Bill Gates Reveals The Next Big Thing In AI, Humanoid Robots Get To work

So let's take a look at some of the AI stories that you missed, because there were actually a lot that happened in the past few days. And I got to be honest, there are actually a lot of robot stories. And it seems like the humanoid robot space is heating up quite a bit. One of the first interesting stories is that Sanctuary AI receives a new investment from BDC Capital and NBC, bringing the total investment to date over $140 million. This is pretty incredible. You can see that there is a huge amount of funding to going into humanoid robots. $140 million to date in a humanoid robotics startup is a hefty amount considering that this isn't a truly large, large company by any means. Compared to some of the others, you know, they're pretty small. But if you haven't been paying attention to Sanctuary AI, they've actually got a really nice humanoid robotics platform. This is their Phoenix carbon robot. It's on the carbon platform. And I think that, you know, since we're going to have a very interesting future with regards to how humanoid robots are deployed and used in the future with regards to the elderly in care and in factories, I think it's really intriguing to see more and more investment go into companies like this because one of the things that you might not know is that what you're currently seeing on screen right now is actually fully autonomous and most people don't know that. So you can see the robot is, is really smooth and fluid. And I think it's actually genuinely one of the most fluid robots that I've ever seen. Like, this is something that is truly incredible because I remember the first time I was watching the Sanctuary AI robotics demo, they actually showcased stuff that were autonomous and I genuinely believed that it was completely teleoperated because you couldn't tell the difference. So this kind of footage shows you that once we do have robots that are able to move this quickly and this fluidly by themselves to do a range of different tasks, I think the world is going to look radically different. In more humanoid robotics news, you can see figure and BMW groups, Spartanberg plant is actually moving towards real world use cases. So you can see right here that this is a fully autonomous end to end. It's AI driven vision model and it's got neural networks for all grass. So you can see that this is 100% autonomous. You can see right here also that this is, you know, AI driven manipulation. So the reason that I think that this is so crazy is because in January 2024, they announced a commercial agreement with BMW manufacturing. And this is truly, truly impressive because navigation is purely happening from object training in simulation. So you, you can see right here that for those of you who thought that, you know, maybe humanoid robots are going to take a longer time to get not acclimatized, but to get, you know, put into real world use cases. One of the places that I do know that humanoid robots are going to be is they're going to be in factories first. Like you won't see them working in stores, you know, doing your groceries and stuff like that. Although we do have those checkout robots, I guess if you want to call them that. But, but what we will see is we will see humanoid robots actually be used in car factories because the work here, you do need repetitive stuff. And the repetitive stuff that you need to do, it's somewhat precise in a really strange way that a robot that learns based on its neural network, whatever policy it chooses to do, it's going to be able to do that a lot more effectively than potentially some humans could. Because of course there are, are things like, you know, the danger aspect. Sometimes you're dealing with extremely sharp materials, extremely hazardous materials. You know, you've got all of these different factors. So once these robots do get a lot smarter and you can train them to be a lot more efficient, I wouldn't be surprised that if we do see these humanoid robots there being used for, you know, whatever tasks that they might be. So this, of course, right now in its current phase is not something that, you know, small companies can even dream about. These humanoid robots are extremely expensive and of course they cost a whopping amount of capital and labor to produce, considering all of the technical details that go into producing a robot like this. But I think stuff like this does show us that we are actually in the future. I mean, if we actually truly take a look at exactly what is going on here, you can see that we have a robot that has learned how to place these things on this, you know, machine or whatever it is, and it's able to, you know, pick and place it and it's able to do bimanual picking, which is really cool. It's able to grab two with two hands and then it's able to place them, you know, solely just learned from simulation. So I think this is going to be really intriguing to see how these things move forward because we know that not only have figure been doing this, but also we did see Tesla experimenting with, with their Optimus humanoid robot platform. They actually also did it really, really well as well. So I'm wondering if this is going to be the future of factories. Now, as the future of factories continues to go on, what you guys need to know is because these factories are getting increasingly more robotic, the future is not human made. The future is going to be just completely automated. And I'm going to show you guys why. So there was this tweet that was going viral and it basically talks about how Amazon now has over 750,000 thousand robots deployed and each robot packer at a warehouse can literally replace 24 workers. Packers cost 1 million each day, which only take two years to recoup the cost. So this was something that was going very, very viral. This is something that, you know, I was speaking about in my post AGI economics community. We were actually talking about the economics of Amazon and why it's