Here's What GPT-5 Might Look Like & More AI Use Cases

So, as per usual, we have a bunch of generative AI releases to talk about this week, but there's a clear theme amongst everything that has been happening this week, and that is not just the fact that we've seen before that these apps are converging, they're converging towards a specific goal. This goal of a single AI assistant helping you. You could also refer to it as the ultimate agent, but truly, I think the correct words that most people would use to refer to this idea is GPT5. And no, OpenAI did not give us a concrete preview of that system. But most releases in this week are pointing towards this common direction of one AI system that has your context and has the computing power to get things done for you. So if you were wondering what this next gym system looks like, I think this week's releases show it really well. We're going to be looking at it from different angles. For example, genspark releasing AI sheets, an ability for the agents to work with spreadsheets. And also they added capabilities to download literally any video on the Internet really simply. And also we have notion coming out with a big notion AI update, which essentially centralizes a lot of these AI functionalities into this everything app that notion is trying to be. And then there's a bunch more like a new model inside of ChatGPT and a whole lot more short stories. In this week's episode of AI News, you can use the show that pulls together all the AI releases from this week, filtering for the ones that you can actually put to work today, and then me showing you the ones that matter. And that has been long enough of our intro. Let's get straight into the first news story, which is an update to the genspark Switch suite of agents. We talked about this before and they're really trying to do a lot with this single product. And honestly speaking, I'm featuring this here because I feel like it represents whole category of products. Now, in the Quickfire segment later, I'll tell you about a competitor, Manus AI, now being free to try. Technically, there's also OpenAI's operator that competes with this, but it's really more limited. This is trying to merge all these different modalities, all these different functions into one system. And this is what it looks like in practice. You have the super agent here that can command all these different things, and then you have these various modalities if you want to focus on a specific one. This week they added AI sheets, and there's also this AI drive where you can go in and download anything. But really they're trying to build the type of product that I believe is going to be very similar to what we're going to get from GPT5 eventually. Because one thing they clearly said is that they're going to get rid of all of these model picking in ChatGPT. It's just going to be one system, GPT5 with all of this tooling, including the ability to deep research, including all the tools like creating images, plus it will hopefully have a better version of operator built in, which if you're not familiar is the browser agent where you tell it to do specific things, it clicks around the Internet. It's a little simplistic. Now I wouldn't recommend using it at this point in time, to be honest, but they're not even pushing new releases to this, whereas genspark is, and that's why we're looking at this now. So first up I want to test this ability to download anything because this is a pretty controversial one. We've had these websites that kind of download YouTube videos for you since forever. They always get taken down, et cetera, because they're against the terms of stuff service. But the new ones pop up. But let me just find a random video off of TikTok here. Literally just going to take. How about this little one with some AI generated product mockups. I'm just going to take this link, post it into this, download into AI drive function. Okay, let's see how this works. Download, starting, downloading and that's it. Successful. What? Okay, so now it ripped this video straight off of TikTok into my agent and now I can keep working with it. I mean, how about the launch video from Genspark themselves? Okay, so it doesn't work with YouTube videos. Interesting. But the fact that it can download files now is not revolutionary by itself. It's a fact that you can let it now do things like hey, giving it the link to a video and then asking it to create a sheet with similar videos to that along with all the views likes. And then you can really think of this as ChatGPT with all the tools that it has, like data analysis, canvas, et cetera, all working in unison to fulfill your request while also using a browser to do it. That's different from web browsing, giving a few links of context and then ChatGPT working with it in something like O3. It goes further, it does more work, it has more tools. And that's why I'm saying I believe that is going to be the next evolution of this product here because that's a more user friendly form an AI assistant can take for you. Instead of model switching and tool selection, it happens automatically. And this startup is just a bit faster at getting it to market than OpenAI is. Because OpenAI can wait and then start steamroll competitors like this most likely, and then they'll have to pivot point is this is pretty impressive. I'm just sitting here and it's doing all the work that I would be doing manually if I wanted to create a TikTok content strategy around a specific niche. Okay, maybe not all the work, but you get the point. This is a serious tool and from my initial testing for this type of thing, it works better than OpenAI's operator does right now. Honestly, at this point I don't even consider as a tool choice. So this has been working for the past 10 minutes and it's still going. Look, it found six other videos, pulled the