New AI Tools That Are Actually Useful

Straight up, generative AI is on fire right now. There's no two ways about it. We just keep getting new applications that are actually useful left and right. And look, the word useful is a very subjective word, but I think most people watching these videos would agree that ChatGPT is useful. That is the one app where most of the population would probably agree that there's some everyday use cases for that. That just really makes sense. Well, this week we have a ChatGPT competitor that is legitimately better in certain use cases. That's exciting. And then also Google went ahead and upgraded their assistant on Pixel phones, so they effectively have an AI powered assistant in their phones already. Independent of what you think about the recent product releases, this is interesting. And then there's many more little things like this spy software. You can just put in an image and it uses AI to detect where that was taken. Or this image generator that actually generates images with transparent backgrounds, something that was just not possible up until now. You had to have some level of photo editing skills to get those results. So, without further ado, let's dive into another week of AI use cases. These are all the apps that came out over the course of the last week that you could be putting to work today. So let's just start talking about Claude, because this is probably the biggest upgrade of the week, as most people agree that ChatGPT is the number one app they use and the number one app that is useful for them when it comes to AI tools. Well, Claude 3 claims to be a better GPT4, but so do many apps, right? Is it actually, yeah, it kind of is. Not across the board. It lacks a lot of the features of ChatGPT, and with certain use cases I would still stick to GPT4, but in many use cases that really matter, it is better. Now, I'm not going to go into depth on this. I created a dedicated video on that. You can check out that video over here. If you just want a summary, then I would say these benchmarks generally don't matter. What matters is the use cases. And the use cases that I tested really impressed me. Particularly the image recognition is best in class. Whenever you're prompting with images, you want to be using cloudfree and its ability to generate ideas and brainstorm is excellent. That is one of the things I go to ChatGPT for all the time. And from my initial testing, which I've tested, done a lot of cloth just seems to be better across the board at that. The recommendations make more sense with other use cases of mine, like Creative writing or coding, it seems to not be that clear. In some cases GPT4 is better, in some cases Claude is better. Again, if you want a more in depth discussion on this, check out the dedicated video on this. One last thing that I want to point out is that there's actually a site where you can use it for free. You can try it for free and Compare it to GPT4 and that site is chat.lmsys.org we talked about this before. This project aims to crowdsource the ranking of different chatbots and as a part of that you can actually go to the website. Go to this arena tab on top and switch the model to Claude 3 Opus. This is their flagship model. Now on the other side you can switch it to GPT4 1106 Preview and you'll be comparing the best version of Claude to the best version of GPT4 today. Down here you just enter the prompt, then it runs it in both and you get the results for free. Now the whole point of the site is for you to rate it. So you know, if you have a preference, go down here and pick which one you prefer. But there you go. That's a free way to try Cloud 3 and GPT 4. As you can see here, sometimes it is overloaded, but yesterday I used it all day and it worked perfectly. Alright, let's move on to the next use case. Alright, next up, very briefly, we have some updates to Microsoft's copilot. We talked about this before, but some of these updates are really nice. And by the way, this is completely free. But before I show you the updates, I'm gonna head up to the settings and change it to the dark mode. Way better. So what I want to talk about here is the Notebook feature, which actually allows for a whopping 18,000 character prompts. Good stuff. Okay. It actually took me a while to understand this feature, but what this is really good for is refining prompts because it takes so much context. You can copy paste a long document into here, you will get the output right here and then you can refine it in here as you go. It's really all about the big input window. And the second feature they added here is these copilot GPTs. So as of now there's only a few presets, but they're opening this up as they go. And I just found this direction very interesting. That's why I wanted to share. So you can go for one of these use cases like a fitness trainer for example, and then you have prompt recommendations and it takes on the Persona of a fitness trainer. It's really nice to see you integrate these different things that we've seen in other apps. And this is all free, but now let's move on. AI tools like this are great, but they can't do it all for you. To get the most out of a tool like Midjourney, you have to know a little bit about graphic design, photography or painting. And that's exactly where Brilliant.org comes in. The sponsor of today's video, Brilliant is an interactive learning platform where you can acquire the tools you need to truly get the most out of AI and achieve your goals. Answer a few simple questions when you sign up and Brilliant will help you find the learning path for you. This includes courses and Brilliant has over 100 courses available for you to try for free right now. And every month they're adding more. One thing that's really great is that they add these case studies like this one in the data analysis learning path called Topping the charts with Spotify. It uses real Spotify data to teach you about advanced probabilities, making the concepts relatable and genuinely fun to learn. If you ever felt like learning isn't for you, let's be real. It's probably because of traditional education. And that's not the case with Brilliant. Their bite sized interactive lessons get you hands on with the concepts, a technique that has