such an interesting company with regards to the innovations it's doing. And I think here you can see that Amazon is taking a very innovative approach in order to making sure it is remaining as efficient as possible. And all of these robots are operating in this grid form. They are all managing to, you know, locate all the packages, ship them out. Of course you've got humans here in the loop, but it is something that is, you know, robots and humans working together that seems to be a lot more effective than just traditional humans in the factory. And of course these robots can work for 24 hours on end. So the, this is likely going to be something that continues to grow as the robots get a lot better. So it wouldn't be surprising if we see these robots more and more in factories. And I wouldn't be surprised if some of them were just completely robot automated. Now this is the startup that I really wanted to talk about before, but I didn't really have the video to where I could even talk about this startup. So if you've ever seen Westworld, and I say that as someone who hasn't even seen Westworld, but I know the concept of the show so I know kind of what they do. But essentially this is clone robotics and this is so incredible because what they're doing is building arguably the most human like robotics ever. Now their website is under construction. They've been essentially just building in stealth for quite some time. But this is one that I've been, you know, really trying to, trying to pay attention to because they seem to be taking an approach that nobody else is taking in regards to humanoid robotics. Now I'm going to show you guys some of their demos because this animation, whilst yes, it is, you know, truly outstanding, it doesn't show you exactly how well their robots are. So what you can see right here, and I'm going to go over some of their tweets because they've basically been showcasing demo videos. You can see here that they are, you know, working on creating robots that have truly lifelike movements and lifelike hands. So what you're seeing here is their hands. Now, I do think that they are probably working on full humanoids, but I think that what they're showing us here is that they are working, of course, on actual hands first. Now, this first hand that you see here is one that they, you know, is kind of like a prototype, I guess, or whatever. And we can see a lot of the wires back here. You can see that all of these wires back here. And the majority of this stuff is covered with this, like, you know, I guess. I guess you could call it not foam, but like this fabric or whatever that they're probably going to be using to, like, you know, grip stuff, I guess, you know, potentially. Or maybe that's just the base material of the robot. But when I first saw this, it was truly impressive because it's the first robot that didn't actually look like a robot. Now you can see some of their, you know, newer tweets. Here you can see about, you know, you know, the ultimate tool user. This is one that I retweeted recently. You can see this is their humanoid hand. A lot of the different applications that they're going after show us that a lot of the ways that they're looking to implement this kind of, maybe it's just going to be a hand, humanoid platform. A lot of the ways that they're looking to implement this would allow for a lot of interesting applications because we know that the human hand is very flexible. It's got huge degrees of freedom, and you're able to do just remarkable things with. With this hand. Like, the human hand is an incredible feat of engineering. And you can see some of these other pictures here. Now, they recently released this pronation and supination thing where they talk about how the hand is going to move. And this one did seem a bit rigid, but there was a tweet from the humanoid hub. This is a great Twitter account, if you're trying to follow humanoid robotics. But he says the movements appear choppy and lacking fluidity. While what is the core issue? How significantly does it impact precision and fine manipulation? Someone that works there, the CEO, I'm not exactly sure who this is, but he says there is no issue. There is an open loop sequence with only four muscle contractions. We'll release the fine manipulation MPC demo soon. So this is just, you know, showing you guys that they're able to do this smoother. And they did. They showed us in another video here where you can see that this hand is moving, you know, incredibly smoothly. So I'm not sure what kind of, you know, engineering is going on here. Know this is definitely a lot more different and it looks a lot more complex than some of the other things that we have seen. But I do like what I have here. I mean, maybe it's just the Westworld in me or the singularity in me that just really wants to believe that this is possible. But it does look like these guys have been doing some very fascinating stuff, some, you know, amazingly impressive, you know, feats of engineering. And I think, you know, maybe we're going to get a demo scene, honestly can't wait. But they've been installed for quite some time, so whatever it is that they've been working on, I think if they do ever do some kind of demo, it probably is going to surprise us. And there's probably going to be definitely like a lot of applications for this kind of technology if they manage to get it working truly effectively. So I'm not sure if I actually managed to keep this clip in before, but what I did want to show you all was this, where Bill Gates was actually talking about metacognition and stating that the next big frontier is, is metacognition. And that's just basically where AIs can think about the reasoning. Honestly, I don't remember if I covered this before, but I do think it is a pretty important post. And there is something else that I do want to show you guys as well, because there was a recent interview post with someone that I wanted to show you guys, but this is important. But we'll get into it where by.

