ChatGPT Takes a HUGE Step Forward & More AI Use Cases

Welcome to yet another week in AI and this one was particularly interesting and a lot of things were released that will have a large impact in the future and I think you should know about it now. I'm talking about native app integrations in ChatGPT in a way that we haven't seen before. OpenAI releasing an agent builder and updates to how SORA handles copyright, and an update to Sora and so much more. In this week's episode of AI News, you can use the show that pulls pulls together all the AI releases of this week and we filter by the ones that are available and useful right now. And the lead story also, clearly this week is everything that OpenAI released during their dev day this week. And I think two things really stand out here and I want to dedicate separate segments to them. First of all, apps inside of ChatGPT and the SDK that comes along with it. And secondly, the OpenAI agent builder. Let's start with the ChatGPT piece, as this is a familiar environment for most of the viewers and frankly most people in the western world. I mean, it has over 800 million active users. And rather than me telling you about this, let me show you because they've released apps in various forms over the past few years, namely GPTs and the plugin store. And then they also had connectors. But this is different from all that it's available now in the US and they noted later it's coming to the rest of the world. But right now, me being on this vpn, I can head on over to my settings, go to apps and connectors that up here you can see the connected apps and this is what's new. These apps, these connectors were there before, but what really changed now is these applications. And now you might be asking like, hey Igor, didn't we already see this? Don't we already have connectors? Didn't they try plugins before? And then the GPT Store kind of did the same thing with actions and you would be right. It looks very similar on the surface, but a few details are very different and I think the details really, really matter here. So I think the closest comparison we can draw here are the plugins that launched in early 2023, I believe, and I believe they were phased out like half a year later. It was some of the similar applications that now under this app's title, Booking, Expedia, Canva, all the usual suspects. But the plugins completely flopped. As I said, they literally removed it from ChatGPT. That doesn't happen often and I Think the reason for that was that the functionality was not that deep. Often the plugins used to just display results from within the app in ChatGPT and then you still had to head out. Plus, and I think this is the detail that really, really matters, you had to enable the plugin before actually using it. Whereas these apps, if you enable them, ChatGPT makes its own decision on when they should be activated. So for example, let's go through the process of connecting Canva. Right here. I happen to have a paid Canva account. Me activating this for the first time. I have to log in to authorize the app. That makes perfect sense. And now that it's authorized, you'll see it appear under my active apps. And now if I go into a new chat and say something like create a presentation with canva for my AI news, you can use show that compares Google versus ChatGPT. And here I explicitly mentioned Canva. So it should use that app to actually invoke it. And as you can see, immediately it accesses my canva account. And here's the interesting part, it starts doing the work right here in ChatGPT. But these are the two biggest changes here and I think they matter. First of all, I don't have to enable the plugin. Once the app is on, it just works. And secondly, it does the work inside of ChatGPT and I don't have to leave. It actually created multiple presentation drafts for me. Let's just say I like this very first one here. I can open it up and the whole presentation is right here. Again, I'm still inside of ChatGPT now. I could follow up and edit it and say something like, the first one is great, change title to the information rivalry and then it should adjust this one accordingly. I didn't even read the whole thing. It asked me, would you like me to go ahead and update the title slide? So let's just say make the edit. Okay, so it made the edit. Would you like me to commit the changes to the presentation? Of course. Okay, so let's have a look inside of Canva now. Yeah, and there it is. So it can edit that way. So this is a new way to work and it's probably not going to be perfect as of now, but intuitively I really think and feel that this just makes sense. Working inside of ChatGPT rather than it being this bridge to all these other apps is what you want to do. Now I really want to show you one more example, and that's a Spotify related one because that works in a different way, and I genuinely love it. But before I do that, I want to give you a few more details here because these apps are limited as of now. As you can see, there's five of them right now in their blog post and they have a bunch of partners that are coming soon. But the vision here is what matters, because along with these partnered apps, they also released the apps SDK where developers can actually build their own apps and integrate them into ChatGPT. And then the vision is to eventually launch something like an app app store in ChatGPT, where you just activate all these various apps and all of a sudden we move away from this model where on your phone you need to select applications to do things. ChatGPT will hopefully do that for you. But I really want to highlight this last paragraph here that says what's next for these apps in ChatGPT? So as I said, developers will be able to open submissions to publish their apps inside of ChatGPT. And then if they meet OpenAI standards, they will be listed and they will also share details on monetization soon with a new agentic commerce protocol. One side notice that the agents SDK actually uses the open MCP standard. So a lot of the existing MCP servers that exist will just plug into this. So essentially people will be able to build their own apps, but for now, we're limited with these. And now I want to show you the second example, which is Spotify, because you could say something simple like create a workplay list with Studio Ghibli music, hit enter, and now it should engage the Spotify app without me telling it, because, hey, that's where I do my playlists, right? And in no time at all, it loads various tracks and Spotify creates a custom playlist in in your account. I did connect the app in advance, but you have to do it once and voila, if I click the playlist, it opens it up and there we are. A wonderful custom playlist that I could now keep editing inside of ChatGPT. So if I follow up with something like mix in Lord of the rings Film Music, ChatGPT is actively editing the playlist for me that will then show up