompt library and they're all set up based on use case with variables that you can change. Then the emphasis on custom instructions. Now the GPT builder. Anyway, just wanted to inform you on this and I'll definitely be creating tutorials on this once it ships. And just a reminder, everybody will have access to GPT soon as this model, including these GPTs and the new features will be available to everyone. Okay, let's move along. If you're using the phone app, the voice Input still uses the old Whisper. So all of these voice assistant features, both the voice input and the voice generation, are the old models, Whisper or TTS1, respectively. I suspect that this is the one that will take the longest amount of time because this comes packaged with the new iPhone or Android app and the new Mac app. And yes, there is actually no Windows app for now, there will only be a Mac app. This Mac app you can actually download already. I have it on my laptop, but but I do not have access yet. So I downloaded it, but when I log in it just tells me, hey, you don't have access to this yet. You'll have to wait a little more. And that's what you need to know. I'll keep you updated on my Twitter. And by the way, one more thing. We're actually running a challenge this week. This is the first time we're doing this where I'm essentially challenging everybody in the public, everybody watching this video, to submit their favorite GPT4O use case. And then the winner gets a yearly subscription to the AI Advantage community where we do challenges like this every single week. So if you ever wondered what people like you watching these types of videos are doing with something like ChatGPT4 oh, we essentially created a crowdsourced database of all the different use cases that you could be applying to your everyday life too. Oh, and one more thing. For all of you developers building with GPT4O, OpenAI released this brand new and updated cookbook. This is how to implement the API and use some of the new modalities. So if you're building with GPT4O, you definitely want to check this out. There's some new things to be aware of as the image processing, et cetera. It's all on this page they put together. I'll link it below. All right, enough on this topic. Let's move on to the next one here. So we clearly cover a lot of super interesting and cutting edge tech is which but a lot of the tools that we show off can create something incredible, but it never leaves the digital realm. And that's why I'm super excited to show you today's sponsor of Script. They made it their mission to actually take some of these incredible creations, namely the visual ones, and they turn them into products. You heard that right. You create something with a tool like Midjourney and then they make it their mission to turn that into a physical product. And they do it by empowering creators and their ideas. So how does it work? Well, they have an iOS app that I'm going to show you now briefly. And basically there's two main functionalities. One is you can judge other people's creators creations and decide if this is worth turning into an actual physical product. Like, is it just me or have you ever looked at these AI generated images and you thought to yourself, wow, it would be so cool to have this in person. And that's what you're doing here. You're basically swiping left or right on these different mock ups and when it gets enough swipes to the right, they make it happen and you can purchase these products. So this is one aspect of the app. Whoa, this jacket is amazing. Look at that. I could actually buy it right now. This is too much fun. Whoa. What about this Medusa Lamp? This would look fantastic in the background. So you get the point. If a product gets enough votes, they partner up with the creator of that idea and they take care of all the design, manufacturing and shipping. Now here's the second part to the app and that's the creation, because you can participate. If I go to this middle part, you can actually generate brand new products inside of this app and then submit it. And then other people can vote on it. And if it goes through and they sell it, you get a revenue share of the final product. And the whole thing here is quality. So a lot of these are not the cheapest version of that product that you can find, but they sure are extremely unique. So let's just make a super quick idea here happen. I'll go over here, I'll pick something from the catalog. Let's say we want a rug that could look good in the background of the video and you already know it. We're going to do cats with hats generated like so. And I'm just speed running this. Obviously you would want to create more detailed prompts for your generations. All right, this should do for a quick carpet. And after filling out these fields, I can submit this. And now it's available in the app and people can vote on my concept. And that's basically the whole idea. And all you need to do is Download the free iOS app, log in with your Google account, for example, and in seconds you can be up and running and looking at some of the AI generated product. So I personally think this is absolutely amazing because offscrypt is really taking care of all the hard parts of this process like designing it, manufacturing it, shipping it, selling it, marketing it. All you need to do is you need to come up with an interesting idea and then get enough people to swipe Right on the idea and they'll take it from there. So if you ever had any product idea, why not take it the next step? And you can do that by downloading the off script app today and they might just bring your next idea to life. Alright, let's get back to the next piece of AI news you can actually use. Okay, now that we talked about the releases from OpenAI, let's switch gears and talk about Google's releases. And look, this is not going to be a video summarizing all the things they announced. There's a lot of interesting things in there. If you're interested