This Was an Insane Week for AI Use Cases

Okay, so this week in AI has been packed with brand new use cases. It took me way longer than usual to actually go through these because most of these are actually novel because we have players like Adobe integrating AI into Acrobat. There's an AI that you can use today to create surprisingly good songs. Plus there's a brand new search engine that I actually started using daily. So let's unpack all of that and even more in this week's AI use cases, the weekly show that shows you all the things that either you or others are doing with AI today.

Okay, let's get into it. First one, Adobe making moves, integrating generative AI into their own software. And the timing of this is a bit peculiar, right? Right after the Sora announcement, friend of the channel Bilal Wali here on Twitter shared the fact that Adobe formed a new 50 person team that works on AI research. And now a few days later, we get this AI assistant inside of Adobe Acrobat. So this feature is in beta right now, but as you can see in the blog post, it is available to all Acrobat standard and pro individual and team subscriptions, including me, I have an individual subscription. And if you update the app, like I did here, you're going to be faced with this AI assistant button. So I opened up the little prints PDF just because it's lengthy and I know the story. And then you can go up here and click the AI assistant button. And you get all of these options that you would have previously had to do manually inside of something like GPT-4 in here. Now, as you can see, asking questions, generating summaries and key takeaways, the typical use cases are recommended to you rather than you having to know about them. And this product really follows what we've been talking about so often on this show, which is that a lot of these little demos and use cases that we see pop up every week over time will be integrated into the big apps like Adobe Acrobat, because PDFs matter. Did you know that there's approximately 3 trillion PDFs in the world? No. Well, now you do. And you can use AI on top of them natively like so. So you have some prompt suggestions right here. But I could also go ahead and ask, summarize the story in one sentence. Okay, that took a few seconds. That's great. He travels from planet to planet encountering various characters and learning important life lessons along the way. Yep, that's the one sentence summary that works. And then you have these follow up questions. So great AI integration right here. But now the question presents itself, why should I be using this over GPT-4? And the answer to that is the same as with most of these dedicated tools. This thing has been built to interact with PDFs, whereas with GPT-4 you can interact with PDFs, images, code, text, and it's built to be a helpful assistant that also gets things done for you. That's very different from this tool in here, OpenAI is going to have a hard time to give you prompt suggestions and to achieve speed like this because they're trying to do so many things. In other words, specialized tools like this are more focused and fine tuned for this specific job. So really, if I do a side by side comparison, I upload the PDF, I use the same prompt, you're going to see that by the time the first letter comes out, Adobe Acrobat already had the answer. But here out of fairness, I do have to say that the GPT-4 answer was actually a bit more accurate. It's still one sentence, but it went into more detail. But if I follow up with saying what are the lessons, it immediately lists these and it's super fast at that. And it's included in the software. So I expect all the apps to integrate something like this over time. But for now, let's move on to some product upgrades by OpenAI themselves.

This is going to be very brief because it's not a huge use case, but basically two things happened. First of all, they went ahead and announced that they're closing down the plugin store and all the plugins in ChatGPT. So if you were used to any of these plugins in here, well, that's a thing of the past. You can forget about them. Starting April 9th, you will be forced to use them through a GPT. And hand in hand with that, they came up with an update that upgraded the GPT store and the depth of the GPT profiles a little bit. I really like little updates like this. But the main thing here is that the message from OpenAI is clear. Forget about the plugins, we're all in on GPTs. Just to show you the concrete change, if I go into this AI Advantage digital marketer that we freely provide, you can go into the About section, and you can now publicly see all the chats, they included ratings now, and you can link social profiles here on top. Now, what I found for myself, but also what I heard from the AI Advantage community is that the core usefulness of these GPTs are not the public ones, they're the ones that you build for yourself, the personalized ones that you don't even publish. So just keep that in mind when you interact with the GPT store. And while we're here, check out the digital marketer. Prompt number four here, for example, generates 30 prompts for digital marketing. And then you can just go ahead and copy paste these and run it with the digital marketer. But obviously, this is going to work better if you personalize the instructions or the prompt here by yourself. So in summary, plugins no more, GPTs are the future, and the best ones are the ones you build for yourself.

