Huge AI NEWS : OpenAI Gives GPT-5 To US Government, Stunning New Humanoid Robot, Metas New Model..

So with an absolutely incredible past couple of days, here are five of the key stories that you most likely did miss and are really pivotal in the AI space. Coming in, one of the first stories that I saw this was Brett Adcock. This is the CEO of Figure, and he said, only recently has time opened a window of opportunities to scale to billions of intelligent humanoid robots. This was not feasible five years ago. We are so fortunate to be in 2024, the first year in history when this is available. Now, what he's referencing here is the Figure product. So I'll let them take and I'm going to show you guys why this is crazy. Foreign. So there's a few things that I do actually want to talk about when talking about this robot, because one of the main things that people haven't realized about this robot is that this robot is actually an electric upgrade. And a recent industry trend that we've seen is that humanoid robots are increasingly becoming, you know, and adopting electric power systems over traditional hydraulic or pneumatic systems. And this shift is essentially driven by several key factors that essentially make electric motors more advantageous for the development and deployment of humanoid robots. Now, it was very hard to see, but you can actually see that it does say electric humanoid right here. So this marks this entrance into the electric humanoid space. And this is rather fascinating because it was also Boston Dynamics that announced their recent robot, and in that video, it did show us an electric humanoid. So this is clearly an industry wide trend. And of course, something interesting that we did see here was that we did see a much more upgraded hand that seemed to have this kind of increased grip area, which kind of looks like the tactile sensing that we did get from Tesla's Optimus robot. So this was one that I really did like here, because although the fluidity of the last spot was nice, and this one also looks rather impressive too. Now, electric systems tend to be less costly and easier to maintain, more than hydraulic systems. And the problem with hydraulic systems is that they're complex, prone to leaks, and require regular maintenance, which can be expensive and time consuming. And in contrast, we've got these electric humanoids, which are cleaner, simpler, and generally have lower maintenance requirements. And this makes them more practical for widespread use in various applications, from industrial settings to home environments. And one of the things you also have to think about really is the noise and environmental impact these robots. If they're supposed to be in environments such as home environments and work environments, the impact of the noise reduction is going to be rather important. Now for this robot, this is going to be actually released in, I think around four to five days, depending on when I release this video. This video should go out just after I'm recording, recording this. But, you know, seeing some of the comments from the founder, Brett Adcock, it does show us that this is very likely to be a remarkable engineering feat. Now the reason that I think this is so incredible is because if we actually take a quick look back at Boston Dynamics robot, this was the scene of their recent electric humanoid, where we can see that it looks remarkably flexible in terms of how it's able to stand up and how it's able to move around and navigate the environment. So this is going to be a very interesting piece of future technology because I do wonder how it compares to Boston Dynamics recent Atlas robot that they're currently developing to, which is, you know, the successor of their previous system, which was, you know, loved by many. And we saw that it was remarkable in terms of how human looking it was. And it was able to do, you know, a remarkable level of things in terms of the flexibility and agility of that robot. Now, one of the key things that he said here is that this has been engineered for over a year. Figure 2 is the most advanced humanoid robot on the planet. And the thing about this company is that I gotta be honest, guys, as someone who is watching the humanoid robot space, I've not seen a single company other than figure make as much rapid progress as Brett Adcock has in the short space of time that the company has been dedicated to its mission for. I mean, if we take a look at some of the other companies going on in the space, it's honestly no shade to them. But we have to really take a look at and really understand that this is remarkable speed in terms of managing to develop a next generation humanoid robot that is not only going to be ready for when AI systems get remarkable upgrades, but being engineered for over a year and already coming out with your second robotic demo that looks to be, you know, industry defining, it truly is remarkable of what you can do when your company is in the hands of a determined entrepreneur. You know, it was only four months ago that we did get this robot where we saw that there was the speech display, speech update, which definitely shocked people because we realized truly with our current systems, with the current architecture, with the current hardware, the kind of things that we can currently create. And I think this is really good because people constantly say that this entire technology space is decelerating. We're going through this trough of disillusionment, yada yada. Yada. But I mean, if you really do pay attention to exactly what's going on, you can see that the technology here is increasingly getting more and more capable in terms of how good things are. And I got to be honest guys, looking at the entire humanoid space, this robot was remarkably agile, it was remarkably flexible. Being able to talk, one of the first robots being able to do that, this