GPT-5‘s Biggest Strength & More AI Use Cases

So, let's talk about the most significant releases in Jana over the past week, including the big GPT5 story. That release got a ton of backlash, and I think it's a really mixed bag. On one hand, GPD5 sometimes really underperforms the previous models. I see that. But on the other hand, the Pro model is better than anything I've seen before. This is the best model in the market right now. In my opinion, and I'm going to show you why I think that. We're also going to be looking at some data which shows the actual usage of Gen AI today and how it developed over the last year. We have some competitor platforms like Gemini and Claude introducing memories, the favorite feature of most chatb users, and so many more quick little stories in this week's episode of AI news you can use, the show that rounds up all the generative AI releases and filters for the ones that are available today and that actually

matter. Let's begin by talking GPT5 because this is the number one story on most people's minds. It's the number one model most people use. And I want to start by reviewing this data and leading into a discussion on what to actually do inside of ChatGpt. Now that we have more options here and how to turn those on. So this data comes from similar web and it paints a clear story. One year ago, Chach5 roughly 85% of the market share for Genai. Today 78% and the trend is clear. They've been slowly but surely losing some market share to competitors. Deepseek came in strong, but now it's becoming less and less significant. And then Gemini has been growing in popularity now being at almost 9% of users with the surprising one of Claude being just at 1. 6%. what in my circles I would say it's like a 60 2020 between chat GPT and then Claude and Gemini but the internet traffic speaks a different story and I think this change in usage is what inspired GPT5 and that leads us into our next story which is the