Accessing video data and getting very good at synthetic data that we can scale up probably two more times. That's not the most interesting dimension. The most interesting dimension is what I call metacognition cognition, where understanding how to think about a problem in a broad sense and step back and say, okay, how important is this answer? How could I check my answer? You know, what external tools would help me with this? The overall cognitive strategy is so trivial today that it's just generating through constant computation each token in sequence. And it's mind blowing that that works at all. It does not step back like a human and think, okay, I'm going to write this Paper. And here's what I want to cover. Here's, okay, I'll put some facts in. Here's what I want to do for the summary. And so you see this limitation when you have a problem like various math things like a Sudoku puzzle, where just generating that upper left hand thing first causes it to be wrong on anything above a certain complexity. So we're going to get the scaling, but at the same time the various actions to change the underlying reasoning algorithm from the trivial that we have today to more human, like metacognition. That's the big frontier.

So basically what he's talking about in this is that the way that humans think is a lot different to the way that AIs think. When we ask an AI system something, we're asking it zero shots. So for example, when we ask it, hey, can you write me this essay? It will write an entire essay from start to finish. And, and it won't essentially pause, reflect, rethink, think about its response. And it doesn't reason, it just generates. Okay, so those are different things. Now, essentially what Bill Gates is saying here is that, yes, certain things happen with scale. You can improve the model via just increasing parameters, data size, just scaling up everything, and you do get more precision with the model. But what you need to also do is you need to add in an architecture which is, you know, quite similar to neurosymbolic AI in the sense that you take advantage of the different aspects of what makes, you know, humans good at, you know, just general decision making and, you know, being able to truly understand the world. So it's basically thinking about how, you know, you can invent new architectures and new ways for these systems to think into, that they get a lot more effective at what they do because currently the methods of reasoning are extraordinarily limited. And, and I think that's just something that's really important. So this is basically what he's talking about. And I think it's important. I think it really is important because if we just keep thinking, oh, okay, just scale, scale, scale, scale, we're not really tackling the problem. But I did say to this that it is very likely that companies like OpenAI and Google are actually working on these things because I've seen a ton of open research papers and they seem pretty promising already with ways to be able to increase reasoning capabilities. If you didn't check out my recent QStar video, I covered videos on how AI systems are going to be thinking in more, I guess you could say, smarter ways that allow them to think a lot more intelligently as a human would. And here you can see state of the art LLMs are four to six orders of magnitude less efficient than the human brain. A dramatically better architecture is needed to, to get to AGI. So he says you just need more data and more scale. The data will keep the thing getting better. But of course we're hitting the limit where we've used all the data on the Internet or in the world or whatever. So what you think the odds are that the wall exists versus us just finding a way because we're humans and this is our way to find through. And I think that this is where the true like insane gains are going to be made. Because you know, like, think about it like this. If you, if you remember what AlphaGo was like, how good it was when it was trained on humans, everyone was like, oh yeah, that's pretty crazy. But when it got really crazy, like when it actually went insane was when it used an algorithm that was trained on itself and it just used, I guess you could say, kind of like a new architecture that wasn't just fundamentally focused on human data when it was just trained on synthetic data. That was when it just got superhuman. So I think now that humans are then going to move towards this area where we're like, look, we've maxed out the data, we've maxed out the scale, this is the limit. That's when things are going to get really, really crazy. Whilst yes, there might be like a. Not pause, but some kind of period where there aren't major, major breakthroughs, at least on the external, I think they're definitely going to be made on the internal. So he says the reason he's doing that is because state of the art LLMs are about four orders of magnitude less efficient than your brain. If you think about the basic input as energy and the basic output as computation, your brain is 10,000 times more efficient than a state of the art LLM. Now Andrej Karpathy has had the same observation, but he thinks that it's six orders of magnitude. So one way or another, these models are dramatically less efficient than the brain, which is of course, you know, I