AI NEWS : Googles Gemini 2 , New Incredible Humanoids, AI Gets UNLEASHED, And RoboTaxis

I wanted to show you all was basically this robot here. Now the reason I want to show you guys this is because not only did I cover this incredible humanoid robot from a few days ago, the fact of the matter is that when I published this video, something shocked me. A lot of people in the comments, like a lot of you guys didn't believe that this robot was real. Like a lot of people saw this robot, saw the demos, saw the trailer, and they were like, this is CGI now. You know what that makes me believe? That actually makes me believe that AI robotics, humanoid robots, that entire industry is moving so fast that even people who are in that area, in terms of their interests and in terms of their hobbies, even they cannot literally believe what they are seeing. Now of course, maybe it was just a demo that looks like cgi, or maybe it was just a poorly recorded video. But upon further evaluation, we actually did get more high quality videos that show us that incredibly, it isn't actually cgi. For example, in this video right here, this, you can actually see this humanoid robot walking around and looking extremely realistic. I think the problem here is the fact that a lot of the humanoid robots we've seen previously in earlier iterations essentially just don't look as good as this. And because of that, most people just can't believe that this thing currently exists. I know for one, when I did see the other engine AI robot and I did see this, I was really initially skeptical. Not just due to the fact that I think technology is not advancing that quickly, but one of the main reasons I think people are skeptical is because some demos have actually been CGI in the past. One of the Apollo humanoid robots were actually cgi. But of course this one you can see right here is not CGI at all. Now remember, this is a company called xpeng. They've been actually working on this kind of humanoid robot for five years. So that means that there isn't. So that actually means that this kind of humanoid robot isn't their first model. This actually means that what we're seeing is likely the work of several iterations. Remember, companies like Boss Dynamics have worked on their robots for many, many years and they've shown us the updates time and time again. And this company decided to just show us one of their updates out of the blue, showing us that, look, as long as you work hard, you can make progress in robotics. And remember, it wasn't that long ago before these humanoid companies started started doing work figure AI. They only started, I think two years ago. Tesla's humanoid robot program. It only started in what, 2019. So when you actually think about how much time these companies have had to actually make their products, it hasn't actually been 10, 20 years. It's only been three to five years at max. So the rate of development we're seeing here from idea to inception is actually pretty good. In the grand scheme of things. 2024 has seen high unemployment rates and even higher layoffs. And that's even if you have a well paying job today, you could be jobless tomorrow. This is pretty scary when you think about it. But let me tell you, there's a way to protect yourself and your family, which is by building multiple passive income streams. I'm talking about learning and using AI as a skill to earn an extra $10,000 every month. I know who can help you here. Growth School is offering a three hour hands on AI training where they teach you how to use 25 plus AI tools. You can register now with the link in the description and trust me, this training is game changing for people who want to grow in their career. And here's the best part. This is normally a paid training and but this is exclusively for free for the first 1000 AI grid viewers who join using the link below. Yes, for free only for AI grid viewers. People who register will also receive a free $500 bonus worth of extra resources. It covers everything from job hunting strategies and salary negotiation techniques to mastering Excel and creating content. This AI training isn't just for tech folk. Whether you're in finance, sales, marketing, hr, a student or entrepreneur, this training is definitely for you. Grow School has helped over a million people upskill across the globe and this is your chance to join that top 1% in an AI first world. So click the link in the description and snag your free spot. And this is the only chance to learn those skills that will secure your future. And don't forget to join growth schools WhatsApp group for regular important updates. So then we got an update from the Nadia robot. Now Nadia is a humanoid robot developed by the Institute for Human and Machine Cognition in collaboration with Boardwalk Robotics. It features a high power to weight ratio it and 29 joints providing exceptional mobility