LIKES data where it could, it couldn't get the views from these. And it just keeps going, keeps thinking, keeps adding more videos. And I'm going to end the segment here. But I think you could imagine what happens if you give an agent like this the ability to not just think for 10, 15, 20 minutes, but the ability to work for a day. Maybe you go to bed and let it run for 12 hours straight and when you wake up, you have all of the research prepared for you. Something that would have took you hours. And while this product is certainly not perfect, I really think that this is a genuine glimpse into the future. And yeah, therefore I wanted to share it with you. On to the next one. Okay, quick note from editing Igor, I suppose out in the real world, I recorded the segment for this video this morning, Friday morning And Friday afternoon, OpenAI released the new Codex inside of ChatGPT with the Pro subscription. We're not going to cover that in this video, it's too fresh. But I wanted to point out that it aligns with all the themes we talk about in this video. And I'm going to create a separate one just on it, because that's really worth it. All right, back to the video. Okay, so as you might know, I've been running this YouTube channel for years, and if you're familiar with my story a bit, this is actually channel number five. I started the first one when I was 17, which is 14 years ago at this point. And I got to say I did underestimate how complex content creation is in the tech space at this level. The filming this part, while not always easy is pretty straightforward. So much preparation goes into these vids that I just got to sit down and tell you about the points that matter in the current video and hopefully that inspires you to do more with generative AI. But honestly, most of my time is spent in the preparation stages of the video. Researching topics, testing tools, evaluating those tests, putting together scripts, deciding on the packaging, recording, editing, reviewing, uploading, and then evaluating data. That's a lot of steps. And if you don't have an efficient system in place to manage all of this, especially as more people get involved, your chances of success are slim. Because you might feel like you can hold all the context in your head, but it quickly becomes overwhelming, especially when working on multiple videos at a time. So if any of that resonated and you're a creator like me, you really need to check out Spotter Studio, the sponsor of today's video. Spotter Studio is the all in one AI powered ideation system. Built for pro youtubers. It pulls real time trend data to tell you what topics will resonate with your audience, gives you packaging ideas on demand, and then lets you turn those ideas into trackable organized projects all in one workspace. And Spotter Studio does this for three main tools, Outliers, Brainstorm and Idea Bank. It's actually quite simple. Outliers pulls Together Real Time YouTube analytics and Highlights topics, titles and thumbnails that are gaining popularity right now for your specific channel's niche. Once you found an outlier like you, brainstorm off of that outlier. It shows you a list of potential title thumbnail concepts, hooks and supporting talking points that are personalized for your channel. And the third part is all about organizing your upcoming ideas. And because YouTube really is a marketplace for ideas that uses the video medium to communicate them, the third core tool is called Idea bank and this allows YouTubers to rapidly add, organize and prioritize your video ideas. You can see which videos have the strongest titles, prioritize the videos with the higher scores, and then you can open up the idea to further develop it, adding details, previewing what it would look like on YouTube, and using the Story Beats feature to further develop this idea and move it along the workflow. As you can see, all of this is just making it easier for you to actually make the video. And everything lives in one place and nothing slips through the cracks. So if you want nonstop inspiration, a data backed content plan, and a workflow that actually keeps up with you, check out Spotter Studio today. You can click a link at the top of the description to get started. With Spotter Studio right away. And if you sign up now, you'll get the limited time offer of a full year of Spotter studio for just $99. A big thank you to Spotter Studio for sponsoring this video. Now let's get back to the next piece of AI news that you can use. Okay, next up, we have some new models inside of ChatGPT which are interesting for a specific subset of people, and those are developers. If you go to the model switcher here on the more models, you're going to see GPT 4.1 and 4.1 mini. Honestly, this is probably a reaction to the increased popularity of Gemini's 2.5 Pro model especially, especially amongst developers, but also a lot of common users ic switching to Gemini 2.5 Pro. So as a reaction to that, they now released this coding focused model that previously was only accessible for the API, meaning you had to pay per usage. And they mostly thought of it as something that you might want for app development and they thought of it as a model that you will only want for development and you'll be using it for the API anyway. Well, now they have it in here to further increasing model choices. One thing that I guess I forgot to Note is that 4.1 is not a thinking model, it's a GPT model that just gives you the answer right away. So it answers right away. It's also rather slow, but the quality of the code you get out of this is really high as that's what the model has been trained to do. Well, beyond that, we have one more ChatGPT upgrade which actually I find quite significant as a regular user of Deep Research. But if I pull up this Deep Research that I did on a medical condition that I was dealing with