Major AI News : OpenAI Leaks GPT 4.5 Even More Humanoid Robots, AI Safety GONE! And more..

So in this week in AI, it actually all kicks off in robotics. And you can see right here that Clone Robotics, the company that is building a Musco skeletal Android version 1, this synthetic humanoid robot is just completely different than anything, you know, any other company is building. They actually finally released a video on the full body. Now I got to be honest with you guys, I didn't think this was coming this soon, but it just goes to show how quickly this company is working. These robots that they are building, they actually don't use metal motors like most robots. They actually use artificial muscles, not like, but called myofiber. And these ones actually try and work like real human muscles. They stretch, they contract, they can actually lift real things. And this is pretty, pretty crazy because they are essentially trying to build the Westworld robot. And that is something that I think for the most part is quite terrifying or quite exciting, depending on where you stand in the AI space. And entirely what we've got here is a robot with bones, muscles, a fluid powered circularity system, and an AI brain that lets it respond and move like a real person. And I think right now, essentially what we're seeing is the real proof of concept that look, this thing does move and it does contract and the muscles work. Can you imagine, once this thing is embedded with a super fast AI, it's able to talk, it's able to respond and it's able to walk around. What kind of future are we going to be when that becomes reality? I really do think that right now we're in that transition period and, and you know, companies like Clone are building stuff that really would have thought to be, you know, maybe 50 years away if you didn't show people this video. So this is pretty crazy. You know, this entire thing has got like 206 artificial bones just like a human. And honestly don't know how they're even doing this. This must be an incredible engineering feat on their part because I know that humanoid robots are pretty difficult, but if they pull this off, man, oh man, the entire market is probably going to be theirs. Now of course, in terms of other humanoid robotics, we have the HMND01, which probably stands for. No, it does stand for humanoid. And this is a British humanoid robot from a British AI robotics startup founded in 2024 by Arkham Sokolov. And this company specializes in creating modular AI driven robots to enhance industrial automation and address workforce challenges such as labor shortages and aging populations. Now this robot stands 175cm tall, which is around 5 foot 9 and weighs around 154lbs, 70kg. And it features a modular architecture that actually allows for customization of its hardware and software for specific tasks. And it mimics human anatomy with 41 degrees of freedom, enabling it to have high mobility and actual precise manipulation. And the performance here. Now they also talk about the fact that, you know, like I said before, it can walk at 1.5 meters per second, 5.4 kilometers per hour, also has a payload capacity of 15 kilograms, 33 pounds. You know, it has swappable batteries, so it has minimal downtime. It's going to be able to run pretty much 24,7. Apparently this robot also excels in pick and place tasks, material handling and navigating complex environments, and actually uses advanced AI for real time decision making, object recognition with a 98.7 accuracy and adaption to uneven surfaces. So overall, this humanoid robot, they are targeting industries like logistics, manufacturing, retail and warehousing. And the humanoid 01 can perform tasks like goods handling, sorting, packaging and kitting. So overall, what we have here is a situation where another company has built their humanoid robots. And I do want to state that, like this robot that we can see here with those modular batteries, I think this is of course the future we're seeing. Tons and tons of companies finally start to build out their humanoid robot platform. And I think what we are seeing here is of course humanoid robots actually start to become potentially part of the future economy. And if these robots do work, and honestly, if you've seen any of the demos this week, and there is likely a demo dropping today that will showcase just how far we've come, it is going to prove that this is most certainly the future. If we're going to expand the economy at an exponential rate, we're definitely going to have to double the effective human population. Not in terms of actual people, but of course actually make up for certain industries where we have labor shortages, certain industries where maybe it's too dangerous to work, but just a lot of jobs that essentially should just be automated, that can really be automated for a fraction of the cost, that is going to have tremendous boost because humans have, you know, a pretty unique set of capabilities and I think, you know, they can definitely be utilized elsewhere. So this robot is definitely one I'm excited for. I don't think I've seen too many companies coming out of the uk so this startup is going to be one I'm going to be paying attention to. And hopefully, you know, if you guys are watching this, please let me visit your factory. I'd love to. You know, look around