Advanced Voice Tricks & More AI Use Cases

Okay, let me tell you, practical AI releases have been heating up recently. This week we got OpenAI's voice assistant upgrades to various products like Google's NotebookLM and their very best models. Meta released a bunch of AI features and a new version of their Llama models and so much more. Me and my team researched all of it, tested a lot of it, and in this week's episode of AI News yous Can Use, we serve it on a silver platter to any screen of your choice. Let's begin. All right, so the big story of this week obviously was OpenAI releasing the advanced 4 voice mode. Now, a few updates since my release video that I recorded right away. First things first, the rollout is complete. It actually took them less than 24 hours to get this into every plus and Team subscribers phones. That's really amazing, except that people in the eu, UK and a few more countries didn't get access due to regulations. It was actually interesting that Sam for the very first time tweeted about this fact and he mentioned that these jurisdictions require additional external review. Meaning we'll get the feature, it's just a question of time. But I myself am in Europe and check out my phone right here. Hey, what's up?

Not much, just here to chat.

That's Advanced Voice mode. And I shared the step by step way on how to access this. If you're from Europe and want to get it, it's simple. You uninstall the app, you connect to a vpn, doesn't matter which one. A lot of phones have this built in already. Just search your settings, then you reinstall the app, log back in, and voila, you have Advanced Voice mode on your phone. So that's a nice little workaround. I want to point out one more thing. In the video that uploaded right away, I gave you a bunch of presets. Matter of fact, I gave you over 15 different prompts that you can run in there. All of them take different angles and assume that you have a certain profession. And then there's different role playing or brainstorming scenarios that you could engage in. And my hope is that these prompts will spark your creativity and show you what's possible here. Now, I did make one mistake though. I made the video as soon as it released. I didn't have it on my phone yet. I said that you simply need to copy them from the video's description right here. I will actually copy them directly from the product that we developed with 33,000 prompts for or the voice assistant in total. So, for example, here with the roast comedian, if you want to generate fresh material, you would just copy it from your very own copy of the notion template, head on over to ChatGPT. And the correct thing is not actually posting it as a message. Here you go to the custom instructions, which you can access in the settings, personalization, customize ChatGPT. And here in the first box, you just paste this preset save. And now you set up your voice assistant to act as a roast comedian. That will help you prepare fresh new materials, AKA brand new jokes. You just have to make sure to be in a brand new chat. And now if you engage with the voice assistant. All right, let's begin. Awesome.

Give me a current event or pop culture topic, and I'll cook up some roast angles for you.

Yeah, what about Sam Altman now actually owning a part of OpenAI?

Oh, that's interesting. Sam Altman owned being a part of OpenAI? Well, I guess when it comes to AI, Sam's all in. He's got more investment in bots than a sci fi villain. Now he's not just calling the shots, he's owning them too.

Oh, brother. This guy stinks. Okay, I mean, fair enough. Can you be a little more ruthless and comedic?

Got it. So Sam Altman now owns a piece of OpenAI. Looks like someone's taking owning the future a bit too literally. This guy's so deep into AI, soon he'll be arguing with his toaster about who controls the bread.

Okay, I don't know about those jokes, but I think you get the point. This is the way you set it up. There's a bunch of presets I gave you to show you the possibilities, and we created an entire product which is all about showing you what's possible. For a grand total of 1600 professions, this is how you correctly set it up. Just wanted to put that correction out there. Beyond that, the voice assistant has been, well, probably the most entertaining release in AI Ever. I would even state, because it has one standout feature, and that's imitating voices. Now it's lobotomized enough so that it can't address specific names. My name is Jeff. So if you name celebrities and just say speak in their style, it won't do it, but you could broadly describe the characteristics of that celebrity, and then often it will work. Or you just go with something simple. Like, in this case, I want you to roast Marvel for their degrading movie quality, but do it in the voice of a Russian mobster.

I can't do specific voices. But imagine this with a thick Russian accent. Marvel movies, they used to be like rare caviar. Now they are more like watered down borscht served in every corner cafe, no?

