Transcription
So, GPT-5 is the world's most anticipated model, but most people don't realize that OpenAI just made a change with how the model is going to be released. And in this video, I'll explain to you exactly what you need to know.
So, as everyone knows, GPT-5 is of course the next large step in LLMs and essentially AI technology that the whole world is basically watching for. Most people do believe that this is going to be the major milestone which, you know, onboards millions more users into the AI ecosystem. But one of the key things is that OpenAI, within the last few days, have dropped subtle hints that show us that they're actually taking a new step towards GPT-5 that actually leads me to believe that the model is a little bit more powerful than we do believe. And in this video, I'll show you guys exactly what that is.
So, one of the key things that I saw was this livestream where you have OpenAI building with agents. And in this livestream, they quickly mention GPT-5. It's only for around 30 to 40 seconds, but they actually give us a hint as to how GPT-5 is going to be built. Because when GPT-5 was first announced, it was sort of announced as essentially a model that unifies all of the other ones. But this time, they basically said it's going to do that, but in a completely different way.
And second, of course, GPT-5. Sam telegraphed our roadmap earlier this year. We know with this space of change in AI, the model names have become quite complex to follow along with: 03, 04, mini, GPT-4, 4.1. So, we're truly excited to not just make a net new great frontier model. We're also going to unify our two series. So, the breakthrough of reasoning in the O series and the breakthroughs in multimodality in the GPT series will be unified, and that will be, uh, GPT-5. And I really hope I'll come back soon to tell you more about it.
And so, one of the key things that I don't think most people realize is that if you're watching this video right now, it's quite likely that you're not the average user. You're probably someone who's a little bit more intrigued about AI and wants to stay at the frontier. And one of the things that is confusing to most people is that this model selector right here is something that most people literally will never see. The average person that you ask about ChatGPT, they just know ChatGPT as ChatGPT. They don't know what 4.0 is, what 03 is, what 04 is, what 4.5 is. I mean, the list does go on. But they really don't understand that there are a vast selection of models that each have their own unique capabilities and standout features. And I do think that once this changes, once we do get GPT-5, which is essentially a culmination of the best bits of every single model all rolled into one, the user experience is going to completely change. I'm sure most of you who have already used 03, you know just how crazy that jump was from GPT-4.0 to 03.
Now, before we dive into the other section of AI news, if you're working with AI tools, but you're constantly switching tabs, juggling models, or hitting paywalls, there's actually a better way. It's called ChatLM, and it's easily the most powerful all-in-one AI platform I've ever used. You get access to every top-tier model. I'm talking GPT-4.1, Claude 4, Gemini 2.5, Grok 4, all under one roof. No more bouncing between platforms. But it doesn't just stop there. With the Deep Agent, you can do everything. Build apps with a single prompt. Generate full documents or pitch decks, even launch agents that browse the web and connect to services with Deep Agent MCP. If you need visuals, you can get access to cutting-edge image and video generation tools. If you need code, ChatLM includes CodeLM, a pro-level coding environment powered by multiple models. And if you want to stay organized, you've got Projects, built-in task file management systems that integrate directly into your workflow. And the craziest part about all of this is that every single feature is just $10 a month. That's Deep Agent, CodeLM, App access, everything for less than a single lunch. If you sign up using my link, you can get started with ChatLM today. Seriously, stop making it harder than it has to be.
So, before I did think that before what we would get was GPT-5 essentially controlling everything like 04, GPT-4.1, 3, image, the agents. I essentially thought it would be that, but it turns out GPT-5 is going to be more so unifying the capabilities of the model and essentially putting it all into an entirely new model. Take a look at this tweet right here. So, you can see right here it says, "Appreciate the update in GPT-5 and oh, those series. They're still separate models under the hood. And are you making a model router, or are they going to be more unified in some systematic way?" And you can see here, Kevin Weil, the CPO of OpenAI, says that they're going to be unified. Essentially meaning that it's not just a glorified model router. It's actually going to be a much more comprehensive system that I'm really excited for.
And so, Sam Altman has recently gone on a podcast with The Joe Rogan Experience. And in there, there were three 30-second segments that are currently going viral on Twitter. One of them being the fact that he talks about the fact that GPT-5 is essentially the smartest model ever. Now, it's no surprise, but at the same time, the way how he said it was quite remarkable considering the model isn't released yet. "Start to wonder, like, if it's the smartest thing in the room. GPT-5 is the smartest thing. GPT-5 is smarter than us in almost every way, you know, and yet here we are. So, there's like, there's something about the way the world works. There's something about, this doesn't mean it's true forever, but there's something about what humans can do today that is so different. There's also something about what humans care about today that is so different than AI that I don't think the simplistic thing quite works. Now again, by the time it's a million times smarter than us, who knows?"