and a wide range of motion. Now this robot is designed for tasks in complex environments and she can actually perform dynamic activities like boxing and opening doors autonomously, which is the update that we got today. So this was one of the updates that I think is really cool because for those of you, as like I said in the previous video, for those of you that think you can Escape humanoid robots, they're teaching them to unlock doors now. So it's going to be pretty interesting with as to how they manage to do this because I think in the future humanoid robots are going to be a lot more advanced than a majority of people. And when I say that, think about it like this, okay? They are going to design these robots to have the fastest speed, the most agile hands and the most degrees of freedom rotations that you can't even imagine. And when you think about the average person, average person doesn't even go to the gym. Now I'm not, I'm not saying it like I'm some, you know, famed bodybuilder, but I'm just stating that we're basically going to build super athletic, super smart, super crazy beings that I think are really going to be interesting when we finally get a look at how this all manages to evolve in the next 20 years. Seems like opening doors will be the least of these robots problems as they manage to develop a lot better. Next we actually got some news from the LMSYS arena and people are now reporting that they are currently getting models called Gemini 2. Now I did report on this I think around one to two months ago but apparently this model is apparently outperforming 01 mini, which is rather surprising considering the feat that open AI have achieved. So I mean I think this is going to be something that happens a lot more because I think Google haven't had an update in quite a while and that's respective to the AI industry because you know, six to seven months is not that long when you think about a major AI overhaul. So I think the fact that you know, Google are testing their second model, I think it's going to be really interesting to see what they have. And of course this wasn't the only type of information indication we got that Gemini 2 is around the corner. We also got this screenshot from testing catalog basically saying that look, Google is preparing to launch a new model, Gemini 2.0 Pro Experience 0111. That does seem like some internal name for their second iteration of the model. And this actually did get a lot of attention and it seems like this user is able to get the experimental model. I personally don't have access to this. So if you do have Google Gemini and you want to test out, I don't know if this is genuinely number two as in they've released this just yet. Google haven't said anything yet but this testing catalog page is definitely very, very reliable when it does come to AI news. So I wouldn't be surprised if this is real, but I will wonder how they're going to finalize the launch, because of course, I think what they're doing right now is essentially just testing it out on users on a broader scale so that they can sort of fine tune the responses and see, okay, this response was good, that response was bad. This response need a little bit tweaking. The real model, you know, it performs well in this area, performs bad in that area. And of course you can see it's just an experimental model, it's not an official Gemini 2 model. So I think it will be interesting to see what Google manages to do on those benchmarks and see how the AI industry reacts once it's able to do that. And if you do remember, it was recently that we also got Google's Gemini 1.5 Pro 002. So there is actually a lot of indication that we're moving towards further iterations of the Gemini model series. So I wouldn't be surprised if this model does come out fairly soon or maybe some kind of iteration. Maybe it's Gemini Ultra gemini advanced version 8. I don't know. I. I'm trying to name names, but I really can't name names. I'm just trying to say that, look, we're definitely entering that era where Google are showing us that they still have a lot of the secret sauce. And I would say for those of you who find that maybe you have certain specific tasks and that LLMs currently struggle with, do try and see if these models do manage to succeed. Because I still currently use three different models. I use Google Gemini for a lot of anything that's related to YouTube. It just performs exceptionally well compared to models like Claude and chatgpt. And I think it's simply based on where these models get their training data from. I don't think it's a surprise that Google models perform better on social media considering the fact that it was largely trained on YouTube data, considering Google owns YouTube. Next, what we also got to see here is, was the Zoox Robo Taxi spotted for the first time ever in San Francisco. So I think this is a huge milestone