Powerful o3 Prompts, iPhone Hack & More AI Use Cases

It's so good to be back home in the studio reviewing all the various use cases and news releases in Generative AI that have popped up on our radar over the past week. The theme this week is a clear one. It's less about brand new things that were launched, but more about squeezing the tools that we already have that are super capable like 03 or chat image generation or the speech to text models to get more out of them for yourself. But hey, it's Genai. Of course, there's some new stuff, too. We'll cover all of that in this week's episode of AI News. You can use the show that rounds up all the releases and new use cases for you. And then I have the pleasure of showing you what happened in Genaii this week. Okay, so first and

foremost, an update to something you've already seen surely and that's the chat GPT image generation. It's now available for the API. But even if you don't plan to build your own app on top of this, we really need to talk about this because the API release made this thing so much more usable, you need to be aware of this alternative workflow that doesn't use chatbt. Let me show you what I mean. So this is what I'm talking about. The image generation that you know and probably love from chat GPT is now available for the API which means you can call it programmatically. Now this is not subscription based like chat GPT but usage based. So you pay per picture some of them costing 25 cents per image. So it's not cheap but powerful and there's actually a fantastic way to use this even if you're a non-dev. Okay. So what you want to do is you want to head on over to platform. opai. com/playground/im images link in the description below. And if you go to that link, you're going to be faced with this brand new interface which is similar to Sora if you've seen that one before, but even better. And it allows you to do things like generate 10 images in just one prompt. Okay, so here's a simple but very powerful prompt that you could use in here where it would make a lot of sense. Generate realistic and clean corporate headshot images based on specific userdefined characteristics. These headshots must reflect professionalism and clarity. And then here I have a few additional guidelines to make these really consistent. Now we can put this in. And then what I like to do is really up this number of images that it creates. Of course, that's going to cost more, but I want portraits in a high quality here. And then what I'm going to do is give it multiple images of myself. Oh, that's not me. But these three should do it for me. Good lighting. Very, very important in the images you actually give it. And then I'll just hit generate, and it will start generating 10 images at once instead of doing them one by one. And I can already move on to the next prompt and do a lot of things in parallel. Here, there's also a lot of fantastic presets. So, this one that you see right here is the magazine cover preset where I added one image of myself. And then as you can see it generates 10 alternatives right away and I could pick my favorite one. Okay. So using it here for the playground is one option that you need to be aware of. But then there's also a second option which has been around for a while. I also want to show you that it's not as speedy but you don't pay per usage and that's if you go through the chat sheep interface and when you head on over to Sora if you're on the plus or the pro plan you can also use a similar interface where you're limited to four variations per generation. So as you can see I ran the same prompt in here. have some similarish looking headsh shot, but this generation was way slower. Plus, Sora has been down quite a few times over the past week. So, it's slower to iterate. It's still better as just using Chat GPT. But the main part and the main takeaway here is if you go through this interface, you're not going to be paying per usage. This is included with the Chat GPT subscription. If you go here, it's going to be way faster and it has extra features like pulling in context from some of these images like so. And then, I mean, look at that. It's generating 20 images at a time and you can just keep on adding. And to round the segment out, I want to point out that the fact that this is available for the API is going to make it available for various apps that you might already be using. One of them that OpenAI shared was the Figma integration right away. So a lot of people start their workflow in Figma. Now you can generate images and edit images with OpenAI image in there. And the same thing can be expected amongst many other apps including some of the video generators now. So for example, a few weeks back we talked about Higsfield which is best at creating humans and human anatomy. Well, now you could start the journey with OpenAI image gen. Edit the image to your liking and then use Higfields to create the video for that. I guess the overarching point here is that they released a Photoshop API that anybody can use sort of. And I can't wait to see all the applications that get built with this because I feel like at this point we don't even exactly know what the killer use cases for this are yet. We'll keep an eye out for that though. Okay

let's move on. Okay, now we're going to talk about my favorite tool that has been released recently and that is OpenAI's O3. Particularly, I want to show you two prompts that you might find useful. I really like these two. But before we do that, I want to walk you through some chatb upgrades that happened over the last week cuz I just know that happens to be the daily driver for most people watching this content in terms of AI tools. So, first things first, they changed the limits on a bunch of things. Okay, deep research limits are now higher than ever. You can see on the pro account I now have 250 per week. They also update on the plus tier and now they introduced a thing called light deep research, also available to free users. So, if you're not a paying users, you can now also run deep researches. They will just use the O4 mini model instead of the 03 model. And if you use up the deep researches in your account, it will switch on over to

this light version. Okay. Next up is a change that they didn't announce, but I think this is important. If you head on over to the settings and you look at personalization, you probably know about the memory feature at this point. And this used to be split into two options here. One of them was, hey, do you want to use memories aka chatb automatically gathering context for you and then using it in all future chats? And then a new one got added recently. Really important to be aware of that. And that's the fact that it looks into all your previous chats and also uses that as context. They used to be separate, but they silently merge them into one option. So it's just, hey, do you want to use memories plus all the previous chats or not? That's the choice you get now. They're not split anymore. I know a lot of people actually had memories on, but the chat history off, well, that's not possible anymore. It's merged and you should probably know about that. And