been proven time and time again to help you learn more efficiently here on this channel. We're all lifelong learners because that's what you need to be if you want to stay on top of AI. Most of the things we cover here are brand new, so by the very definition, if you want to use it, got to learn it. And Brilliant is here to support you on that journey. Go to brilliant.org deiadvantage and start your learning journey today for free for the first 30 days. And the first 200 people that sign up will get 20% off the annual subscription. Thanks again to Brilliant for sponsoring this video. But now let's move on to the next use case, which is actually Google's Gemini that replaces their Assistant. And before the comments fill up with Google Gemini, hate, as it usually happens recently, hear me out. This is actually too good not to cover. I think this just makes so much sense because essentially what they're doing is they're replacing Google Assistant on Apple, that would be Apple Siri with their Gemini large language models. So it just becomes so much smarter. And if you paid close attention, yes, they did this right on release, but it couldn't do a Lot of things. It couldn't create tasks or look into your emails. Now it can, but as an iPhone user, I didn't get to test this yet. Luckily, one of the AI Advantage team members, Daniel, has a Pixel phone and he tested this, this out. So here's a little screen recording on how it works. You can use voice input to tell it things like do you have any unread email messages? And then it uses Google Workspace to actually look at your emails and it gives you all the subject lines and then you could prompt on top of that, summarize all of them, which ones are important, related to my work. Whatever you might be doing, set a reminder to take the trash out or ask about upcoming football matches. Now, if you look across the Internet, some people claim that they might have jumped a gun on this, that it's not a fully developed product. But I really like the fact that they took this bold step and this is one of those use cases where it's just a no brainer. Would I want a better Siri in my iPhone? Absolutely. Will it come for sure when? I have no idea. It might be 2030. We shall wait and see. It just makes so much sense though, right? Because usually these assistants, they're pretty two dimensional and with a large language model in there, especially one of the better ones. Plus the ability to actually interact with some of my apps, eliminating the need for me to go around and, you know, press buttons and spend time setting it up. That's a real no brainer. So I'm glad Google took the steps step here. So, yeah, if you have a Pixel phone, you can switch out your Google Assistant for Gemini and you'll officially be running around with a supercomputer in your pocket that has an AI assistant that runs on it. The future is here, I guess. Okay, let's move on to the next one. And this one is actually super fun. Okay, so this is the TTS Arena, AKA Text to Speech Arena. And if you've been following the channel, you might have seen me talk about Chatbot arena, where you run a prompt and it gives you two outputs with different large language models and, and then you get to vote which one you prefer. This effectively crowdsources the ranking of these chatbots instead of relying upon these benchmarks that have kind of lost their credibility across the last few months, if you ask me. So this is the same concept but for speech synthesis. In other words, I could go in here and say something like, welcome to the AI advantage, and then I could synthesize this and it's going to generate this with two random synthesizers. And honestly, when I was testing this out, I just found myself stuck in here doing this for 10 minutes. It feels like a game. And the game loop here is really spinning spot on because you test out one. Welcome to the AI advantage. And then another. Welcome to the AI advantage. Yeah, so in my opinion, clearly the first one is better. And then you just go into the next round, can randomly generate phrases like this, the lathe to make brass objects. That's terrible. He used the lathe to make brass objects. It's a bit better. And then you just keep doing this and then you can tab over here on the leaderboard and you'll see the best speech generators according to the site. Maybe less of a use case, but more of a resource nevertheless. A fantastic site that I think could be useful to many of you. All right, moving on very briefly, this just popped up on my radar a few minutes before I started recording this, literally. And what this is is a new interface that you can use inside of automatic 1111 that generates images with transparent backgrounds. So if you're not aware, this is not possible with any diffusion model as of now. Dall E Stable Diffusion Excel midjourney. It doesn't matter. None of them can do transparent backgrounds. That is actually a bigger deal than you think, because once you have a transparent background that is actually flawless, you can composite it really easily with other things. Like, think about a lot of the problems we have with AI images now, like the text not working out properly. Well, if you could generate two images, one where it's just the text stylized just like the background image and then the background image, you could connect them really easily and you wouldn't have to create everything in one process, which seems to be a big problem with these diffusion models. Another great thing about this is that it eliminates the need to to remove the background in some workflows. So people are building these workflows where they link a lot of nodes together and oftentimes one of them removes the background, and then everything can happen in an automated fashion. But the problem is if you have automated background removal, it's just not going to work every single time, especially in cases like this. When you're trying to remove the background behind hair, for example, that's particularly tricky. And in Photoshop there's an extra tool for that. And then you have different sliders that you adjust to make it really look good. And if you use an automated tool, it's never really going to be perfect at this. Well, you avoid that Entire process if you generate the