and see what you guys are building now in terms of other robots, okay, that are actually in practice, that are actually in the factories. We've actually got to take a look at this robot right here. And I might play the video, but of course, I do want to give my two cents on this. And this is Helix. Now, I don't think this one broke the Internet. And it didn't break the Internet because the demo wasn't as good as the previous one. And when I say it wasn't as good, I don't mean that this isn't a bigger breakthrough. This is by far the biggest robotics breakthrough that we've seen. But it just doesn't appeal to the average person, which means the average person doesn't really see this. But essentially in this demo, what we got to see was a Figaro would actually just identify any item and then pick up that item without seeing it before. Do you know how insane that was? This is something that, you know, we've struggled to have robots do for such a long time. And having a robot see something that it hasn't seen before and to be able to actually use that item, be able to pick it up and put it in a fridge, that is something that is actually a real world use case. Because a lot of the times, even with LLMs, people are like, yo, this thing, when it, you know, like, does it generalize out of its training distribution data? There are even tons of research papers that I read where they're like, yo, these models, these generative systems cannot generalize out of their training data. And basically all they mean is that, look, when they see something that they've never seen before, they get confused, they get thrown off and they have no idea what to do. But these guys are seeing these items for the first time and they're literally able to pick and pack and just do all that kind of stuff. Which means this onboard system that they built is incredible. And remember, guys, this onboard breakthrough that they've made, it was so good that they decided to terminate their partnership with OpenAI. And, and I'm guessing that they decided to do that because of course they're thinking, look, OpenAI is building their own humanoid robots. Do we want to share our secret sauce with a company that is potentially competing with us? We just developed an algorithm that is probably a lot better than theirs and we kind of want to take as much of the market share as we can. So OpenAI, it's goodbye to you. And we can use any other voice AI company because to be honest with you, I think a lot of other companies have caught up to OpenAI's advanced voice mode. So that partnership, if it was just purely based on that, it may not be that strong anymore. But this by far is just absolutely insane. Later today we're probably going to get another more impressive demo. So I'm going to play part of this so you guys can actually understand what's going on without my boring commentary. But honestly, it's just pretty insane at how fast the AI space is moving.

Hey, Figures. Can you come here? Okay. I'd like to try something new. Even though this is the very first time that you've ever seen these items, I'd like you to use your new Helix AI and try to reason through where you think they belong in the scene and then work together to put them away. Does that sound good?

Now, as the AI space is moving fast, there is also Deep Seek. And Deep Seek is essentially, of course open sourcing more stuff. So this week they are basically going to be open sourcing more information. You can see right here. Let me just unfreeze this. But you can see it says we are Deep Seek. You know, a tiny team exploring AGI and this is putting more pressure on OpenAI. And of course they're warming up for their open Source week. They've actually done a few things, but essentially they're going to be open sourcing 5 repos. These aren't crazy LLMs and stuff, but these are basically just small repos that of course do different things within AI. Of course the average person isn't really going to use this, but of course you can check it out because deepseek are definitely going to be open sourcing some stuff. And as Deep Seek put pressure on OpenAI, they've actually sought to raise the stakes and of course release GPT 4.5. So at the time of re releasing this video, GPT 5.5 might be out. But recently I'm not going to spend too much time on this, but they're currently readying the server capacity for GPT 4.5 and GPT 5. And Sam Altman, of course, he recently acknowledged that this model will launch in a matter of weeks. And apparently this week they're probably going to be hosting the model. So watch out for it. This week and even today we actually got this where they said, try the GPT 4.5 research preview. Also, I think this is really interesting. We had Sam Altman say that GPT 4.5 has been much more of a feel the AGI moment among high taste testers. Than I expected. Now, I think this statement is a little bit bigger than we do think. Because if you feel the AGI thing is usually what happens when there is a model that exhibits some kind of capability, some kind of ability that I don't want to say wasn't foreseen, but something that makes you feel, wow, this is like I'm feeling the future. And we did have that with like advanced voice mode. We had that with GPT4 and of course GPT4O. So I'm guessing that this is probably going to be bigger than most people think. I don't