New GPT-4o Voices & More AI Use Cases

Ladies and gentlemen, this week in AI news you can use. We have voice filters and a bunch of interesting releases, including a new GPT4O use case as released by OpenAI. Give me like a nice evil maniacal laugh. An app that creates music for you one stem at a time, a bunch of updates and some interesting niche apps and wait lists. This is going to be a good one with a wide variety of tools. So let's dive right into this week's AI news you can use. All right, so first up, I want to address something from last week's video because I had a line in there where I said, hey, the voice assistant from OpenAI is coming next week. The new voice assistant that will come over the next week. But I was actually saying next week, so I just rushed it a little bit. Sorry for any confusion there. Although it might not be that inaccurate. If you're not aware, on Monday, June 10th, we're going to have the WWDC keynote from Apple and apparently they're going to announce a partnership with OpenAI. Here's some of the rumored announcements coming up if you're curious about that. But we'll covering this in a separate video once the keynote note happens. So that would be a fantastic week to release the voice assistant. But just to be clear, we don't know. What we do know is that they released a new use case demo here. And as so many applications that come out these days seem to be mimicking what OpenAI is doing with this, we absolutely need to look at this. Be the lion. I really want you to like embody what that might feel like. Who goes there? So what happened is they uploaded a singular video on their YouTube channel named Character Voices with GPT4.0 voice, basically showing how adaptable the voice inside of the new voice assistant feature, again, that is not out yet is now. This is definitely a response to the whole Scarlett Johansson drama, but we're not here to cover that. We're here to talk about these capabilities. Now, I encourage you to check out the whole thing. This is one of the most amazing demos. Now let's think about what the villain might be. I don't know what animal would work best, but let's start with like some kind of laughter. Give me like a nice evil maniacal laugh. I wonder why they kept it in their pocket for so long. Maybe a well thought out marketing stunt with the ScarJo thing. I don't know. See, so as you can see, the voice is super adaptable and we actually have some other releases linking to this. So we'll cover this all together. Okay, we have 11 labs coming out with a sound effects generator. Up until now, we only had Meta's Audiobox, which we covered a few months back when that came out. And that is really good at generating sound effects. Okay, so now we have 11 labs for sound effects, 11 laps for voices, and ChatGPT coming out with this model that kind of includes it. All right, you will be able to generate voices and sound effects with the OpenAI voices. It's all packaged into one. It's just not accessible today. But I want to share one more thing about the voice assistant with you. So it seems like they started rolling out new features related to the voice assistant. As Alex friend of the Channel, tweets here on X, some devices got a brand new menu feature inside of ChatGPT. Now it's rolling out gradually. I don't have this on my phone yet, but what does it do? Well, it's pretty simple. It gives your ChatGPT app the access to your microphone all across your phone, independently of what you're doing. So if you leave the app, ChatGPT is still going to be on recognizing your speech. Speech. And this is really what enables this voice assistant concept, right? You want to be using it with other applications, not just while you're in chatgpt. So again, another hint at the voice assistant coming soon. And I just want to Note here that OpenAI is not the only company attempting this AI assistant idea. Many others are going after it. One great example of this would be Nvidia with something they called Project G Assist. And this is a personal assistant targeted at gamers. You can ask it questions you often find yourself looking up online. Like, what's the best early game weapon? The best early game weapon is the Spear. Unfortunately, this is also not accessible today. I just wanted to briefly show you that this is definitely happening. And basically, it's a little chat interface that can assist you in games and you can ask things like, do you see any problem with my display settings? And then it looks at all your computer settings or your graphic card settings and gives you tips accordingly, even charting something like latency for you as it changes the settings. My point, AI assistants are absolutely happening. And they're about to happen now. Not in 2025, but now let's move on to something you can actually use, which is 11 lab sound effects. And this is sort of a subset of the voice assistant functionality, because as you just saw in the new demo, the new voice assistant can change the voice depending on what you ask it for. And that means it can essentially generate various sounds from scratch. They're just showing it off on human voices. But a subset of that is creating sound effects. And 11 labs released their new sound effects generator. Now, this