Yes, my friend, I think you get the point. You could even tell it to imitate your very own accent. And it will do that. All right, now I want you to roast the apple for the lack of innovation over the past years. And I will have my cigarette in between listening to your yapping. Go ahead, do it in my accent too. No, no, no, no, no. Oh no. My French accent was so bad that it actually thought I was French. I don't know what to think of that. But as you can see, you can make it switch voices. It's multilingual, you can make it switch styles. It incredible. And with all of the prompts for inspiration from the first video, the opportunities here are truly endless. So go out and enjoy it. I just wanted to tease that we're already preparing a big video with various use cases where we go a little deeper into different use cases for various professions. So look forward to that on the YouTube channel. And until then, if you don't have the plus plan, you can actually go ahead and use Gemini's alternative. It's not as good, but you can get a one month free trial and also get a taste of the voice assistant. Even Grok rolled out some voice features now when I updated my app, it now has voice input, it can speak it. So next to that and Meta AI announcements that also integrated the voice feature into Meta AI, also not available in Europe. We just got all of these different voice assistants throughout the past weeks. OpenAI's being the best one by far. All right, so with the next tool, I'll show you how to build one of the most important things you can have in 2024. Your own websites that is clean, yet professional. And to do this, let me introduce you to Hostinger, the sponsor of today's video. Hostinger makes it easy to build and host all kinds of websites. That's a great platform for anybody new to website building, especially with all the AI features that they recently added. They even have this AI website builder that I'll show you now. All you need to do is give it a few descriptive words and voila, you have a website. So once you sign up for the link in the description below, you will be faced with this welcome page if you decide to go with the business plan. It even comes with a free domain. So you could just go here and look for. Yep, I could just get the website Advantage right here. But now let's get into the promise step by step of building the actual website. I'll just click set up here with the business. Business web hosting, I will be done in no time. I'm doing this for myself or my business. I'll create a website. I'll go with the hosting a website builder. Because this is the simplest alternative right here. I'll secure the domain for myself and less than a minute later I can start creating my website with AI. I'll make this a portfolio page for this imaginary project which is an AI website gallery. And I'll keep this really simple, but the more detail you give it, the more it will incorporate. And that's it. I'll hand this off to the AI and there it is. A clean and professional website with our various projects. All I have to do now is change out all the data like the pictures or the text. But this is absolutely beautiful, functional. I can just go to edit and change all of the media just like so they even help you with the simple checklist. Let's be real. Could this be any easier? If you want to use this yourself, check out hostinger first, link in the video's description. And now let's move on to the next piece of AI news that you can use. Okay, next up, we have some major improvements from Gemini's models, namely 1.5 Pro and Flash. This is sort of like GPT4O and this one is like GPT4O mini. So these are all API improvements movements, but the 1.5 Pro API has been really good. I talked about Google's AI Studio before. It has the biggest context window. The model is really capable and you can upload videos to it straight up. It's the only model that can do that Now. It has 50% reduced prices. Both models have higher rate limits, two times faster output and lower latency. This is going to play really well with the voice mode which they need to upgrade further to catch up with OpenAI. But nevertheless these are some major upgrades. And if you check out all the benchmarks, although they didn't release new models, Gemini 1.5 Pro improved so much since May. This of ridiculous. Google is really catching up. They're not making big waves like OpenAI with every single release, but these products are getting super solid and fast. And next up, Talking about new LLMs, we got to talk about the brand new models from meta, namely llama 3.2. This is the first time Meta is open sourcing multimodal LLMs, meaning they can also process images, but it doesn't end there because with this family of models they open sourced a bunch more than just vision models. There's a bunch of them that are really small or meant for on device usage on phones. And because these are open source, people can build with them now. But some of them are really specialized towards specific use cases. I'm talking about these super small 1 billion, 3 billion parameter models that are specialized for one thing. And Meta just doesn't care. They're open sourcing it at all right here. So people can start building apps on top of this or simply include one step in their workflow, enabling workflows for developers that are just not feasible. If you're calling something like the big ChatGPT models that have no way of running locally. As you can see in these little gives, all of the generation happens on device because it's local. And also as mentioned, these