Sam Altman Just Revealed MORE DETAILS About GPT-5 and AGI

So in a recent interview, Sam Altman let some new details about GPT5 emerge and it leads us to believe that GPT5 is going to contain more capabilities than we originally thought. So in this video, we'll dissect the Sam Altman interview and he touches on some rather fascinating points that many people hadn't discussed before. But without further ado, let's dive into the first clip in the interview because it is really interesting.

Look, when we decided to launch ChatGPT, we thought it was going to do well, but we had no idea how well it was going to do. We knew they were going to be great eventually, but we didn't think they were good enough to resonate like ChatGPT and GPT4 have. This technology, even with all of its current limitations, is far more useful than we thought and can integrate into our lives in, in a much more valuable way than we thought. Now that we know that, as we think about launching the next much better models, we come with a different perspective. I learned something about important and urgent problems and not letting the important but not urgent ones fester. Here's the thing, I'm going to go get done in 2024.

Now what I found fascinating was that that clip was actually circling around Twitter and I think people did kind of miss the mark on what Sam Altman meant when he was talking about this because there have been some key changes in how Sam talks about the AI. And when I say AI, I'm talking about GPT4 and GPT5. And that is because I think what Sam is trying to do is downplay GPT4 capabilities. Now the reason that I think he's doing this, and this is just a hunch, but hear me out. I think that Sam Altman knows that, you know, these open source companies are catching up to GPT4's level. And meta has even said that they publicly, you know, are going after the GPT4 benchmark. Google has previously gone after the benchmark with Gemini. But I do believe that Sam Altman saying that, you know, like for example, he was saying here that GPT2 was good, GPT4, you know, was okay, is basically saying that, you know, GPT5 is going to be pretty incredible. But of course, as we do know, something that you do need to remember is that it might be a GPT 4.5 release. Because I believe that if these systems are as even good as some of the internal people have claimed to be, then I think the shock would be a bit too much for the industry. And essentially what that would mean is that they might face scrutiny for, from regulators and AI safety teams. Because of course, as you know, with every iteration in the models, there's a significant jump in ability and also emerging capability. So I'm gonna just play this bit again because he says even with all of its current limitations, and I mean, think about it like this, right? If you're a company of an AI company, like you're the CEO of an AI company, all right, you're not going to be out there publicly saying that, you know, even with all of its current limitations. But GPT4 does have limitations. And you have to remember this is a model that has been out for quite some time. So I, I think the fact that he talks about the fact that it's far more useful than what we thought and this is why I say, okay, it's pretty different because he said that now that we know and we think about launching the next much better models, we come with a different perspective. So that is why I say it's going to be more fascinating. And even in the recent video, we did talk about the fact that Sam Altman, when he talked to Bill Gates, he was essentially saying that GPT5 might be a more personal AI system. Now, if you, if you don't believe that that is the route that they're taking, you may have missed the secret update in which they did a memory update. Now, if you don't know what that is, I'm going to show you guys a quick screenshot. Essentially, if you've used ChatGPT for quite some time, you may or may not have seen this screenshot right here. Now, this screenshot is not a leak, it's not a false, it's not fabricated, this is 100% real. Because even Greg Brockman has tweeted about this. The OpenAI employees have tweeted numerously about this. And, and essentially what we have here is the memory update. So essentially it says your GPT can now learn from your chats, keep the conversation going. It's going to carry what it learns between chats. So even if you start a new chat, it's going to know what happened before. It's going to also improve over time, which means it's going to remember details, remember preferences on how you like to be spoken to on certain things. And of course you can reset your GPT's memory and the custom instructions. So, you know, it just goes to show that Sam Altman and OpenAI in the future, what they're planning is essentially more personal based AI systems because I do think that a personal based AI system is definitely going to be more useful and definitely get more engagement from the user because I think when people have seen how effective the Rabbit demo was for having an AI agent that is personally custom to you, I think people have realized that especially with character AI as well, how crazy that platform is taking off. It's slowly catching up to ChatGPT's users and all it is is a website where people can talk to AIs. I think the level of a personal AI and I also think that Siri and Alexa and Google's, I don't know what it's actually called, but Google's home assistant have actually lost the mark in terms of not being able to catch up to what is state of the art right now. So I think this memory update for ChatGPT and Sam Altman saying that in future models they're going to be focusing on personalization and having an AI system that helps you out. I think that is going to be really where the key focus is. Now there was another clip in the interview that Sam Altman did actually talk about which was rather interesting.