discussion around the GPT5 release I don't want to spend too much time on this I could be talking about this for half an hour but you probably caught that the backlash against GPT5 was drastic and most people were very disappointed with the new model and I think the disappointment came from one place in particular and That is if you're on the automatic model switcher and you ask certain questions, then it automatically picked what level of thinking it wants to use. And what I've seen happen in practice is that they introduce sort of a new tier in the model that is even below GPD40 previously. So a lot of people just tried GPD5, they got routed to that version and they received inferior results to what chat GPT was giving them before and they were furious. Rightly so. And I have to say I try to recreate that behavior that we saw in the first few days and I haven't been successful at that anymore. They ironed that out over the last week. Now what I found is that most of the GP5 responses come in at least at GPT4 level. But nevertheless, to give some power users the choice back, you can go to the settings here and enable show additional models, which is off by default. And I do recommend turning this on. Especially if you follow this channel, you probably know your way around this model selector at this point. I talk about it regularly. and it gives you these extra choices. So, I think I really have two important messages. First of all, the hate that was directed at the weekend of the GP5 model, I feel like has been ironed out. I think they made that decision to save costs because the popularity of Genai has just been rising and now they're serving 700 million monthly active users, many of those on the free account, and they just wanted to improve the economics of the company, which resulted in a worse experience for most people. That being said, I think at this point, what I said a week ago in the release video actually stands. I think for most users, this is a better experience now. they don't have to deal with model switching and that weaker model to me seems to be history and for anybody who really wants the personality of GPT40 which I agree is incredible. It just feels more like a friend than any other model out there. You get that choice now. Okay, but here on this channel we're about the practical applications. So, let me switch gears and actually talk about how you can get this to do more of what you want. I think I have two points that really matter here. The first one is a realization that some of these thinking models before they did not use memories. Okay? And now all the GB5 models do. People don't realize this. They didn't just improve the model switching. They improved also the integration into the entire ecosystem of their features and apps. So let me demo this quickly to you. So if I save a memory here that just says that I love roller skating and then I go into a brand new chat and I open up one of the old thinking models like 04 mini. What do I love? And it doesn't know because it doesn't have memories. But if I open up the new GT5 and go to thinking mini new chat, what do I love? You love roller skating. This was not the case before. Most people didn't realize these. And memories are arguably the most powerful feature for most users that do not manage their context manually with custom instructions or projects and custom context files. Things that we teach in this channel and the community and now even hold webinars and workshop for companies that teach how to engineer the context in a way that empowers people who don't know how to craft the context themselves. Chat GPT solution to this are the memories and they didn't work with most of these models. I think that's a really important point that has been completely missed over the past week. GPT5 has it on all of them and I think that's great. And the second practical point I want to make is GPT5 Pro is insane. I do not say this lightly for anybody blaming me to be a OpenAI shell on this channel. Guys, I'm not affiliated with anybody. I haven't even been in touch with OpenAI. They didn't give me early access to the models. They never sponsored me. I'm just here to find the best product in the market for myself and then I want to make honest recommendations to you. Let me tell you this five pro model. It just works better than anything out there for the general purpose queries especially business related tasks that I love for writing and coding. I myself am still torn between claude and chat GPT with Gemini being super strong too. But when in doubt GPT5 Pro, oh my god, enough talk. Let me actually show you. So this is an amazing prompt that I kind of discovered recently. What I do in my free time, I often watch psychology related lectures and read books. And what I love is as that knowledge of mine kind of builds up, a lot of it translates into AI prompting. And sometimes I just come up with prompts like this. What are the components of some of my favorite activities, hobbies that I love so much? And can I find those components in other areas of life? Essentially asking of what other things in life would you love based on what you already adore. This is just the first version of the prompt. I'll be crafting this further. But then I proceeded to give it some activities that I love and some of the reasoning behind it that I see there. And then I ran this through GPT5 which was a good answer. It told me, hey, with kite surfing, these are the components. There's technical mastery, there's freedom, there's adrenaline and risk, self-reliance, and more. With paddle, it's about tactics, friendly competition, social connection. Entrepreneurship, it's about growth and experimentation, problem solving, multiskll integration, impact creation, and more. And then it gives me recommendations of other activities where I could find this. Like similarly to kite surfing, I could try backcountry snowboarding, which is actually my second favorite sport on earth next to kite surfing. So great recommendation there. FPV drone flying, which I do too. And then rock climbing and sailing are the ones that I'll keep for the later years of my life. But I've been very aware of these, let me tell you, like I do enjoy rock climbing and sailing is kind of on my bucket list for things to kind of explore later in life. And overall, these other answers also kind of map. I guess there was one surprising one which was like volunteering to lead big initiatives in nonprofits but many of these other ones like running events or community groups. I mean we run a community like these are obvious to me that I also enjoy these and would be good at this and then it gives me a little analysis of everything. So this is the GPT5 response. Okay. Now if I switch over to the GPT5 pro response I think it really illustrates how damn powerful this thing is because this is like having multiple PhDs in your pocket. But again this is the pro model. This is behind the $200 pay wall. Sam forgot to mention in the presentation that this is the model he was really talking about, not the one that's going to be available on the free tier. And I think that discrepancy between the expectations is what caused the outlash. I just wanted to show you that the pro model, it delivers and it does it in a way that I haven't been able to get this level of quality out of the other models. Prompt adherence is also unbelievably good. It never forgets any detail of what you tell it. Okay, so to this answer, I'm going to scroll through this. You can have a look at it if you want, but it takes a completely different approach. It takes an approach as if this was like a research paper on your interests with the goal of identifying the patterns and showing you what you will love in life. And if I had to score this, I would say GPT5 is like a level of insight and depth that is maybe like a five out of 10, whereas this is like a eight or a nine. Like it analyzes all the ingredients independent of the activity. It gives you ways to integrate it into your everyday life for each one of them. And then it doesn't stop there. It comes up with patterns that you can reuse in different areas in your life. So it comes up with this concept of a kite window. freedom first scheduling where you predefined a wind window each week that is a non-negotiable autonomy block where you just channel creative energy in a blocked off window just like I would with some of the other activities I love. I thought this was genius. Or this 12 experiments per quarter rule where you start with a hypothesis, give it a time box and a single metric and then it gives you implementations of starting a ship club ladder where you get together with some people and you assign points for ship lessons, toolkits, community contributions, kind of integrating everything I knew with what I actually do. A multi-week implementation plan and then a lightweight way to finish it all up. And it even offers to create a template, which is something I find GBD5 Pro does forever. Now, I could keep on showing you examples like this to prove the point. Okay, I just picked one that I thought just really illustrated it well. For anybody who's still skeptical, I have another one here where I basically told it about all the current opportunities I'm looking at with the channel. So, this one really is quite personal. Let me even showing you this. It's actually putting myself in a very vulnerable position cuz I'm considering some opportunities right now between speaking education and what we already do. And I was just blown away by how it structured the output. It gave me different points of view. One was focusing on max profit. Another the maximum amount of freedom and so on. And it even created an opportunity scorecard and it ran different scenarios for me and save those in a CSV file. Are you catching this? Chat GPT literally thought about various scenarios about how the different things I'm thinking about could play out and created multiple documents to help me decide and to help me see into the future. And from my experience in this world, I would just say that it maps onto this IQ score of the models. Honestly, I'm not exactly sure how trustworthy this is, but it has been going around social media where GPT5 Pro actually reaches 148 IQ, far surpassing all the other options. And 5 minutes later, we have the template that we let it generate. Let's have a look at what it did here. Oh, that looks very nice. And actually, yeah, I do want the CSV. So, I'll just go to Thinking Mini here. I'll just say yes, CSV for Notion import. And then this is just going to run some Python code as all the tools seamlessly integrate into this. And let's see if I open up notion here. Say import and give it the daily activity logs. There you go. That's a way to score it everything based on a custom methodology it came up with. And it did the same thing for some of the other scoring methods that invented here for me. And I personally just find that really impressive and effective. But you do need the thinking or the pro mode. When it comes to default GPD5, it's I don't know. I would say essentially the same as the models before. And I get why many people don't like that with all the hype that was happening, especially if sometimes it underperforms. But yeah, my point being open was hyping up this whereas most people got this. But understand that on this channel we're always talking about maximizing the opportunity here which hides down here. So naturally, I'm going to be judging the best version, not the worst. And I think that one is incredible and I use it all the time now. Okay, let's see what's next. Okay, so in case you missed

