GPT-5 Confirmed, Huge ChatGPT Upgrades & More AI Use Cases

In this week's episode, we finally have information on OpenAI's GPT 4.5 and GPT 5. Some much requested ChatGPT features, documents mapping out various AI use cases amongst all industries and more releases, inspiration and things you can put to work today. In this week's episode of AI News, you can use the show that looks at all the AI releases and filters for the ones that you can use right now. Or I guess this week we'll be talking about a few more that can really help inform your choice of tool or what to plan for in the coming months. All right, let's get into it. Starting with the GPT 4.5 and GPT 5 roadmap. As tweeted by Sam Altman on X, this has to be the most requested thing that people have been asking for since GPT4 release pretty much two years ago now. Wow, time really does fly. Damn. So the first big point on here is that the next Release will be GPT 4.5 that they internally were referring to as Orion. There were a lot of rumors around this model. Now we finally concretely know what it represents. It it's the very last non thinking model that they will ship. After that they'll be progressing to unifying their offering into just one experience rather than what most AI assistants look like today, which are these model pickers where you kind of got to know what these are good for. I mean GPT4O GPT4O scheduled tasks 103 mini 03 mini High 01 Pro more models so Sam Altman here is basically saying this is coming to an end because what will be referred to as GPT5 will be a system that will integrate all of this tech into including their state of the Art O3 thinking model into one focused experience where ChatGPT will figure out if you need a thinking model or not, if you need one application or another one by itself. Now I thought this comment that I also found over here was important to add into this because this is not going to be a model router that will be deciding if you're going to be picking GPT 4.0 or 01 or 01 Pro. No, it's going to be a unified experience, one big model that does both of it. Interesting. And when asked about the estimated time of arrival on these, he said weeks, months, which in my books means before summer essentially that'd be my guess, maybe even sometimes before April. Who knows. Okay, one final note on pricing plans because that was also touched on here and in some other tweets so plus subs will be able to run GPT5 at higher levels of intelligence than free users. So free users will get unlimited chat access to this GPT5 model that will be coming at the standard intelligence settings. Plus subs will get a higher intelligence level, and Pro subscribers will be able to run it at an even higher intelligence level. Essentially the same thing they have right now. Free subs get GPT4O pace ups get 01 and then pro subs get 01 Pro and then, sure, now we have 03 Mini, which complicates this whole ordeal, but you get the point. And as of Deep Research, they announced concrete numbers. It's going to trickle down to ChatGPT plus users, which is super exciting because I think everybody should at least give this feature a shot. It's super, super powerful. I still think it's the most powerful AI tool that is shipped recently. Still playing with Operator, but Deep Research is just so much more convenient for most people. And two users per month in the free tier. My guess would be that this will ship together with GPT5, so most plus users will have to wait for another month or two or three. Okay, and now let's get to the concrete updates that have shipped inside of ChatGPT this week that you can actually use today. So my favorite is this one right here. OpenAI01 and O3 Mini. And O3 Mini High now supports both five file and image uploads in ChatGPT, meaning you can reason over your documents. This is the most convenient thing. I have so many markdown files with context on various target audiences and my company's roadmap and things like that. No more copy pasting of that. And if you want to add research papers, company PDFs or even complicated diagrams and infographics to your thinking models, which, by the way, they really shine at working with. We'll look at that in a second here. You can do that now. Another minor feature upgrade is that you can now create projects from your mobile phone. And that's just all of these features kind of spreading between the mobile apps and the desktop apps. Usually it arrives in the web version first, and then in the coming weeks, it trickles down to the rest of the platform. So in case you're missing a feature on the mobile app or desktop app, just give it a little bit of time. They'll probably follow up with it soon. All right, then, I want to round out this OpenAI update segment out by pointing out some best practices they published for their thinking models and how to use them, how to prompt them, what what to use them for. I figured that this would be worth pointing out as the central theme of this show is what to use these tools for, right? And one of the examples here they bring up is it's really good at reasoning over complex images like this architectural drawing. And if you open up this link that I'll include in the description below, they give you tips on how to choose the best model. And I thought this was most interesting. When to use our reasoning models. Here's this example with the visual reasoning over this architectural plan that includes a lot of details. This is the type of thing that if you tackle this problem with GPT4O, it will you would often miss crucial details and you'd be like, why is it missing it? And I remember myself being in that