previously said that, you know, robots, you know, just like require so much energy, electricity, whilst humans can literally eat like a packet of Doritos and be fine for the entire day. You know, technically speaking, of course you still need some actual nutrients. But you know what I'm trying to say, like humans, the output ratio is just insane. So the reason that's important is that nature has shown us that a better architecture exists. And I have to imagine long before we get to anything that is universally agreed upon as AGI, we're going to end up with a dramatically better architecture that's going to be far more energy efficient. And it's not going to come out of just optimizing transformers. And it's probably not just going to come by putting some planning and reasoning on top. It's probably going to be some different base architecture. And at some point I'm sure somebody will figure out what that is. So I do think that this, this is going to happen now and I do think that this entire scale up is a little bit different because previously there was of course an AI race. But I think with OpenAI now, what's kind of happened is you've had that situation where this company has ignited the fuel and now there's so many different people working on this stuff. There's so many more investors in the space, there's a lot more eyes on the space. So it's like this has just accelerated the development even further. Because when you have, you know, an entire ecosystem of everyone sharing work, that's fine. But when you have people competing against one another and you've got, you know, financial incentives, it's really, really going to be a lot more, you know, intense with as to how many breakthroughs are going to be, you know, had. So this is something that I think is, you know, remarkable when we actually think about the future of AI. So that brings me to this part of the video where you can see that there is actually going to be a Gemini update. So apparently the Gemini update that was scheduled for 6 June 2024 has been delayed to June 18, which means we could be getting some very important updates from the Google team. Now this brings me to an important point where I'm talking about Gemini, because one of the things that Gemini actually did update with was the ability to create images. However, if you do type this prompt in, or any image prompt, for example, you'll see that you get a response stating that, I'm sorry, I just can't create images yet. However, this is where today's sponsor comes in, because you actually need to use them. You actually need a VPN. So the sponsor of today's video is actually NordVPN, but I'm going to show you why you actually need to use them if you want to access advanced AI features. So what I'm going to do is I'm now actually going to switch my location to America. And you're about to see something crazy. So now that I've taken a quick flight to New York, we're going to go ahead and see this prompt again. Let's see what Gemini says when I say can you an image of a dog. Now that I've applied the vpn, would you look at that. NORDVPN allows me to actually access advanced AI features that I literally otherwise wouldn't be able to. And this is why today's sponsor of the video NordVPN is offering you guys a really cool exclusive deal to the channel. NordVPN actually allows you to instantly switch your location to anywhere on the planet. And it's something that is really, really useful. And if you want to be able to securely access different locations and of course access AI features that just simply aren't available in your region yet due to whatever restrictions there might be. Now one of the coolest things about this offer is that every purchase of the 2 year NordVPN plan will actually receive 4 months bonus on top. And on top of that you actually get 20 gigabytes of saily data vouchers. And Saily is basically a new E server sim app that is brought to you by NORD Security. And it literally just allows you to have Internet wherever you are on the planet pretty much. So if you travel or stuff like that, this is definitely going to be useful for you. Now don't forget NordVPN do have a 30 day money back guarantee. So if you have any issues you can always get a refund. And my exclusive link where you get the deal is actually nordvpn.com theai grid but with that being said, like I was talking about Google, you definitely do need to access some kind of VPN if you want to access these kinds of features and VPNs. And this is not even like part of the sponsored video but, but VPNs like NordVPN have actually been pretty useful because we do know that all of these AI companies, for whatever reasons and restrictions, regulations going on in the eu, sometimes they do just roll it out to America first where they will test the product and then roll it out to the eu. Now this is pretty crazy. We got an update from eleven Labs. I personally wish they would just focus on making the like emotions in their speech models. I don't know when that's coming but I think that they should do that. But essentially they introduced a voice isolator, remove unwanted background noise and extract clear, crystal clear dialogue from any audio to make your Next podcast or interview sound like it was recorded in the studio. You can actually try this free.

