GPT-5 is a Game Changer for 99% of Users

So today's a big day. It finally happened. GPT5, you can see it right there, is finally out. It's available in chat GPT. It's available through the API. And I've been waiting to make this video to tell you all about it and to discuss what this means for most users since well, early 2024 really. That's the first time they kind of teased it. What we're going to talk about today is first of all that this is the brand new model that is going to be the default in chat GPT. Matter of fact, they nuked all other models. This is the only one you get to choose from here on out. No more thinking selection. And also, it's available across all plans. So, if you're a free user, this is probably the happiest day for you when it comes to AI releases ever because they're making this available to everybody, ranging from the free all the way to the $200 pro plan. What my plan is for this video is I'm not going to run elaborate benchmarks. not going to be building esoteric games and physics simulations and seeing if it's incrementally better than what we saw before. I'll have multiple examples in the end where we'll be comparing it to the competition. I think that matters. But mostly I want to answer this one question that I have written down on my paper here. And that is what does this mean for most people? your mom? What does this mean for people at work? What does this mean for students? What does this mean for most people that don't spend their entire free time nerding out over AI and figuring out which model is the best one in this particular week? And I think there's a lot here, okay? Not just that this is available to everybody, but it also brings new capabilities and merges a lot of old capabilities into one system now. And my hope is that this piece of content is going to help you navigate that. And it's and also I hope that this is going to be a piece of content that will help people who might have been sleeping on the releases of the past few weeks or months catch up and really get up to speed so you don't feel behind and you realize where we're at in AI space and how to use this to your very own advantage. So with that being said, let's dive right into it. I'll start out with summarizing what this release is

all about. There's a 90-minute live stream if you want all the details. I'll give you a quick summary here of what matters. Then we'll be moving on to some of the facts around it, like what's new and how that realistically compares to some of the competition and some of the competition's best releases. Matter of fact, this week, if you're not familiar, we had two big releases with Opus 4. 1 anthropics best model coming out and Gemini deep think finally coming out. They announced this um a few months back. It's out and then OpenAI came out with this model which is better at most things. That's kind of the summary from what we've seen so far. But in the end of this video, we'll be going in and we'll be practically looking at a few examples that matter to me. We won't be looking at some of these esoteric benchmarks or like test cases that nobody would really be doing. I want to look at a few things and that's what we're going to be doing. I want to look at the writing style. I want to look at its ability to generate ideas and recreate one of the demos, the one with the castle we saw in the live stream. I thought that one was really impressive. I want to see how well Claude does at that. We're going to compare that. So, without further ado, let's dive right into kind of the summary and the overview of what GPT5 is about and what you can expect. Before we do that, I want to say one last thing and that is where I'm coming from as you know your presenter here today. And I want to point out that I'm coming at it from a power user perspective. At this point in time, I did the rough calculation. I have around 2,000 somewhere between two and 3,000 hours just inside of chat GPT. and its competitors like Claude or Gemini. That's not even including the development, the API side and usage of these models. I've been doing this every day since the release day of chat GPT. I really I'm an obsessive type and I kind of went all in on this. So, I'm going to be giving you my perspective but with a lot of empathy for somebody's perspective who might just be coming in or might just be catching up with this topic. Okay. My primary use cases, I think this is important, are not development use cases. They're not medical I'm not doing physics with this stuff. I'm writing, I'm planning, I'm ideulating, I'm using it for entrepreneurship, marketing, I'm using it for psychological tasks. I love that. Those are my personal interests and that's where I find it really empowers me. So that's kind of the angle that I'm taking here. a power user who is a more common perspective on this rather than one of a developer or one of a physicist or mathematician evaluating these models. So first of all, this is the flagship model from OpenAI and it's going to be replacing all other models in Chat GPT. Again, no more other models. That mess that we had there with the model selector and thinking models, non-thinking models and coding specialized models, and this one is good at psychology, and this one uses web search. Well, no, that's all gone. GBD5, that's all you get. And it automatically selects between the various tools it