in AI and if you want to explore what direction Google is taking, you can check out the full keynote. But this is news you can use. These are the releases that you can put to work today. So if you want to check out one thing, it would be Project astra from Google DeepMind. It's basically their version of what GPT4O promises to be when the voice assistant ships. So I'd strongly recommend you check that out. But beyond that, I have an exciting freebie here for you. Because the number one question I receive with all of these Google AI products is how am I supposed to make sense of their entire lineup? There's like four different versions of Gemini. There's smaller models, there's enterprise models, they have offerings across Google Workspace for private consumers, for enterprise consumers, there's developer interfaces. Google Search now uses AI, it's included in all their little apps and so on. There is just so much, matter of fact, I counted it. There's A total of 444 AI tools and offerings that Google has right now. So what the AI Advantage team did here is we actually went ahead and created a full overview of all of their offerings and we decided to give it out for free. So if you care to gain an overview of everything, you can check out this free resource. I will also link it below. But look at this. Basically here's an overview of all the different Gemini models, what they do and how to use them. Consumer products, business facing products, business and developer facing products, all of their AI related research projects, new features that they announced but that are not available yet. And we even compiled all of this into infographics. I might create a separate video where I take you through the full thing. But for now, here's the resource. You can check it out, you can use it, you can share it with your friends and family because they have a lot of goodness when it comes to AI tools. It's just not very clear how it relates where to find it and which ones are the tools and offerings you might want to consider for yourself. But now let's talk about what actually shipped from Google this week, because there are some things that are available already and a waitlist I want to point you towards. The one big thing that shipped is a Gemini Advanced update, and the main change here is that they made their Gemini 1.5 Pro model accessible through Gemini Advanced, that is their GPT4O competitor that is accessible through a simple web interface. And yes, that does cost $20 a month, but it includes a million tokens of context, which is 1500 pages versus GPT4.0. That right now has 32,000 tokens of context. Good enough for most use cases, but here you get 50 times more context. Now, GPT4O is better in most other categories, so I would usually recommend that. But if you want to upload a thousand pages, this is what you would want to use. They also expanded the accessibility to many new countries. By the way, this is a common thing amongst many tools. I'll show you some others that did the same thing throughout this week. OpenAI really pushed them to do that. But again this is shipped to Gemini Advanced and the big thing here is that it supports document uploads, meaning that if you have some business use cases where you want to give it a lot of data and then talk to it or rework it into other formats. The Gemini 1.5 Pro model inside of Gemini Advanced is the simplest user interface I know of today if you want to do it with a thousand pages. Now I do have to point out that usually this doesn't work as well as people expect because the data needs to be labeled. You can't just dump all your info in there and expect the AI to make sense of it. It needs some context. By the way, if you're familiar with Rack, this is the same problem there. You can't just click give it everything, it won't make sense of it. A little tip that I learned from building chatbots is that the best thing you can do as a beginner is actually restructure the data into question and answer pairs. But that might take a lot of work if you have a thousand pages. Oh, and just to round this out, one more very important fact is that they actually offer a two month free trial now with the 1 million token size in this web interface and you can upload Google Docs and PDFs to the model now. Plus, one of the announcements was that there's going to be to be a 2 million token size window, meaning you're going to be able to add 3,000 pages. I'm not exactly sure who was asking for that at this point, but there you go, that will be coming down the line. So Google definitely making some moves, but most of the things they announced were simply announcements. They weren't shipped products. But one of these was really exciting. It was Google's new video model, a direct competitor to OpenAI, Sora, Runway, Gen 2, Pika Labs or all the other video models. Now they call it veo. And look, the quality was not on SORA level. That's the simplest way to express it. It's very good. It seems to be better than all the other generators. But even the examples that they showed off which will obviously be cherry picked, those will be the best of the best. They weren't on the level of Sora examples that we also don't have access to. So AI video right now is just a space that has a lot of promise, but we don't have access to the very best tools. The ones we have are kind of meh meh. But why am I bringing this up? Because they opened up a wait list for this very tools. So you can head on over to this link as per usual, it's linked in the description below and you can actually sign in here with Google, pick your country and you will be added to the waitlist of this brand new video tool. And let me tell you from experience, once Google does a waitlist are usually pretty fast to roll these out. So I would expect this to be days or weeks and not months. But again that is just my estimation based on all the other Google AI tool wait list that I've been on before. And if this releases