Okay, so next up, we have Suno V3 that released an alpha. If you didn't know, Suno is the best in class music generator, in my opinion. It's really unparalleled when it comes to creating music with lyrics, because it writes a text, generates the music, puts it all together. And to use V3, it's not free right now, they had a Valentine's Day promotion where you could create songs for free on Valentine's Day with V3. But I gotta say, this is such a big upgrade from version two, rather than me just talking about it, just have a listen to a country song that I just created here just by saying create a country song about a generative AI that replaces all YouTubers. Are you kidding me? That is so good. And here's another one about this weekly show. Okay so I tried different use cases here. It seems to be really good with country and drum bass. The pop was okay, and the grime that I tried didn't work too well, but go explore yourself. This is getting ridiculously good, and I think I'll just have to create some songs that I'll be using from here on out.

Okay, moving on to AI image generators. Some very interesting news this week, especially when it comes to generating images with text in it. Because up until now, we got half-baked versions. Last week, there was the stability AI stable diffusion free announcement. It's not something that's usable, so we didn't include it in the show. But as a reaction to that, Ideogram officially came out with Ideogram 1. 0. And if you didn't know, they had a sort of preview live. This model had one really big strength, and that was creating text. And now that they released, you can for free try this out. I just did a standard benchmark prompt with a cat with a hat. So as you know, that's just my go-to prompt in image generators, and then I judge the results based on me running this for every other image generator. And I'd say about the style of this, it's very storytelling fantasy-esque, which is neither a good nor a bad thing. That's just what it is. That's how the model is trained. But as I mentioned, the big strength here is text. So I'll say a logo of a pet store named, and then I'll just say a cat with a hat. And as you can see, that's pretty damn good. A lot of these are actually perfect. Matter of fact, three out of four were perfect here. Let's just regenerate to make sure. Four out of four. Something that is unthinkable in Majorny or Dali, although they did get way better with text. So there you go. That's a free tool you can use.

Moving on to the next one. So in last week's episode, I showed you this example of the NBA using generative AI to transform highlights from matches, almost in real time, into a stylized version that looked just like the Spider-Man comics. But that was NBA's proprietary model that they're developing. We don't really have access to this yet. So this week, I'll be showing you how you could create something similar by yourself. And I'm happy to say that this segment is actually the sponsor of this week's video, DomoAI. And they just launched this brand new video-to-video feature inside of Discord that transforms your videos into stylized versions of it. And it actually works really well. Let's have a look. So as of right now, this is only accessible for Discord. So what you need to do is join their Discord server. But luckily, using it is really simple. You just head on over to one of these generate channels, and then all you do is say slash video. So what I'm going to do here is I'm going to use one of the Sora examples, because I think the implications of this are really interesting. AI generated video and then AI remixed video, a lot of possibilities open up there. So I'm just going to take this man on the cloud, and then I'm just going to copy over the prompt that was used to generate it and hit enter. And then here are all the choices. From my testing, I do prefer the illustration styles over the anime styles, because the anime styles keep changing the scenes a bit too much for my own taste. So to do this, you will need the starter plan. But once you have it, I'll just pick a 3D cartoon style on this one and generate a three second video. Let's do it. So while this generates, here's an example that I tried beforehand of this animated Sora clip transformed into a comic style right here, then a Chinese ink painting or paper art. Look at that. Or here's a clip of J. Cole rapping in pixel style. Now on play this for copyright issues. But the cool thing is that Domo does maintain the audio under the clip. So if you're making music, you can just run it through Domo and turn a simple phone recording into stylized music video. That's a great use case right there. There you go. It just did it. By the way, I really like what it did to the clouds here. If you want to try it yourself, it's the first link in this video's description. Thanks again to Domo for sponsoring this video. And now let's look at our next use case.

