MASSIVE AI News : AGI Secret UNLOCKED! o3, GPT5, OpenAI Moves Faster And Stunning Humanoid Robots

So one of the first pieces of AI news, because I really just want to get stuck right into this, is essentially this piece of news right here. Now, I know that this tweet is essentially a large part speculation, but I can't help but think that the timing of this tweet is quite specific considering we've heard time and time again many, many sparks of ASI from OpenAI employees. So basically, and I'm going to link this to another point because I was supposed to do that in another video, but actually did manage to forget. But it talks about how they got, you know, some read into what internally is happening at OpenAI and Holy Mother of God, I don't know how to express my feelings without sounding like hype. I don't know what to say, but I'll share this. The innovators are coming. The problem is we don't know how they got here. Now, that essentially refers to the levels to AGI. Most people actually do forget that this is the entire roadmap for OpenAI in terms of what they are setting out to accomplish. I think it's really important to know where the company's headed so we can actually understand where the industry is going as a whole. But essentially level four is the innovators and that's of course AI that can aid in invention. So this is going to be something that's rather, you know, fascinating because this is of course AI that does change the game. And it's going to be pretty interesting because of course, you know, right now, this is the year supposedly of AI agents. I've heard various things from different laboratories, but yeah, I mean, I mean, it's super, super interesting that now this is the innovators. Now I'm a little bit skeptical, I do have to be honest, because AI that can aid an invention. I mean, you know, at the first glance you would have to be skeptical, but we have seen inklings of this. When we actually think about how AI systems have been developed in the past, when we look at things like AlphaFold and stuff like that, when we think about how it manages to make stuff, I mean, I guess you could say invention is basically just search and just testing stuff out. So I mean, potentially I don't think that would be that hard considering the systems before this, the reasoners are actually built on a search based architecture. So maybe that could be there, but I don't know. It's a pretty crazy tweet, honestly. Now the tweet goes on to say that this is not 01, it's not 03, it's not 4.0, it's not GPT5, but that it could be an iteration or new version of an existing thing. And that's actually making me think that it is Somehow similar to AlphaFold in a sense, that it's able to create new things. And this says being vague and sincerely simultaneously is very difficult for y', all, yada, yada yada. But I think, you know, this is like somewhat true, because now that I'm thinking about something, I do remember that Sam Altman actually said something rather specific. He actually did say that one of the things that is going to be coming in the new year is something that a lot of people didn't think about. And that is something that I do think is along the lines of using generative models to potentially invent certain things. And I think that potentially OpenAI are working on that in secret. So if I had to guess, like, if I had to put my best guess forward, I would say that this is potentially something that is along the lines of something that uses a generative model to potentially use search, maybe to come up with new solutions or new existing potential. I don't know, ideas maybe. I mean, there's a lot of stuff going on. There is actually a new Microsoft model that does something along those lines. We've seen the iterations of AlphaFold. Actually, I'm going to be covering something later on in the video that is like that. So it wouldn't be surprising if OpenAI kind of had that kind of model. I don't think something like that would ever be released to the public, but it definitely is something that is super interesting. Now, I think I do have also something that I want to share with you guys as well. There was a really weird tweet that I saw, but it was, you know, super speculative that I'm going to show you guys that to I'm going moment. But amongst the AI hype, I did want to talk about the fact that, you know, there is, of course, some truth in certain statements. So Noam Brown, the person who, you know, worked on reasoning and OpenAI actually said lots of vague AI hype on social media these days. There are good reasons to be optimistic about further progress, but plenty of unsolved problems, you know, in research remain. So this is a good tweet and a lot of people were praising this tweet because whilst right now you have pretty much not every OpenAI employee, pretty much, but a vast number of OpenAI employees talking about how they've maybe achieved AGI or how maybe they've figured out exactly how to achieve asi. No, I'm brown the one really spearheading the reasoning of OpenAI is actually talking about this vague AI hype. But of course we're optimistic. But there are still a lot of unsolved research problems that still remain in the space. So it's definitely a sign that whilst yes, things are moving ahead, of course there is still realism that people are encountering. He says that between the 01 announcement, the 03 announcement and various podcast talks, I think we've said a lot. We believe 01 presents a new scaling paradigm and we're still early in scaling along that dimension. It's definitely very true. The scaling paradigm that the