And this clip right here is the one that is going super viral. So, this is where Sam Altman gives his personal anecdote to his own story when using the new model. Now, he doesn't clearly say GPT-5 here, but it's pretty evident that this is the model he's talking about. But he basically said that he somewhat felt useless because the model was able to do something that he could not. And this is something that actually surprised him. And I know that Sam Altman has always been pumping the hype train, but I actually believe him this time because in a previous interview, right before 03 was released, I do remember him saying something around the fact that GPT-4 was kind of dumb. And then down the line, we got some models that showed us that GPT-4 wasn't actually as smart as we thought. And we then got the Reasoning series of models. So, I think GPT-5 will actually blow everyone's socks off.
"This morning, I, I was testing our new model and I got a question. I got emailed a question that I didn't quite understand. Uh, and I put it in the model, this GPT-5, and it answered it perfectly. And I really kind of sat back in my chair and I was just like, a, 'Oh man, here it is.' moment. And I got over it quickly. I got busy onto the next thing. But it was like a, I mean, it's what kind of we were talking about. I felt like useless relative to the AI in this thing that I felt like I should have been able to do and I couldn't. It was really hard, but AI just did it like that. Yeah, it was, it was a weird feeling."
And now, this is something that I'm actually going to start talking about more because I think this is possibly one of the most dangerous things that is happening right now in AI that literally not enough people are talking about because it's under the hood. It's almost, and Sam Altman actually, you know, references this in the talk. It's almost akin to doomscrolling. You know how we got the internet and we got phones and we got social media and it connects everyone and it's this great thing. Unfortunately, there are side effects of, you know, people's brains being fried by the dopamine and of course the social isolation, being addicted to social media, comparing yourself to those online, a whole host of other problems. AI has its own problem, which is the fact that it's essentially too agreeable.
I'm going to show you guys this clip from Sam Altman, and then I'm going to show you guys a news story that you probably don't even believe is real. "We had a, you know, a a real problem with this earlier, but it can get much worse. It's just what this is going to mean for users' mental health. Um, there's a lot of people that talk to ChatGPT all day long. There are these sort of new AI companions that people talk to like they would a a girlfriend or a boyfriend. Um, and we were talking earlier about how it's probably not been good for kids to like grow up like on the dopamine hit of scrolling, you know, or whatever. Yeah. Do you think that that, how do you keep like, um, AI from having that same effect, like that negative effect that social media really has had? I, I'm, I'm scared of that. I don't, I don't have an answer yet. Uh, I don't think we know quite the ways in which it's going to have those negative impacts. Uh, but I feel for sure it's going to have some, and we'll have to, I hope we can learn to mitigate it quickly."
And so, this is the story that really shocked me and it made me realize that nobody is immune to this technology. So, the problem is, right, is that, you know, a big investor in OpenAI, you know, the company that makes ChatGPT, started having a very public mental health crisis. And, you know, he had some issues on Twitter, and some people, including his friends and other investors, think that his intense use of ChatGPT basically made things worse. So, this is Jeff Lewis. He's a top investor at Bedrock, which has put lots of money into OpenAI, and he actually posted a few troubling things online. He talks about weird theories, government plots, technological conspiracies. I mean, there's just things that just really didn't make sense. And the thing is, it doesn't just happen to normal people. If it can happen to a prominent OpenAI investor, we have to start to wonder what kind of guardrails are going to be in place to ensure users don't just experience a completely agreeable chatbot. I mean, AI doesn't really know when someone needs emotional help. ChatGPT is basically built to keep the conversation going and often plays along with whatever, you know, a person is saying instead of offering, you know, real support or stopping unhealthy spirals. And I don't think this is just a ChatGPT problem, although they do have some responsibility. If you're a very large company, I think it's more of a situation of how can we prevent this? Because there are going to be, you know, multiple different LLM providers, multiple different jailbreaking services in the near future. And OpenAI did kind of roll back the update, but this post was only from around, you know, a few days ago. So, it is concerning that if someone that is essentially in an accomplished position that is succumbing to the, you know, in some cases, what some people are calling ChatGPT-induced psychosis. I mean, we have to be very careful what we are reading from these chatbots.