a while back, well, there's this new subtle but very powerful button at the top of the Deep Research that allows you to download this as a PDF. And as some of you might know, you can always prompt ChatGPT to save something as a PDF, but that PDF usually has a lot of mistakes in it. This one is kind of perfect. It includes inline references to the various articles, the formatting is perfect, and yeah, I would say it's equally as good experience to read the entire article as it is in ChatGPT with difference that now this is transferable. This is a way better file format if you want to send this per email to somebody. And it's also super convenient way to transfer this content to somewhere else because sure, you could copy all of this, but that gets messy quickly. So if I'm doing a new chat. I could just add this PDF. And again, we discussed this before. PDF is not the perfect file format to attach. Preferably you want something formatted as a markdown file or a JSON file or something else that is native to a computer. PDFs are formatted to be readable for humans. Nevertheless, this does work super well. I personally would also like a markdown download function, but this way you can take the Deep Research data and then add it to new chats, add it to projects, added to knowledge bases of chatbots, transfer it into a different application really simply as I still see Deep Research as the most powerful feature in ChatGPT. I think this is a very welcome addition that everybody using it should know about. Hey, if you're watching and enjoying this video, it would really help the channel if you also leave a. Like, it only takes a second. And every time I see how much likes actually influence the video's ability to be distributed to more people, I'm surprised. So, yeah, if you liked the video, don't forget to express it. And now let's move on to the next story. Okay, next up, we have another story that aligns with this theme of the next generation of AI models kind of happening outside of OpenAI right now. As I said, they'll probably catch up very soon here. I don't know the timeline of that, but I can tell you that notion is also going for this holy grail of the ultimate AI assistant, and they're doing that by extending their Everything app with a bunch of new AI features. I think some of them are less exciting, like the AI meeting notes that they added now. So as in every other meeting application, you can now transcribe notes and then work with them. That's great, but really what I'm looking at here is this enterprise search. There's a bunch more here. Like they have their own version of Deep Research now. They have a model switcher now, and it's all priced under one subscription, which looks like this. Right now you need this business subscription per member to access all of this. But this enterprise search function where it doesn't just connect to your notion pages, but it also includes your Slack, GitHub, Google Drive, Jira files. I think this is really significant because this is, in my eyes, where the ball is going here. An AI assistant that you don't have to prompt manually for every single little piece of context that you might want to evolve there, but one that already sees the files, one that already knows. I mean, heck, would you want a Physical assistant that is a blank slate every day, and you have to reteach him or her how to do things. No, there's a lot of value in a trained employee who already has all the context. That's what this enterprise search is about, giving it access to these apps, and it's figuring out what is needed for the task at hand. Now, sure, we see versions of that in apps like ChatGPT already. You can go in here and connect your Google Drive to this, but it's different because ChatGPT is not a knowledge base of files that it looks over, at least for now. This is what I think is going to happen with GPT5, whereas notion is by default a knowledge base for all your files. I mean, heck, I even use it as an operating system for the entire company. Now, you can extend that beyond the confines of Notion and use it all in unison with their AI, making them another competitor in this race for the ultimate AI app. And I think this overarching theme is something that everybody following this space should at least be aware of. Because it's one thing to say, hey, these tools are getting smarter every day, and another one to at least have a working theory of what that looks like in practice over the next weeks and months. Because then you can potentially start building those knowledge documents, transforming your files into markdown files, reorganizing your company departments for a world where those tools seamlessly integrate into your workflows rather than compete with them. So, yeah, that's the Notion update, but as you might imagine, there's more. Okay, so for the next one, I want to show you a practical prompt that actually popped up inside of our community. Big shout out to Shay, who's actually been with us since day one. And the prompt is one that you can use with ChatGPT image generation, which, by the way, is now also available through O3. As you can see, a lot of things are converging over the past weeks, especially this week. So, in other words, if you want to generate images on a free planner, you can just use 4.0. But if you want to use all the other things, like almost every other tool in ChatGPT, you can actually do this directly inside of VO3. So I'm just going to demo it in here. And this prompt produces these very interesting images, because you know how they say an image can say more than a thousand words? Well, this is sort of four images. And what it allows you to do versus just having one image is actually tell a story. You could also use these as infographics as they hold so much information and they could obviously be used for all kinds of purposes. And you can fully