This Weeks Most Interesting AI Tools You Can Use

Hmm. Well, this wasn't possible a week ago. This week we got a ton of new AI use cases and today I'm going to dive in each and every one that you can use right now. They range from brand new open source LLMs to image enhancers that a week ago were paid and now you have free versions and so many new apps that we simply didn't have access to a week ago. So with that being said, let's dive into the newest AI use cases that you can put to work today. And we'll be starting our little journey with a tool that last week the comment section seemed to absolutely love. It's AI Image Upscaler. But up until now they were powerful but quite pricey. But CREA AI that we also mentioned last week came out with the beta version of their brand new platform that is as of now, completely free to use. So if you head on over to CREA AI and simply log in with something like your Google account, again, as per usual, all the links will be in the description below. You get access to all of these tools for free, which is absolutely incredible because if you look at this upscale and enhance tool, the this is very similar to Magnific that we've been looking at last week, but this is completely free for now and the way to use it is you simply click this plus button at the bottom left and then let's test it on the same image as last week. I'll just keep the default settings say enhance and then it will upscale it. All right, so let's have a look at this before, after. Okay, so it just made a human in there disappear, which is interesting and might be what you want. Now let's just turn down the AI strength and see what we get like so, so as I turn down the AI strength, you can see it actually preserved the human face in there. So exactly as expected. Whenever you make it less AI, you can expect more of the to be preserved. So as you can see, Cray offers many other tools on the site. You can turn logos into illusions, or you have another version of this live canvas where you can just move things around and draw on it. Just like the Leonardo one we looked at last week. If I just say eagle on top of a blue mushroom and I change mushroom to tree, then we get exactly that. All right, the next up we have Tesla coming out with Optimus Gen 2, a humanoid. Hold up. Well, this is a fascinating innovation. This show is all about things that you can actually put to work today. You teasing me? You naughty, naughty. Let's move on to a use case that you can add actually put to work now. Okay, so now let's have a look at a scary topic, but also extremely entertaining one. Because if you head on over to File AI, you get something that I would consider the most entertaining, deep fake type of app that I have seen yet. Because what happens here is it allows you to pull in the camera from your computer and once you turn on the camera. And there we go. George Clooney in a suit. So a few things I want to point out. Look at the little chandelier that it adds on top and then how it turns my sweater into a suit. And then obviously my face and my hair is George Clooney. So look, it's a fun little web interface. It's not perfect. Expect it to be laggy. But this is kind of incredible because you could change this text prompt to whatever, look at me presenting as George Clooney. And let's see, let's do something a little less popular than George Clooney. How does this perform if I change the prompt to Borat? What? Okay, this is next level. Great success. Again, it's not perfect, but it's something you should try out and maybe even show to others. Okay, next up we have a very interesting use case. And this one might not be so useful, but so interesting that I just had to include it. Look, somebody rebuilt YouTube with fully AI generated videos. Okay? So every single video here is 100% AI generated. And I just thought this was too interesting not to share with you. It's kind of an interesting thought experiment of how the YouTube landscape could look like once AI content creation really takes over. For now, obviously this is nowhere close to good enough, but yeah, you could just go in here and watch a video on the llama championship. I just really wanted to share this as I found this very inspiritive as to where social media platforms might be heading over the course of the next years. So next up, I want to tell you about a data analysis tool that is actually also the sponsor of today's video. I'm talking about Julius AI. And you might be asking yourself, Julia Igor, but you've taught us about the code interpreter inside of ChatGPT that already does data analysis. And you would be right. But Julius AI takes it a step further. You cannot just analyze data sets or create visualizations. You can even train machine learning models with only text prompts. And the whole point here is that Julius is purpose built for data analysis and offers some user friendly features that ChatGPT is missing. For example, you get secure file management, data cleaning functions and even intuitive system prompting that teaches you how to use Julius as you go. And additionally at each step of the process, Julius suggests prompts that allow you to dive deeper into the data in a user friendly way. Plus Julius also has vision, so whenever you have a complicated equation or task, just take a photo of it and it will write and execute Python code snippets to solve the problem. And here's an additional recommendation. Whenever you would be considering Excel, you can now use Julius and then create graphs and charts in seconds. This AI powered workflow really does save time in a lot of use cases where usually Excel would have been the go to. And the best part about it, you don't need to know a thing about coding or data analysis to start using this tool. It makes data science available to everyone and you can check it out in the first link in the description under Julius AI Today this week was full of new large language model releases, including a brand new tiny model from Microsoft and Mistral coming out with models that are now best in class at their size even punching above their weight. Okay, let's have a look at this and let me show you how you can actually use these and what these would even be good