O1 series represents, I think it's more significant than people are giving credit for because I personally think that this is going to truly lead to some superhuman performance in certain areas quicker than most people do realize. And that's just based on the past things we've seen with other prior systems that are pretty much neuro symbolic that have achieved really amazing things. So before we dive into any more AI related developments, let me actually tell you about something that is revolutionizing the AI art space. So have you heard about Recraft? It's actually the new king of image generation and I'm actually not exaggerating here. Their V3 model is currently at the number one spot on hugging faced text to image benchmark, outperforming giants like Midjourney and OpenAI with an impressive 1139 Elo rating. You might be thinking, well, what sets Recraft apart? Well, we're actually talking about photorealistic images that will make you do a double take, perfect text generation for all your design needs. And the most anatomically accurate AI model out there. Over 2 million creators, including teams at Netflix and Airbus already using it. And here's the best part. I've actually got a special deal for my viewers. Use my promo code to get $12 off any plan. You can do that by using the code grid 12. Whether you're a casual creator or a professional designer, they've got options ranging from 12 to $48 per month for unlimited access. Check out the link in the description to try the revolutionary V3 model. And of course if you don't want to check out the link in the description, you can simply scan the QR code on screen now to get my exclusive discount. And for all you developers out there, they've got a powerful REST API that lets you integrate both raster and vector outputs directly into your projects. It's super straightforward to set up. Just grab your API key and you can start generating images programmatically in minutes. Now, Sam Altman actually gave us some information on O3. This is the highly anticipated model that is currently state of the art across a variety of different tasks. You can see here that one person commented, I want a model smarter than 01 Pro. I'm willing to pay. I'm using it all the time for coding and it would be awesome if it could generate tens of thousands of lines of code at a time. It can do a decent job generating only about 1,500 lines now based on experience. And he does say that O3 is now much smarter and we are turning our attention to that now and O3 Pro with an explosion emoji. So I think this is just going to show us that O3 is probably a really capable model. And honestly them releasing it around the February mark is going to be super interesting because I personally didn't think that that was going to be the case. I thought it would be a much longer time considering the fact that O3 is going to be really expensive. And from what I'm hearing that they're still losing money on O1 Pro mode. So that inference cost at the moment, however doing it, it's probably not that efficient. So I'm surprised they're releasing O3 as quickly as they are, but who knows, maybe they're not going to release it straight away, but There is an O3 API that should be coming on that same date. Now they also spoke about GPT5. I did speak about this in another video but but since this is a news video, I'll quickly introduce, not introduce, include what they said but they basically talk about how for the estimated time of arrival and the performance they are still figuring at that out but of course they do think we will be happy with those results. Now we actually also got an incredible update from the unitary robot. This was something that genuinely surprised me because I knew these robots were incredibly flexible at what they were able to do. But when I saw this video it just surprised me because it just goes to show that with a few upgrades, a few software fixes, a few more trainings or whatever it was optimizations that we're able to do to make this robot a lot more effective that with the same platform that they released I think about a year ago now that they're able to make a robot that looks borderline human jogging down the hill right there. That was very uncanny. I would definitely believe that that is cgi. If I wasn't in this space, if I saw that robot with walking by me, I would be like, what the. What's going on? When did these robots just ascend and become very human? Like, so I think this is showing you that the robot race is really, really on. And I think unitree is one of those companies that they definitely must be working seven days a week, 24, seven. Because I've never seen a company move this quickly in terms of their hardware and in terms of the iterations to the software, in terms of how that's making the hardware move. Like, the gate on this is incredibly human like. And it's really incredible because prior to, you know, the. I don't even remember what it was called, but the engine AI robot, prior to that robot, most robots just walked in a very strange way. So, I mean, you know, walking is definitely one small thing, but when you do have robots that are just walking around like that, it's definitely going to, you know, maybe not feel like a cyberpunk world, but start to feel like, okay, maybe this kind of technology is actually real. And I think most people also underestimate how. How difficult it is to get a robot to not only run up a hill, but also jog down a hill in a way that looks human like, but also manage the stability, the uneven terrain, and to do it in an environment that it hasn't been prior, you