situation a lot, just looking at the results and being like, why did it miss this detail? It clearly shows the abbreviations. Why can't it use these on this image? And sometimes it works, right? You run it 10 times. Sometimes it works, sometimes it doesn't. These thinking models are really good at doing that consistently just because they loop over themselves and ask themselves, okay, what else could I do? Did I consider everything? Did we consider all the abbreviations? Plus they include some tips on prompting the reasoning models effectively. In my eyes, there's nothing groundbreaking here. I think we've talked about most of these either in videos or live streams on this channel. We had some thinking model focused live streams, so I guess most of these were covered in there. But there you go a lot from OpenAI this week. Love to see it. Now let's move on to the next piece of AI news that you can use. Okay, next up we have the first public release of VEO2. Now I think this is generally considered the most capable video generation model that has been teased by Google but never properly released. And the first time and seeing the light of day here is through a YouTube feature which unfortunately I don't have access to yet and I haven't been able to make it work on my phone. But if you're in the US you can just open up your YouTube app and you should have access to VO2 in their so called Dream Screen feature which allows to create AI generated backgrounds for your shorts. And friend of the channel Bilawal actually put this to the test so I just took the freedom to borrow some of his footage of a ginormous Chihuahua running across Times Square I guess. So again, just to recap, this is the most capable video generation model in the world today. I think a lot of people would agree on that statement and they're shipping it in a way that I think is sort of usable for many people. If you're a short creator or you've been looking to create more short form content, you can now empower it with these AI generated backgrounds. You can even prompt from multiple clips and stitch these together. This is only available in the U.S. canada, Australia and New Zealand at the moment, coming to more countries in future. So, yeah, there you go. I guess that means we're going to see more AI in your Instagram shorts, TikTok YouTube shorts feeds, because people will just be using this and re uploading to others. Not exactly sure that's a good thing, but I guess we'll just see. So ever since I started my first YouTube channel, which is 13 years ago at this point, the younger version of myself had certain dreams, like getting a video with a million views, hitting 100k subscribers, and most importantly, getting a sponsorship from Raid Shadow Legends. Ladies and gentlemen, we made it. Let me tell you about Raid Shadow Legends. So if you're not familiar, Raid has been a gaming phenomenon for years now, and they just keep adding more and more to make the game better every year. This game has AAA graphics, endless content, and you can play it anywhere, whether you're on your PC or your mobile device. And what I really like about it is the PvE mode where you can play through the story that includes all the campaigns and dungeons. And all of this strikes a great balance between challenging gameplay and pure fun. But you can also play it with multiple people. So if you want to team up with your friends and family, the clan system will help you get connected with others. I even started an in game AI Advantage Clan. Now you can do that yourself for your own family or friend group. Heck, you could even join the Advantage Clan under the name the AI Advantage. And Raid is actually running a special event right now. It's called Alice's Adventures. Oh, and it's inspired by the classic Alice in Wonderland fairy tale. So if you ever loved Alice in Wonderland growing up, this. This is a fun twist of epic Raid gameplay. There's five new legendary champions, including Alice the Wanderer, the Mad Hatter, the good old Cheshire Cat, and the Knave of Hearts. If you're playing, make sure to challenge the Queen of Hearts in the event dungeon for rewards, like the pinpoint gear set, available until March 5th. And if you log in for seven days before March 26th, you get Alice herself for free. So one little tip is that having Alice in the Beginning is a real lifesaver. I wish I had that when I began. So go ahead and download Raid Shadow Legends using the link in the description or scan the QR code that is on screen right now. New players get a ton of bonuses like two exclusive Epic champions. First of all, Drake in the Lizard manufaction immediately and Knight Errant from the Banner Lords faction after hitting level 15. One more thing is that if you actually sign up now, you can use the promo code Monkey King to unlock the legendary champion Sun Wukong 2. That's one that you would really want if you go into the PvP arena. So including Alice from the events, that's two epics and two legendaries right from the start of the game. In other words, there's never been a better time to start. So use the QR code or click the link in the description to get to download Raid Shadow Legends today. All right, There you go. YouTuber achievement unlocked. Now back to more AI news that you can use. Okay, so this has been shared by Anthropic and I really, really like this. It's not an application, but it's a mapping of all the different ways AI is being used in the world. We're talking real world tasks here and I love to see data like this because it contextualizes so much of the stuff we do on this channel. There's a lot of interesting stuff in this article like the median wage of the various jobs that are really affecting affected by AI usage. I mean, CLAUDE literally mapped out how many percent of the conversations were within a specific profession. So for example, almost 2% of all conversations were by copywriters with copywriting related tasks. And as you can see, the median wage for all the tasks that people run inside of Claude is around $60,000. Super interesting data in my opinion. But this graph here at the top, I thought this was the most enlightening to me because it showed the different sectors titles, especially the various tasks that people use the with AI assistants. Look at this. Computer and Mathematics leading with 37%. 