has. It's still it has all the tools now. So you get uh web browsing, you get custom instructions, you get the canvas, you get the code interpreter. All of that is built into one thing. If all of those don't mean much to you, don't worry about it. You're kind of just by using this product, you'll kind of learn about them. For example, the canvas is like a word editor inside of chat GPT and custom instructions is sort of like giving it an extra prompt in addition to what you're already sending it. But I think the massive thing here is and we'll get to kind of the benefits of this whole thing for the users here but I really want to point out I think for people who are not extremely versed in how to use these models this is going to be so much more intuitive right you don't need super long super precise prompts to get good results out of models like O3 previously and this just builds on that okay this is not release chat this is not GPT3. 5 from 2022 where it really makes a difference if your prompt is this long or this long. Um, and the shorter prompts are just not going to give you the results. This works differently. So, I think it's a huge upgrade for free users. And that's going to be kind of the story of this entire video. I'm going to be kind of pointing that out from multiple perspectives. Yet, power users are also happy to see this because one amazing thing is there's no more thinking. There's no more selection between, hey, do I want a thinking model? non-thinking model? Matter of fact, this is not even too new. If you paid close attention inside of chat GPT over the recent months they had this change where even GBD40 was thinking at times it never spent minutes thinking but it often spent like 10 15 20 seconds thinking. So they already implemented this ability to automatically decide if it should just give you an answer right away or if it should think for 15 seconds before it gives you an answer. Now this is fully integrated and you don't even get a choice. It just decides on its own if it should be thinking or if it will just give you a quick answer. If you ask it's a fact of like how many people live in New York like it shouldn't be thinking for a minute or two. Um some of the previous models like 03 Pro that question right there would have thought for like five minutes and then it gave you the number that GP40 could have given you right away. So now it's all merged into one thing. You just get the answer with hopefully the right thinking depth. Now let's talk benchmarks. I pulled up the releases

from this week that matter which is uh Entropics. These are the main two competitors. Thropic and Google Gemini, their benchmarks in some areas are still better. Uh GPD5 like long story short, GPD5 wins on most of them, but on specific benchmarks, these other models are still better. For example, on AM 2025 mathematics, well, deep think reaches 99. 2. If we look at uh competition math right here, you'll see GBT5 with no tools is at 94. Excellent result. But yeah, Gemini wins on that one. Agentic tool use um anthropic wins on one of these and loses on the other one of them. Okay. And then there's a few that um open didn't even include like terminal bench which is the terminal usage there. Um I suppose Claude would just win otherwise they would include it. And then there was one more actually I pointed it out MML which is multilingual QA. So like if you want to do multiple languages, these other models might be performing better than GP5. I mean there has to be some trade-offs, right? So they didn't even include those. On some of the other things, a lot of the coding benchmarks, it just straight out wins. Okay, but again, we're not here to talk about benchmarks. I just wanted to highlight that they're best on most of them now, but not all of them. And some of these competing models when it just comes to benchmarks still win on some on certain measurements. So what really matters is how it performs in the real world. What really matters is, hey, are you going to be gravitating towards this product? Is it going to be making your life better? you more productive, freeing up more time to do things that you actually care about rather than maybe some data analysis or some research and tedious tasks at the computer? And that's what we'll try to answer in the last segment of this video where we look at multiple examples. But now I want to actually talk about some of the features and some of the things they pointed out during the live stream. Okay, so I have like three bullet points right here that I want to kind of touch on and discuss and then we'll get into the demo part. So one of the things that this will be doing is that it's supposed to have way less hallucinations. I think that's actually really exciting and that's the type of announcement we haven't seen out of any competing company yet. I haven't seen Google come out or anthropic come out or any of the Chinese models come out with like an announcement that hey this model hallucinates way less. It's more reliable. Open AI is kind of a first on that. And when I talk to people, it's one of the top things that I hear from people as a concern about AI. Like it hallucinates a lot. How can I trust it? I can't trust it, right? It's just going to confidently give me wrong facts. I still need to like get the right facts. I can't just like it. I don't want it to lie to me confidently. So, it's