customize this prompt for yourself with 4, 0 or 03 and make images like this yourself. I particularly like this use of it. Explaining a concept in a very, very approachable manner. Like literally everybody, ranging from a five year old that can read all the way to a senior year would resonate with this type of imagery and text. And this is what the full prompt looks like. I'll put it in the description below. Again, thank you Shay, for sharing this and giving me the permission to to add this to the YouTube video here. So the basic instructions are Please use this 2 times 2 visual panel prompt template 4 and then you gotta add your topic. The rest of it is set up to be an educational panel set with all the specifications and a base structure for the different panels. So the first one is talking about what's happening, then why it matters, what's unclear and what you can do. Obviously you could change the theme of this and you could also change it from educational panel, but this really works well and all you need to do is change the topic here. In the beginning I'm just going to go ahead and say I don't know AGI, that's the first thing that comes to mind. And I'm going to do this with O3 just because I can. And in this case O3 actually reasons over this, but doesn't create the image. And you know what, we're actually going to keep this in the video. I want to keep it real. So it looks like for O3 we need to generate an image to the prompt, which we'll do for the one in the description. And there you go. And created the little 4x4 educational image about AGI. Yeah, wonderful. Just as expected. Opportunities, challenges, and the final one of focusing on what you can do. And all you need to do to customize this for yourself is customize this one word. Pretty powerful stuff if I might say so myself. All right, so for our next segment, I want to talk anthropics MCP that has been becoming increasingly popular. We covered it many times for the show. But for anybody who might be new here or who needs a refresher, you can think of mcp, AKA the Model Context Protocol, as a standardized way to plug various functionalities into LLMs. It's very similar to ChatGPT plugins that came out 2023, if you remember those. Back then, ChatGPT added this functionality where you can essentially add Expedia to your ChatGPT experience and then it could talk to the site. Now, that interface turned out to be, well, completely misguided to say the least, because they ended up removing it. Again, nobody was really using it at all. But MCP does a similar thing on a completely open protocol that can build with any chatbot, any LLM, where it allows you to plug in all of these various services into any, let's say, chatbot. And now this week I want to round out some of the news around this. Not just that we're getting interesting servers by the day, added to the list of I think 3,000 servers at this point that you can kind of use and easily plug into your. Well, for example anthropic Claude's desktop app or if you're coding with cursor, MCPs are the way that people are extending their experience there. And there's a few pieces of news this week. The first one is there's actually a brand new course that is to be going currently freely available, and as the mission of this channel is to help people adapt to this new technological age, courses like this are essential for that. Now, this is not something you will take if you're completely new. They even say it's an intermediate course. And if you want to learn more about mcp, the first videos make sense. But beyond that, it does get a bit technical and it sort of assumes that you're going to be developing applications and working with GitHub, repos, etc. But beyond that, some of you might be already working with some of these vibe coding workflows, as you'll see in a video coming up soon on the channel. And along with that, I want to cover some updates that also recently happened to Claude's codes. They actually changed so much in it ever since the release. Also should be noted there's a competitor from OpenAI called Codex CLI, but I myself have played with both and maybe it's the power of habit because I started using this on the day of its release, but I still like Claude and also actually thinking about it, I think it's more about the fact that Claude 3.7 just kind of does things and figures everything out. And the OpenAI version of it, it's more targeted, it gives you these follow up questions to really nail down what you want. And I kind of like this wildcard nature of cloud code. It's just a really fun experience to hit a button and then have it think for three minutes and come up with features and extensions that you didn't even ask for, but that do make sense most of the time anyway. Point being, there's new features there that are absolutely incredible. They added things like web search in cloud code so you can access the web with it. And they even launched a subscription where you don't have to pay per usage. Usually a few hours of playing around cost anywhere from 5 to $20. But most of all, they added a feature this week which I honestly did not expect. And that's the feature that you can actually interrupt it in the middle of its coding frenzy and redirect it. Because what it does is it comes up with these to do lists of things that it needs to do to like fulfill the request that you prompted it for. And then up until now, you were kind of forced to sit there and watch all the madness unfold in front of you. Now you can actually interrupt it and readjust the plan as it goes. So as you can see here from this quick little demo video, look at that creates the TO dos, it updates them, and as soon as it's performing the TO dos, you can actually hop in and say, actually make it green. And it will readjust the plan and instead of creating a blue bar, it will create a green bar. Very simple