image without the background from the get go. Same thing with Glass. Notoriously tricky. And you need to know the specific Photoshop technique to do this well and automated apps weren't good with this. So I'm excited to see where this leads. We're going to cover the workflows people built with this when something new comes out. But I thought this was significant enough to include this and to let you know about it as in my opinion, this is a bigger leap than most people realize. AI just becoming better every single week, huh? Oh, okay. Moving on to something we are going to try here. It's a new Release by Stability AI and in my humble opinion, as a non 3D artist, this is probably the best image to 3D model that I've seen yet. And again, this is just my very brief take. I tested a few images but it's actually quite good. And you can try it out over here on this hugging face space. Okay, so what we're going to do here is we're going to screenshot one of the Sora generated videos. So this takes images as an input. So we need an image. And what I'm going to do here is I remember this little cat character down here. I'm just gonna take this guy. Oh yeah, that's the one. I love this little dude. And what we're gonna do is we're just gonna screenshot him. That's important to get the full body. And then we're gonna just upload him over here and look in real time. I'm just gonna say generate. And if this model is not overloaded, it's actually surprisingly fast. So let's have a look 17 seconds later. Not bad, right? Okay, it looks like somebody punched him in the face or something, but this is pretty good. And I found this result to be consistent across multiple examples. Obviously some of the showcase examples are some of the best ones you'll find. Look at all the detail on this robot. Not bad. So yeah, Stability AI pushing the envelope on what's possible with these models. And again, I understand that most people won't be using these, but yeah, I find it interesting that these use cases do come up and over time we'll see these integrated into apps we know and love. And then next up, a new feature in the AI video. World Collapse has shipped something for everybody where it takes a video and it syncs the lips to the text text that you provide it with. So if you generated a synthetic voice, now you can make the character in the video match that Pika now supports lip sync and it is truly spectacular. Wait, hold up, hold up, hold up. These things can talk for real. Now let me tell you, I went ahead and tested this and it's really hit or miss. Now. I gotta say, the voices are really good because they partnered with Elevenlabs to generate those. But the lip syncing, it's okay. If you have animated characters, it works really well. I wouldn't even bother trying it on photorealistic things that doesn't. It doesn't work at all. Or I think it would be more accurate to say that it would never fool a human, that that's a real video of a human talking. But for animated stuff it can be really useful. And if you're trying to tell a story and your character talks, it's a no brainer. You need this in your workflow. Okay, then I have one more use case for today and this one is really. Let's just say I was very hesitant in putting this into the video. But you know what, I think you should know about it. It's trending all across Instagram and it's this app called Geospy AI and essentially you put an image and it tells you the geolocation of the image basically based on what it sees. Now the thing is not fully rolled out as you can sign up for early access to the Pro version which will be able to provide you with the exact coordinates of the image. And this stuff is a bit creepy and if I had a choice if this thing should exist and be available to the public or not, yeah, I would go with not without hesitation. But that's just not how it works. This thing exists and I feel a responsibility to actually share the existence of it with you. In my testing it has been quite good, but as of now, it can only locate the scene city, not the exact coordinates. Okay, so if I head on over to Google Maps and I go into Street View, I'm just gonna screenshot a random part of this. So I don't know if you're not a geoguessr pro, I don't think you could tell that this is a specific city. I don't know, maybe. And all I'm gonna do is upload this and this tool is free right now, so you can try this on any image on your phone. I'm just gonna upload this and we're gonna see if it identifies correctly. And yeah, there you go. Country? Portugal. City? Lisbon. Estimated coordinates again, these just identify the city, not the exact location that is supposed to come in. The pro version. Either way, I think it's better to be aware of these things rather than shutting your eyes and pretending like these things are not happening. Also, in our weekly newsletter we did multiple segments on deepfakes and fake telecalls because I think it's really interesting for you to know about this stuff. The only way to protect yourself is to be informed. If you're just going to shut your eyes and pretend like AI technologies are not progressing while you're just opening yourself up to people abusing technology, technologies like this, you know, now that you heard that app like this kind of is under development and other people are going to do that too, and in a few years it might be quite common. You might just reconsider the next thing that you share on your Instagram story, or you might make your profile private, whatever it might be. I think education around these topics is the only way forward, and I hope I can contribute my part to that. And with that being said, those were all of this week's use cases. If you want to learn more, we have a playlist with all of the previous videos. So just because something came out a few weeks ago doesn't mean it's not super useful. Particularly particularly. Last week's video seemed to be really useful to a lot of things. I still use Gigabrain for research most days, and this week I've been using Claude every single day. So new and useful stuff is coming out all the time. If you want to stay on top of it, subscribe to the channel and I'll see you next Friday when we look at next week's use cases. Alright, that's it for now. See you soon.