across all my devices without me ever leaving ChatGPT. I could just open the playlist and that's it right there. Once stuff integrates like this, it sort of just works. But the real power of this will become evident once you get multiple applications that use regularly involved inside of ChatGPT. And then time will show if people actually use it. Okay, let's See what's next. Okay, so real quick, I'm going to show you a platform that we've always used to teach and execute automations internally at the AI advantage. And that is make, the platform that lets you automate just about anything under the sun. And also a big thank you to MAKE for sponsoring this video and making our productions possible. They really are the perfect partner for us to work with because we actually use MAKE all the time. If you're not familiar, it's a visual orchestration platform that lets both individuals and teams accelerate and scale with AI. With a combination of MAKE and AI tools, you can orchestrate things visually. You can build and scale up agentic automations, monitor everything in real time, quickly gather insights and implement the right changes. But ever since we started using make, the platform has developed a lot because they've added a ton of useful features and updates. Two of my personal favorites they added recently are the AI Agents and the Grid View. So first up, AI agents, if you're watching this channel, you might be familiar with that concept. And in the context of make, they're basically a dynamic way to adjust a workflow autonomously without relying on a specific rule that routes the automation the same way every single time. The decision making becomes more fluid and intelligent through the use of AI tools by plugging LLMs into the middle. And because make is a visual first AI assisted local platform with over 2,500 pre built applications, you can automate things away quickly without getting caught up in custom development work. So these AI agents in MAKE pair nicely with their new Grid View tool. This is basically a new way to view automations and also edit them. Grid View shows you a real time interactive overview of your entire AI and automation landscape and you can easily monitor in real time. It also makes it way easier to find bottlenecks in your automations so you can remove them or troubleshoot issues all from one place, which allows you to scale with control using MakerID, MCPS and analytics. So if you want to start creating intelligent agentic automations, sign up for MAKE Today with the link at the top of the video's description. New users will receive one month for free on the Pro plan, and that plan includes 10,000 operations per month, amongst other perks. All right, now let's have a look at the next piece of AI news that you can use. And this next story really shook up the entire startup space in AI. Because they shipped their very own no code interface that directly competes with, well, I would say most AI startups out there these days and look, overall, we're not going to be doing a deep dive on this tool. I just want to show you that if you go to platform.OpenAI.com agent builder, well, they now too have a automation builder. But I think again, just like with the apps, this is a case of, hey, they ship their own product, but they're nowhere close to the depth of some of the other tools that have been at this for years. I think this is serious competition. But other tools, even like make, who sponsored today's video, it's just not even close in terms of all the integrations and documentation and options you have in some of these competing tools. Whereas this Agent Builder has a lot of things, you need a lot of the essentials. But there's some major, major limitations. One of them is that if you want all these integrations, well, there needs to be an MCP for them. So you can add these MCP nodes here in the middle, and then you could connect up either one of the MCPs here, or you could do a custom one right here. But the beauty of using something like make or one of the other products is that the connections are all just there. You just log in via OAuth and you don't need to be using the MCP server. Everything that you see here and all of these nodes, they're really robust. But I do have to point out the limitations here, where another one is that you're strictly limited to OpenAI models, obviously, whereas with these other tools you can use any models. For example, I myself love to use Claude for writing. Not possible here, which might be okay, it's just something to be aware of. But yeah, other than that, classic automation tool, which could also be dubbed the Agent Builder, there's a trigger here, there's actions here, conditions in the middle, and when you add these agentic notes, it basically lets you add instructions. You can use one of the OpenAI models. So, as we'll talk about in a second, we could, for example, use GPT5Pro here. Now you could add tools to the agent via the MCP servers and then there's more customization. But that is kind of the gist of it. If you want to learn more about this, the documentation is really the place to do that. And they also shipped the chat interface along with that. So now it's really easy to create a chat interface to engage these workflows just like the apps. I feel like this is the same story. This release kind of echoes the same theme. As in like, this is an earthquake that happened, but it Might take weeks or even months to see if this produces a tsunami that hits the shores. As of right now, it's a nice development and we should definitely know about it and start using this stuff. But it I don't think this is a complete game changer as of today. But again these are dev focused releases and people have to start building on top of this for it to really matter. Okay. And there's one more big OpenAI related story that we have to cover and that are the changes to Sora. I'll keep this rather brief. There's really three bullet points I want to hit here. The first one is that they shipped an API. So at $0.10 per second you can now generate Sora 2 videos through the API. I did the quick math. A future length movie would therefore cost around $700. Fair enough. Just to be clear, that's what it would cost to generate it. You're not going to get a movie of just one prompt, but if you just accept the first generation, that's what it would cost. Secondly, they updated the moderation policies so you might have noticed that on Sora 2 release you could just use everything. All the different IPs were in there. You could essentially generate most shows and historic characters, celebrities, all that, no problem. Now they're starting to reel that back and Sam Altman published this blog post where they said that hey, this is a new category. They're learning, they're making changes quickly, they're giving the rights holders more control over the generation of characters and they're going to introduce a way to