Transcription
July 2025. Remember this month, the AI landscape as we know it is about to change forever. A recent tweet from an OpenAI insider hinted that it's surreal watching people use Chad GPT today, knowing what's coming next. They've definitely seen something that we haven't until now.
Today, I'm diving deep into the most detailed breakdown of Chad GPT you will ever find. Release timelines, capabilities that make Chad GPT4 look like something ancient, and why even the creators are absolutely stunned by what they have created. If we haven't met yet, my name is Serban Sida. I'm your friend here at Taft. We're fighting to give you the best and the newest AI tools out there. In this fast-moving future, only firsthand insights will keep you ahead. Taft is the world's leading platform for AI tools with a newsletter that's being delivered to close to 2 million people. This newsletter is the best you have ever seen. Join us by clicking the link in the description and you will not regret it.
So, you might be wondering why July 2025. Well, while some people predict GPT5 in late 2025, inside whispers and even Sam Altman himself point out to mid-2025, more specifically summer 2025. This is not just another update. OpenAI does not do slow rollouts. Think back to Chad GPT4. Remember at first it was silence but then boom, a revolution overnight.
So what makes Chad GPT a gamechanger? First of all, we have advanced reasoning. Every OpenAI model now has step-by-step reasoning baked in, including Chad GPT4, 4.5, even their coding tools. The reasoning traces are shrinking, which means that the AI is actually working more efficiently. And number two, we have coding dominance. Here's a little-known fact for you. OpenAI's very own team uses their coding tools more than any other external tool. This says a lot about it. When the creators of this tool prefer their own AI as opposed to any other tool out there, you know you've crossed the threshold. At number three, hallucinations might be solved. GPT4 hallucinates around 30% of the time. The public thinks that this is unavoidable, but this is actually just an engineering challenge. GPT5 is expected to slash hallucinations to under 15%. And at number four, we have the true Omni AI. And here we have real-time two-way audio, meaning that it will actually record audio. Right now, you cannot speak with Chad GPT because it doesn't understand sound. It just makes an assumption of what you're saying and it translates it into words. With GPT5, you're going to be able to communicate with him just like you're communicating with any other person out there. And then we have ultra high-definition image processing and also video processing and potentially very, very likely generation. Its native voice will sound so smooth that Siri will sound just like a toy. Think about it this way. If it's digital, GPT5 will master it better than any human out there.
All right. Next up, I want to talk to you a bit about the staggering scale of GPT5. GPT4 runs on 1.4 trillion parameters. GPT5 is rumored to have more than one quadrillion parameters. Now, there are a lot of experts who say that it's most likely closer to 50 trillion. Still, it would be a lot more impressive. This is 30 times more than GPT4. And Sam Altman himself said that the era of scaling parameters is over with GPT5. This suggests a major shift in how models such as GPT5 are built. It doesn't mean that bigger equals better anymore. It also has to mean that the model should be also smarter and not just larger. And he explained this by using the diminishing return problem. Research shows that after a certain point, adding more parameters will just give us smaller results compared to the cost of computing. And Sam Altman basically saying that OpenAI has reached that point. And from this point forward, in my opinion, I think that the biggest shift will be reasoning over memorization.
So by mid-2025, we are probably going to enter an era of fully independent AI agents. These agents will help us with things such as seamless workflow automation, real-world API connections, and computer operations faster than any human out there. As for the market potential, I think it's basically limitless. It's infinite. Every computer out there, every laptop user, everybody. And I think this changes everything.
Now let me give you guys some leaked benchmarks which I would say they are beyond human perception. First of all, we have the MMLU, which stands for Massive Multitask Language Understanding. It basically tests reasoning across 57 subjects such as STEM, humanities, law, mathematics, physics, chemistry, biology, engineering, anything you can think about. A human expert benchmark will have a score of around 89%. GPT4's score is 86%, and the projected score of GPT5, it's around 95%. Next up, we have the SWE Bench, which is for software engineering. It's basically checking the ability to solve real coding issues. Now, the GPT4's score was 32%. Claude 3 Opus was 44%. Now, as for the GPT5, it's rumored to have an 85% score, which will surpass any human developer out there. Next up, it was tested for advanced mathematics, which is pretty much PhD level. It basically tests problems from very, very complicated math equations. GPT4's score was around 25%, and the projected score of GPT5 is around 50%, which will pretty much compete with any human with a PhD. Next up, they tested the multimodal tasks. It pretty much tests image and video understanding, for example, charts, diagrams, video reasoning. Now, Chad GPT4's score was around 75%, which is pretty good, to be honest. But then we had Gemini 1.5, which got the score up to 85%, and the projected score of Chad GPT5 is more than 90%. Which means it's pretty much near flawless when it comes to cross-model analysis.
So, as you can see from these numbers, this is not just better AI. It's an AI that outperforms almost all of the humans on planet Earth in pretty much any cognitive task. Now, there's a secret that keeps researchers up at night, and that is solving the scaling laws. Right now, they know exactly how much compute and data is needed to reach artificial superintelligence, aka ASI. And right now, that's not speculation anymore. It's math. So, Ilya Sutskever, this guy right here, he is OpenAI's co-founder. When he left in order to focus on ASI only, it was a silent signal. That means that we are closer than anyone admits.
GPT5 will not be just a better AI tool. It will basically redefine AI as we know it right now. AI agents are going to be working while you sleep. Coding will replace entire development teams. Reasoning will surpass experts, and we will have a multimodal AI that sees, hears, and creates. Just like a digital god, every AI today is the worst than it will ever be. Every AI tool that's in existence today is the worst that it will ever be. Let that sink in for a second. The future definitely belongs to those who adapt first.
Now, while a lot of people are debating if AI will actually change the world, we, the team behind There's An AI For That, we are already building for the post-GPT5 world. If this video opened your eyes, hit that subscribe button and show this video some love by giving us a like. It would greatly help us with the YouTube algorithm, and you're going to be the first who will be notified when we are posting new videos. Before signing off, I just want to let you know that the future is much closer than you think. Until next time.