Need to remove background noise from your video.

Use our new voice isolator model for crystal clear audio every time. That's pretty crazy. That is actually pretty, pretty crazy. Like imagine being able to do this and you can literally just pretty much record anywhere you want. I mean, that's a game changer for content creators or for people that don't have an environment that is, you know, really effective. So you can pretty much record anywhere. Just clean up the audio and then you're good to go. I mean, this last piece of audio sounds every time super, super clear. I mean, it's just incredible what technology is able to do. There was also this. I was actually meant to get to this earlier, but I didn't. It says Eureka. Introducing the world's first AI doctor. Eureka can order labs and give care in the real world. It is covered by health insurance just like a healthcare provider and is already working with thyroid patients in the United States. It's 90 times faster than most care in the United States. And nine out of 10 users want to continue with Eureka's recommendations. Eureka thinks like a doctor and reasons like a detective. Before any care starts, a board certified physician reviews Eureka's recommendations to make sure that everything is in order. And Eureka currently specializes in an endocrine conditions like thyroid and diabetes. So you can see right here, this is the video in which they discuss how this system works and how effective it is.

I'm Sina and I'm excited to introduce you to Eureka. Eureka is the world's first AI doctor that can order labs and deliver care in the real world. Let me show you Eureka in action. Eureka gives you fast and personalized care for your conditions. Eureka has unlimited time, attention and patients. When Eureka learns that I have hypothyroidism, it knows exactly what questions to ask next. Eureka can do this because it thinks like a doctor and reasons like a detective. In a few minutes, Eureka discovers that I'm taking too little medication and then recommends an adjustment to improve my symptoms. Eureka then orders follow up labs for the future to monitor my improvements before any care starts. A licensed physician reviews your recast recommendations to make sure everything's in order. Working with Eureka is almost 90 times faster than traditional and specialist care in the US and users love Eureka. 9 out of 10 users want to continue with Eureka suggested care. And maybe that's why Eureka also receives three times more excellent care ratings than the average care visit. We built Eureka to take symptoms seriously. And to give incredible care 24 hours a day, seven days a week. Eureka currently specializes in endocrine conditions. And if you have diabetes or a thyroid condition, you can get care now by visiting eurekahealth.com There was also this.

Rather interesting perspective on how AI is going to be in terms of dominating the human race in certain aspects. Now many people think that superintelligence is the biggest problem that we need to face. I don't want to say I disagree, but it's not something that I currently think about. I think about stuff that are more near term. And he. Ilya Polasukin says that there's no economic incentive for AI to wipe out humans, but there is a risk that we become so addicted to the dopamine systems they feed us to keep us in tame that we end up in a deocracy. And I agree with this, because if you don't know already, like even this video that is, you know, that you're watching right now, this video appeared on your feed because of the algorithm, the YouTube algorithm. The YouTube algorithm is AI, you know, determined. You know, AI determines, you know, what it's going to show you. And essentially that means that AI is already deciding a lot of the things that you watch. Because humans are intrinsically lazy. We pick up our phones, we don't go on the subscription feed. We just look at what's first on our phone, and that's what we watch. And because of that, my video is here. Hopefully it was. And I think that in the future, things like that are going to get a lot more exacerbated as these models become even better. You know, just being more engaging, not a human.