over the next weeks, they will have the best video model in the entire space until SORA comes out. So consider yourself informed. Sign up to the wait list. And just one last quick note about this website. It's actually an incredible website. We covered this on this exact show a few months back when it released. This is what they call their AI test kitchen and it harbors a bunch of amazing creative tools. Some of them are super unique like text effects that allows to create alliterations and explode words and acronyms. It's really good for lyricists or anybody who wants to juggle around words in a creative way. But as you can see on screen right now, I don't have my VPN activated, meaning this won't work as I'm sitting in Europe. But if you're into creative and fun things with AI, I highly recommend you revisit this. Although this came out a few months ago, it's a really fun way to explore AI capabilities and completely free. Okay, and there's one more thing that Google actually released this week, and it's this brand new Gemini 1.5 flash model. You can access it through a site that harbors a lot of new chatbots and models like po. And if you watched last week's episodes, you will know that there is this new website that actually benchmarks the speed of these different models. So if what you care about is speed, this is usually relevant for developers, then this site ranks them. And this new Flash model ranks above the 1.5 Pro model that is in advanced. See how confusing this naming gets? That's why we created the resource. Check that out there. It should make more sense. But yeah, this Flash model is speedier than the Pro model by quite a bit. But look, these two models are down here. But if we look at the new OpenAI GPT4O model that you can access freely, that ranks up here, it's twice as fast than their Flash model. So, yeah, there you go. Google announced a lot of interesting things, a lot of inspiratory things that get me excited about the future. But when it comes to what has been released this week, OpenAI does take the crown. And I did a little survey on the YouTube channel. You might have seen it. At this point, over 700 people voted. And I asked, which one of these announcements did you find more interesting or exciting? And OpenAI just won by a landslide because of some of the points that I just showed you. When it comes to what we can use today, OpenAI is the clear winner. But I do have to say I'm impressed by what Google is doing. It seems like they're pulling all of the different strings together and it's just clear that they have all the ingred to compete for the number one spot in this race. But only time will show and I'll be here covering it, just like I'll be covering the next update here, which is the fact that Anthropic actually shipped their model to the entire world now, so all the European users can finally use Claude 3 just as a refresher. OpenAI's GPT4O, Google, Gemini's Advanced and Claude's Opus model are considered the three best AI models available today. And all of this competition between OpenAI and Google actually push them to ship this to the entire world, which makes me wonder what's up with all these limitations on release Usually all these tools come out and it's not accessible in the European Union, the UK and a few more countries. Now that the competition releases their tools to everybody, they do it too. I don't know, I don't fully understand that. Maybe somebody can clarify in the comments. I thought it was like a legal barrier that is unsurmountable, but apparently it's not that hard to ship these things. So Both in the iOS app and in the web version, no Android app available yet, unfortunately. You can use this from all around the world now. But yeah, now that GPT4 is better at vision and free, why should you pay $20 per month for an inferior model that is slower? To be fair, some people do like the writing style of plot, but I think that would pretty much be the only reason. And talking about AI models that are inferior to GPT4O but now open up to the European Union, Hey, I now have access to Twitter's Grok without using a vpn. So to update you on this one, it's pretty much a consensus across the entire space that there's no real reason to use this over some of the the top models, especially now that GPT4O is free. Again, I can't overstate how bold of a move that was by them. But the one thing that Grok does really well is that it actually pulls in the Twitter feed. So it's super up to date. It doesn't need to browse the web, it pulls in the Twitter feed and it's just aware of all the latest happenings in the world, as Twitter is the place where a lot of news breaks or arrives at first, and Grok has access to that, but the model really is not that great. And every conversation around the best AIs, it usually doesn't even come up. And that's for a reason. So yeah, that's what happened in the land of LLMs for this week. Let's move on to the next category, which is other companies that came out with interesting updates this week and they were completely overshadowed by all of these massive announcements between Google and OpenAI. But this one is actually really interesting. This is Stable Artisan by Stability Eye, the company behind Stable Diffusion. And what they did is they created a Discord interface where they actually did something surprising which is pull together multiple models they have. So they took their different image generation models, their video generation model and their music generation model, and you can access all of this through one interface in Discord. So look, in practice it's very similar to Midjourney, but It has the ability to create videos and sounds too. And look, before we give this a shot and try this live here, I just want to point out this is a paid tool. It starts at $9 a month, very similar to midjourney, but you do get three days for free. If you just want to try this out, just watch out. They do make you commit with a credit card and then it just auto renews after the three days. And for that you get 900 credits. And these get used up as you use the tool. So obviously generating video will take up more credits. 