it, I recently posted this video showing you how I built a real mobile app with AI that we even published on the Google Play Store without needing to code. Turning our most popular GPT into a mobile app was a lot of fun and now anybody on Android can download and use it. So if you want to build a real mobile app, too, you should definitely check out Bubble, the platform on which I made the app, Sam the prompt creator on and the sponsor of today's video. So for me personally, this was the first project I ever did on Bubble and I went into it hyperfocused on creating this app. Once you get into it, you start realizing how much is actually possible with the different tools. It is very customizable and they do give you a ton of resources to help you get started. But from all these resources, by far my favorite are the templates, preset projects that you can customize. And from all the different ways to learn, this is usually my favorite. I like to take a starting point that gets me right in the middle of the process. I see how that works, edit it, add something, break something until I feel confident to actually start from scratch on a new project. And that's exactly what these templates let you do. or just take them, customize them a bit and use them as is. So, for example, there's this free one which allows you to build a dashboard with OpenAI integration. This is the most popular template and allows you to turn data into dashboards that you and your team can use to visualize data. And obviously, this has been made for you to customize it to exactly what you need. Or how about this other one, which is a multi-purposebased template where you can do way more. And this one allows you to build websites in no time that also transfer to mobile really well. And in total, there's over 1,500 templates in Bubble right now. So, if you've got a project in mind, you almost never need to start completely from scratch. And if you've ever done a project like this before, you'll know that editing is so much easier than building completely from scratch. So, if you want to build real mobile apps without code and get them onto the most popular app stores, go ahead and try Bubble today. They're giving away a free month to all users right now. So, check out the link in the description to get started now. All right, now on to the next piece of AI news that you can use.

So, now we're going to talk about clawed memories cuz they finally added the popular feature in chat GPT to claude. You can enable it if you go to your settings. It's enabled by default, but you can check it if you go into your settings and this function here, search and reference chats. Allow cla to search for relevant details in past chats. And there's this simple prompt that they make you run when you log into Claude for the first time. Hi Claude, what were some highlights from our recent conversations? And as you can see, it pulls up some of my recent conversations here. I really like how it shows you exactly what chats it references here. That's wonderful. And then it summarizes it back to you. So yeah, these are some of the recent conversations I had with Claude here. But it made me really curious about one particular question and that is how does this feature interact with projects because that is my main use case. So I'm going to just pull up my work context here where I have my company mission vision values and overview of our entire company ecosystem into markdown files here as context with a little bit of custom instructions. And I want to see if this is going to pick up on this chat that I'll just start in my general folder, which is I love rollerblading. I'm just going to say that straight up. And then we're going to check inside of the project if it knows that. So I'm going to go back into the work context and ask what do I love? And sure enough, it looks into the context documents, but there's no mention of rollerblading. So as they should, projects are still self-contained folders. But it still leaves me with one more question and that is if I say I love kebab which I actually do in Austria that was one of our go-to dishes after school. So it acknowledges all the context I gave it and then it says okay and noted on the kebab appreciation. Okay, I love it. Love it. So what if I now say what do I love? Does it reference the old chats in the project? Cuz this I would actually like to see. And sure enough no mention of my kebab love. Here — they ask you how you are. You just have to say that you're fine. — So just like in chat GPT, this feature only works outside of projects and you can always see that here at the top through the breadcrumbs. This is just a general chat. Whereas this kebab love is inside of the AI advantage work context. And this is the number one way I recommend people actually work. Build your context into markdown files which you attach to a project and then group conversations by topic. You don't want all context for all conversations, especially as you get more proficient with these tools. But most people won't be doing that. most people will be just starting new chats and for those referencing previous chats is actually super helpful in giving more relevant answers as most likely their requests won't be that precise. Now on the same

