AI Use Cases Are Evolving At The Speed Of Light

All right, it's been another week in the world of generative AI, and the releases have not stopped. I have a bunch of applications to look at this week, but I just gotta be honest, over the past two weeks, it was way more announcements of future stuff that you can use rather than stuff that you can use today. And as you know, that's what we're focusing on this show because we did get new ChatGPT upgrades from OpenAI. Microsoft announced a plethora of things. Some of those you can actually put to work today. We'll be looking at them and there's a few AI features and upgrades from applications you might already be using that you might want to consider. And with that being said, it's time to have a look at all the AI news you can actually use, starting with ChatGPT. And I barely missed this in last week's video. It came out on Thursday, we shoot on Wednesday. But I do want to highlight these. These interactive tables that are now available through the use of the data analysis tool. If you're not familiar, that's basically ChatGPT's way to do math reliably. And there's a bunch to talk about here because they did change the code interpreter quite a bit. And that's why this week I uploaded another video where one section of it talks all about the new data analysis tool. So if you want to learn more about it, that's the place to look. But let me just tell you, this is a really fantastic feature. And if the product keeps developing in this direction of more reliable, editable, customizable features and outputs, well, you might already know this. But that does get me very excited about the future of this tech. And beyond this new feature in ChatGPT, there's actually a hugging face space that has been going absolutely crazy. People seem to love this. Now, here's the problem. I tried to test it all day today. So many people are using it that it's overloaded. I can't even get a basic prompt to work. But I think that the main pull of this is that it's free GPT4O, which, yeah, you could have that by yourself. But as you can see here in these tabs, they plugged various modalities into it. So, one, you could obviously upload images, but then you can use voice, which also works in ChatGPT. But then there's this live chat where you can upload images from your camera directly. And to me, this is the one unique feature that this has. If you want to play around it, you can see if it's available, it's completely free and then you can prompt on top of the images. It's a public and free site. I just wanted to highlight it before we get into the Microsoft stuff because there's a lot they announced. But this show is about all the announcements that you can actually use today. And to be frank, Microsoft did not release a lot of stuff that is available today. Now. Nevertheless, if you're curious about those announcements, I would recommend looking into particularly the Windows Laptop and the core copilot that they have on it. Now people have been freaking out about this because essentially what it does is it takes screenshots of your screen regularly and then it uses them as context. Anyway, that is a story I want you to follow. But what we will be talking about today here more in depth, are first of all their version of GPTs, their co pilots and their co pilot Studio. This is what their presentation was all about. So let's start here because we are at a very interesting point in the AI space if want you, if you're a user that is looking to get more out of these tools, you will be familiar with GPTs at this point, right? They're the specialized versions of ChatGPT set up to do one particular thing. There was a lot of hype around the release, but turns out most people don't use them regularly. You need to be intermediate to advance to actually make use of them. You need to know how to structure a prompt, maybe include an interaction sequence, maybe include extra documents in the knowledge base. And if you want to use the actions, you need to know how to host the API endpoint and then turn that into OpenAI's key schema that then you use in action. Simply put, building GPTs is just not feasible for most people. But nevertheless, the entire industry agrees that this is the direction the product will be evolving in. Because Microsoft is leaning further into this direction of doing copilots. And they have Copilot Studio where you can build them, which is like a GPT building environment for corporations. And there's a thousand features they shipped along with it. It's going to be able to facilitate your team meetings, it's going to be able to manage your projects and so much more. But the reason we're talking about it here is because you can access these, you can access Microsoft Copilot, you can subscrib to Copilot Studio. By the way, this is something I didn't know for a lot of corporations. They blocked chatgpt.com you're not allowed to use that. Only AI that you're allowed to use is Copilot because the companies might have an office subscription and then this is an add on that you can use. So as a consumer who always wants the best tool, like me, you might be wondering, why are people using Copilot if I have GPT4O? Well, you know, GPT4O might be blocked in your company. That's why you would be using it. But here's the reason I wanted to highlight this. Because the interface moved forward, forward. And now we have actions that don't require you to host API endpoints. We have actions just like I expected them to involve little consumer friendly toggles like this, where it allows you to log into