20, as you can see, versus using stable diffusion Excel, which is around half a credit. And yes, this also includes access to Stable Diffusion Free, which they recently released. This is their best model, but it does cost 6 credits per generation. Oh, and one more thing that I should point out here is that upscaling is 25 credits, which is quite a bit. So just be careful with upscaling. Only do that on pictures that you actually want to use. And with all that being said, let's get into this. And here's an important note. If you want to use this tool that as of now, only works in Discord, you also need to sign up and subscribe with your Discord account. And once you do that, you can head on over to the Stable Diffusion Discord server, go into one of these artisan rooms and say dream instead of imagine imagery. And you already know what we're going to prompt first. Cat with a Hat. Let's go. Let's see what this gets us here with stable diffusion 3. All right, very nice. I like this first one. And then we can keep working with this. As I mentioned, this is a combination of multiple tools. So let me do some outpainting on all sides sides where I add more cats with hats on all sides. Excellent, Excellent. Okay, that didn't work. Let me just try some of the other features here. Let's turn this into a video and do the creative upscaling too. Okay. And we have to prompt it while upscaling. So let me just repeat the prompt here. Keep the creativity at the default setting. Again, this is just a first look here. Okay. And let's review the video that it created here. Yeah, there you go. This is a typical Stable Diffusion video where it does slight motion well, and then at certain certain points, it just morphs into unusable things. But if you want a very slight animation on something, that's where this actually works. It's just. Yeah, it is what it is. But the upscaler on the other Hand. Look at that. This looks excellent. The original image at 350 kilobytes over here and then the new upscaled image of 2.5 megabytes over here. Wow, look at the difference. Yeah, day, night. So look, I think it's really nice that they combined all of these tools in one interface. Obviously stable video is what it is. If you're familiar with the tool, it's just not that great. But the upscaler here is actually really impressive and it's really convenient to have all of this in one interface. So if you're looking to create many of these, this is probably the most efficient workflow you can have with all of the tools including upscaling and video generation in one chat interface. Look, even though it is discord, it is the most user friendly way to generate these. Rather than having multiple websites and having to download and re upload files across the place to generate videos. It's just a welcome addition that brings together multiple of their tools. One thing that I am missing here is the audio that they promised on the sales page right after a little review. They actually did not promise the audio in the blog post, but it is included in the announcement video. So that's probably coming soon. And while on the topic of AI audio, I just quickly want to point you towards this announcement by 11 labs. They're working on a music model which is not available today. It's just too good not to show off. Just listen to this. As they're super good at recreating voices. Their rap abilities are best in class. Have a quick listen. All I ever crave was to shift the paradigm. Boldly advancing, no fear, just climbing prime. I don't know about you, but to me this does pass the Turing test. This sounds like a human being yet again. It's not available today. I just wanted to bring it up as we're talking about audio. Okay. And I got one more tool for you this week and that is this tool called IC Light, which comes with a hugging face, basically so you can really easily try it. And basically this allows you to relight images with AI. So basically input something like this and say sunset oversea and it changes it into this. This is not just image generation, but we're starting to get image editing capabilities with AI. Few more examples of a husky. Aw, I love huskies. Turned into a sci fi RGB glowing magically lit husky. Or even better at turning something simple like this into a beauty photo shoot. Now let me briefly try this myself because these examples are usually cherry picked. Let's take this high quality Instagram worthy picture here. And let's use one of the prompts that they use in their examples as I want to keep it fair, I don't want to switch it up too much. I'm just going to change the first part to man and then keep everything on the default settings and I'll just say relight. I'm super curious to see what we get here. First try no editing, no two takes. Okay. Ten seconds later. Okay, that's not that bad. I'll slightly vary the prompt and run it one more time. Not bad. Look at that. It put me into forest. It adjusted the lightness and the colors of the image to actually fit it. It perfectly color matched it. I actually really like this result. Look at that. This is just a demo, but soon we will have these tools built into interfaces like we saw with stable artists and bringing it all together. And when we combine something like this with GPT4O's new image generation capabilities, you're not going to need Photoshop for most use cases this anymore. You're just going to generate exactly what you want with the correct text with the character consistency of it just by uploading one image of yourself. And then you're going to be able to relight it. With tools like this that eventually will all be baked into one tool, the future is going to get interesting to say the very least. And with that being said, I hope you have a great day. I'll see you soon.