know, tested on. So I think that just goes to show how robust these robots are getting and with all the kind of policies that they're probably, you know, running there. I mean, it really goes to show that this team honestly just doesn't make great hardware, but, but also makes great software to really boost the performance of these robots. So this is something that's genuinely surprising me. I thought robots would take a lot longer, but it's clear that even on the software side right now, we're getting tons and tons of investment, and there is clear incentive for these companies to actually speed up the robotics process, because we know we're at this cross section of where AI and robotics kind of collide, and that's just going to further push the industry in terms of what is potentially capable. I mean, it's definitely really going to be an interesting future. Now, if we're talking about humanoid robots and the unitary, you know, G1 robot, and this robot is pretty small, but I can imagine a huge robot running around. I want to show you guys the engine AI robot. Now, this was one that broke the Internet, and I didn't even include this in the previous video because I actually forgot, but it was crazy because I remember when I saw this one, a lot of people were basically stating that this was cgi. And I was like, okay, people are freaking out because I remember when I produced another video, okay, on Engine AI, on a different kind of robot, they were also thinking that that one was CGI too. So I think technology is starting to reach that point where we're starting to get to that point where we actually may have a situation on our hands where a lot of the technology just doesn't seem real. Like people who are in this space will know that yes, this is of course realistic. They know that reinforcement learning and by changing the policies of how the robot walks and updating it through many different training cycles and that kind of stuff, you can achieve this eventually. But people that are outside of the community would just think that this is 100% computer generated because of course they haven't been paying attention to the updates. And this is, you know, I don't want to say surprising because I understand that, you know, it looks super superhuman, but I'm starting to think that if people think that is cgi, what happens when a robot can like run around a corner, can, you know, you know, jump up and down and do other human like movements and it's completely autonomous and has real time audio. Like people are going to be, I don't want to say losing their minds, but they're definitely going to be more surprised than they are now because this is just a human like walking gate. So if people can't believe this on face value, when it does seem to be pretty realistic, it's going to be quite the shock when you do have fully autonomous robots that are out there doing a variety of different tasks. Now there is actually something super interesting. There is a new company trying to achieve AGI. So we can see here that Francis Soleil is, is joining forces with Mike Noop to start ndir, a new AI lab. And their focus is deep learning, guided program synthesis. And they're betting on a different path to build AI capable of true invention, adaptation and innovation. So if you are familiar with these people, well, if you aren't familiar with these people, that person is Francis Soleil. And this is the guy that created the infamous ARC AGI benchmark that is, you know, highly heralded as the key to AGI because if you can solve it, then potentially you could solve AGI with that same kind of framework that you use, because those problems are essentially supposed to be resistant to memorization. So that is the claim there. Now this is super interesting because I always love when people are taking different approaches to AGI because we've all seen how OpenAI are taking their approach to AGI and how Google are taking their approach to AGI, but I think we're actually starting to get many different approaches. Like whilst yes, LLM Search is putting us into a situation where there are some really, really impressive discoveries about how these systems scale differently to traditional the GPT series of models, but I think that when we actually take a look at other companies that are looking at different architectures and different ways to scale these models, I think that is where we're going to get a lot of innovation. Because I think it was Yann LECUN that said OpenAI kind of started this off ramp where everyone is just focusing on the LLM architecture because there's so much hype, there's so much money, there's so much investment in that area and it kind of, I don't want to say ruined the space because obviously it didn't. It just actually sped things up quite a bit. But he was saying that not many companies are actually looking at different hybrid architectures and that of course, you know, that is something that Meta is doing. But now that these guys are doing it, and since they created that crazy benchmark, I would not be surprised if they managed to do something crazy that's able to learn in a remarkable way. Now what's crazy about here is that they say that, you know, current deep learning based AI, while impressive and economically valuable, is ultimately constrained by its inability to efficiently learn and adapt. It excels at known tasks, but crumbles when faced with open ended problems. It reflects back on us only the knowledge programs and abstractions found in its training data. Like the man behind the curtain in the wizard of Oz, it hides the fact that human