was all surprising. So if they can do something that's even better than this, I'm truly going to be impressed and surprised. Now, Meta are coming at us again with another open source model release, which is rather incredible. They're introducing Meta Segment Anything Model two, which is sam. Now this is a model that is rather effective at segmenting any precise selection in any video or image. This is really crazy because it has many different use cases. I mean, you could use this on a variety of different things. But if you remember SAM one and the kind of things that people were able to do with that software, I think this is going to be truly game changing. And you can see right here that using SAM2 you can select one or multiple objects in a video frame right here you can see the boot being selected and you can see the different objects. And of course you can see here SAM2 is capable of strong zero shot performance for objects, images and videos not previously seen during model training, enabling use in a wide range of real world applications. Now one of the most insane things about SAM was the fact that it has real time interactivity and results. It says SAM2 is designed for efficient video processing with streaming inference to enable real time interactive application. So imagine being able to track things in real time. I mean, this is going to be something that opens. I mean, I can't imagine the kinds of possibilities with this stuff because you're going to be able to, you know, track things in real time, see where certain objects are. I mean, you could use this in, you know, augmented reality. You could use this in just, just a variety of many different things. And of course you can try the demo yourself. For example, what you can do here, it says you can, you know, create effects right here. So if you click this, try it now, button, click, accept, you know, it manages to load up the demo. You can see right here, you can pretty much track anything in. So for example, let's say I wanted to track the ball, I could just click the ball right here and you can see that for the rest of the video. If I just click track objects, you can see right here that for the rest of the video it is going to Go ahead and track exactly where that object is. And it's pretty crazy. I mean, it's just, you know, completely remarkable at how AI systems are able to do this. Like literally being able to track someone's foot here is. It's weird. I mean, it's definitely weird. But I think, you know, it's interesting. You know, I think this kind of technology is going to be used by some creative people to do some really genius stuff. Now, an interesting thing right here is that OpenAI has pledged to give the United States AI Safety Institute early access to GPT5, its next model. It says OpenAI's CEO Sam Altman says that OpenAI is working with the US AI Safety Institute, a federal government body that aims to assess and address the risks that in AI platforms on agreement to provide early access to its next major generative AI model for safety tests. Now, if you aren't aware of why this is such big news, it's because, number one, the next frontier model is supposed to be a model that is much more capable in terms of its reasoning abilities and in terms of its abilities to produce factual responses. Now this is something that many people, including people like Sam Altman, have said that they're always going to need to test these models thoroughly before public release. And it seems like this is probably going to be the first time that they're going to collaborate with an external government body to see exactly how well these AI systems truly are in terms of their safety and reliability. Now this is an interesting move because OpenAI has been under so much criticism recently because of all of the people in their departments, just basically, and I do want to say it's of course not everyone, just basically just a named few, but there's some pretty prominent figures that have left OpenAI citing safety concerns and basically stating that, look, we're not getting enough compute to do AI research here and if I want to do meaningful research, I'm going to have to go somewhere else like anthropic if I want to get it done. And you can see right here in May, OpenAI effectively disbanded a unit working on the problem of developing controls to prevent super intelligent AI systems from going rogue. Now, I do wonder, and this is just a wonder, pure speculation, that are OpenAI going to truly give them the most capable model? Because you have to understand that for OpenAI, they are a product company, meaning that what OpenAI do want to do here is they're in the business of making money and a lot of money. Now of course, what they don't want to do is they don't want to put themselves in a position where they're giving their models to the US government. And then the US government's like, hey, maybe you can't release this. It means that OpenAI is not going to be able to release a product or whatever. I mean, you do have companies like Anthropic, Google and Meta that are hot on the heels. And right now, currently, I would argue that OpenAI's GPT4O is not the model leader when we've got models like Claude Sonnet 3.5 that are just so effective. And I would say I do wonder if OpenAI are actually going to stick to this if other companies manage to get ahead of them in the AI race. As you know, Google is really working hard on their next model and some rumors are basically stating that it's going to be better than GPT5. So we will see if that is going to be true. Now, there have been some incredible demos from the GPT4O advanced mode, and this one actually blew my mind. Essentially. ChatGPT advanced voice mode. He asked it to count as fast as it can to 10 and then to 50.

Hey, I want you to count from 1 to 10. Really, really fast, as fast as you can.

1, 2, 3, 4, 5, 6, 7, 8, 9, 10.