So GPT2 couldn't do very much. GPT3 could do more, GPT4 could do a lot. GPT5 will be able to do a lot, lot more. And, and the thing that, or whatever we call it, and the thing that matters most is not that it can, you know, have this new modality or it can solve this new problem. It is the generalized intelligence keeps increasing and we find new ways to put that into a product, we find new ways to use it. But that's, that is the high order bit, I think that dominates everything else in the importance is that the overall capability of the model, its overall intelligence, its ability to do longer, more complex problems, more accurately, more of them, that is increasing across the board. And that to me is one of the few things that make this totally different from any previous kind of technology.

So yeah, you can see right here he actually spoke about GPT2 could do a lot, GPT3 could do a decent amount, GPT4 could do a lot, but GPT5 will be able to do a lot, lot more. So, you know, publicly speaking on GPT5 and saying it's going to be able to do a lot, lot more means that I think, you know, especially since if you have looked at the recent benchmarks, you'll know that, you know, these models, I wouldn't say they're capping out at their capabilities because the capping out the capability is 100 and the last time we checked, I think the last MMLU benchmark was 87%, which is very good. And of course, Google got 90% on Gemini Ultra. I think the point Sam Altman is trying to make here is that these models are going to become more accurate, less hallucinations, more effective at retrieving information from documents and long pieces of text. All of this stuff is going to be increasing over and over and over as they move forward. And I think with that generalized intelligence and being that smart, I think it's really going to usher in a next level in AI. And like Sam Altman said, you know, it's what they focused on. And it's not about the new modality, which could be photo, it could be video, but having a model that is really, really smart and having that general intelligence. Because if you remember what Sam out once spoke about was the fact that they are trying to create, of course, artificial general intelligence. And if you want to get to AGI, you are going to have to, of course, have a system that does have different modalities, but you will need to have a really sophisticated system that doesn't fail, you know, as much as these systems do, and they don't feel that much, but the instances where they do mean that they can't be used in certain industries. And of course that is a hurdle that they need to get over. So I think that really fascinating stuff from Sam Altman discussing the capabilities of Future models like GPT5, or like he said, whatever they might call it, goes to show that we are going to be in for perhaps a very good shock in terms of the model's capabilities. And I would presume that they are going to release GPT 4.5 just after the Jumbilite release, because I think once Google gets the spotlight, OpenAI are going to quickly take it back and then Google's going to have another mountain to climb. Now, Sam Altman did also talk about something that I thought was rather fascinating because I think people were going a bit too crazy because essentially what Sam Altman talks about in this clip, the interviewer just asked him, you know, what's going on with the military use? And some. Altman in this clip basically says that, look, okay, I don't think it's wrong for the military to be able to, you know, reorganize some files using our, you know, LLMs if they want to. So I don't know why people were going crazy with that because as always, sometimes people run away with a narrative. And I think this Clip clears it up. But like I said, it's rather interesting to see how we, you know, navigate.

This question that allows open AI models to be used by militaries. Why did you make that change?

Because what we tried to do is say here's what you can't use the models for rather than here's this group of people that can't use it. There were parts of the Department of Defense that had, I think super legitimate, super good use cases for our models and a blanket saying, you know, anyone whose address ends in dot mill can't use this. I think it's bad. I'm like very proud citizen of the us I'm a huge supporter of liberal democracy continuing to do well. We don't want our models being used to like make kill decisions of course, but there's a lot of other stuff that the military does that's quite important.

And you've got some great use cases that you talked about helping, you know, prevent suicide by veterans. Obviously if someone giving a speech at West Point wants to translate it into Swedish, that's probably not a high risk use. Then you have stuff at the other end, you know, developing a nuclear weapon, clearly against your policies. But there's so much in the middle. There's so much that is not building a bomb, it's not destroying property, which is another one you have. That's allowed. That's not allowed rather. But that could be really harmful. I mean these AI engines are incredibly, you know, capable persuasion engines. How do you draw the line of what a military can and can't do do? And do you think that's the right place?

One of the things that we believe very deeply is that society and this technology have to co evolve. We believe in iterative deployment a lot for the obvious reason, which is that people need time to gradually update and think and figure out what the rules should be. But there's another part which is so.