general intelligence actually created that data. And if AI cannot efficiently adapt, it is forever constrained by what humans teach it. And to compress the next 100 years of scientific progress into 50 or perhaps 10, we need general intelligence, not task specific skill. We need computers that can pose problems and explore new territory, not just apply known solutions. We need computers that can innovate. Now, the path to AGI is not through incremental improvements to existing methods. The problem with deep learning are fundamental and cannot be addressed superficially. It's time for a new paradigm. And the good news is that we now know what this new paradigm is. Program synthesis. So this is where they explain that this is the key to AGI. And I might make an entire video on this if you guys want me to, but it's actually super interesting. Okay, now I'm going to just basically simplify this. So this is a different focus, and it's basically how most AIs learn by guessing patterns in lots of examples, like recognizing cats from millions of cat pictures. But this approach struggles to fully understand things, and it needs way too much data. Like, in order for humans to recognize a cat, you don't need to look at 50 million different pictures of cats. You only need to see a cat once or twice and be like, okay, I know what a cat is. So program synthesis is, you know, I guess you could say it's like human learning, because it's like teaching the AI to write small programs that perfectly explain how things work. So, for example, instead of showing the AI millions of math problems, you know, it just learns the rules of math, and then it can solve new problems with just a few examples. So this is pretty crazy because this allows for generalization power. So program synthesis helps AI understand the big picture better, so it can solve new problems even with little data, which is how humans essentially learn. So this means it doesn't need as much data. And this mixes program synthesis with deep learning, which is pattern recognition, which means that it could become way smarter than humans. Now, this is a new and growing field, kind of like where deep learning was back in 2012. But researchers are starting to realize that this could be a big part of the puzzle for building smarter AI. So it's actually a pretty big moment for AI, basically saying, look, we actually need to take new approaches to AGI, and this is going to be one of the key ways that they do take that. So another thing that I forgot to add and cover in one specific video, but this is why we have the news videos, is that OpenAI shared their vision for shared prosperity. So this is where they talk about shared prosperity is near and and measurable as the new jobs and growth that will come from building more AI infrastructure from chip manufacturing facilities and power plants. And as our CEO Sam Altman has written, AI will soon help our children do things that we can't. Not far off is a future in which everyone's lives can be better than anyone else's life is now. And with such prosperity in sight, we want to work with the policymakers to ensure that AI's benefits are shared responsibly and equitably. And this economic blueprint that they were outlining was designed to support the entrepreneurship and individual freedoms that have long been at the heart of the American innovation ecosystem. So overall, they're basically saying that, look, they want shared prosperity for everyone. Now, I gotta be honest with you guys, this was actually met with a lot of backlash from, you know, various different parts of the Internet. And I probably, I'm not using the best example of person to, you know, quote this, but Gary Marcus says that's awful. If they make a lot of money, they aren't going to share it and they will likely destroy more jobs than they create. And this is, you know, one of the times where I see a lot of people agreeing with Gary Marcus, because if you aren't familiar with his reputation, some would say he has a reputation for just disagreeing with everything in the main AI space. I honestly do think Gary Marcus has a lot of valid points about AI, but I think sometimes his valid points get overshadowed by the fact that when there is a genuine breakthrough in AI, he just simply disregards it at the NASA, you know, hype train or another thing that's just too hyped up. So for me, I do think that sometimes it just goes a bit too far. But the point is that I did see a few posts about Reddit saying, you know, we're about to experience a very different world in the future, but most people aren't even prepared for it. Now, if you want to talk about AI doing some, you know, things that a lot of people wouldn't want it to do, which is, of course, no more hires. This is where we have the Salesforce CEO Marc Benioff, actually talking about how he probably won't hire much more software wear developers this year because of the productivity gains they're gaining from AI. Now, at first glance, I think this is pretty crazy because it's one of those situations where it's like, okay, oh my God, all the software engineering jobs are gone. Like I said in another video, I think that is not the case. I think it's actually quite the opposite. But I do think that what AI is enabling, it's enabling for us to create more companies at scale. And what I mean by that is that what you used to need maybe 200 people to do, you now probably can do with a 25 person lean team. And, and I think that's what this is saying is that maybe we're not going to fire people, but maybe we're just not going to hire any more people.