make money for video generation generation. This is still very vague, but in the end he said this. Please expect a very high rate of change from us. We plan to do our iteration on different approaches in Sora. So the product is going to evolve change, they're going to test things just like they are in chatgpt with the apps now for example and then only time will show where this will land eventually. But the virality that this cost is definitely an indicator that the category of AI generated video for entertainment is just a thing from here on out forever now. And the first related thing that I wanted to tell you about is just a simple GitHub repo from Matt Schumer Sora extend where you can take clip and artificially extend them when you're calling the API. So you could create clips that are longer than the time limitation in Sora tool. And that's a Sora related updates this week that you should know about. And amongst all of those OpenAI releases there were even more, but I think it's enough if we kind of just list these others rather than going into detail. So first of all, they made GPT5 codecs widely available now, so it's not in Research Preview anymore. If you're not familiar, that's the coding specific model and they added a few new things like a native Slack integration or again an entire SDK where people can just build really efficiently with the model. Then also GPT5 Pro is now available for the API and I think this is significant. I don't know where, but for me personally, GPT5 Pro is the go to model if I really care about a question. It's gated behind the $200 ChatGPT Pro plan, but now you can make API calls which are pricey directly to the call and just pay for what you use. They also ship the smaller voice model GPT Real Time Mini. It's 70% cheaper than advanced voice mode and that's pretty much all from their dev day. Next up we have a short story, but one that nevertheless I think you should know about and that is the fact that there's a new top contender for best image generator out of China. It's fully open source, a very large one, so you won't be able to run this locally. But Hanyuan image 3.0 didn't just surprise me, but also everybody in our community. This open source model placed at the top level along with midjourney Gemini and even beats out some of the other Chinese models. And this isn't just kind of a personal take in lamarina's text to image rankings that actually got first spot. But rather than talking about it as per usual, we'll just show you some example images. I think this one actually excels across all categories. I mean the logo looks really great, the pictures look detailed but not too detailed, the graphics look great and especially this comic when compared to some of the other image generators came out super well. And the thing is fully open source. So you know we get all of these big stories out of the US and OpenAI but China is not stopping, they're just pursuing a different strategy which is a very open source focused 1. Perplexity's Comet being available to everybody for free. If you're not familiar, that's their AI powered browser and up until now you had to pay $200 and have an invite code to access this and now it's just there. But honestly with next quick hit from Gemini and what Claude did with the browser extension, I think Comet just doesn't perform as solidly as some of the competition here. So even though this might be free and you can try it now really easily, I'm not sure it's worth your time. But hey, I know a lot of people use it, so leave a comment if you have something specific where it's better than Claude browser extension. I haven't found an example like that yet, but happy to learn. And the next story sort of ties into this because Gemini 2.5 computer use came out and it's a new computer use model and you can actually really easily test this if you head on over to gemini browserbase.com they hosted a version of it and what you can essentially do is give it a task and inside of this virtual machine it's just going to perform it. The benchmarks on this model look really impressive, but I think this integration where it's in a standalone computer rather than in your browser is just inferior. At least for me it is, because it doesn't have all the login credentials. It's going to lose the cookies. It's just cumbersome. Look, even now if I gave it this basic task, I need to kind of go down and accept these cookies so it can proceed. But if you've never experienced one of these computer use models, this is your opportunity. You can just go to this link, give it a task and see see what it does on this site. It has never been easier. Half a year ago you needed to run it through Docker and install a bunch of stuff. Now it's just you go to a link and give it a task and this sort of works. Although I do think that the Chrome extension is superior just because it is everything that your browser has and that's what you care about. But you can see it working here and it's a great thing to play around with if you haven't explored this category of AI at all yet. In this week's quick hit section, it's just Xi introducing Imagine V09 and you can test it out right here and it sort of has this infinite scroll interface. So if I just go ahead and say a cat with a hat, as I usually do, then it's just going to generate an infinite amount of cats with hats if you keep scrolling. If you just hit play on maybe this wizard cat right here, it's going to turn it into video. The model is good, but nothing special. They're just throwing their hat into the ring, no pun intended there, and this is what the video looks like. That's fine. Then we have also some updates to Google Opal and this was a thing that I personally found really interesting when it came out. Google product that makes these no code interfaces as simple as they could be. And they added a bunch of things to this. So first of all, it's globally accessible before it was gated and you needed a VPN. Now 50 new countries have access and can play with this. And they added features that were really missing, like debugging. So if you want to get into building these no code workflows, I think this actually might be the best entry point. Just because it's so damn simple and completely free from Google and it's available everywhere now. And there you go. That's pretty much everything we have for this week. I think a lot of these releases, especially the OpenAI ones, will need some time for people to build on top of them. These are developer focused releases and as you saw, the ChatGPT apps are very limited for now with large projects plans for the future. So I think over time these will be really significant. But as of now, I would mostly say it's a good moment to learn about this stuff, to start experimenting and using it, but the real impact of this will only be seen over time. All right, and with that being said, my name is Igor and I hope you have a wonderful day.