GPT4o Voice DELAYED! Amazons Secret AI System, Actual Terminator Robots, GPT-5 PHD Level + More

So the crazy week in AI. Here are some of the best stories that you did miss and you do want to take note of. If you're someone that does work with Excel Tables a lot. There was this new tool released that is called Otto. Now, I've seen a similar tool to this before. I'm not sure exactly what it was called, but I will leave a link to it in the description. But this one has been widely accepted as something that is very, very good because it's multiple AI agents in Tables. They do have a short demo here that I'm going to show you guys and then we're going to get into today.

I'm excited to share with you, Otto a new way to interact with AI. With Auto, Tables become the interface to AI agents, where columns are the building blocks of powerful workflows. Here I decided to ask Auto to help me research the hottest AI startups. I'm looking for a lot of different information, like the total amount raised, company description, the CEOs, the founders and market verticals. To start, we actually use Generative UI to help kickstart and build a table for you. And if you really want, you can go in and configure the columns with the prompts and the tools the way that you want it. So here I actually want the total amount raised in a slightly more readable format, so I just give it the instructions to do so. Before we start, we need to add some entities to the table. There's two main ways to do this. One, you can add any data you might already have, either from a list or an external data source. Or you can browse the web with an AI and assistant and it can try to extract the relevant entities for you. Then, for every single row and column, we spin up a brand new agent. These will all run in the background asynchronously, and once they're done, you can check in on them for every single cell. You can actually verify where we got that information from clicking in, clicking sources and seeing exactly where we sorted that data from with Auto, we decided to take a very different approach. The majority of the AI tools that you use today are all chatbots with Chat ux. Now, while that's great for kind of answering quick questions or conversing and ideating, they're actually really not that good at doing a lot of work and automating a lot of repetitive tasks that we spend the majority of our time on. That's why we ditch the chat in favor of a table. Now, this isn't just an ordina TABLE Every column is actually the instructions for an AI agent to do very specific tasks. The added benefit is that we can automate hundreds of these tasks all at the same time. So I've recently been wanting to take a deeper look at the crazy financials Nvidia the last few quarters so I just asked Auto Help me analyze Nvidia's quarterly finances and it gives me a pretty good preview table with columns to find things like net income, revenue, total assets and liabilities. I'll just go in and quickly modify one of these. Then I'll just upload a bunch of Nvidia's financial docs that I just grabbed from their site, which is just from their last few quarters, and I'll upload them and I'll let it get to work here. I noticed that the documents didn't have EPS inside of it, so I decided to add web search as a tool and then use the quarterly data as additional context to help the agent find that number for me. Even with ChatGPT, you would have to sit there for every single file and talk with it, ask questions and wait for his response. Now you can do it in a matter of minutes and the data is formatted exactly the way that I want. While Auto is great at doing market research, browsing the web and web scraping at its core is just an AI agent with a set of tools. That means you can give it files, images, URLs, or even access to your own APIs. By giving them very specific tasks, it becomes a lot easier for us to guide them. So instead of trying to think about the market vertical, the CEOs, the founders, it only needs to focus on one specific thing at a time. Every single cell becomes its own AI agent. Not only does it make them more accurate and reliable, it also doesn't blow through tokens because it doesn't need to solve multi step multitask plans. We hope you're as excited as we are. We're moving fast and we have a ton of features planned. If you want to be part of the first people to help us shape the direction of the product, you can sign up for early access on AutoGrid.