doing better on that. They quantified this in some numbers. But I think this is one of the things that only time will show. I don't dare to kind of like judge this today from the first results from the release. But that's really an exciting direction that OpenAI is taking. I hope we see more of that. Secondly, they're saying it's more emotionally resonant than before, meaning it has very high EQ and meaning it will pick up on the user's behavior. Matter of fact, there was a release um an announcement from this week where they're actually including a little box that pops up if you're talking to chat for too long that says like, "Hey, are you sure? Like you want to keep chatting? It's been a while. Maybe take a little break. " And this in combination with this announcement or that GPD5 is supposed to be even more emotionally resonant than before. I'm really excited for it. One little side note is they had this um model called 4. 5. Most people didn't use it. I actually used it a ton. I really liked it for all my psychological use cases where I kind of like simulated different scenarios or I had different smart people discuss an idea that I was interested in or I used it as a coach on several like entrepreneurial decisions on all those. The 4. 5 model was the best until Claude for Opus came out. Then I switched to that because I really just like the human aspect of it. The way it sounds, the way it gives concise advice and doesn't like overexlain. I really like that. But 4. 5 was amazing and this is supposed to mirror those capabilities. When they used this language, when OpenAI used this language for the first time, it was the 4. 5 release and now they said it again, emotionally resonant. So you can expect that it's going to be a better friend. u human. It's going to be a better AI. It's going to be a better companion for humans. I think that's the right wording right there, isn't it? Okay. So we can expect that and we can look at some of that in the writing example here in a second. But I have three more bullet points here. One of them is um also an interesting one that we've never seen before. It's supposed to be less deceptive. Meaning that when it doesn't know the result, it's not just going to stand there and pretend like it knows it every single time. Sure, still going to happen a lot and it is a big problem. Sometimes it just doesn't know how to get to the right answer and it just gives it to you and it's like, "Hey, here it is. " And then you're sitting there like, "Okay, cool. Thank you. " You work with it. One time I pulled up a quote that I was even using in a lecture and then after doing some research and like trying to find the source of the quote I realized it completely hallucinated that thing and it confidently presented it to me as if it was like it made like it used a real institute of like some standardization and whatever but like the quote was made up so that happens and now it's supposed to be better one more thing two more things that are supposed to be better um it's way better at medical use cases apparently this is something I can't test right Now, I don't have the expertise, but we'll put it put together a set of test prompts and then either in tomorrow's news you can use video or in next week's video, our show that we do every Friday that I do every Friday. U me and the team research like all the AI releases and show you the ones that you can actually use and then u practical applications for them. Shout out to news you can use. You can check that out. Subscribe to the channel to check that out every single Friday. But medical use cases we'll be looking at there. I think you kind of need to get a bit more nuanced and test a bit more in depth to really have an opinion on that. And then lastly, um I want to point out that actually only the pro plan is going to have deep thinking. Now, if you're not aware, deep thinking is like this capability that you can enable in other assistants or you could already enable. I mean, Gemini literally released a model called deep think now. So, that's what it does by default. and Claude, you can kind of go in here and yeah, do extended thinking. They had this capability already. Now only the pro plan is going to have that. So that's a $200 plan and it's sort of the equivalent of previously uh 03 pro if you're familiar with that. Also only available on the free $100 plan, $200 plan. So that's where it like thinks for five minutes before it gives you a result. Probably only relevant if you're doing maths and physics, maybe development. I think on development it can be actually really useful but nobody really used OpenAI models for development up until now. They used Gemini or claude. This might change now because like they made a big point out of development being a big thing. And those are all the kind of features you can expect. Now two more unexpected