because one of the things that has been long discussed in the AI industry was the fact that, look, autonomous vehicles just simply aren't coming. There have been so many different articles, so many different people just simply hating on Elon Musk and stating that, look, every year you say next year, autonomous vehicles is just simply a fad and it's never going to be a reality. And in the last six months we've seen Waymo reach 150,000 trips per week in I think San Francisco or somewhere in America. And that number is steadily rising. And now for the first time ever, we see Zoox, which is an autonomous vehicle that currently is only testing for the employees of Amazon. But we can see that it's currently being tested in the real world. And I think this is going to shake up a lot of industries because if travel is completely automated, then that frees up humans time to do a lot of other things. I do remember watching so many different TV shows where you could literally see characters in autonomous vehicles just doing different things. And it did seem so futuristic at the time. But now the future is literally here. So we can see here from this article from the San Francisco Chronicle. It says Amazon owned autonomous vehicle maker Zoox is is planning to offer rides in San Francisco after years of testing, but only to employees and in select locations to start. It says. Zoox Co founder and GTO Jesse Leventon made the announcement on stage at the TechCrunch Disruption Conference in San Francisco on Wednesday, the outlet reported. And I think this is a major milestone because if not only Waymo are starting to get into the area where of course you have self driving cars, I think it just shows us that these companies are finally starting to iron out all, all of the previous kinks that we thought we had with these cars. And then of course, I mean, we can imagine what society looks like in 10 years. Of course some people are still going to be driving. But you have to understand that this is a major piece of technology that literally saves people time. Imagine instead of driving to and from somewhere an hour every day, imagine you could use that hour to potentially get extra sleep. You could use it to study, you could use it to read a book or learn a new skill. I mean, honestly, the possibilities are endless and it does sound crazy, but this is what technology acceleration feels like. So we also have this news, which, I don't know how to say it, I mean, I think this one has the AI community divided. I think it's like 80, 20. And this is the kind of news that I don't think is essentially positive news for most people, because the Godson's truth is that I'm not entirely clued up on the details, the specific details of what Palantir does or any specific wars going on at the moment. But what I will say is that it seems like these companies are embedding their technologies into the government for war. Now, it's not something that I want to cover too much because it's not exactly the best kind of news. But it does seem like this kind of technology would eventually be used by the US government at some time. So I don't see this as something that is surprising. I knew it was going to be some kind of company, but what I will say is that it is surprising to me that Anthropic is going to be the company that manages to do this, because I'm guessing it's due to the connection between Amazon and Anthropic. If you aren't sure with what's going on there. Amazon actually have a huge investment in Anthropic. So I'm guessing that with Amazon web service maybe they have some deal with the US government, but overall they're basically saying that they're going to be using Claude to summarize a lot of data and for a lot of different things like US defense and intelligence communities. And of course I'm guessing that they're going to be developing some kind of models that are really secure that they can use in without having their data compromised. Of course security is going to be one of the main things that you need to focus on, especially when dealing with things like US defense. So I'm pretty sure they're going to be developing some kind of AI system that is off grid that the US government can use and engage with. And I do remember that there was something like this with ChatGPT, although I can't remember the exact story right now, but I think that is the extent to what this is going to be. So one of the craziest statements I actually heard from the Sam Altman interview was this statement about what abundant intelligence and abundant energy will lead to. Now. Now I saw a lot of people giving this statement on Twitter. Some flack because of Altman was talking about Dyson Sphere, but essentially in theory what he's saying genuinely does make sense. Although it does sound crazy right now. If you do get abundant intelligence and abundant energy, literally anything is possible.