then obviously we had the release of O3, which was absolutely massive. And with more and more benchmarks coming out, it just kind of proves what everybody has been feeling in terms of vibes. And that's the fact that this model is absolutely incredible. On ARC AGI, 03 here has become the industryleading AI reasoning system by a large margin. Double the score at 120 of the cost compared to next leading chain of thought systems. If you're not familiar, this is one of the hardest benchmarks out there. And 03 absolutely smashed that. Plus, also in LM Marina, it leads in many categories. Now, if you're curious about all the details, you can head on over to this link, which will also be in the description, and you can pick the category. So, I don't know if you want to look at German performance, well, 03 is not even close to the top. But maybe you care about coding, and in that case, 03 leads the pack here. It's just an incredible model and has become my new go-to. So, at this point, I want to show you two prompts that I found super useful over the past few days, and both of them happen to be of Twitter. The first one is sort of a brainstorming prompt, but a very interesting one. Look at this. You have consumed more information than anyone in the history of the world and you've demonstrated an extraordinary ability to make connections among them. And then the prompt proceeds to ask about most important non-conensus or even not yet hypothesized things. So basically picks out really unexpected ideas. Shout out to Py McCornic for this one. And if you run it, you will most likely want to add a topic underneath of an extra building block. Like so here we're doing generative AI for product marketing. And if you just look at these results at all free surfaces, the thing with this model is it can really surprise you. so smart that it comes up with these conclusions sometimes that just leave me staring at the screen. It's really good. I mean, look at this first one here. So, it says, "Most teams still treat Genai as a content engine. " That's fair, right? The real unlock is feedback, letting models both write and instantly metacritique variations. What a great insight. This instantly sparks some ideas for some workflows that you could develop and improve with AI. How about using O3 to critique your content rather than just generating it? And there's so much more in here. I mean, if you want, you can pause and read through some of these ideas here. I particularly like this one, atomic content distillation. Just feel free to try this prompt for any topic that you're interested in, and chances are you might just be surprised. And another one is this trend analyst over here from Amin over on X where he uses it to spot trends before they hit. He tells it to analyze Reddit, Twitter, and Product Hunt. And then basically he asks for different tool system or services he could build to solve those. Now, you could easily take this and customize it to what you do. So in this case, I ask it to suggest a YouTube long form or Instagram real idea covering this trend that it finds on these various platforms. So if you look at the thinking, it goes ahead, it looks at those platforms. And I love the fact that it does that and then finds new search terms that might not be obvious that surfaced within its initial searches and it looks at those, looks at more, brings it all together for you. And then here are different pain points marketers are facing with the underlying emotion. And then it even comes up with some content ideas right away. I like this one. and free signs. Your AI content is killing your brand voice. It already drafts for you, which is nice. But you could do that on top of every one of these ideas. And I just found this approach to giving it specific websites and then telling it what you expect to be super powerful as you give it enough room to explore the internet yet you tell it where you want it to go and it uses intelligence to deliver that for you. So there you go, two super powerful prompts that you can customize to your own liking. And yeah, straight up this and deep research are just the most powerful things in the AI space right now. So go have some fun

with that. Okay, so the next use case is something I've been really excited to try myself, and it's implementing AI voice recognition as an option on my phone instead of the built-in dictation function, which just makes so many mistakes. We'll do a quick comparison once this is set up, but essentially this tutorial comes from a team member at 11 Labs shared this on Twitter. And I figured doing a quick segment where I show you this tutorial. Obviously, this only works if you're on an iPhone, and it works best if you have one of the newer ones with the action button. I suppose alternatively you could also set this up with a double tap on the back. I think that's another way to trigger shortcuts. Anyway, let's get into this and let's implement the 11 laps speech recognition. So, moving forward, I can use my voice to dictate anything in nearperfect quality as AI delivers it versus the built-in tool. Okay, so I'll link this below, but first step, you just need to get the shortcut that he linked. That could be any simpler. You simply click this, add the shortcut to your phone, and then when you double click it, it will open up this editor. And then we just need to link a 11 Labs API key in this step and a OpenAI step. So when I click on show more and head on over to my 11labs account, I can simply copy the API key into here. And then I'll do the same thing for OpenAI. Get my API key. But here I only replace the word OpenAI API key. And it's important to maintain the space between the two words here as he points out in his little tutorial here. And then that's it. The shortcut is set up. All I need to do is head on over to my phone. Make sure the action button here on the side activates the shortcut. So, in my settings, I'll just look for the action button. Right now, I have start voice conversation. I actually stopped using this recently. I love to prompt with voice, but the voice assistant I don't use on a regular basis anymore. The dictation feature I use all the time, though, so can't wait to actually switch this up. And now you can see here under my shortcuts, the 11 laps transcription has been set up. So, how about that? Let's try it out. Press the action button. Okay, there's some permissions I needed to give it. Hopefully, this is just the first time while using it. Okay, here I am testing my brand new shortcut. The quality of this should be way higher in theory. I'll use this everywhere if it works. And I can paste this in. Wa! Look at that. That's actually perfect punctuation and everything. So, just for comparison sake, let me try out the built-in dictation here. The quality of this should be much higher in theory. If it works, I'll use it everywhere. And yeah, it's minor, but look at that. The punctuation on the first one is better. It's clearly a separate sentence. And okay, comma is also superior. So there you go. That's a practical little AI hack that if you're iPhone user allows you to use these super accurate transcription models in your everyday life. Okay, that was a great little segment. Let's see what do we have next. Next up, I want to point