note I want to point out that also Gemini shipped a memories feature into their platform but that feature seems to have not rolled out to my account as of yet. They also just tweeted this 16 hours ago, but yeah, three out of the four big platforms. The fourth one being Grock, kind of the new kid on the block. Very smart, very controversial, bit behind on many features. Doesn't have memories yet, but all the other players have this core feature. And I can't wait to see how this is going to develop because I think long-term memory is actually going to be one of the key ingredients to making these assistants not just helpful, but truly indispensable in this modern world. Next

up, I want to talk about my favorite app in the entire genai space recently, which is Cloud Code. I think it's unbelievably capable and this ability for it to spawn multiple agents, use them subsequently or even in parallel, each one of the agents essentially being a powerful prompt or persona that you have is unmatched. Especially when you start adding MCPS to the specific sub agents, it's just unbelievably powerful. And I've been really obsessing over this stuff. Like I find myself staying up way too late just trying out new think experimenting, really obsessing over AI again. Like I haven't for a while now. And they've been shipping a lot. So, I want to give you a quick roundup on what they released recently because I think what's absolutely amazing here is that the community comes up with new and better ways to use it and within a few days they ship features that do exactly that. One of those examples is the core agents feature that I just talked about that people used to do it manually through scripts in our community like two or three months ago. Now there's a agents feature for two or three weeks now. Then there's this other workflow where people have been using Opus to actually create the plan for what clot code is going to do. and then sonnet the cheaper and faster model to execute that plan. Now this was cumbersome because you had to switch model you had to do the plan then you had to switch model again execute the plan switch model to continue planning and so on. You could also do this through custom files but it was tricky. Now they have a feature where it does just that. It plans with opus every time and it executes with sonnet. Plus they shipped the upgrade to the sonnet API where it can now handle up to 1 million tokens. So that's up from 200,000 tokens before. And all of this stuff compounds. Okay, so if you have multiple agents working in parallel, before the plan probably came from Sonnet and they just had 200 tokens each. Now the plan comes from their smart orchestrator Opus and they have five times more context each, meaning you can run five Sonnet agents simultaneously with a million tokens of context each. And then they deliver their plans to orchestrator, which is Opus, to analyze everything and create the next step of the plan. This thing is becoming a powerhouse. And I'm not quite ready to show you some of the things that I've been building and playing around with. Don't worry, I'll do that soon enough. Right now, I'm actually holding a community event later today, showing the community some of the tricks that I learned in Claude Code and some of the things that you can create with these parallelized agents. But yeah, I just wanted to start to sneak that into this new show because my personal feeling is that this is the next big phase of AI. I don't think it's going to be GPT6 that's going to completely change the game. I think it's these applications that run multiple agents in parallel and that actually work and can actually manage millions of tokens and you get to lean back and let them work instead of just AI assisting you as you work. It's still a long way, but I think clot code is the closest application that we've seen so far to actually delivering that. And while it's focused on development, people are bastardizing it already for writing life management and all other types of use cases. If you're curious about that, I would recommend you check out this claude code page on X. They share different projects and releases from the community and Enthropic. They're not affiliated with Enthropic, which at this point, I also want to point out that neither am I. They did sponsor us for the Claude 4 release, but honestly, we would have covered that anyway. And I've been very consistent in communicating that Claude are just my go-to models when it comes to development ever since Sonet 3. 5 really. But you make up your own mind. I'm just here to share what I find, and I always want to be transparent about any type of bias or involvement that I might have. I'm also just human. But yeah, this is just the number one app that I've been using over the past weeks. Hey, quick note. If you're enjoying this content, don't forget to leave a like and subscribe to the channel. Genai to my surprise is still a lightly technical and very niche topic and every single like helps out the channel. And with that being said, let's get to the next piece of news that you

can use. And then we have one more story which is this tiny tiny texttospech model. It's called kitten text to speech and what it does turn text into speech. Nothing revolutionary you might think. Let's have a little listen. — Hey, have you heard about kitten TTS? It is a realistic open-source texttospech model with — not that impressive, right? Considering some of the competition and now there's one reason I'm actually highlighting this and showing you that this exists because this thing has been going crazy on hugging face and the reason for that is one this thing is fully open source and two it's around 20 megabytes. Okay, what would I even compare that to? What is 20 megabytes comparable to? It's five MP3 songs. It's four to five highresolution photos on your phone. It's a 10-second video. It's nothing, okay? And it's fully open source, so anybody can build this into their app, run this offline, whatever they want to do. I just thought it was really impressive that a model that is, I don't know, okay, not very good to be honest. But this is the level we reached like 3 4 years ago. The only thing that was free was the Google Translate robot, which sounds like this. — My name is Eagle. How are you? And honestly, while this might not be much better, it's tiny and fully open source compared to the Google product, which I believe is only accessible through the API, right? For this search, I actually want to switch to O3 because it searches the web incredibly well. Yep, as I thought, those are cloud only. So really, I just love this trend of AI democratizing a lot of these technologies and making them available to anybody to build on. I think people are widely underestimating how much power AI is going to give back into the people's hands through software releases like this and many others that we cover over time. Sure, the Frontier models are always going to be in the hands of the big labs. And there's going to be certain things that only those can do. But as they raise many of the crumbs