things that they mentioned and I think most people overheard. So I think these are good and then we'll get into the practical part here, right? That most people are probably waiting for. So first of all they said there's going to be GPT swift voice. So if you're not familiar, GPTs are their chat bots within chat GPT. You can give custom instructions. They can do actions. Matter of fact, in our community, we've been building interesting things with GPT since the day of release. And even now, we're doing things where like you can daily journal with the GPT and then it saves it to your own note takingaking app or database and then you can pull in like previous notes and like work with them. You can do really cool things with GPT. GPTs, that's just one example. They're getting voice mode. So if you want if you have a daily journaling voice node uh GPT that is going to be saving uh your journal entries to your very own database your very own notion for example or whatever you might be using now you can just turn on voice mode and like talk to it. You can just blabber as I am right now. You can kind of just do that. For me and many others that is the most natural interface. Um one thing about talking that I love is you can talk faster than you can type even if you're a fast typer. And a lot of times when I type, I feel like my thoughts get ahead of my uh ability to type. So talking is huge. And with GPTs, we're going to have GPD5, the connectivity, and the voice mode, which is amazing. I think this is really going to like revive GPTs a bit. I hope. I love GPS. I still use them. They're so useful. And talking about connectivity, this is my very last unexpected point that some people might have missed. Somewhere in the middle of the presentation, they just quickly mentioned this. They're gonna have calendar and Gmail connectors in GPT5. So, you're gonna have a true personal assistant where you connect it to your calendar and your emails and you're going to be able to check in the morning. Maybe set up a custom GPT and just ask, "Hey, what's up? What are we doing today? " And it just gives you the rundown autonomously right there inside of Chat GPT. Um, for anybody thinking right now, for any power user thinking, um, I'm going to go over it's not shipped in my uh in here, we're going to be using it through the API. But for anybody thinking, hey, Eigor, don't we already have those connectors? Yeah, we do for deep research and the agent, but not for chat GPT. If I look at connectors right here, uh, where is it? Data controls connectors. You can see Gmail is connected, but it's only available for deep research and agent mode. next week. So that is like mid August. It's going to be shipping to all plans including um well the pro and the plus plan, not to the free plan. So if you're paying $20 a month, you're going to get to connect this to your calendar, to your emails, and use some of its capabilities.

capabilities. But now let's finally try it. Let's give it a shot. Few things I want to try here. Okay, first of all, I always really care about the writing style of these things. I feel like if the model sounds bad, I'm going to be using it way less just because so much of what I do is related to writing. And even when I just read the results and they just sound like crap, I don't like it. That's why I was or like I am a big Claude fan because it just sounds human. It sounds natural. It keeps things concise. I like that. With GPT, I really liked 4. 5. I liked 03 for its web search. But yeah, let's just give this a shot and have a look at a basic writing prompt that I just ran right in here. Um, I always use this one. If you follow the channel for a while, you might notice already. Write me an email to my boss about the broken coffee machine in our office. And then I just ran this in here and I'm going to put these up side by side. I'm going to have the Clawed Opus 4. 1 result that released a few days ago. And I'm going to have the GPT5 result here on the left. And this is going to be my very first look at the writing style. So exciting to see how this actually performs. Okay, so let's start by just reading this broken coffee machine in the office is the subject. And again, this is just one example. Please make up your own mind. I am going to give my opinion because that's why I'm here. No. Hi, boss's name. I wanted to let you know that the coffee machine in our office is currently not working. It seems to have brief briefly described the problem if you know it. Um, could we look into getting it repaired or replaced? Please let me know if you'd like me to contact maintenance or the supplier to arrange a fix. So, first of all, I really like how concise it is. I really like that there's no not a lot of EM dashes in this. I like the um it gave me a variable here though, which I don't know if I'm a huge fan of actually. I am. I think AI writing should be a bit more. It should be AI plus human. Most people that just copy pasted GPD4 outputs straight to jail with those. Jesus. Like nobody wants to read your generic like email on like oh let's dive into the problem matter or like here are three bullet points describing the like I don't want to hear none of that. This is actually good. I'm a fan of it. I had to think for a second but I actually like that it's including variables now. Um this sounds super natural and yeah otherwise super concise. I like it a lot. Let's look at claude here to compare because this was my go-to before GB5. Let's see. Coffee machine repair needed break room. Okay. So it just makes things up which is fine. like it often um like that can be useful. I think practically dayto-day this might be more useful. Let's see. I wanted to bring to your attention that the coffee machine in the break room has stopped working. It appears to have malfunctioned this morning and is no longer brewing coffee or dispensing hot water. So, it's