another app. So this is how this product is evolving. We're going from a generalized model to more specialized GPTs, co pilots. You know, in Google's case, they call them gems. It's all the same thing. It's a specialized version that has the tooling to do a specific thing. And the reason I'm focusing on this in this episode is because they introduced something, something new that up until now you might have known from applications like Zapier, Make.com or other no code automation tools. Because these copilots now allow you to create workflows. And these workflows are essentially no code automations. There's one trigger that sets the automation off and then there's multiple actions which can be a simple prompt that is just ran in ChatGPT, in this case in a copilot. And then you can build a whole flow that triggers every time something happens and you get to set up what that something is. That's all that basic automation is about. Here's the thing though, just to have access to this Copilot Studio, you pay $30 per user per month. So this is completely independent of your OpenAI subscription. And I took this time to highlight the point that the GPTs that we have and use, this is something you can use today. They might not be at a point where they're changing the world right now and when they're so damn useful that you just couldn't be living without them, but they're clearly a step on the way to this future, which is going to be dictated by more of these agents that go out and do stuff by themselves, by them introducing these workflows that automatically happen in the backend, where the copilot just works for you while you sleep, for example. That is the definition of an agent. It can think and work independently of you. So in other words, no code automations are simple versions of agents, right? And if that's a fact, then this leads to the point that if you're watching this video, you probably care about implementing these tools into your life, but you might not be sure where to start. So my recommendation is really start with building GPTs and learning. No code automations and prompting. If you can learn these three things, all of these future product roadmaps independent of the company, whether it's OpenAI, whether it's Microsoft, whether it's Google, whether it's Anthropic, whether it's Meta, or any one of the other players, they're all going to be following this direction of building personalized agents. And by you learning these more primitive versions today, you can get a head start. And once these features come out, you can dive into the deep end of the pool and look at the advanced features. Just like when GPTs came out, all the ChatGPT nerds were looking at actions and how to use them because they all already knew what to do with instructions and the knowledge bases. So you might not be able to use a bulk of these features yet, but you can learn how to do it now. How do you learn? I actually held a live stream this week that I was very happy with. So I just want to share the fact with you that there's a segment in here where for 15 minutes I teach you how to build GPTs in one prompt. This is something I covered on the channel about four to five months ago, but it's one of the prompts that just stuck and I keep using it regularly. So I showed you in here how to do it. I gave it away for free. Plus, we evaluate the results of the GPT4O use case code contest where I picked the three winners, the three most useful use cases for GPT4O as submitted by you guys. So if you haven't seen the stream, I strongly recommend it. In the second, I'll move on to the more fun stuff. But before that, I do want to cover the new models that came out. The five free family is actually quite incredible and I'm just going to glaze over this, but the point is it includes multiple models that are very small and they really punch above their weight class and all benchmarks. Okay, they're available today. You can build with them today. And the one that really caught my attention is this vision model, because this is a natively multimodal model which is with 4.2 billion parameters. If you're not familiar with parameter sizes, Facebook is pretty much building a model now that has 400 billion parameters, and usually the 7 to 14 billion models are the small ones. So this is even smaller. And as it is natively multimodal, it has these vision capabilities in a model this small. So why is this interesting? So, first of all, it has a very large context window. Secondly, you can use it today. You can try it out here, start building applications with it, plug it into your own chatbot that you might be using, whatever the case is. But thirdly, and most importantly, it's small enough that we're going to see this on phones soon. So this is how these things work. They come out like this, then people test them, people build on them, and then the good ones will get integrated into hardware over the course of the next year, I suppose. But the problem, up until now, we never had models this small with a large context that also had vision capabilities that are decent. All right, then, as promised, now we're going to move on to the fun stuff. And this is a very simple app created by Hume AI. This one is called Chatter, and it's an interactive podcast experience. And it's exactly what it sounds like. You just get to go in and you'll get to participate in a podcast where the counterpart is an AI. Let me just show you. This is surprisingly fun. Me and the team have been playing with this throughout the last few days. And, yeah, let's just click start.