is not a first in the space. We already have Meta's audio box that is really good at generating sound effects. And look, today if you need sound effects, you're probably, probably a video creator that wants to include them with the edits. But very soon this is going to be an essential modality when you combine it with something like OpenAI Sora or similar. So let's give this a spin. Let's create a whoosh to bas it. This is a sound effect I used to use all the time in my edits back in the day. It should sound something like. So essentially it's this airwash with a base hit in the end. Let's see how well 11 Labs did with this. Okay, that one's okay. Maybe add one more word, see how this goes. You know what I gotta say, it doesn't really do what I want it to do here. And look, we actually went ahead and did some further testing on this run in dozens of different sound effects in here. And the overall conclusion is this. If you keep the prompt short, it actually generates usable sound effect. But as soon as you start getting a little more specific and you venture beyond 4, 5, 1 words in your description, the results are often not as good as you would want them to be. And even with something simple like a car crashing into a tree, there's just always the same screeching tire and then not even a proper crash. There's one, but it gets cut off in the end. So look, I'm a fan of 11 Labs products, but this one is not too good. I would still recommend the Audiobox this demos where you can create sound effects just like this. The only problem is it's not always accessible. It's a research preview, so at times it's just not available. Like right now, they ask you how you are. You just have to say that you're fine when you're not really fine. Okay, so as you might already know, having skills to go along with some of these tools is invaluable. And I wanted to share this fantastic resource with you today that helps you do exactly that. I'm talking about Brilliant because in this fast paced world we all live in, continuous learning is key. And finding the right resources to learn from makes all the difference. And the reason I think Brilliant is great is because they're an interactive learning platform. So it's not just prerecorded content, but there's quizzes and little exercises in between. They offer thousands of lessons in math, data science, programming and of course, artificial intelligence. But as I mentioned, what really makes them stand out is their hands on approach to learning. Instead of monotonous lectures, you really get to engage with the material and build your problem solving skills in the process. And I actually have a concrete recommendation for you. I've been more and more a fan of learning the basics of Python. Even if you're not planning to program or do anything with it. It teaches you how logic works and that helps with prompt engineering. It helps with a lot of these AI tools and they have a great course called Practice Applied Python. A lot of times getting into coding can be a big roadblock for people and I think this is a great way to overcome that. Why is it more important? Well, some of the more advanced techniques and workflows do require at least some knowledge of Python. As I mentioned, just understanding the basics like variables and data structures goes a long way because it transitions into no codes, tools, it transitions into prompt engineering and more. And this course is fantastic for getting you up to speed with the skills that you need to thrive in this digital era. So go check out Brilliant to quickly learn about Python, LLMs and all kinds of other math and science type topics. To try Brilliant for free for a full 30 days, head on over to brilliant.org daidvantage or click the link in the description. Plus, if you decide to stick with it, you'll get 20% of an annual subscription. A big thank you to Brilliant for sponsoring this video. And now back to more AI news you can use. Okay, next up, we have something that has gathered quite a bit of attention across the Internet. This one is called Tombcrafter. And what it does is that if you give it two images of a cartoon, it will generate all the frames in between. So here you have one frame with the candle not burning here of the second one. And look at this flame coming up. This is not just decent, this is very, very usable. And that's why this got a lot of attention. This is by far the best tool to do this that has come out yet, and you can use it today. So we tried it out on a few examples of our own to see if this actually works. Here's image one and here is image two. Here's what Tooncrafter produced. Okay, that's not terrible. This is one of the trickiest things you can throw at it. Let's maybe look at something easier yet crucial to a story. This image of a character. Image one, image two, where she's smiling. And here's what Tombcrafter created. Look at that, the eye movement, the smile. This one is really good. One more example. Image one, and then image two, where the hair is in a different position. And here's the toecrafter result. It works as expected. You even have the leaves flying through the air. So could the resolution be better? Could we have more frames? Sure. But this is the very first