making a bunch of assumptions here. Yeah. Since many of our team members rely on the coffee machine throughout the day, this is affecting productivity and morale. Okay. Reasoning. I like that. Several colleagues have mentioned having to leave the office to get coffee, which takes time away from their work. Okay, look. So, like at this point, I'm kind of like I'm a big fan of the fact that there's a variable here to actually fill in the things that GPT doesn't know from the concept. And here it's just making it up. Maybe let me run it one more time and see if it does that every time to give it a fair shot. But I think as is, I just prefer GP5's result here, especially with the variable. It's just going to make the internet a better place honestly because people will have to think a little more. Okay, let's see. I wanted to bring to your attention that the coffee machine in location. Ah, so here it's doing the same thing. So I suppose yeah, thing that I was a fan of. Okay, so I guess it just depends on your luck of the draw. Here's the optional description. Let's see how it rounds it out. The human factor. Please let me know if you need any additional information or if there's anything I can do to help resolve this. Yeah. So I think this is identical. This is very identical. So on this I would say the first one GP5 was clearly better. Now I feel like it's just a tie. So let's let me try one more. Um more chat prompt here with GPT5. Uh the chat model. So I want to see if it does the bracketing thing again or how this performs. So on the second try it also did the bracket. So I think this is a huge upgrade. We're just going to enjoy reading uh Twitter and LinkedIn and everything on the internet more because people are going to have to actually write stuff themselves from here on out rather than GPD40 like writing it in that horrible robotic style every time. Yeah, I like this a lot. So, writing first test I would say passed equally as good as Claude at the very least if not better. I think it's kind of on par. I'm really happy about this for the entirety of the internet. Seriously, like this is great. Okay, so that's one thing that I wanted to test. Secondly, let's do idea generation. Okay, to prompt this, I write a weekly AI newsletter for 70,000 subscribers where I cover the latest in generative AI tools and apps and prompts. By the way, if you care link in the description, haha, little call to action. Generative five, generate five distinct, varied, unique, and interesting articles, article ideas that blend emerging AI trends with actionable use cases for creators and businesses. And then I want a specific format for each one of them. So I want a title in quotes, a two sentence description of the concept and its subscriber value and at least one AI tool or new source to feature. Okay, we're going to run this. So basically five article ideas and then some uh specific set of formatting. Let's see how that does. And then I'm going to run the same thing in cloud and we're going to compare. Okay, hope you're enjoying the video so far. If you are, maybe take a second to leave a like. Um that really helps out the channel. But now let's compare some of these ideas. I'm just going to give you kind of my judgment as a content creator in this space that you know does this for a living and spends a lot of time with the tools. Let's see how good these ideas are from my perspective. So from inspiration to income, how AI is redefining creativity as a service. Um subscriber value readers will learn practical strategies to monetize AI content at scale while protecting their brand and IP. Okay. Using midjourney opens GPT. That's decent. What else is What else do we have? The rise of microAI. Why small models could outperform the giants for your business. Okay. GPT4 and Claude make headlines. Micro models can be trained for niche tasks. Eh, it's okay. Problematic setup. Like, okay. Not bad. Um, I'll give this one like a three stars. This one is maybe like 3. 5 stars maybe out of five. Something like that. prototype using AI to build and test products in days, not months. Oh, that's actually good. Complete mockups. Yeah, this one is really good. Figma's AI features. Galileo AI for not familiar. Runway for visual prototypes. Okay, that's good. I was thinking this was going to go more like builder. So, these are solid ideas. Um, I wouldn't say best I've ever heard, but pretty good. Deep fake ads and synthetic spokes people. This is really good. This is I'll give this one like a four stars. AI as your second brain building knowledge bases that learn with you. This is fantastic and maybe 4. 