We are going through a global labor shortage where we see all these declining birth rates. We understand that it is harder to hire, especially people here right in The United States and sales and service, you can see it. And this is going to give us the ability to do more. And as an example of that, look at engineering. I think in engineering this year at Salesforce we're seriously debating maybe we aren't going to hire anybody this year because we have seen such incredible productivity gains because of the agents that work side by side with our engineers and making them more productive. And we can all agree that software engineering has become a lot more productive in the last two years with these basically these new models.

Now software engineers aren't the only ones getting the short end of the stick. We actually take a look at the Wall street job losses. They actually might replace 200,000 roles with AI. So we can see here that it talks about how back middle office roles are. You know, these are the ones that are at risk. And of course the bank's profit could surge due to the improved productivity. So it says global banks might cut as many as 200,000 jobs in the next three to five years as AI encroaches on task currently carried out by human workers. According to Bloomberg, intelligence and chief information and technology officers surveyed with Business Insider indicated that on average they expect a net of 3% of their workforce to be cut. So customer services could see changes as botch managed client functions while noyo customer duties would also be vulnerable and any jobs involving routine repetitive tasks are at risk. So it's going to be a really interesting time as the role of individuals of certain companies change. We're going to see new management roles for people managing ZI teams. We're going to see different roles emerge. I mean it's going to be a rapid, rapid transformation as companies try to, to, you know, reap the benefits as quickly as possible. And I do personally think that the majority of these companies probably won't be hiring people again. They're definitely going to be trying to, you know, make as much money as they can because I think people need to understand that most companies just exist to make a profit. I know that some companies exist out there to help the world, but when we're talking about banks, their job is literally to make money. So it wouldn't surprise me if they didn't hire much more people. But I do think there will be some new roles that are going to be super interesting and I think it will make a lot of the, you know, human to human interactions with clients a lot more valuable and interesting. And I do think that there's going to need to be a lot more people working in security because with all of this, you know, AI impersonating people's voice, it being able to talk exactly like you, it being able to, you know, take an image of you and then immediately start a zoom call. I mean, it's going to be pretty crazy now if you're thinking what kinds of, you know, jobs and stuff people are going to be doing. Maybe you might be managing an AI agent swarm. As Satya Liddell has said at a.

Recent conference, the real breakthrough for me is can we demystify creation of these agents and make it simple? Just like how we say, you know, create documents or we create spreadsheets, we should be able to create agents just like that. In fact, the way to conceptualize the world going forward is every one of us doing knowledge work will use Copilot to do our work.

And.

And we will create a swarm of agents to help us with our work and drive the productivity of our organization. And so to that end, we are building Copilot Studio. So think of this as simple as if you can create a spreadsheet or a document, you should be able to create an agent. And that's what Copilot Studio does. So here I'm building. I happen to be a frontline worker. I want to create a field service agent. I have a field service system. So all I need to do is to be able to give it a prompt, right, that says in instructions that say, I want you to behave like a field service agent. And then I give it knowledge sources. In this case, there happened to be a SharePoint site with a lot of documents as well as my field service applications. In this case, let's say it's a Dynamics application. So I give it the two sources of knowledge. I give it the instructions, click, click, click. And I have an agent that now can dock with copilot. So I can go into my copilot and say, at field service agent, I can now start talking to my field service agent. So that's the simplicity or that's the ease with which one can start creating agents. These thousands and hundreds of thousands of agents that will be there inside of an organization, just like documents and spreadsheets.