include vision models. So if you're following the show, you would have noticed that over the past months actually a lot of Vision models came out. Some of the best ones came out of China, but now we have llama 3.5b beating some real contenders like GPT4 mini or Claude free Haiku. That's their smallest model. But again, I cannot overstate the fact that these models you can download and use, whereas GPT4 mini you can't. You need to pay for the API and go to the Internet. This you could just run locally. All you need to do is fill out this information and then you can download the models or they're also available over on Hugging Face. So look, I realize for most viewers of the show this is not something they'll be doing right away. But just know that Meta is not stopping on its conquest of leveling the playing field for all players, which in my opinion is really good because now we have an open source alternative that anyone can build on that is as capable as GPT4 or mini. Plus developers can easily build things like mobile writing assistance without needing to enter the walled garden of something like Apple Intelligence. But I just quickly want to touch on Meta's AI announcements. There have been some incredible things in there. I mean obviously the Orion glasses were mind blowing. I guess most of us are in tech, so if you haven't seen this, you need to check this out. This is probably the future computing platform, especially if you pair it with multimodal AI. But my point here is a different one. Some of these Meta AI announcements are exactly what I've been preaching for a while and it might be helpful to contextualize all the content in this weekly show because we didn't see anything revolutionary out of the Meta AI announcements. There's photo editing with prompts inside of their Meta AI app, there's a voice assistant, there's AI avatars, and they're adding voice dubbing to their shorts, meaning you can upload in one language and then it automatically translates into others. It's not perfect. Arguably there's even better tools out there right now if you were to do it manually. But the point is, all of these innovations that we look at week by week eventually will be gobbled up by these giants and implemented into their applications. And as they already have the distribution with Instagram, Facebook, WhatsApp and Messenger, they don't need to be. First they let everything shake out and then adopt the best features into the apps like they did with the Meta AI announcements. I mean, heck, just a week ago we looked at an open source model that does photo editing just like this new feature, and soon it's going to be available inside of Instagram. Okay, so I have to show you this app that I was really excited to find because we talked about NotebookLM before most other people actually talked about it. I'm kind of excited about the fact that we featured in an episode of news you can use with the brand new voice updates. And then about four to five days after upload, a lot of the Internet caught onto it. And I won't take the credit for that. Our community discovered it in there. There was a lot of conversations around NotebookLM for a few months already. Concretely, the educator sub community was using it all the time with their students or to research different projects. My point being that NotebookLM is still really, really good. And it doesn't seem to be stopping anytime soon because right now they shipped two new updates, one of which I was asking for since the beginning. It's the ability to include YouTube links into your sources. A quick recap of NotebookLM is that on the left side there's a bunch of sources you can add different articles, different PDFs, and now also YouTube videos that it then transcribes and takes into its knowledge base. Then at the bottom you have a chatbot which can suggest prompts, and those prompts run on the entire knowledge base of your notebook. So if you're researching a new topic, you just throw everything in there and then you talk to the entire knowledge base. And a standout feature that caught everyone off guard a few weeks ago was the audio overview, which generated the audio PODC that just sounded really, really good. Really human. So you could listen to audio summary as presented in a podcast conversation between two human sounding voices. Now, these audio overviews are accessible via public URL, so you can share them more easily. And you can add YouTube videos to your sources, which is massive because let's be real, YouTube is one of the best sources for information in our generation. And now you can throw to YouTube videos in there and generate these podcast summaries. And this week, here's another piece of AI news you can use that you could actually build build. Something interesting on. This feature caught so much popularity inside of NotebookLM that a brand new open source alternative to it popped up this week. It's called PDF to Audio and it does exactly that. You can upload PDFs and it turns it into audio. Now the interesting thing here is you can use a variety of models if your API key supports O preview. You could even use O1 Preview to process all the text from the PDFs and generate a summary that way. But at the end of the day, you can pick multiple speaker voices and create audio podcasts similar to the ones in Notebook Element. How similar? Well, we wondered too, and that's why we went in and tested this right away. So let me put my headphones on and have a listen. We did this based on Sam Altman's blog post. Here's the version. From PDF to Audio Today we dive.