That part right there. The reason I stopped that clip is because like I said, okay, we don't want to essentially shock the entire, you know, industry or the entire population. We with a new crazy model that's able to do 100 million different things. And I know that's a vast exaggeration, but the point is I'm trying to get you all to understand that if OpenAI came up with GPT5 tomorrow and it was an AI agent, it had 90% on all benchmarks, I think people would be freaking out about this rather than being astounded by just how good the model is. Because I think People would realize just how quickly we are accelerating to a point where humans are no longer needed. And that wouldn't spark inspiration generation, it would spark fear because the timeline would be too quick for humans to, you know, get their brains wrapped around the fact that, you know, in just a couple of years they might not be needed for any, you know, real tasks. So I think what Sam Altman is saying here is that, you know, humans do need time to adapt. You know, you do need to be able to, you know, process the emotion and the feeling of, you know, societal change. You know, societal change can't just happen overnight because that just isn't good for anyone. And you know, I'm pretty sure that they know this and they understand that. Look, even last time when GPT4 is released, there was a huge letter that all of these, you know, company owner signed. You know, Elon Musk, Steve Wozniak, all of these people, you know, did sign because they were like, whoa, okay. We went from GPT 3.5 to GPT 4 and that was a huge, huge increase, you know, in such a short amount of time. So they were like, whoa, whoa, whoa, whoa, whoa, we need to slow down, we need to stop releasing anything other than GPT4, GPT 4.5 to GPT5. So Sam Altman's basically saying that, look, okay, we're going to make sure that we need to be able to co evolve and of course iterative deployment, which essentially means that the next model is likely to be GPT 4.5 and after that it could be GPT 5. But I think, okay, and don't quote me on this, but I do think that they might limit some of the capabilities and I think maybe that's why they nerfed GPT4's capabilities. Just because I think the adjustment period is definitely something that would be better for humanity in the sense that people do need time to adjust. Because if we go from, you know, GPT 3.5 to GPT 4 and then people are like freaking out and then we go to GPT5 and GPT5 is just crazy. Like people can be like, oh my God, like, you know what I mean? So I think, you know, iterative deployment is definitely something that they are leaning to and of course giving society the ability to co evolve with the software and make that transition is definitely a very good point. But here is the rest of the clip because it is, you know, rather fascinating.

Society and this technology have to co evolve it. We believe in iterative deployment a lot for the obvious reason which is that people need time to gradually update and think and figure out what the rules should be. But there's another part which is it's it that you can't separate the technology from the world. You can't just, even if you get everything magically right, you can't build it in secret and then put it in the world all at once. Because the world is going to keep changing with each iteration. Which means on those middle cases I don't know what the right answer is yet, nor does anyone because no one has really like thought through and seen how the, the institutions and the world and society shift and reshape in, in response to this. So there will be a lot of things that we'll have to start slowly on and iterate as we go and there'll be a lot of middle cases. But we do want to support the U.S. government and other governments too. And like I find it odd that everyone thinks this is a big gotcha question where they're like going to say wait, are you saying you support the US government? And my answer is now essentially the.

Clip did get cut off there, but essentially Sam Altman's art answer is yes, of course he does support the US government because he's a US citizen. So I think that is a complete given. Now does he support everything the US government does and you know, watches all their political moves? I don't think anyone does. So let's just not get into politics. We are staying on the topic of AI technology but like Sam Altman said that society shift, you know, they're going to start slow. So like I said, I think they might even drip feed us some of the capabilities because I think in some cases it is best to do that because like as Sam Eltman said, we do want the ability to get everyone up to speed as things do change. Because I don't think you're one day GPT4 and then the next day everything is automated. And I don't think that's how quick things are going to go. But I do think that in response to GPT4, the way society reacted does show us. Okay. And I think the fact that right now what he's saying about this and saying the fact that we do need iterative deployment does actually show us that they did make a giant leap. Because with Q Star and all the leaks and the fact that Sam Altman said it was an unfortunate leak does show us that they are moving faster than they are letting on. Because think about like this, right? If you're letting, if you're moving really, really fast. Okay. And you know, you don't want your company to be regulated into the ground. You're definitely going to go in small, small steps because you want it to be a gradual transition because that's good for everyone. But if you're moving slowly, you could just release the next model and the next model and you're moving gradually anyway. So I think this is obviously the chance here, but this does show us that AI is moving rapidly. And that's of course, just an inherent nature of the field.

And of course it allows open AI models to be used by military.

Then of course, there was something here. And I think this just goes to show us on the back end, like, I mean, in four to five years when this thing is built, how crazy OpenAI is actually going to be in terms of its company size. I think it's going to be a monolith, because if they're able to get this off the ground, that would really change everything if it is as effective. So take a look at this because this is where they talk about raising for other stuff. But Sam Altman basically said that what I was doing was an OpenAI thing.

I think there was this misrepresentation, not hard to guess by who, but that I was like off in the Middle east raising money for this chip effort. And then it was somehow like a SAM project. It was like an open AI project that the board had decided was a clear strategic priority. And it's not a separate thing. So, you know, I'm not going to go try to, like, fight everything in the press because you all will do your thing and I respect that. But that was like, wildly misreported. These are like, these are critical to OpenAI efforts. I don't invest much in personal startups anymore. Like, I don't invest personally much in startups. I do support the ones that I invested in previous to this things. Like, in general, if there's like a startup that I supported and they're raising more money, I try to, like, continue to support, but, like, I'm not. I don't think of myself as an investor anymore. That was our thought. I mean, that was that. That is the whole reason we have the strategy of iterative deployment. And I think, I mean, I think we've been right so far to do this.