Hello.

Hey, what's up?

Welcome to Chatter, an interactive news podcast. The date is May 22, 2024. The top stories today are quite impactful.

Just cut to the chase. What are we going to talk about today?

All right, let's dive right in. Today we've got some fascinating stories. First, a staggering amount of rent debt is plaguing millions of American households. We're talking over $9 billion owed.

Oh, no. What are we going to do about that? What's your opinion? What should we do.

And ensure proper investigation?

I absolutely agree, but, bro, could you be a little more relaxed with me? Talk like a human. Include ums and ahs. And you know what, actually throw in a funny random word here and there, because why not?

Oh, sure thing, buddy. So, yeah, addressing the rent debt, it's, like, super important, right? Some folks think, like, expanding rental assistance and maybe providing more relief funds can to be, you know, helpful. And as for the tragic accident, it's, like, really important to support the families in your proper investigations. Oh, and banana hammock. Just to keep things like what?

Banana hammock. Let's go. Okay. Thank you for this fantastic conversation. I could keep going. It's really fun to trip this thing up. And here's the thing, the underlying tech is actually interesting. Their whole point is that they have an AI that includes more emotional awareness and intelligence, so it can pick up on your speech patterns just like GPT4O voice assistant promises to. But we still have to wait days or weeks for that. We are not exactly sure. While this is available today already, I felt like a lot of people might enjoy this. The latency super low. And also I always like to highlight if there's an app that would be really good to show off to your friends or family, this is the one. Like just pulling this up and starting a little podcast conversation could be a fun little party trick for your next dinner party or bar night, whatever it might be. Now you know about this and we can move on to the next one. If you're finding this interesting, leave a like on this video. It really helps the channel. But now let's get back to the next piece of news you can use, which is actually a new Adobe tool and Adobe has been shipping. People are not talking about it all the time, but let me tell you, I used to be a power user of Adobe Lightroom and I used to take photos of various events back at the beginning of my freelance career around 10 years ago. Just if you're not aware, Lightroom is their bulk photo editing software. So Photoshop is really good if you're doing one image or if you're doing doing basically more advanced and fancy stuff. And Lightroom is what you want to use if you have 300 images and you got to edit all of those. But it also includes a lot of tools. You can retouch the skin a little bit, you can manipulate the lighting just like in Photoshop. But probably the number one feature that I always missed in here was selecting a part of the image and then removing that. You always had to do the round trip to Photoshop and back to do it. And then you had to know what you're doing in Photoshop. This was still cumbersome. We didn't have generative fill back in the day, right. But now they brought this feature into Lightroom. So if you're a photographer or if you know any photographer, bring this to the attention because this is actually a big deal. You can just brush over a subject in the image and it will instantly remove it. And these generative tools inside of Photoshop work super well. So I expect this to work the same way. So yeah, I really wanted to highlight this, because I just remember the days where a feature like this would have saved me hours regularly. And now it's included in the Creative Cloud subscription. And now we get to the last use case of this week. And I'm really excited about this one. Nobody really talked about this. Nobody really created YouTube videos about it, tweeted about it, wrote about this in their newsletters, none of that. But 11 labs released a new feature where you can embed a player on any website that is going to read out the entire website for you. Now, this is not a revolutionary concept. You've seen these players before, but they're usually terrible. They're so bad that when I see one of those players, I don't even bother clicking it. Brother. What's that? Because I know there's going to be some robot guy jumping out at me and trying to read me the article. Like, Audio Native is an embedded audio player. Like, I don't want to listen to that. I want natural human languages. If the people who are reading audiobooks would be reading these articles out to me, I'd be using that feature all the time. And that's why I think this release is really great. Because 11 labs, you're probably aware if you're watching this, they are the state of the art when it comes to AI audio. They have the most realistic sounding voices. Now, at this point, a lot of players caught up, but I still think they are the best. And you could implement that into any website. Now their voices will auto read it for you. All it takes is one embed code that you can add to every website. Build a tool, doesn't matter. If using Squarespace, Wix, WordPress, everyone can add one embed code to their website. And if not, you always have chatgpt to assist you. My tip would be using screenshots along the way to assist you in implementing it. And there you go. Those are the SWIX releases. As mentioned, there weren't as many because most of the stuff coming out out have been announcements. We'll be covering all that once it rolls out to users. All right, and if I missed anything, you can leave a comment below and share it with the other viewers of this video. And with that being said, I'll see you next week in another episode of AI News. You can use.