Find SD. art and this is provided by civet. ai, which is a lot of stable diffusion models that you can use on their site. And the whole trick here is you can upload various images where you're not sure what model was used to do that or what fine tune of the model was used to do that, to be more precise. So there's many popular fine tunes, but one really distinctive one is this juggernaut XL, which is trained on beautiful people's pictures, I guess. That's the whole concept here. You'll only get images of people that look like supermodels. And if you wanted to recreate that, it would be kind of tricky to find out what that is if you're not aware that juggernaut XL does exactly that. Now with this tool, I can just upload an image and for free without logging in, it's just going to find what stable diffusion models could probably be used to get this result. So let's see. This one has a 100 score. Let's open up all three of them and let's review. Okay, so as you can see, the 100 score actually got it wrong. It says dream shaper, but on the other one, it correctly identified that it's juggernaut. And these results are actually consistent with our findings while I was testing this tool. It's really good at pointing you in the right direction, but don't expect it to be spot on 10 out of 10 times. And to be honest, that's really all you need. You need the right direction, something that's similar enough or something that gets you that result. So yeah, probably a really good tool to have somewhere in your bookmarks bar. Now you can reverse engineer AI images really easily.

All right, next up, we have a really big one. Everybody has been talking about this throughout the week. It's Mistral coming out with their large model, which is essentially a GPT-4 competitor. And yes, that's Mistral, the French company that has been coming out with open source models left and right. But it turns out that this model is actually not open source. And matter of fact, they removed committing to open models from their website. So yeah, it seems like they used open source as a marketing tactic to lead up to their big announcement of this model. They released many models up until now, including small 7B models and their Mistral model, which had really interesting outputs calling upon various experts to gather those. But none of that really competed in terms of answering questions, reasoning or code completion with GPT-4. Well, now they are competing. And as you can see in this graph, Mistral is actually doing super well on all the benchmarks that these are measured at. So well that it actually beats out Gemini Pro and Claude's tool here. It has a 32K context window that matches GPT-4. And if we're just judging by the benchmarks, which is usually not the approach you want to take, but when it comes to coding, it actually turns out to be pretty reliable in my experience. You wouldn't want to use this model for coding. For that, GPT-3. 5 still performs better and GPT-4 just smashes it. But that's basically common knowledge that for writing code, GPT-4 has been and still is king. But they did something here that they haven't done before, which is release a chat interface. Because they are French, they call it le chat. It's a dark mode by default. It's le chat noir. And then when you make it white, it's le chat Mistral. Excellent. And you know what? This thing is actually completely free to use. So you can just head on over to chat. mistral. ai. Give this a shot yourself. Let me know in the comments what you think about this. And yeah, I'll be giving this thing a proper spin over the course of the next week. Very interesting development. We're getting a lot of models here that are almost as good as GPT-4. Just keep in mind that it's almost a year since GPT-4 released. So it's about time OpenAI make their next move, which I'm excited for. But for now, we'll be looking at the next use case that we can actually access today

which is my last one for today. And remember in the intro when I mentioned that there's a new tool that I actually use daily right now? Well, it's this one. It's called the GigaBrain. And what it does is really simple. It's a really well functioning search engine for Reddit. That's powered by AI. But if you ever found yourself on Google searching for a keyword and adding Reddit to the end, well, you might know that this is decent, but it's not perfect. And the Reddit search function, it's too messy. It pulls from all the subreddits and none of these ways of ranking it really work. And you don't even see inside. You can't see the comments. I personally never found this to be too useful. But the GigaBrain actually solves this. Because if you go ahead and search for something on here, it's going to show you the sources. It's going to give you AI generated TLDR summaries, and then you're going to see all the source threads right here. I really enjoy this interface. You can see all the replies right here. No need to open up the sub pages, right? And if you see one where you're like, yeah, I really want to read this, but this is just too much right now. Well, just toggle this little summarize button. And there you go, you get a nice little bullet point summary of everything going on. And that includes the comments underneath. I mean, how great is that? And it's free. All I needed to do is log in with my Google account. So there you go. I think this was the most colorful week in terms of AI use cases since the beginning of the show three months ago. You can check out the entire playlist if you're looking for even more use cases that came out in the past weeks. And other than that, leave a comment below sharing which one was your favorite today so we know what to cover more. There's a lot to try here until we do this again next Friday. I'll see you there.