animation tool where I'm like, wow, you can actually tell full fledged stories with this. If you take your time of creating the first frame and the second frame and then you can combine it with tools like the ones we looked at. You can create your music with AI, you can create the sound effects with AI, you can create the voices of the characters with AI. This is one piece to the puzzle that has not been solved yet. And it looks like Toolcrafter is really on that edge of just barely being good enough to be used in the real world. Fantastic stuff. So our next tool here is actually quite impressive. What it does is create songs one stem at a time, meaning it separates out the different instruments for you. And then you can extend them and alter them based on your text prompt. Let's just have a practical look at what the interface looks like here. So this is Frederic AI, and basically you get this chat interface on the right side and then it will generate the stems for you. It's a very, very beginner friendly interface. So if you ever played with GarageBand and had some fun in there, chances are you would enjoy this too. By the way, GarageBand used to be the very first creative software I ever tried on a computer. And it got me hooked on the idea of, wow, I can just create stuff and learn this software and become better at creating stuff. I thought that was absolutely incredible. And that really was the first spark that I'm learning. Photoshop, Premiere Pro, After Effects, Python, the basics of Unity, and so much more. Often a little inspiration goes a long way. And this might just do it. Because look, if you want to do a rock track, I can just add this to the project, like so, and I can go to various stamps and just click. And then I have these loops that I can easily extend. And look, I never learned too much about GarageBand, but I do know that if I shift these up, the pitch, pitch should change, right? Then I can add effects, like so. And then a new project, just using this rock preset, by the way, you get 10 snippets like this a month for free. So you get to play around with it for a few minutes. What I could do then is simply loop it like so. And let's see what we got from this preset. What is this? And yeah, the free version is limited to these preset snippets. So if you want to generate your own, you do need to subscribe for currently $10 a month. But I feel like this might be a great way to get started with music production or see how these AI tools will integrate and into music production, which might be very relevant if you're really interested in something like OpenAI Sora because essential piece to creating with that will be the underlying music and the sound effects. So the tools that we just looked at here because video is really only 50% of the story. So if you better be creating Sora, better get a feeling for the audio part also. This might just be a good way to do that. Oops, had to run to the hairdresser. Bit of a changed look here. So let's get back to the next piece of news you can use, which is this new LLM leaderboard by Scale AI. And this one is different from a lot of the ones that have come out recently. The problem of ranking different LLMs has been something many people try to tackle. There's benchmarks like mmlu, there's Chatbot arena where users rate the output and then an ELO rating system ranks them. But there has been a discussion around how reliable these rankings are. Some of them can be manipulated by including some of the questions or the user preferences into the training data. So Scale AI is attempting a new approach. They have new methods to rank these where they don't discover close what exactly is being asked for. And it results in these various leaderboards for coding, math, instruction following and language understanding. You can look into all the details of what they disclose in here, but most importantly, you can check out their leaderboards that are supposed to be independent. Again, this just popped up, but I gotta say, these rankings look very reasonable to me. The one that I personally care about the most is instruction following. And yeah, for that GPT4O is fantastic. Also remember when Llama Free came out, a lot of people were excited about the fact how strictly follows the prompts that you give it. This is something that doesn't work as well with something like Gemini. And look, I gotta say, when it comes to instruction following, my personal anecdotal experience is lined up with what this leaderboard says here. So I'll bookmark this review. It over time. But as of now, this is a great leaderboard to look at as opposed to some of the benchmarks as they're published by the model makers. In my opinion, those hold less weight than leaderboards like this. Okay, our next piece of news you can use is actually a quick update to Udio. The change is very simple to explain. Up until now you had a 30 second limit when generating a song. Now they upped it to two minutes. As we talked about many times before, YouTube is absolutely fantastic for generating songs. And now you can actually create proper songs rather than 30 seconds and then adding 30 seconds at a time. And along with that, they included wave downloads. If you're not familiar, this is