That are there now. Something that was also rather interesting was Sam Altman actually spoke about how he believes that we're actually going to be moving even faster. And remember how at the start of the video, I spoke about, you know, OpenAI are talking about how things are moving pretty quickly. In a recent interview, this was a statement that was pulled out because it is a change in the tone of his voice. And I don't really usually see that. There's usually like a consistent. I wouldn't say hype train, but there's usually consistent theme of, you know, we're going to do big things, we know where it's kind of headed. But recently it's like, okay, things are moving even faster. And what's crazy about this is that, like, you know, when I've seen OpenAI's track record on the things that they've done in the past, and I'm actually, I don't want to say surprised, but I'm actually somewhat excited because I do know that whatever it is, it probably isn't as hypey as most people think. I know all the time people hype up products and stuff, but this is truly like next level stuff when you actually think about what we're dealing with. They're dealing with super intelligence, AGI and that kind of thing. So the implications are definitely profound. And if that stuff happens faster, the implications are going to happen much faster, too. What's something you've rethought recently on AI or changed your mind about? I think a fast takeoff is more possible than I thought a couple of years ago. How fast feels hard to reason about, but something that's in like a small number of years rather than a decade. Wow. What do you think is the worst advice people are given on adapting to AI? AI is hitting a wall, which I think is the laziest fucking way to try to not think about it and just, you know, put it out of sight, out of mind. Now there was also this thing called evolutionary scale model 3. So this is basically about proteins and how proteins are essentially tiny machines in your body that do super important stuff like building muscles, fighting germs and basically keeping you alive. And scientists study these all the time to make new medicines, cure diseases, and even design useful things like glowing proteins. Now, there's this fancy new model as I'm talking about, called ESM3, and this is basically the ultimate protein expert. And it's trained to understand what they're made of, how they fall into shapes and what they actually do. And it can actually create new proteins and invent brand new ones. And it's designed to invent a glowing protein that is totally different to any that exist in nature. And it basically simulated 500 million years of evolution in a computer. And, you know, they've essentially decided to open source this technology. So it's pretty, pretty crazy. So I do think that this is going to be the kind of thing that OpenAI released sometime soon. I'm not entirely sure what kind of, you know, avenue they're going to go down because there's just so many different verticals that they could challenge and they could take on. But it's going to be super interesting to see where this stuff does go. We also got to see the crazy effectiveness of AI in learning. So there was a new randomized controlled trial of students using GPT4 as a tutor in Nigeria and six weeks after school tutoring. AI tutoring equaled two years of typical learning gains, outperforming 80% of other educational interventions. And it helped all students, especially girls, who were initially behind. So I think this is one of the things that, like a lot of people are caught up on. LLMs and large language models and AI for this and AI for that. And there are so many different verticals where this is truly helping a lot of people. And I think a lot of the times, like all the stories I do cover just cover like really, really intense AI news. But we're seeing that AI has a wide range of use cases and there is actually a video I do need to include where someone is essentially blind, but they are now using the meta glasses and they can simply say with a prompt, exactly, you know, hey, what's in front of me? Hey, is my stake done? Hey, what does the sign say? And I mean, I can't imagine being blind. And then this technology gets invented. How much of a lifesaver that is going to be? So it's just something that is just super, super incredible and it's pretty wholesome. So another thing as well, there was also a crazy study about 01 and I covered this in detail, but it basically is not good in terms of the reasoning stuff. So maybe that's why they're including that new architecture. But I think it was super interesting because they talk about how, just like they did with the Apple paper, there was a 30% reduction in accuracy when math problems were slightly variated, which kind of suggests that maybe the architecture that we're using isn't the most efficient. Even if we can get the solutions, there's a 30% reduction in accuracy when the problems are slightly variated, which is not good. So basically saying that, look, if these LLMs are so reliable and they're so good, why is there a 30% reduction, especially for these reasoning models on similar questions where we only change the numbers and the text? So it's definitely a little bit of a red flag, but it was something that I covered that was super interesting. Now something else that was rather concerning was the fact that we're now starting to get more and more conversations about AI and warfare. Not in the sense that, you know, we're going to be using ChatGPT to, you know, power killer humanoid robots, but in the fact that, like, AI is quickly becoming a national asset, like a national thing that countries are really starting to pay attention to, because eventually AI will most likely be used to create weapons of national defense in terms of the inventions, in terms of what it's going to be capable of doing and managing systems. And in this short clip, Yoshua Bengio actually talks about how data centers in the future might actually become military targets for foreign nations that want to take out a country or catch up in the AI race. So it's quite interesting because this is something that I wasn't really focused on. I didn't know that this is a thing that will exist. I remember reading the very long winded document called AI the Decade Ahead and I was thinking about how the role of AI is going to quickly change and this is going to be something that countries have to face and.