for. So first of all, Mistral announced multiple new models and the one that has been talked about the most is this small model, this 8x7B. And the reason this got so much attention is because it's really punching above its weight class here. And yeah, in a second I'll show you how to use this. But first let's talk about why you would want to use this. So as you know, there's these benchmarks that these large language models are measured on. Every large language model that comes out gets run through these and that allows us to compare them reasonably well. So as you can see, the small model got 70.6% on the MMLU benchmark. Now just for reference, GPT4 achieved somewhere around 89%. But we shouldn't be comparing this to GPT4 because this model is fully free and open source, so you can download it, run it locally without the need for Internet connection, and you could potentially build your very own apps on top of it. What we should be referring to is the previous leader in this category which was Meta's Llama 2 model. So if you look at this Microsoft article that they just released this week, you will find that under language understanding, which essentially means they just averaged out all the language benchmarks, we get a score of 63.7 for this Mistral Tiny model. Plus, Microsoft also had a release where they announced their V2 model, which also achieved a 62 in language understanding, which is better than the Llama 2 model at 13 billion parameters. But the Microsoft model only has 2.7 billion parameters. So what's going on here? We're getting releases that are 20% of the size and perform equally as well as something that came out a few months ago. And why does this matter? Well, on a phone you're not going to be able to run a 70 billion parameter model, but a 2.7 billion parameter one might just be able to run on your computer or even your phone, because not to get too nerdy on this, but the rough math on this works out in a way where for a 7 billion parameter model, you need about 28 GB of RAM. So that leaves us with two questions. One is, how does this compare to the free version of ChatGPT 3.5? Well, according to the D4 technical report, it scores a 70. And I can tell you that in practice it performs very similar to the 70 billion parameter model of llama. So essentially we just got models that are almost as good as GPT 3.5, but they're only 5% or even less of the size. And they're open source, which means you can use them and start building your very own apps on top of them. And so can everybody else. So that leaves us with the last point. How do you use this? Well, there's this one site called Together AI, and if you just sign in with, for example, your Google account, you're going to get your very own API key here and you can simply get started. And this is just a fantastic way to test various prompts with different models, because as you can see here on the right side, you just get to pick the different ones. So we just talked about this Mistral Media model, right? It's right here, Mistral 8x7B. And now you could test your favorite prompts in here and see what model fits your needs the best. And maybe you want to use it for creative writing or marketing. And then it's just fantastic to have all of these different models in one interface. And as of now, they give you $25 in credits for free to start off, so you can just start using it by logging in with Google. Isn't this great? So, I don't know. I couldn't really judge if this is any better or worse than GPT 3.5. It's just different, but it's faster. And you could use this locally without needing an Internet connection. That's a slightly different workflow than just using this website. This is really the simple way. But yeah, there you go. So with this Mistral 8x7B, we have a new champion in terms of performance versus size. And it's open source, so you can use it today. All right, Thiago Forte on X shared a use case that I absolutely love and he's using GPT Vision to analyze handwritten text. And the reason I want to feature this is to not because there's no other software to analyze handwritten text, but there's no other software that actually understands handwritten text. Whenever I tried this with something like Adobe's Acrobat, certain words are just not readable and then it doesn't have the ability to fill in the gaps. Whereas with GPT you can totally do that. So let's test this briefly, shall we? All right, so as you can see, I wrote it in two different styles. I usually write in the first one. Please don't judge the quality of my handwriting too harshly. As you can tell, I'm a computer type of guy. But I just took a low quality screenshot of this recording and all I'll say is the same thing as Tiago, which is, please transcribe the handwritten text. Let's see how well this works. This is not a high quality image, this is not very high quality writing, and it did it in both cases. So let's see if I open this in something like Adobe Acrobat and I say recognize text. It's English. Yes. Let's see how well it does. And there you go. It didn't really recognize text at all. Now look, not to bash Acrobat, that's obviously set up to read computer generated text, but if you have human written text, even in the scribbly low quality writing style as I have, then there you go. That's another use case for ChatGPT that you might not have considered yet. Right? All right, let's move on and let's completely switch gears here because we'll be heading into the territory of AI audio generators. And before I lose you, because you might figure that this doesn't concern you, you don't really care about what kind of music AI is creating. This is not that this goes a step beyond. And yet again, this is a meta research project. But I think this one might be interesting to absolutely everyone because it goes a level beyond anything that we have seen yet. And I'm not just talking about the quality Here. Because Audiobox really is about combining multiple tools into one. So if you try these demos, you will quickly realize that, sure, you could synthesize voices as many other tools do. You could restyle voices, you could generate sound effects, or you could erase noise from speech, or you could head on over to the Audiobooks