think this is just going to be another LLM upgrade. Some people are saying it's a magentic framework, some people are saying it's a new crazy voice mode. There's a lot of speculation right now, honestly, there's been so much on the timeline, but I will say is that it will be excited for this and I do think that it might come on Thursday or Friday. The only reason it didn't come is because of course this person said, steal the show tonight, man. Livestream at 7:30 and chill. Of course he said, that wouldn't be very nice. That is, of course in response to the Grok tweet. So I am thinking that potentially this is, you know, they didn't release it because of course yesterday they released Claude. The day before that, you know, a few days before that there was other AI releases. And I'm guessing they just want to have enough coverage of theirs because it's pretty crazy with as to how much hype AI is getting now. Now, of course, if we actually go back to the robots, there was actually this clip that was going viral. I think this one was pretty funny. That's why I'm including it into the video. But essentially people were freaking out on Instagram and Twitter and stuff and saying, yo, this humanoid robot, it attacked a lady. And funnily, it does look like it attacked a lady, but it doesn't. Okay, this is just a unitary robot. This is like the H1, which is essentially their earlier platform. So I don't think that that is the actual G1 model, because this G1 model is silver. I'm actually going to show you guys which model this is. But yeah, you can see here on Google that this is the unitary H1. So the H1 is the black one that's a little less flexible in terms of how it's able to move its body. And. And I'm guessing this was some kind of demonstration of robots or whatever because it Seems like the robot is dressed up there. Just a bit of a malfunction, but hopefully we never have that kind of scenario in the future now, of course. And I actually recently relaunched my AI Grid Academy after coming out of surgery. I'm actually back now posting in there and I'm actually going to be documenting zero to a million dollars in 18 months with AI agents. I recently outlined my entire plan. I've done it with Claude and and I think I actually have a really decent plan that I'm going to be sharing with my community on how I plan to get to a million dollars with AI agents. I know that does sound pretty crazy, but it's something I'm willing to give a go and everything I learn along the way will definitely be shared there. And next week I'm actually going to be building out the distribution for this company, making myself some AI agents that can essentially automate the social media channels that will be promoting this. So of course, if you're someone that's trying to build with AI and you want to know what I'm up to, this is just something that I'm going to personally be working on every single week, every single day, and I think it might be of some value to you. So that link will be in the descript. Also, if we're talking about future crazy stuff, there was also the Google AI scientist. Honestly, honestly, honestly, I think this is absolutely insane. I saw some videos debunking this, saying that this isn't true at all, but I do think that this is how we're going to get towards some kind of scientific discovery. This AI co scientist system is pretty incredible when you think about what it does. It aids scientists in terms of doing research, generating novel hypotheses, research proposals and experimental protocols. This one is built on Gemini 2.0 and it acts basically as a collaborative tool rather than like some kind of AI agent. This completely takes away the job. So this of course generates a hypothesis. I've already made a video on this, but I am just doing a quick, you know, summary of what's going on here. Then it, you know, researches and then it creates detailed overviews. Now this is something that I personally think is quite underrated. I didn't get a lot of views on my channel, but like I said before, this is something for, you know, people that are in, you know, the research industry. So I'm pretty sure this is something that, you know, rocked their world, hopefully. And I honestly can't wait for wild scale applications of this because one thing that I always Think about is that it's not that you need a billion different breakthroughs, you only need one or two small breakthroughs in order for a huge wave of advancements to come from that. And what I mean by that is that you only need one thing to kind of shift the paradigm of your understanding for multiple more inventions and other things to be built on top of that. So I do hope there will be a day when the AI is able to of course find something novel. That is going to be a very, very important day because that is going to be the day that people are like, whoa, if this thing found something new, then maybe we should take this even more seriously. And then of course there was basically this article on Google that basically talks about how this thing kind of works and basically the co scientists managed to find something that this person was working on. So you can see right here, it says a complex problem that took microbiologists a decade to get to the bottom of has been solved in just two days by Google's new AI tool. Professor Jose and his