Okay, now even faster. Okay, now louder and faster and count up to 50.

1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 40, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 42, 33, 34, 35, 36, 37, 48, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50.

And I think this is a little bit uncanny as to how it managed to take a breath in the middle to pause. I mean, when you think about it, it's a robot. It doesn't need to take a breath. It's not a human. It's just mimicking what humans do, which is, I mean, strange. I mean, this is so, so strange. I mean, I don't even, you know, have the words to describe the kind of technology that we're seeing. And if this is as bad as things are, I can only imagine how crazy AI is going to make certain technologies and certain interactions with our current tech. I mean, honestly, this kind of thing just makes me wonder what the world is going to look like in terms technology world, for how we interact with AI systems that are lifelike and really Truly human. Like, because if someone was on the phone to me and I told them to count to 50 and then they did that, I personally would not be able to tell that it's an AI. Now, in some more OpenAIUs they actually did release, well, not release unfortunately, from OpenAI, but they spoke about GPT4O long output. So it says OpenAI is offering an experimental version of GPT4O with a maximum of 64,000 output tokens per request. And we hope this experiment helps you explore new use cases that are unlocked by longer completions. And Alpha Participants can access GPT4O long output by using the GPT4O 64K output alpha model name. You can see the pricing right here. And this is quite a small announcement. So most people did miss this, but it is rather fascinating to see how this works because one of the things that they did Talk about on GPT4O was its ability to just generate pure completions of tech. So I'm wondering if this is just going to be able to generate entire essays from scratch, entire books. Of course, it's a lot more than 64 tokens, but I do wonder on how that coherence is going to compare to some of the other models that exist in the space. Now. Most people did miss this and I nearly did because the announcement didn't get that much coverage. But I saw this and I was like, I have to cover this because it's absolutely incredible. So this is Flux Point one, a new era of creation. And basically this is a new open source image generation software that is unbelievably realistic. Now when you see images like this, this of course looks, you know, really realistic until you look at the badge and you think, okay, then maybe it was AI generated. But the thing is, is that whilst, yes, you know, a few years ago this would have blown any person's mind, I still think that, you know, the kind of realism that we're looking at is here now. So I'm going to show you guys a picture that it recently generated and you're going to say, wow, okay, this is incredible. So this was an image that was on their page and this is an AI generated image. Now I'm going to show you a few more in a moment. But when we start to look at images like this that are iPhone style, that look like they were captured from someone who literally just took them on their phone, I think these are the kind of images that are the ones that are probably going to cause the most problems in terms of the trust levels for Individuals actually believing, you know, what they see online. Because if I were to see this image and I was browsing on Twitter, there would be no immediate red flag that this is AI generated at all. I would look at this image and I would think that, okay, this looks, you know, immediately realistic. But one of the things that you're about to see is that this model can actually generate completely new photorealistic images that are just iPhone only. So you can see right here, these are AI photos made with the Flux Point 1 Pro Boring Snapchat photo circa 2015. I mean, if I were to see this image on someone's old Instagram, I wouldn't think it was AI generated. There's no obvious tell here. Like, genuinely, there is no obvious tell. I mean, I guess maybe in the back the car kind of merges with the house. So I guess if you're analyzing every image with a microscope, maybe it's something that you could see. But even then, most people on social media look at an image for maybe just two seconds, and in those two seconds you have to kind of tell yourself exactly what's in the image. You're like, okay, it's just girl standing near some houses. You're not thinking, wait, is this AI generated? Yada yada yada. So this kind of image and this kind of software is probably going to bring more credence to the dead Internet theory. I mean, if we take a look at this image right here, this image just doesn't look AI generated at all. Of course, if you read the text, you'll see that it is AI generated. Of course. And of course, if we look at this image right here, I mean, this is probably the most shocking one of all, because if I saw this picture, like I said before, there's no obvious tell that this is AI generated. I mean, maybe over time we're going to find some things like T shirt patterns or whatever, but you could always prompt the model to have a plain T shirt with just an apple on it that it would still get right. So maybe if someone's T shirt doesn't have text on it, maybe then you could say it's AI generated. I don't know. It's a very strange area that we're moving into. But let me know what you think about Flux 1.1. Do you think this is good? Do you think this is bad for the misinformation? But I do think that this is the kind of, you know, photorealism that is, you know, really dangerous.