maker and use all of this at once. Okay, so they have a little animated tutorial, but we'll just head right into it. Because if you pick one of these presets, like let's say we're fixing a rocket engine, right. It will open up this interface where you have a multitrack interface with different AI generated sounds. Okay, so I think the best way to show this off is by simply showing you the engine isn't supposed to sound like that. Have you tried tightening the Quantum Flux modulator? Let me try that. Nope, no dice. Okay, so as you can see, this goes beyond just voices. It generates a sound effect. The voices, it combines it all, and you can play around with it in one interactive interface, which is really fun. But let me try and add something new here. Let's do some crazy sound effect. And this is what I find impressive about this. It does some of the ones that you might consider impossible. So let's just try this with something unusual, like a laughing bear. Let's see what we get. Not bad. So if you ever produced a video ever, you know that getting sound effects is a big part of polishing a production. So you could just generate them yourself. You could register new voices, add speech, and combine it all into one interface. And another interesting thing here is this voice restyling. So if I go ahead and just record my very own voice, the audio generation technology produces high quality speech and music for enhancing presentations. The audio generation technology. And by the way, that was recorded on the internal mic. Let's just see how this works. And I'll just turn that little recording from the internal mic of the laptop into a different phrase in an excited and happy style. Let's see what we get here. Don't forget to like the video if you are enjoying it. Okay, the laugh was interesting. Let's do it again. Don't forget to like the video if you are enjoying it. You know what? I do want to give this one more shot where I get a little closer to the source. The audio generation technology converts speech into music, creating a sound that's both melodic and voices. Okay, so that seems like more of a fair test there. Right? It's done. And let's generate again. Don't forget to like the video, if you are enjoying it. So as you can see, these widely vary, right? And honestly, it's not really close to my original voice. Not great, not terrible, but nevertheless, this looks very capable. And getting a whole suite of tools like this for free, it's not a bad deal, right? So just be aware that this is a research preview and when you do this often enough, it will just cut you off because there's just limitations they had to put on the usage. It's time to stop. So again, a new set of audio generation tools. Very interesting stuff. All right, so let's look at Microsoft Designer and this is a very quick one because it's essentially Dall E Free that is freely available. And some AI Advantage community members have found that this works better than the Bing image creator. It's not overloaded, it's more reliable and it's completely free. So you don't need ChatGPT plus to use Dall E Free through the Designer Image creator. That's basically it. The link is below. Let's move on to the next one. All right then here I quickly want to return to this image. Upscaling also very briefly, we looked at this last week, but there's this brand new trend of upscaling emojis. So look, if I upload this crying emoji, you can pick all these different styles, right? And these have been going quite viral all across social media. So just want to show you how to do it. So for example, if I pick science fiction and horror, I bump the creativity all the way up. Well then I will get this result that is not the original emoji anymore as you can see still. So look at that. Oh my God. What appeared in its mouth. This is horrifying. But yeah, some of these have been going absolutely viral across social media. You could use different styles and you could use Create AI to do this for free yourself. You will just have to prompt it a little bit in the process. All right, next up, let's talk about AI video. And I want to quickly address this clip that has come out and that went viral across social media because somebody made a brand new news show where the audio and the video are 100% AI generated and it's better than anything that we've seen yet because they actually took care and put some work into making this good. So here's a quick clip. If you haven't seen this. Hello and welcome to Channel One. A new way of consuming reporting and thinking about the news powered by artificial intelligence. It's not fake news. There Isn't a computer somewhere writing its own news stories about things that haven't happened? All right, interesting experiment, but what we care about on this show is how did they do it? Well, across the past few months, we highlighted some of these tools, but I just want to reiterate, the best way to generate AI video today is Heygen. And I reckon this is what they used to because as you can see, the level of animation is just next level. And for voices, they probably used 11 labs and the new feature that I highlighted a few weeks ago where you can use your very own voice to affect the intonation of the synthesized voice. If you do this carefully and rerecord multiple times, you can really get to a level where it's almost indistinguishable from humans. And to me, it seems like that's what they did here. So I'm not going to do a full tutorial on how to do this A to Z. This could be a separate video if you're interested in that, as per usual, you can leave a comment below and into the most popular ones, I will dive deeper, deeper and create dedicated tutorials. But there you go. This would be the tech stack that I would recommend. Use Heygen and Level Labs. They're not the cheapest choices, but they are the best. And if you want results like this, you'll have to use the best. Okay, and that wraps it up for this week's use cases. If you want to check out more of these, we created the same video last week, which you can check out over here and subscribe to receive weekly updates on what's new in the world of AI and what apps you can actually put to work. Now I show you outside from my houses T shirts.