your attention towards a update from Midjourney, the Midjourney 7 release that was mostly underwhelming. They now have a new UI that now gives you more options in terms of editing or really dialing in what you want. It integrates the previous paint tools, but now also has layers. And as you can see from our initial testing, you can bring in various images, combine them in a way that you like and then keep working on top of it. If you have a specific vision, this new web interface that is now rolled out to all subscribers, not just the yearly ones. That's how it started. It can be a great addition and definitely gives you more control than what most people are used to from these generative AI tools. Although I do have to say Photoshop is obviously layer based and they have their generative AI features since what is it one and a half years now. Me and the team use that all the time by the way, pretty much every single week on the thumbnail designs and the video graphics that you can see in these videos. But if you quickly want to get things done within my journey, well now you can. Okay, so next up we have a

release from Gen Spark. We covered this company a few weeks ago. I actually made a mistake. I said it's a Chinese company. It's actually a US company and they released a super agent which is sort of a version of operator and then manus that we saw which they released with some free credits and now they added a brand new feature which is specifically focused on creating slides aka powerpoint presentations. Now spoiler I don't think this works as the app that most people agree is the best at this which would be Gamma and it works so much better than all the co-pilot PowerPoint presentations but this is a new approach which is a bit more agentic in a chatbot style interface. So let's give this a quick shot. I'm just going to give it some bullet points from the notes on this application itself and let's see what this comes up with. After answering some follow-up questions, it starts thinking and working through this as you would expect any agentic tool to do. Let's give this a minute and I'll see you once it's done working. Okay, after about 6 minutes, I got four different slides. I'm just going to stop here cuz I feel like this would go on forever. By the way, I'm on the free plan here and this ate up all my credit. So, I guess you could do a few slides like this for free if you want to try it out. But look at this. I mean, this is sort of interesting. Even though it says slides, this is not really a PowerPoint presentation at all. It's more like a infographic package that's a landing page, which is not a bad thing. I mean, look at all of this interactive content here. There's like audio players built in there that don't work. But essentially, this thing just vibe coded multiple landing pages visualizing all the info I gave it. And I have to admit, I kind of rushed answering these answers just because I'm doing a live demo here and I kind of misunderstood the purpose here, but it doesn't really matter. I just wonder how you would really use this. I guess you could use it to enhance your presentations and you could customize every single element in here as this is all HTML code with some JavaScript in between for the interactive elements. Anyway, not sure how practical this is, but certainly kind of a new agentic workflow that caught my attention and I just wanted to show you. Talking about new agentic

workflows, there's actually one more that I want to feature this week. This one you cannot try right now, but that's okay. Nevertheless, I want to spotlight the new feature from Dcript here, which is essentially creating a gentic video editor. No worries, video editing team. This thing is a long way from actually replacing skilled human editors, but this is really the first big step we're seeing in this direction. And you can fill out a form to apply to test this early. But essentially, if you're familiar with Dcript, they built a suite of tools that is for people who don't want to use complex editing software or doing more editing of just talking head videos like podcasts by not even dealing with a timeline, but essentially just a text editor. and they added a bunch of AI features over the past years, but now with this new release, they're looking at removing the AI features and letting a agent control those features in the background and you're just talking to the agent and the entire editing process just happens. You tell it to make the video more concise. Well, it uses all the underlying tools and prompts to do that. And it really is quite interesting. And in this announcement video, they showed multiple demos of this. I think mostly this makes sense for content that is what either educational or a podcast, things that are not that complex to edit. And yeah, future video editing, at least some of the very simple edits, might soon look more like a conversation where you just say something like, "Can you edit this down? " And then it just does it just like you would with a freelancer, I suppose. And again, it's sort of the same idea as Genpark over here and all these other agentic products where it's not you working with AI tools, but you working with an agent that is working with AI tools. And video editing and PowerPoint slide creation are just two new categories that popped up on my radar for the first time. So, I wanted to show you. Okay, that's what we got for this week. I hope there was something that resonated with you that you're going to be able to put to work yourself.