I think these are like the key to the two key inputs to everything else that we want. There's a lot of other stuff of course that matters, but the unlock that would happen if we could just get truly abundant intelligence, truly abundant energy, what we'd be be able to make happen in the world, like both, like come up with better ideas more quickly and then also like make them happen in the physical world. Like to say nothing of it'd be nice to be able to run lots of AI. And that takes energy too. I Think that would be a huge unlock. And the fact that it's, I'm not sure to be whether, like, whether to be surprised that it's all happening at the same time or if this is just like the natural effect of an increasing rate of technological progress, but certainly very exciting time to be alive. Solar plus storage is on a good enough trajectory that even if we don't get a big nuclear breakthrough, we would be like, okay, ish. But for sure, it seems that driving the cost of energy down, the abundance of it up, has like very direct impact on quality of life. And eventually we'll solve every problem in physics. So we're going to figure this out. It's just a question of when. And we deserve it. There's, you know, someday we'll be talking not about fusion or whatever, but about the Dyson sphere. And that'll be awesome too. Yeah, this is a point in time. Whatever feels like abundant energy to us feel like not nearly enough to our great grandchildren. And there's a big universe out there.

With a lot of that. We actually also got a fascinating demo slash use of the CLAUDE computer agent, which is where CLAUDE takes screenshots of your computer and then manages to do certain tasks on that. So this is where Ethan Molik basically fed a video in to the CLAUDE computer agent. And you can see that it is taking screenshots and then it is looking at what is wrong on the site and then it's writing down any site violations. And in the Excel spreadsheet, so you can see here, it looks like there are some immediate safety violations. You can see you're supposed to use color codes for this. You're supposed to do that. You can see inefficient material, staging, poor tool organization. And it's pretty crazy. Like, I don't know if you would say that this is dystopian, but having an AI agent look around at security cameras, maybe, maybe, guys, we could even have a humanoid robot that is able to walk around certain sites, but look around, screenshot the data, put it into an Excel table, and you literally sit back and relax as you have some kind of manager humanoid robot, walk around construction sites and look at everything that is going right and everything that is going wrong and updating it all in real time to some kind of Excel table. And this is kind of the future, guys. Like, this is the kind of thing that we will see in the near future. And I think this kind of thing is, you know, sped up a little bit in the sense that a lot of the times a Computer agent is quite slow. The responses from Claude 3.5 Sonic do take a lot of time to get back and, you know, to execute one response to another response. And of course, for certain moves it does take quite a while. But with all of that, I do think that it's going to be pretty crazy to see how this computer agent manages to evolve over time. So now I also wanted to show you guys this because this was going absolutely viral on social media. But essentially what we have here is, is Volkswagen's tool that allows you to autonomously save your life potentially. I know that sentence was pretty awful, but basically this just saves the driver in case of any emergency, maybe if they pass out, maybe if they lose consciousness, maybe if something is happening in the car, the car is automatically able to get to a safe spot and pull to a complete stop. And I think this kind of thing is, is really important, perhaps, probably should be a safety feature in all modern cars after a certain year, because I do believe that, you know, instances where people do unfortunately lose control of the car happen more often than not. And you might be thinking, okay, but what does this have to do with AI? I do think that it is some kind of AI that is able to detect when a user is not present. So I do think that in the future, when we do have even more advanced AI, this kind of technology that's going to be able to, you know, perhaps look and say, okay, with a model like GPT4V or Claude Vision, be able to say, okay, I can see the user is not well. I can see that there's a rash on the user, yada, yada, yada. And maybe what we could even have is an agent that's able to call the police station and actually inform them of exactly what's going on. They can say, yes, the user has stopped here. I didn't hear them breathing, I didn't hear any kinds of thing. We broke the car multiple times times. And it could even give the police or the medical services updates about where that user is in terms of exactly what's going on with the car. So I think that kind of thing is going to be pretty crazy. Like, imagine you got into a car crash and your car was stuck, but then you hear your AI agent calling the police, and it's able to accurately describe exactly where you are, what's happened, and it just tells you to sit tight. Now, of course, I hope none of you ever get into a car crash, but it is something that I think the future of AI is when we, of course, are only paying attention to the technology in terms of these crazy updates that we're getting. I think when we look to how this AI can actually, you know, save lives and improve lives in terms of safety, I think it's going to be really easy to implement some things that most people wouldn't think about and that would help us in a variety of different scenarios. Now, a crazy thing as well is that of course, most of you guys would have known by now. In fact, literally everyone knows by now that Trump is now the President of the United States. 47 President of the United States. Now, what's crazy about this is not the fact that he's present. The fact is that he's planning to unleash AI by repelling restrictions. Okay, so it says we will repeal Joe Biden's dangerous executive order that hinders AI innovation and imposes radical left wing ideas on the development of this technology. Now, I think you guys need to understand the implications for this. Basically, the executive order was something that I guess you could say slowed down AI progress. It was just, you know, a bit, somewhat restrictive. So the fact that Trump is saying that, look, we're no longer doing this, we're taking this thing off. And for example, it even goes into how it says one lays out new requirements for how tech companies test and conduct risk assessments for their AI models, a practice known as red teaming. And under the Biden provision, companies developing large AI models are required to share all red team test results to the federal government for review, assessing things like how vulnerable the AIs are to being hacked. And this is something that is no longer going to be there. Now Trump is basically saying that this is a radical left wing idea. I don't know which way you guys swing on this. I'm not someone who really involves myself with politics because I know that it can get really divisive at times. All I will say is that the changes that this now is going to have in terms of knock on effects is the fact that like, okay, companies are going to be moving a lot faster with how they develop these models. Which means that if you thought the AI ecosystem was going to slow down, this shows us that, look, for the next four years, regulation is not going to be impacting these models at all. All which does mean that, boy, oh boy, things are going to be moving even quicker. Because one of the things that I genuinely did think was going to happen was that at some time there was going to be some kind of regulation, but now that you know, it's pretty much all systems go this just shows me that, okay, things are about to get really crazy because if there are no regulations saying that look, you don't need to red team, yada yada yada, basically just means that, look, these models are going to be deployed quicker than we initially expected. Now, Epoch AI have essentially produced one of the best benchmarks when it comes to Frontier math. They basically have invented this new benchmark that that is supposed to be resistant to future AI models. As you know, one of the things that Sam Altman recently said was they are going to probably accelerate all the benchmarks in 2025, which essentially means that these systems are getting so smart that we need new ways to test their intelligence. And this benchmark basically has current models at around 2 to 1%. So this frontier Math benchmark is one that is going to really test the capabilities of these models.