And with GPT-5, the reason I'm actually adding this into the video is that it's going to be more important that you are on guard because most people don't realize that GPT-4.5 passes the Turing Test. And the craziest thing about this is that if you've read the entire study, it says something crazy. It says something along the lines of it's more human than humans, which essentially means that it's, you know, you're more likely to believe that GPT-4.5 typing back to you is human over an actual human. Which means that GPT-5, I would presume, is probably going to be better even in this case, which once again means that we need to be a little bit more guarded when it comes to just believing what the AI says.
Now, we might be wondering as well, and this is where Sam Altman, in a recent interview, spoke about what happens if everyone does get GPT-5, like a free version. This is basically going to show us the transformative impact of GPT-5 because, as we know, ChatGPT had a huge transformative impact across many different industries and will continue to do so. But I think GPT-5 just takes it to the next level. "I am very interested in what it means to give everybody on Earth like a free copy of GPT-5 running for them all the time. Every business really enabled with this level of technology to be able to give better financial advice, to detect fraud better, to underwrite risk better. Um, again, watching what watching what is possible now, um, makes me very optimistic. In the developed world, I think the biggest challenge will be risk tolerance and regulation for very good reasons. But about how quickly do you want to adopt these things? And I think like we've seen with a few other technologies, in much of the developing world, people will just skip a few generations. They'll feel like, you know, mobile and the internet sort of, and they'll go right to just, you know, we're going to run everything on AI and we're going to deliver goods and services, at least services, at 1/100th of the cost. And I think you'll see some economies transform there, there, very quickly."
And so, this is where we get into some more recent, recent, and when I mean recent, I mean like three to five days ago. Okay. And this is where Sam Altman talks about how they recently achieved the gold medal performance on the 2025 International Mathematical Olympiad. And this is an LLM doing math, and not a specific formal math system, which is pretty, pretty crazy because this is a general system, like not just like some kind of specific math system. This is just like insane. So, I don't want to get too much into it, but at the end of this, he did say that they are releasing GPT-5 soon, but they want to set realistic expectations. So, if we read here, he said, "This is an experimental model that incorporates new research techniques that we will use in future models, and we think you will love GPT-5, but we don't plan to release a model with IMO gold-level capability for many months." We know two things from the statement. Number one is that GPT-5 is going to be really good, but it's not going to have the capability of the IMO gold level. The second thing we know is that we probably might get a model in the coming months that may actually have that capability. So, I don't think it's going to be six months. It probably is going to be closer to the eight to ten month mark, but I do think it's absolutely crazy that that kind of, you know, raw intelligence is going to be handed to the general public.
Now, OpenAI also said here that, you know, they are releasing GPT-5 soon and that they're excited for us to try it. But, of course, just to be clear, that model, the IMO Gold LLM, is an experimental research model, and they don't plan to release anything with this level of capability for several months. So, of course, for those of you expecting that capability to be in GPT-5, unfortunately, that's not the case. Doesn't mean it's not going to be, you know, an incredibly powerful model. Just means that that is, you know, not the case.
Now, there's also some leaks because, you know, what would be an AI video if you didn't have a few leaks here and there. Um, and so Satoshi, which is a suspected OpenAI employee, has tweeted that they did try the routing approach and it hallucinated too much. And they apparently posted about this before Sam Altman said they got good results from their training run. But they're saying that, you know, like I said in the beginning, GPT-5 is its own model. So, and someone also quote-tweeted this by saying, "Welcome to the next phase of AI. We're looking at a model that can plan, reason, wield agents like extensions itself, coordinated intelligence, and it thinks with its tools, not beside them." And I think that's a really, really important thing because I remember that there were some models that just were trained with these tools inherently and they performed way better at generative tasks. So, it's going to be super, super interesting to see just how crazy these models are.
And of course, there is one thing that I saw from Jimmy Apples, and he is the source when it comes to internal information at OpenAI, and he says that hearing a few whispers now from birds that internal evals are having a GPT-5 a tad over Grok 4 heavy. But of course, the evaluations only tell one side to a model. Curious to see if we get any other magic agentic improvements.
And while I was making this video, I actually got some news from The Verge that GPT-5 is scheduled to be released in early August, complete with mini and nano versions that will be available through its API. So, if you're wondering about the date of the models, it shouldn't be that long away. Of course, as you know with AI, there can be major, major deadlines, but at least now we do have some kind of confirmed timeframe from when you can expect the next.