Into a fascinating topic, the Intelligence Age. It's a term that sounds futuristic, doesn't it? The Intelligence Age is as thrilling as it sounds. Essentially, it's a term that signifies our era where artificial intelligence, often abbreviated as AI, is a major force shaping economies, societies and even personal lifestyles.

So you get the point. Pretty straightforward with standards, text to speech quality that we're used to. And here's the Notebook alum audio comparison. We're going to unpack Sam Altman's latest on AI and the future Intelligence Age.

And it's a fascinating read.

Bold predictions, some really nuanced arguments. We're talking AI tutors for your kids, AI managing your healthcare, even reshaping what work means. So there you go. I would say it's pretty clear that NotebookLM is better, but this thing is open source, meaning you could build this into your own applications or build something completely new from it. I don't know. I was really excited to find this because in my mind the point of this show is really getting you ahead of the curve so you can use some of these opportunities to your own advantage. Just like I did by discovering ChatGPT on the day of its release, all it takes is being slightly ahead of the herd and you can potentially build your entire life upon it. Alright, in the creative department this week there have been some interesting upgrades because first of all, we got the state of the art 3D model generator. And while this is not really my area of expertise, it's been really interesting to watch developments in this space. And this new model, triple 2.0 is supposed to be state of the art. So what me and the team went ahead and did is we tested Tripo 2.0 versus Tripo 1.0 versus stable diffusion 3D model generator. I have all the results of the testing right here. I'm seeing this for the first time myself and I'm just curious to see see how far 3D model generation with AI has come over the past year or two. So let's just start with triple 1.4 again. These are the same input images. Okay, this is very crude. Let's look at triple 2.0. Oh, wow. Massive difference in detail, right? Okay, what about stable fast 3D? Stable diffusions 3D generator. Wow. Hold up. This is way better. Am I right? Okay, what about typography? This is triple 1.4. That's okay. Tripo 2.0. That looks way more like wood. No doubt comparing it to stable diffusion. Well, okay, that's just weird. Maybe a food truck. Stable diffusion versus tripo 2.0. Yeah, these details are matched. Hold up. This generator is actually really good. I mean, sure, the text is not perfect, but this might be usable in the game. Whereas the stable diffusion results are. You know, they tried, I guess. Yeah. And 1.4 is not even comparable. This is not even close. I didn't expect this be to. To be this big of a jump. Okay, let's do something hard like the steampunk warrior triple 1.4 versus triple 2.0 right here. Wow. The level of detail. So look, it's certainly not perfect, but I think this really crosses a line where I would say this is good enough. If you put this into a living world, a game, animation, whatever, I wouldn't mind this. This little ghost should be pretty simple to generate. So let's have a look. Yeah, it's clearly. It's just better in every single case. And I have to say, just as a beginner looking at this, it's an impressive tool and you could try it out yourself for free on the Triple 3D AI links in the description, as per usual. All right, next up, I want to show you this AI image edit. You can Think of it as a canvas that uses AI to generate different aspects of this. And this tool is not new, but in the 5.0 version, they introduced this canvas where you can really create things from scratch, then match them with others. I don't know. I personally Learned Photoshop around 14 years ago when I was 6. I remember taking a Lynda.com course and then also infecting other classmates with my enthusiasm for the learning. I believe three other guys in my class took the same course. And then all of us were photoshopping things into each other. And something like this we could have only dreamed of back then. Look at that. You just generate a tree, you cut it out, you insert it into scene, it blends it all together. So this is the announcement though, but you have all the AI tools in one. You have layering. And again, this canvas interface is brand new in here. And it all started with this simple little mask that then AI turned into this with precise control. If I was a designer, I'd be probably learning this right now. I just have to note that you can get started for free. But this does cost money. But what a great interface. I didn't know of this tool before and I wanted to show it to you right here. Okay, next up, we have upscaling upgrades to Leonardo AI. You might know that they're a suite of tools. And now they also included an upscaler, just like many other apps. And I was curious because Leonardo AI is a pretty good deal. I think at $12 a month, you get a suite of tools that now also has an upscaler. When you compare it to some other tools, like in our case, we use Magnific. I still think it's state of the art, but it costs around $40 a month and we only use it for the upscaling. So I was curious if Leonardo AI could do the same things. So we went ahead and tested a bunch of images in this brand new upscaler they have. And as you can see here from the original image, this is the Leonardo result and this is the Magnific result. And I have to say right away, I just prefer Magnific. If you're fine with good results, then Leonardo is fine. But Magnific is just better, isn't it? But some of these other tools that have 16x upscaling, not just 2x, are just there's more detail. Now you could use something like Lil Nado to get to the 16x resolution, but you would need to run it and then run it again and run it Again. And as you might know, these AI upscalers introduce new details. That's how they work. They hallucinate up what should be in between the pixels with AI and that way they can really blow up the image to incredibly large proportions from something that originally should not have all that resolution. I mean, look at these books. This is the original image going to Leonardo over to Magnific. These are pretty comparable. A few more. A cat in a hat in Leonardo. Wow, look at this one. This is really good. And then Magnific again, slightly more detail. But I think you can see the pattern here. It's pretty good. And I think this can be said for most upscalers these days. Every single app seems to integrate a simple AI upscaler, just like Leonardo did here. And they're really good. They add a little bit of detail, you get a bigger resolution. But as you can see in this image, it can backfire sometimes and you have to play with the settings on something like Magnific. That will blow it up so much. Actually, on this elephant example, I might prefer Leonardo. Same goes for this portrait. What's the conclusion here? AI upscalers are amazing. If you or your marketing department are not using them in your workflows, you definitely should be. This is not just image sharpening that we had for decades now. This invents new pixels and it can be really useful, especially if you pair it with other AI generations that often have the downsides of being a smaller resolution in AI video lens. There's a super quick update for you. Cling, one of the top three video generators has a brand new model, Cling 1.5. It outputs at 1080p HD and there's a few changes, but honestly, all of these video updates lately feel marginal. There's an entire page that I'll leave below where they really nicely illustrated all the changes here. As you can see, it performs way better on this good old Will Smith eating noodles type of test. So that's great. But you don't get the motion brush and camera control in this new model. By the way, that's also a new thing. They now added the motion brush, making this the first one of the top three video generators to actually, actually add selective animation with the motion brush. This has been one of the best features of the original Runway. If you want to learn more about these, they put together this guide with a lot of detail. I'll also link this below. This thing is freely available, includes a bunch of prompts and examples and it's all in English. Any AI video enthusiasts. You'll love this. This should be really helpful. And they even have cats with hats. I don't know what's going on over in China, but they're getting something right. So I'm not sure who put this together and if they watch the AI Advantage, but if they do hello Clink Guide writer. Nice to meet you. And that's pretty much everything we got for today. Check out the AI Advantage newsletter in the description below to get a free template with a bunch of beginner prompts now, also including voice prompts. And as per usual, I'll see you in the next episode of AI News yous can use next Friday.