team at him Burial College London had spent years working out and proving why some superbugs are immune to antibiotics. He gave the co scientist a tool made by Google a short prompt asking about the core problem he had been investigating and it reached the same conclusion in 48 hours. And he said he told the BBC of his shock when he found out what it had done, given his research was not published, so it could not have been found by the AI system in the public domain. I was shopping with somebody, I said, please leave me alone for an hour and a half, I need to digest this thing. I wrote an email to Google to say, you have access to my computer, is that right? And the tech giant confirmed that it had not. So overall, basically the decade of research that this person done was dwarfed because this AI was able to basically reach the same conclusion within 48 hours. So if you remember when Sam Altman, Daou Amade, they were basically all saying, look, AI research, not AI research. But of course, industry ride research problems are basically going to go down in terms of how long it's going to take you to solve them. This is basically the first demonstration of that, which is crazy because we know that in 10 years and 20 years, how much research, how is that going to advance? It's going to be pretty crazy if you ask me. Now, of course, we had Microsoft Research introduce bio emu. So this is essentially something that can generate thousands of protein structures per hour, unlocking new possibilities for scientists in drug discovery and research. These stories, whilst they're not going to break national headlines, they're still important to show you that on the back end, these companies, whilst, yes, sometimes they're focusing on LLMs, otherwise they are still focusing on building things with a real world use case that are going to help society out in a whole. So if you aren't familiar with what this is able to do. Proteins are essentially like tiny machines in our bodies that do all sorts of important jobs, from building muscles to fighting diseases, but they don't just stay in one shape, they like bend, they twist and change depending on what they need to do. And understanding those shape changes are pretty crucial for developing better medicines. And the problem is that like traditional medicines, you have to study protein movements like molecular dynamic simulations. And these take forever because they require massive amounts of computing power. So with Microsoft's BioEmu One, this is basically just an AI model that speeds up the process by predicting thousands of protein structures in just an hour using deep learning. So instead of running like these slow simulations, it learns from existing protein data and generates accurate models of how proteins might behave in different situations. Now, this means that scientists can now study protein dynamics way faster and cheaper, which could lead to breakthroughs in drug development and medical research. And of course Microsoft, thanks for that, they're making this open source so researchers everywhere can actually use this. Now, if we actually take a look at the differences between AlphaFold and Bioemu, we can see some clear differences. AlphaFold is basically where you can predict what a protein should look like in its final stable form. And bioemu1 is better for understanding how proteins actually move and interact, which is key for designing drugs that influence protein behavior. So in short, Alpha, not AlphaFold, but, you know, AlphaFold basically gives you a high quality screenshot of the protein, while bio emu1 actually creates a full animation of how the protein actually moves. So just screenshot this if you are confused on the difference. So next, what we had here is Microsoft's Muse. And this is a generative AI model designed to revolutionize gameplay and digital environment simulation. So basically it's AI generated games generated by AI in real time. So this was developed by Microsoft Research in collaboration with Ninja Theory and other Xbox Game Studios team. And it's built on the world and human interaction model and it uses, you know, advanced deep learning techniques to analyze and generate game visuals and controller actions based on extensive training data, which basically offers new possibilities for game development, preservation and interactive simulations. Now, I did see a lot of discourse on Twitter about how they spent hours and hours and millions of dollars developing something that is so much worse than traditional games. But I do think that these early explorations will lead to some very interesting things in the future. Maybe we don't get fully, you know, AI explored games, but we get different kinds of VR world or something along the lines. I think people shouldn't be so skeptical and so dismissive of new technologies or as even if those don't lead to something, those innovations can be brought onto many other different things. So this is pretty crazy. This can create in game graphics and corresponding actions based on inputs. This is something that's pretty crazy considering that over time this is going to get better and better. And I do really wonder if the future is going to be AI generated games. One of the things that I do wonder about when we talk about AI generated futures, there is always this collective