AI wasn't that crazy in terms of the news but but I did want to include this because it is rather important. So you can see someone here commented saying fantastic. So my cosplay made with the physical labor of my own hands is being labeled as AI generated content on Instagram with no way to correct this tag. Essentially the problem is is that Instagram have rolled out a feature presumably one to detect AI generated content. Now, you might argue that yes, this content does look AI generated, but but that is completely unfair to the creator of the content, who it isn't actually AI generated. The problem with this is that if Instagram can tag your photo as AI generated, even if it is, it kind of raises the suspicion can Instagram basically determine what is right and what isn't? And what kind of system are they running to where they can auto flag certain pictures that aren't AI generated? Whatever system they are using, I would say they definitely need to roll it back immediately. They because this is going to completely ruin the trust on the Internet. We all know that trying to understand exactly what's going on right now is definitely not something that's easy to do considering the proliferation of online available AI tools that are easily able to pretty much generate anything that you can think of. But I'm guessing that what Instagram have tried to do is make the platform a lot better, but in doing so they've kind of ruined it just a little bit. So it will be interesting to see what kind of AI tools they do come up with. We do know that there are certain ways that they can embed the AI metadata into the images, and I've actually experienced this when I've posted a few things on Instagram myself. So I do know that actually does work. But maybe this is a glitch or maybe they are testing a different algorithm. Either way, be careful when posting stuff on social media because you don't want your post labeled AI generated when it actually isn't. Now, interestingly enough, Amazon is actually working on a chatgpt killer. Now I think I know why, and I think that a lot of people have this wrong. So it says the secret internal project named Metis is likely in reference to the Greek God of wisdom, and the new service is designed to be accessed from a web browser, similar to how other AI assistants work, according to people familiar with the matter and an internal document obtained by Business Insider. So essentially Amazon is working on a secret chatgpt killer. Now this Personally, I don't think that it makes sense for Amazon to work on a chatgpt killer, because as you already know, the point right now is not to compete with ChatGPT, but kind of, you know, have a different attack vector in the sense that if you are a company who's trying to develop a chatbot, it would be rather more useful to develop a chatbot that is for a specific application than just your average run of the mill chatbot. Because OpenAI are pretty much, you know, too far ahead. So you can see here that of course this is called Metis, but and of course it uses RAG to provide up to date information and to automate tasks. Now of course this secret internal project named MEFT Metis which is going to be giving up to date responses. I think that this is definitely something that is joining the crowded market. In addition, you can see here that Metis is part of Amazon's AGI team led by Rohit Prasad, it's head scientist and senior vice president with people familiar with the project. Said Jassy last year said that this team would report to him and be responsible for building Amazon's most ambitious AI models. And the vice president of AI has direct oversight of Project Metis. One of the people said Amazon is also relying on Alexa for the development of Metis. Two of the people said many employees working on Metis moved over from Alexa's AI team. And this is of course the thing that I was leaning towards more so because this is where I think Amazon has a real actual foothold in terms of AI devices. You know, the internally dubbed remarkable Alexa that they first reportedly, you know, reported on in January, which is going to be a new paid version of Alexa powered by Remarkable Alexa, a new web based browser service. So I think if Amazon of course wants to actually compete with ChatGPT, they need to focus on Alexa, which is an AI device that I actually see the general public use today. Like often on times, you know, watching a video online, I'll hear like streamers or content creators or people doing, you know, whatever. If they're at home doing like an interview, they'll be, Alexa, do this yada, yada, yada. And I can see with the, you know, already large customer base that Alexa already, you know, facilitates and the large customer base that Amazon already has with the millions and millions of customers, they have a remarkable, you know, area for distribution of their model. So I think Amazon should, you know, potentially try and take that route in and go after that market because it's right now an untapped market. And if you're someone that's familiar with the Rabbit device which was, you know, launched earlier this year, you can see how it was received because so many people bought into Rabbit. They generated, you know, $20 million a sale. And it was literally a product that, you know, essentially just did really, really well, which, you know, shows that there's a huge demand for consumer AI hardware. So that actually works effectively. So essentially what I'm trying to say here guys, is that Amazon has a very, very I know I say the word very a lot, but very, very interesting and unique approach to these chatbots that I truly can't wait to get my hands on. So if they do release a new Alexa, I'll definitely be picking up those devices because Alexa is already pretty decent and if they manage to develop something that is, you know, a lot better than what they currently have, I do think that this is going to take Amazon to new heights. And this secretly makes Amazon a very big sleeper in terms of the AI race. Now here's a piece of news that you probably didn't want to see. OpenAI says we're sharing an update on the advanced voice mode we demoed during our spring update, which we remain very excited about. We had planned to start rolling this out in alpha to a small group of ChatGPT users in late June, but need one more month to reach our bar to launch. For example, we're improving the model's ability to detect and refuse certain content. We're also working on improving the user experience and preparing our infrastructure and to scale to millions while maintaining real time responses. And as a part of our iterative deployment strategy, we'll start the alpha with a small group of users to gather feedback and expand based on what we learn. We are planning for all plus users to have access of this by the fall and exact timelines depend on meeting our high safety and reliability bar. We are also working on rolling out the new video and screen sharing capabilities we demoed separately and we will keep you posted on that timeline. So essentially what what they are stating here is that look, they just need more time to get this thing right because they kind of overestimated what they could ship out in terms of the timeline. And this is something that I iterated myself if you don't follow me on Twitter. This is a shameless plug for my Twitter account. You should