When I was doing a math PhD at Harvard, a recurring topic in the math lounge was artificial intelligence's progress in math. We'd see news about AI solving Olympiad problems and stories of experts in other fields misjudging how soon AI would beat the best humans. And inevitably the question would arise, are we next? How long will it be before AI can research math better than us? Not all of us were concerned by AI acing math contests. Most of our work takes years. I hear everything from humans will be obsolete as researchers in the next five years. The claim is that AI won't grasp the complexity of modern math research within our lifetimes. So how do we determine who's right?

There are benchmarks that assess basic mathematical proficiency in AI, but none that reflect the depth of contemporary math research. At Epoch AI, we're interested in measuring and tracking progress in artificial intelligence. Existing benchmarks for complex scientific reasoning are close to saturating. We need something new to be able to tell how much progress we're making towards expert level ability. And that's why we've built Frontier Math. We worked with over 60 mathematicians worldwide, professors, IMO writers and fields medalists to produce hundreds of original, extremely challenging math problems.

These are actually genuinely hard problems. One of the things that will be, I think difficult is a lack of training material like pretty niche. It's probably not that well documented.

Yeah. So I took a look at the 10 problems you sent. I think of the I could do the analog number, three ones in principle and then the others I don't know how to do, but I know who to ask.

The problems of Frontier Math range in scope from Olympiad style puzzles to research level challenges and span across all the major fields. Of mathematics. They're also beyond the capabilities of current AI. We've tested the most advanced systems available and each exceeded in less than 2% of these problems.

In the near term. Basically, the only way to solve them, short of having a real domain expert in the area, is by a combination of a semi expert, like a graduate student in a related field, paired with some combination of a modern AI and lots of other packages and things like that.

Although Frontier Math measures advanced mathematical reasoning, we use problems with closed form answers like integers, so they can be automatically verified without significant advances in the math libraries of proof assistants. This is a challenging design constraint for the mathematicians writing these problems, whose research is almost entirely conveyed through proving theorems. An AI capable of solving this challenge will have a drastic effect on mathematics. As a mathematician, I am excited to see how AI systems will transform our field, not by replacing us, but by becoming a powerful collaborative tool that will expand what is possible in mathematics. As a mathematician, I am excited to see how AI systems will transform our field, not by replacing us, but by becoming a powerful collaborative tool that will expand what is possible in mathematics.

So one of the things I actually wanted to quickly show you guys was this right here. You can literally see, okay, and this is pretty surprising, is that O1 mini and O1 preview and actually don't solve more problems than Claude 3.5 Sonnet and Gemini 1.5 Pro 002 in Frontier Math. And this, I don't even know what to make of this because we do know that essentially 01 Preview and 01 Mini are touted as the models that are better in terms of overall reasoning. But when we look at this chart in terms of mathematics, it seems that 01 is falling behind. Now, of course, this isn't the full O1 model, and it's quite likely that we're going to have Orion and GPT5.