So, yeah, I think we all need to pay attention to this because essentially one thing that most people don't realize is that the current chips that we do have, the current graphics cards and, you know, processes that we do use to, you know, run and train. These models aren't designed for AI, okay? And there are companies out there that are specifically designing stuff for AI, and if they're able to get that stuff effectively, I honestly, for the life of me cannot remember the company that is working on it because there is so much AI news that I honestly look through. But the company was able to say that they were able to get things going 100 times faster based on, you know, current, you know, small preliminary results. And if they're able, like, think about it like this, guys. Long story short, if they're able to get the training runs down from like a couple of months to down to like a couple of hours, I mean, that would be a pace that is unprecedented. And I mean, that would have serious ramifications. Now, I know that sounds like a dream. It might not even be possible anytime in the near future. But if we do manage to increase the speed of the processing and just the specific AI stuff, that is going to be really, really, really incredible in terms of just the speed, because we already know that the breakthroughs are happening in the research, but the training runs are definitely something that do take a long time. So for that, that would be a major breakthrough. But of course, like they said, this is what he said is an OpenAI priority. So it wouldn't be surprising if in the next couple of years we do get some breakthrough in chip design that allows us to do these training runs much faster and of course, allows us to iterate even quicker. And of course, we do have this statement. So around the Internet and around the AI communities, there has been the consensus that Sam Altman recently has been downplaying AGI. Now, if you know what AGI is, you'll know that it's going to be the most transformative technology within the decade. And like the tweet says, Sam signed a letter saying that AI extension Central rish should be a global priority. And he also said that the bad case is it's lights out for all of us. Now he hints at AGI like it will be a nothing burger. And it's curious how closely his words match. Whatever, what's most convenient at the time. So I think what Sam Altman is doing here, and of course this is just pure speculation, we have no idea why he says what he says. There could be a variety of reasons, but I think closest to the model, you don't want to state that, you know, bad cases, the lights out for all of us, this is an existential risk. But now he's basically saying that look AGI, you know, because basically in the clip, I'm not sure if I'm going to show it on here, but he said that, you know, when we have AI, the world will have a two week freakout and then people will go back to their lives. But before he basically said that, you know, this is going to be a really transformative technology. So either way, I think, you know, AGI is going to impact anyone. Because if he, if the company does succeed and actual AGI, like they actually do it, they get it done like an actual AGI, that's good. It can, you know, do a bunch of stuff, it can do research, yada, yada yada, it can do everything that, you know, the DeepMind says, you know, like a competent AGI could do that's better than 99% of humans in tasks. I think that will change the world fundamentally and I don't think we go back to a world without it. So I'm not sure how people are going to just go back, go back to their lives because their individual lives and sense of purpose is going to be changed. So I'm not sure how that's going to work. But I do find that fascinating and captivating that he actually did say that. So maybe it's just downplaying it, I have no idea. But there was also a comment, okay. From another YouTuber that I do watch that has really riveting stuff, but I would take a look at this because it's rather important. So essentially, David Shapiro, someone that creates some of the best AI content on the web, essentially talked about how Sam Altman said, AGI is coming soon, but it won't change the world. And that really just doesn't make sense because he says, I don't know. To me that's like saying that the invention of electricity or steam power or the printing press didn't change the world much. And his personal read is this. And either he's toning down the rhetoric since the OpenAI drama. So a few months ago he was of course fired and then rehired. And maybe the PR teams are saying, you know, just don't spout off and saying that AI is going to, you know, do crazy stuff. Because to be fair, there are many news articles out there that like run with any, like, any narrative that, you know, AI is going to destroy us. So I mean, it is wise to not go down that route at all. And then someone, you know, the second point is that he's finally realizing the massive rapid occasions of what they're doing. And they have no idea how much is going to change the world and they are legitimately worried, hence the firing and attempt to sync OpenAI by the former board. And it's a combination of these and the other factors. So do you think the AI will change the world much? And then there was a poll and most people do think AGI will change the world much. So I mean by definition it's going to like by the definition, if you've seen DeepMinds, you know, classification for what they view AGI as. So I would say that AGI, of course, if it is real, if it is effective, if it is able to be done or certainly change. But I think Sam Altman is is right to say what he says because companies will honestly run with it and we don't need another PR disaster slash OpenAI board drama. So we'll be interesting to see if he does change his tone in the future and how, you know, certain things do get translated. But I'm really excited about GPT 4.5 or GPT 5, whichever one is released and let me know what your thoughts and comments are because we are definitely moving very fast into an uncertain future and it's important to pay attention.