37% of the use cases with AI span across computer and mathematics. Then you have arts and media, education, administrative, social sciences and life. And then in the end we have business and financial use cases. And I think it's fair to say that on this channel we generally focus on five out of six of these here and there. We do cover very basic computer programming tutorials, but mostly those are targeted at people who've never worked with with code before. This is just not a developer channel. That's no Secret. But if, for example, you're a business person, which I know is a massive cohort within this channel's audience, you can look at this and you can see, okay, these are the top tasks that are being performed with AI as analyzed by Anthropic, which is one of the biggest players in the game. So number one would be analyzing financial data and developing investment and budgeting strategies. Makes a lot of sense. Code interpreter is super powerful and with the thinking models you can really upload sheets and ask for insights that go beyond something that you might be able to pick up on within minutes. Provide personal finance advice and education. I've been there before, asking on how to budget and how to track expenses in light of changing variables within my life and business. Or how about this one under Education and library? Designing and developing comprehensive education curricula and materials. 100%. It's so good at that. We use it to structure guides outlines for courses. We use it to design quizzes from various video lectures we run in the community. The educational use cases here are vast and this show is mostly a first look at the new things that come out and we test and compare some of the releases. But in this case, if you ever wondered what AI is actually being used for, I got to say this article right here by Anthropic is one of the best ones I've seen on that topic. And I'll absolutely be bookmarking these and referencing this in future public speeches, educational materials. I really enjoy this, so I figured you might too. Attention everyone. We are gathered here today to discuss Zonos, our new open source text to speech model. All right, so this is the new application and I wanted to show you this one because it does some things for free that up until now were only behind paid walls. So basically Zyfra here is a text to audio application applying human sounding voices to your text. But you also have this clone voice feature for free that I will now test for the first time. So let's give it a shot and see how this works. I'll need my headphones here. And it should also be noted that they actually published this model under Apache 2.0 license, which means it's open source and we can generate 100 minutes a month here free for free. I just logged in with a random Google account right here. So let's try this clone voice thing now. Okay, so I will just speak some random messages here. We can kind of skip over this in the edits. Just let me speak for a minute so it can get a feeling for My voice and create a clone. I will just read off some random text that I see. Maybe this. Hello and welcome to Zonos. Okay, seems like that is my recording right here. I'll call it Igor 1 and once I select this voice, I'll change the text so it's not too easy. Type out a little message, hit generate. Welcome to AI News that you can use. This is a free voice cloning up. You can now bookmark. Oh, I even misspelled app, but that was pretty good. It might not be state of the art, but it's free. And that was just from a 30 second recording. Some of the 11 labs voice that I have and we sometimes use that. I gave it I think 90 minutes of audio recordings. If you use their tool that also takes under a minute, it will be a similar quality level to this. And this one is free and open source, so that's pretty neat. On to the next one. Okay, so this is just a follow up on lechats that we covered last week. Mistral, AI's ChatGPT competitor. Hit enter. One, two, three. Holy moly. Okay, that was super fast. Just as a summary, they have non thinking models that are competitive and the main thing that sets them apart is the speed, which is unbelievable and accessible to you on the free tier also. So if you want the fastest AI assistance, then this is probably your best choice right now. And now you can also get it as a mobile application application. They prepared a little comparison video here showing you the speed and I demoed it last week and I can confirm that that's pretty much representative of how these applications here behave. As you can see, it's not even close. I mean, this video goes on for a minute and the lechat completion pretty much is done in the first two seconds though. Look at that. So you can get the app for free on your Apple or Android device and you can go ahead and impress your friends with how fast your AI is compared to their slow one. If you have friends that are into