consciousness that we have. And I mean that right now everyone is probably watching the team TV show. Like, I mean right now the TV show Severance is quite popular. There are other shows like Breaking Bad that were popular a few years ago. And it's like, if everything is AI generated, do people still not want to be in that same bubble? Because I get value out of certain things if the wider community is using that. Like for example, this game is AI generated. If my entire experience is completely unique to me, am I still going to be part of the community? Is that something that people do want to have? I do think in maybe some experiences it's going to be pretty useful. But I do think that, you know, not everything is going to be AI generated because a lot of people will still be wanting to be part of the community and of course, you know, share that collective experience that you can all talk about. Now here is some interesting news that I didn't really see anyone talk about. And this is about AI safety. And I know that AI safety might seem something like, it might seem like boring, but until you guys start to see the crazy stuff going on with AI and honestly, I think something's brewing. Like, I've read the research papers, I've seen what people are saying in the labs. I've seen the way that we are being basically forced to rush these AI systems out thanks to the competition from other countries. And AI safety has taken a very, very big backseat. So recently it said that US AI Safety Institute will be gutted. And this is what Axios is reporting now. Take a look at this. Okay. It says after reversing a Biden era executive order on AI regulation and firing staff across several government agencies The Trump administration is gearing up to make cuts to the US AI Safety Institute next. On Wednesday, Axios reported that the probationary employees at the National Institute Standards and Technology, which houses AI safety, is bracing to be fired imminently. Okay. And sources told Axios they are preparing to fire 497 roles, which is basically going to leave them gutted. Now, this thing was created to oversee and conduct AI model testing and partner with developers on regulation efforts. This was signed by many companies, but now it seemed like that might be over. So this is not good because this would weaken the US Government's ability to oversee and regulate AI development. This is what was created to basically ensure companies stay in line with their safety initiatives. And as pointed out by Jason Green from caip, this could actually impact national security, particularly regarding the monitoring of AI related nuclear and biological risks. Now, of course, the cuts to both AI safety and semiconductor production programs could affect the US competitiveness, particularly in relation to China, because this kind of conflicts with the US Government's goal of achieving AI dominance. And I think there's now going to be a gap in public private collaboration on AI safety research and regulation. So this isn't really good at all. You can see right here, it said, because the government has only recently begun to work on AI safety, most of the experts at NIST to understand the catastrophic risks posed by advanced AI have only recently left the private sector. Throwing them out of the government deprives us of the eyes and ears we need to identify when AI is likely to trigger these risks. The savings would be trivial, but the cost to our national security would be immense. And someone else also tweets that to everyone going, but companies do their own testing anyway. This is where people are like, hey, look, these models are safe. These companies have their own internal tests anyways. The private sector cannot test in most areas relevant to national security without the government involvement. And this is because the government information itself is classified. So you need the government to work with these companies if they are to ensure that the testing actually protects the country with rules that are built by the government. So this is something that of course is not good at all. So this is going to be super interesting to see how this story develops. Like I said before, I think AI safety is going to take a backseat with all the competition going on. And I just hope that for the love of God, there isn't someone out there building some kind of Ultron. Then of course we had. Then of course we had Adobe finally release their Firefly video model. This is something that's really useful for you if you are a creator. Then of course we had Adobe finally released their Adobe Firefly video model. So this is something for you. If you are a creator, you can experiment with Generative Video. This is something that I believe is going to be going pretty pretty crazy. So for those of you who are wondering how exactly AI Video is being transformed in terms of actual workflows inside of video editors us, this is going to be something that you want to try out. I do think that maybe right now it might be free. Do not quote me on that because I haven't checked it out in quite a long time. But for those of you who are looking for workflow based stuff, Adobe Firefly is definitely the place to be. I currently do use Generative Fill in Photoshop quite a lot of the times, but this is definitely something that you should use.