become widely available, and that leads into this week's quick hit segment, which are all the stories that I thought were worth covering, but maybe not worth talking about for a few minutes. So, we just quickly brush over them. And I'll start with the first one which relates to the kitten text to speech story because we got a first tease of Grock 5 coming out later this year. Interesting. Can't wait to see that. I mean, just as a reminder, benchmark wise, Gro 4 is still crushing it on certain benchmarks. Just adoption is not there yet. I think that might change with time because of moves like this. As they progress, they made Gro 4 free to all users worldwide right now. So, if you want, you can just try it out right now. Make up your own mind on how you like the current model. And this is what I'm saying. As these labs race each other, all of a sudden the consumer is benefiting in that race.

Another quick hit from this week is Mjourney releasing HD video. This is 720p HD and it can create this type of quality out of their video generator which is actually super good. I think Mjourney kind of carved out their space in the video generation space. It's just got the static quality of the image generator and now you can do higher resolution. Up until now it was just 480p. So that's nice. As we're talking

about AI video, V3 is finally fully accessible through the API. Up until now there was API endpoints but just for providers like foul where you got the whole thing in one and now you can do things like split out just the dialogue or just the sound effects. They have a specific preset for cinematic realism and it just really opens up all the options for people who want to create videos on mass rather than one by one.

And that story also leads into our next quick hit which is Google releasing their own version of a learning product just like Chachip did recently. So, not a model that just responds to all of your answers, but one that helps you understand and asks you questions and doesn't just give you every single answer right away. Plus, they opened up a program where they're investing 1 billion over 3 years for American education. This applies to some other countries, too, and you can apply for a free Gemini account if you're a student or educator or educational institution. And I just love to see that because these days, I really see the discrepancy between people who are just kind of oblivious to these tools and people who are really embracing it. That gap has been there for a while, but it's widening with stuff like claw code happening now. This is what I've been talking about for years. I don't think people can just catch up in a week or two anymore. You need to learn about prompting, which leads into context engineering. Then you need to learn the basics of automations and how different apps interact with each other so that you can understand how MCPs work and connect to the prompting component and then bring it all together in aic system where you create multiple agents and orchestrate them. Essentially, it's a ladder you got to walk up. And I think initiatives like this are going to help with catching some people up. And the advantages mission always has been and is to show you how to not get left behind and just climb that ladder, but get ahead of others through some of the newest apps. So that's what we'll keep doing on this channel by covering the last few stories we have this week, which actually is

another Google release. This is Jules. This is a asynchronous coding agent different from Cloud Code, Cursor, or the others. This is supposed to work simultaneously with you on a cloud server in the background. So, I haven't used this yet. I think this is more for larger repost rather than building your own little apps, but it's fully out of beta and available now if you want to check it out. And then lastly, we have

two quick releases. One from Luma Labs where they have their own video editing feature now just like Runway released last week. It looks very similar in practice where if you're on a paid plan, you can do these edits where you replace the background or add visual effects. This is a product category that is still just shaping up. These are the first iterations of it we're seeing. And then

finally, I want to end on the Matrix game, which is a opensource version of Google's Genie free last week, the world simulator that creates entire worlds and lets you control them with a computer. Now, you do need a powerful graphics card. I did some research. Seems like a 4090 would be enough to run this locally. This thing is fully open source. So, if you have a 1490, you can try running this on your own computer. But, it's funny to see a big release like Genie Free last week and then somebody follows up with a open source alternative. Love to see that. And that's it. That's what we got for this week. A lot of small releases, a lot to discuss. Honestly, I did not expect this much to happen in AI during an August. And if you're still watching this, I would just like to say a heartfelt thank you for your time and attention. I hope you get what you're looking for out of these videos. My hope is that some of this content inspires people to go out there, create, grow, or maybe just create more space in their life to have more time for what they really care about, because I know that's what AI did for me over the past years. Some of these apps make me just as curious as a little child. And there's an infinite amount of things to learn here, which I love. And I just wanted to thank you for making that possible by just hanging out here and learning together. And with that being said, I hope you have a wonderful rest of the week.