a high quality audio format which makes this usable for more commercial purposes, as opposed to MP3s that are so heavily compressed that if you try to do any editing on top of them, they just don't leave you any room to tweak the results. Whereas WAV files are way less compressed, so you can do those edits after the fact. So both Udio and Suno are becoming more mature and more usable for real world purposes by the week. And one more thing, and this is something super unique in the AI space. They actually added a feature where you can extend your very own audio. So up until now you could prompt and generate songs from scratch with their tool, but now you can upload your own song, your own voice, your own sound effects, whatever it might be, and extend it with with their generative tool. Very interesting and unique. You do it through this upload button here on top. And then you can take this clip. And add more duration to it, which would sound like this. And to round things out, I have two more things here. One of them is Perplexity Pages, which Perplexity, you might be familiar with it. It's the AI powered search engine that includes detailed links and references to what it presents to you. And they're coming out with a new product which is Perplexity Pages. And this is essentially Article Writer that is then built into search. So, you know, as of now, a lot of SEO experts specialize in using AI to flood the Internet with various articles. And Perplexity being the search engine, just figured, hey, why don't we just generate those articles ourselves and then we have more control rather than leaving it up to random people. And this is just a very interesting development. So what they're doing here is creating something in between WordPress and Medium where you can create your own articles that are all AI powered and you can publish them to the open Internet. But then they live under perplexity, just like medium. Articles live under medium, but they're still freely accessible. But as opposed to medium, you just know there was more AI involved in the creation. I personally think this is preferable to random websites that pretend like they're not AI written and people project false expertise on articles. And you look up something and at the end of the day you're just looking at a rewritten GPT4 output. I appreciate the honesty behind this idea and I can see bigger players adopt this. We could expect something like this out of Google at a certain point too, because right now a lot of Google search is just AI generated articles. So maybe the solution to that is providing their own set of AI generated content articles. But then you have the bias of the corporation. I don't know. This is a tricky topic, but nevertheless a very interesting development of the Internet as we know it. That perplexity is going after here. And eventually I'm sure we'll see something like this also with video once those tools get good enough. Okay, and to round out today's episode, I want to point you towards something that is not available today, but you can sign up for the waitlist and you should absolutely do it. This is one of the most interesting and inspiring ideas I've come across in AI space recently. It's Showrunner, and basically they created an AI generated South park episode last year with their entire show, Generation Engine. There's a lot of things here, okay, so you could talk about this for 10 minutes straight because there's so much interesting stuff here. But basically this is a prompt to show engine and they're trying to do multiple things with it. Again, this is not out yet, but the idea is you're going to write a prompt and generate a brand new show with it, or you're going to extend one of your favorite shows with a new episode. Or they have an aspect to this where they have a version of San Francisco with a lot of AI agents just living their life. And then from that they generate a show. So you can always follow what the agents are up to, how they're living their life. They're organizing birthday parties, visiting each other. Extremely interesting stuff. Building up on a lot of research that has been published over the course of the last year related to autonomous agents just living in these little cities and then narratives emerging from that. So they're trying to turn this into a media format that you can not just watch, but that you can manipulate yourself with prompts. Super interesting stuff. I'm on the wait list as soon as this becomes available, I'll be covering this more. If you want more details, I recommend this Fred on X with various videos that show off these different ideas more in depth, like the simulation and also this idea behind creating your own TV shows. Very interesting stuff. A lot of content here. And then I suppose one day something like this could be built or a couple acquired by Netflix and you'll just have AI features extending your favorite show with separate episodes and stories that you personally care about and that matter to you. And with that, that has been everything I have for you today. I'm looking forward to the Apple announcements during wwdc. And don't forget to subscribe to the channel for an episode like this every single Friday. All right, see you around.