Right. So unless somebody goes and says, let's kill all humans, it's not going to go. And like, you know, magically do that. In the blockchain world, you realize everything is driven by economics one way or another. And so there's no economics which drives you to kill humans. Like, that's just not like this. Even if it wants kind of freedom, et cetera, well, then it just flies out in the space, et cetera, right? But the reality is what we're doing right now, we're building systems to improve ourselves, to further what our minds can. And yes, we'll have autonomous agents that will have specific missions and goals and we'll be doing things, but at the end, when they are acting in the physical world, it's governed by the same kind of laws that we are governed here. If somebody, you know, uses AI to start building biological weapons, it's not different from them trying to build biological weapons without AI. And so ideally with AI, we actually should be better at finding that out and actually preventing this. It's people who are starting the worst, not the AI in the first place. The more realistic scenario is more that we just become so kind of addicted to the, you know, dopamine from the systems which are just trying to keep us entertained and we are more just stop trying to ourselves become more intelligent and more sophisticated using these tools and using this kind of systems as bicycle for our mind. And I think that's what I would like to prevent is kind of the kind of convergence into like, oh, actually I just can stare at my phone and scroll through things.

Idiocracy.

Yeah, that's what I was referring to.

There is also going to be Grok 2 out in August. So keep your mind out for that because you know, I would say, I gotta be honest, there is a bit of nuance to this because, you know, GROK is, you know, I think I have grok. Do I have grok? Like, it's weird, okay? Like super weird system the way how you use the system. I don't know, it's weird, okay? Like GROK is just a really weird system, if I'm being honest. So I would like these guys to have an actual website, not just baked into the X platform, so you can actually use their models for whatever it is that they might do. But you know, either way, apparently Grok 2 is going to be out in August, so it will be interesting to see what that that is because last time Grok 1.5 vision actually did some remarkable improvements. Meta also published this paper, which is 3Dgen, a groundbreaking technology introduced by Meta for generating high quality 3D assets from textual descriptions. It leverages two primary components, Meta3D asset gen for generating 3D shapes and Meta3D texture gen for generating detailed textures. With 3Dgen, users can quickly produce 3D models complete with textures and physically based rendering PBR materials, allowing these assets to be relit in real world applications. It's pretty fast. Creating detailed and accurate prompt 3D models in literally just under a minute. And it even actually supports retexturing and redoing the models based on new textual input. So it works in two ways. It basically generates a 3D mesh and then with the basic textures from the text prompt using Mesa 3D asset gen and then a second tense. It just enhances the texture quality and then using the details. The Meta 3D texture gen based on either the original or new text Prompt. So this kind of integration, they've kind of combined two things together to make something that is really effective. And it's something that I think is really cool here. You can see that it's able to texture many different things in multiple different ways. You can see that the theme here is some kind of cotton wool theme. So, yeah, I think this is going to be something that slowly, slowly improves. Because when I was first looking at these kinds of papers, it was always kind of interesting to see how bad models were. And every year we're literally seeing small, small and smaller and smaller improvements on how 3D models are being generated just simply based on textual prompts. So this is going to be something that I think once it's integrated into the 3D world, it's going to be something that's remarkably effective for just many different use cases. And lastly, but not least, once again, we've got some information about AI agents taking a much longer time than you think because they essentially require you just a lot more precision. And that's probably only going to happen with scale. And that's something that we saw Mustafa Suleiman say, because essentially, you know, these agents do require a lot more, you know, accuracy if we want them to chain a lot of different actions over a long period.

If you want an agent to act in the world, usually that acting requires you to, you know, engage in a series of actions, right? You talk to a chatbot, it only answers and maybe there's a little follow up. But with agents, you might need to take a bunch of actions, see what happens in the world or with a human, and then take more actions. And so you need to do a long sequence of things. And for that long sequence of things to actually work, the error rate on each of the individual things has to be pretty low. Right? If I'm a robot and I'm like, you know, okay, I'm going to pick up this thing and walk over there and I'm going to pick up that, you know, I'm building the house or something. There's probably thousands of actions that go into that. And so all of this is to say the models need to get more reliable because the individual steps need to have very low error rates. And I think part of that will come from scale. Like we need another generation or two of scale before the agents will really work, if you want.

So don't let that put a dampener on your enthusiasm for AI. Hopefully you guys all did enjoy this video. Hopefully you guys do have a great weekend as well. And I'll see you guys in the next video.