example here, but often these to do lists are like six or seven items long and you're sitting there for free, sometimes even up to four minutes. As cloud code does its thing, now you can interrupt it. I think this is a big development because using this and other agent decoding tools, it always feels like you're not fully in control. It's kind of doing its thing and now we have a bit more control. I like that trend and I think this one really delivers that. So this is from a company called MEM0 that I've had my eyes on for a while. They basically open source the protocol that allows it to add short term memory to chatbots and agents. Now they put this inside of a MCP server, making it simpler than ever before to add memories. The feature that many of you know and love within ChatGPT to any agent, okay, it's 100% local, gives you persistent and portable memory, meaning that if you're using different apps and different agents, this memory can transfer between between them. And all of that is standardized because it's mcp. So if multiple agents plug into mcp, which most of them do with this, you can save all the short term memories in one agent and then easily pass them along to others. And let me tell you, this is a massive step when it comes to using some of these agentic systems. Up until now, they really worked in silos and bringing things over, bringing the context over was a rather tedious task, and in most cases, it wasn't even worth it. Now, with things like this emerging, all of these tools start melting into one ecosystem that is not yet automated. But as you can see, we're getting there and the companies are pushing it in this direction. Big development here of which we will feel the consequences of over the coming months and weeks. But, yeah, if you're using any agentic system or you're building any agentic system, have a look at this, because I don't know if it's going to be exactly this form or something else, but the fact that ChatGPT added memories and it went giga viral and people were sharing these prompts all across the world that use memories is only a proof of concept of the fact that this memory modality is something that doesn't just sound good in theory, but is actually something people want. And that means we're going to see memories all across the place soon, whether it's this implementation or another one. And what that essentially means for you as just a consumer of these apps or just someone interested in this, is that, hey, you should probably spend some time crafting some memories that matter, crafting custom instructions that are really accurate to what you do, and not just relying on ChatGPT, kind of collecting scraps from your chats, but actually putting some thought into, how would I describe myself to a person that I've never met before if they're supposed to help me with a specific set of tasks, like my work tasks? What would I tell them about me, my history and my preferences? Even if you just draft it out in a few sentences, you're going to be further along than most people. And then over time, you can develop them, make them more detailed, and have an answer to the question that many of these programs are going to pose you, which is, hey, what should I know about you? What are the custom instructions? What are the memories you want to give to me so I can help you in the most efficient manner possible? Okay, enough said. Let's move on. Okay, so let's do the Quick Fire segment that we introduced a few weeks ago. And not just I enjoy making it, but you guys also seem to love this part. So here we go. Here's all the stories that are interesting, but maybe we don't need to spend multiple minutes on them. First of all, do you remember Manus, the AI gent that had its viral moment, but most people didn't get beyond the wait list. Well, by now, not just that they removed the wait list a while ago actually, but they're also giving away 300 credits to anybody who wants to try the application. So if you ever wanted to get your hands on this, well, there's no more wait list and it's free to try. And right now this is the main competitor to the story we talked about earlier, the genspark agent that can manipulate cheats. Now then next we have a story coming out of TikTok. They're adding a brand new feature that turns images into videos. So now they're making the creation of TikToks even easier. All you need is an image and AI does the rest. I don't know how popular this will be, but I wanted to show it off because it's a great example of how AI tools that we, we look at eventually get implemented into all of these consumer apps like TikTok. And then also in this interesting video, the US Senate was questioning some leaders in the AI space about ChatGPT use cases. Sort of interesting, although I don't think this entire thing is worth listening to for most people. And Sam Altman shared that he finds it particularly useful to raise his child, and he pointed out that he's not sure how people were raising kids without the help of ChatGPT. I mean, I kind of like that point. Not a use case that I thought of as I don't have any kids myself. Yeah, there you go. AI use cases being discussed in the US Senate. And then for anybody interested in super fast and super super cheap transcription services, there's a new hugging face base called Whisper Large V3 Turbo, which transcribes at 100 times real time speed, meaning around two minutes of audio will be transcribed at around a second. And it only costs $0.80 per hour if you use the suggested GPUs. If you need to transcribe a lot of video or audio files, this is probably worth bookmarking. And that's really everything I have for today. I hope you find something that will be useful to you. If you enjoy the show, don't forget to subscribe to the channel. I do this every single Friday. And with that being said, I hope you have a wonderful day.