that type of stuff, I guess. Okay, next up. This is quite quick, but you might have seen the super bowl ad by OpenAI. Very neat ad covering the history of technology and how we are in this technological age with AI. But the most interesting thing here to me is that Mike Bespalov here on X actually created the little app that allows you to create exactly the type of visuals that you saw in that ad with a tool that he fully built with OpenAI tools, namely O3 mini. Honestly, it had me a little confused why he says Video Sora here. Because there's no Sora API, nor does Sora get mentioned anywhere in his code in here. Maybe someone can clarify that one for me. Either way, he built a tool with O3 mini to create an ad just like the one that they aired during the Super Bowl. And you can try it yourself here. If you just go to this link. All the code that O3 mini wrote is up here and you can just use it down here. So look, I can just change the grid side or even upload my own images. How about we do that? What about my good old trusty picture of Keanu Reeves that I somehow always have laying around on my desktop? I don't even know why, but I guess it's just a thing at this point. And look at that. It's not a video, but like, yeah, that's Keanu in this dotted effect. So maybe change some values here up the contrast. Keanu is sort of getting lost in here, but yeah, that's pretty good. We have some video laying around. This was a little thing we did in the community where we hooked up ChatGPT and text to speech to 3D avatars and a game engine. This one's too long though. How about we go with something more simple? Oh, this could be nice. I remember, yeah, this was one of the tests I ran where I transformed this graphic into a little lens with. I think it was Luma Labs beginning and end frame feature. Okay, let's do this. Oh, wow, look at that. This looks beautiful. Yeah, there you go. That's a camera lens in the end. So maybe increase the grid size a little bit. How would this look? Still looks like a camera lens. All right, there you go. I think you see the point. Just wanted to show you that this is another way you could be using these tools to recreate effects in multimillion dollar commercials. Love it. He even shares the prompts that generated the tool down here. If you're curious, link in the description below. Okay, next up. I just wanted to shine some light on this story because Google published a series of 30 seconds videos showing off various generative AI use cases. And that is kind of the theme of this channel and this entire show. So I wanted to show you this. You can scroll through it. I thought some of them were really inspiring. Some of them were very sweet and wholesome, and you just have these very well produced videos that are 30 seconds long on various AI use cases. So if you want to hear a mini story about translating conversations live, you can just click on this. I actually don't speak a lot of other languages, but Gemini and Google Meet it translates everything live. Honestly, I think this is a pretty neat way to explore a few new use cases for yourself and they obviously they're promoting their own AI tooling in the process, but just thought this was neat and wanted to point it out here. Okay, super quick mention of a new paper that came out. I usually don't feature these because Labs like Google DeepMind keep publishing new and new research, but I thought this one was particularly significant because it focuses on AI for math. And although this is not something you can particularly use right now, I just wanted to break down this preconception of AI not being good at math because that is really changing. It started with the thinking models and now it progresses with things like Alpha Geometry 2 here by Google's DeepMind that is actually achieving gold medalist performance on some of these math tasks. And that's basically the summary of this entire thing if you're in a space that relates to mathematics. I think at least reading the abstract of this and then looking at some of the results this achieved and how powerful AI tool can already be today may not be released, but you can kind of include this in your worldview. But I do like looking at these things that look a bit ahead too because you can already start including them in your worldview before the general public does. And that for me so far in my relatively short 30 years on this planet has been a winning strategy that I can only recommend, but given it takes a certain life set up for you to have the time and the energy to invest in things that might be around the corner and not here yet. But I just wanted to point this out and I'll keep sprinkling in little tidbits like this throughout future episodes because although both scientific research or on the other end of the spectrum, dramatic news stories surrounding AI might not be the focus of this channel and this show. I think some of these are so significant that I'll kind of mix them in here and there with the goal of keeping around 80% of the content really practical and use case focused. That's sort of the formula I came up with and I hope you enjoy that too. And that's pretty much all we got today. If you want to learn more about the Deep research feature, which in my opinion is the most capable AI tool out there today, I created a standalone video on the channel showing you various use cases on how to really put that 200 doll tool to work. I thought that was one of the best videos we've created in recent memory. So go check that out if you haven't seen it yet. And otherwise, I will see you next week for another episode of AI News that you can use.