5 stars just based off the title. Check custom GPTs and vector databases. Well, but like you can't really combine a custom GP with a vector DB. Okay. So, this is just like hallucinating it. So, although it's supposed to do that less. Yeah. Okay. So, not bad. Let's see here. Rise of AI voice agents. I'm just going to kind of like brush over this. I've run this prompt many times in here. I would say generally my feeling is telling me Claude is a bit better at this. It's just a bit better. Just these I don't know. You tell me. But these titles, they just sound a bit more clickable, a bit more interesting, and I think it's even more relevant. So, but it re it's really good. I got to give it its prompts uh props. It's really good at adhering to the prompt. So, when you tell it something, it's actually going to do it. That's not the case with all models, by the way. body spin around the block like often you tell it like hey I want one two three things and it's like okay I'm going give you number one on the first one and then forget about the others for the rest and like it's a mess so prompt adherence seems to be really good here I like it the quality of the ideas eh I don't know I'm a huge fan of claw for ideas I think it's really good at that um first impression of this it's okay it's good um but I think the one that many people want to see is how does it demo with the games and how does a demo with the physics simulations. It looked really impressive in their presentation. But mind you, that presentation was crafted by one of the biggest AI companies in the world and they had they have hundreds if not thousands of testers and prompters and they have access to the user data and they have some of the best examples in the world. That's my point. So what really matters is how it performs for your stuff, okay? And how it performs in practice. Nevertheless, that being said, this damn example with the castle. Amazing. So, this is what it looked like in their example. I ran the identical prompt and this is what we got. I think you can kind of like play it and you can pop the balloon. So, this works. It's just graphically the other one is a bit better with the environment and stuff. I think they did pick an example that it would be exceptional at, but nevertheless, it's clearly better than Claude here. Where are the characters? Are they like inside of the castle? And I can't see them. Oh, I think it placed them like inside, but that's not really Well, how do I think these are the characters. Ah, yeah, that's the noble. But yeah, that's not amazing apparently. So, yeah. So, GBD5 does better on this. So, look, across the board, conclusion

time. Okay, what does this mean for free users? power users? For free users, this is unbelievable. Like, people are going to go from, you know, like the stone age. They only had GPT40 which is like over here. Okay, GPT40. GPT40 and power users with paid plans using this on a daily they were already using um 03 which is you know at its base like way better but then it could search the internet. So in some cases it was like way better. Okay. And now GPT5 is just kind of did this thing. It's a slight upgrade to 03 and all that stuff and it makes it simpler to use. But that combination of ease of use, it's a fantastic illustration. No, that combination of ease of use and its ability to actually um perform better as the base model and all the tools integrated. By the way, I have to move the camera so you can see this full thing. Make it a gamecher for free users. Okay, GBD40 to be fair was still pretty good, but not as good as 03. When you have technical difficulties and it just went into the web and thought for a minute and found like 20 articles and then gave you the exact response, it feels like magic. Of felt like magic. Claude feels like magic. Gemini feels like magic. GBT4 did not feel like magic. This is better than all the competitors and it's simpler to use. So I think for free users, complete game changer. Please show your relatives, get them on this on the train of people using this. I mean, we have 700 million weekly active users in Chat Chipd and now they're transitioning to this insane. Now, for people who follow this channel weekly, for people who use AI daily, for people who geek out over this stuff, this is an incremental improvement that is very welcome. I love that there's no more model selector. I love that it's it looks really solid in coding, and only time will show how good it really is. I'll be reporting back on use cases tomorrow and next week and the week after like heck this is going to be my new daily driver for now. Let's see how that goes. I think for development tasks where you use something like claude code and you really go into the deep end of the pool and you use five agents with multiple MCPs at the same time to build things. I think there like claude is still king just because that damn app works so well. But for everyday just codewriting this might be the best for creating quick apps. helping you with writing and everyday tasks. This is probably the best and everybody can use it on the free plan. So, this is going to drive adoption. This is going to make it easier for every single one of your family members to actually use it. This is going to show people what AI can really do in 2025 rather than showing them a little glimpse without all the tooling and all the stuff that makes it powerful. I think I personally think this is a pivotal moment for AI because the free users are going to be like what? This is possible. You guys were doing this all along. Or maybe they're not even going to say that because they're not aware. But models before they were insane. Most people didn't know that. Those 700 million weekly active users, most of them, they have no idea what the deep end of the pool looks like. They're about to find out. For everybody else, fantastic upgrade. Let me know what you think though in the comment section below. It's pretty much everything I have for today. It's a good little discussion. Go ahead and try it out. Should be available. I'll put all the links I used um in the description below. and I will see you very