probably follow me. I tweet pretty cool stuff, but you know, a couple days ago I said that I think GPT4O voice mode isn't being rolled out yet because once it does it will reach maximum capacity very quickly. That kind of technology will be used a lot more than you can think another ChatGPT moment but for voice. So essentially what I mean by that is that if we think about the amount of applications you can have for a real time AI system that is as good as GPT4O, that it responds immediately and can look and analyze images with remarkable, you know, consistency in terms of the voice and in terms of the inflections that sound remarkably natural I think there is going to be no other product that hits the market like that. Considering it's only $20 a month and can be accessed on your phone, on the web, pretty much anywhere on the planet. I do think that that is going to reach maximum capacity almost immediately. I mean like I did another tweet here where I spoke about how, you know, you're going to be able to use this for pretty much anything if I can even find it. I'm not sure where I did tweet it, but the point is guys, is that it's going to be used by absolutely everyone. You are going to have a remarkable, remarkable shock if you just think that they are able to roll this out really quickly. Because I think the main thing here is that they do need to prepare their infrastructure to scale to millions while maintaining real time responses. OpenAI are working on so many different projects. You know, they've got GPT5 and I think this project itself will likely take up the vast majority of their computer because it's something that's so game changing that I think, you know, the world just isn't ready for that now. I would love for them to ship it out immediately. I would love for to be, you know, one of the people that get it early in the alpha groups. But I do know that this is a project that is very hard to accomplish and it's a lot harder than people think. And eventually let me know down in the comment section below if you're going to be one of the lucky ones to try out advanced voice mode. I'm guessing that it's probably just going to be once again people that are located in the US probably only like a thousand to two thousand people, maybe just ten thousand, whatever, kind of random. I'm not sure how OpenAI chooses it, they don't actually state. But I think it's going to be truly fascinating to see how this actually is responded to by the general public, not just people in the AI bubble, people like you who watch these videos on a consistent basis. I think this is going to be one of those technologies that truly changes the game. And then after that I do think were going to have the other companies like Meta, like Alexa, of course, go ahead and try and catch up once again. Now Unitree Robotics, the innovative robotics company also did a vet. So now what we have here is something that is either going to be fascinating to you or scary to you based on how you view the AI future. These are humanoid robot dogs that are by the company Unitree Robotics, if you remember the previous demo where they showed their AI agent avatar, the humanoid robot G1. That robot was done by this same company here that is showing a remarkable ability with their new robot dogs. In fact, they're not actually new, they just have new capabilities, such as extreme disturbance to remarkable recovery. We can see right there, he's literally throwing them around. They managed to get up within literally one second and are able to walk. So not only is this showing the remarkable ability of the robots to be able to immediately get right back up, but these things are incredibly, incredibly durable. Now, I know you might not have your creative hat on right now. You might just be thinking, oh, my God, what is going on right here? But imagine if someone put these things on with, like, armor. Imagine they were made out of completely bulletproof stuff. They had like, you know, a gun mounted to them. And of course, with robots, we've seen that with the remarkable accuracy of vision systems, that it's not out of the way to say that these couldn't be used as killer drones. And there's literally an episode on Black Mirror. I'm not exactly sure what it's called, but I can guarantee you it is something that looks, I'm not gonna lie, guys, and this is the first time I'm gonna say this exactly like what we're seeing right now, where they have a robot dog that looks exactly like this. Yes, exactly like this. And it just won't stop coming for them. And it has advanced vision systems, it has advanced capabilities. And I think in the future, considering the fact that this is a popular, very, very popular platform for, you know, researchers to do their developments on and of course, to experiment with, considering the fact that how reliable these systems are, I wouldn't be surprised if in the future we do see dogs like these running around at 30 miles an hour with machine guns on their back being the new standard of war. And whilst that is terrifying, I mean, unfortunately it's a situation on our hands where if these are actually effective in war and, and they're durable and they're cheap to deploy, then we're likely going to see this kind of happen. And it's one of those things where, like, you know, if a country does it before you and they get the upper advantage, then you kind of have to do it too, but even better. So this is an example of what you get when you combine, you know, reinforcement learning with some very, very capable robots. And I honestly can't imagine what we're going to get when we get humanoids that are this capable. We're able to throw them around and, and they're able to run around and still get us a cup of coffee and I mean after whacking it with a baseball bat, I'm truly wondering if it's going to even want to do that anymore. So we'll be interesting to see on that part. The humanoid hub also tweeted about Fourier intelligence that revealed their in house built vision based perception network. So this is GR1 using their new Fourier perception network, using a bev, a transformer and an occupancy network in order to navigate the current surroundings. You can see that this is their humanoid robot. We haven't really heard too much about this humanoid robot. It's not as in the press as much, but I think this is a remarkable feat allowing it to be able to navigate the surrounding environments with a new vision system, which is pretty, pretty interesting. I know you guys hate that word. Still going to include it in the videos though. But yeah, I think we can see here the first steps of how these humanoids are going to be able to navigate certain environments. And I, I can't lie, this is one of my favorite parts of robotics because humanoids are something that just fascinate me so much because whilst yes, it is quite uncanny for a robot to be human, like in terms of its movement and of course in its stature just the way it looks, I think it's still pretty interesting to see how we're starting to develop these technologies to the point where they're actually being able to be used in outside and indoor environments. So I think that, you know, a year from now, two years from now, three years from now, I think in the future we're definitely going to get some truly advanced humanoid robots that can probably run around do dodging humans and just doing a whole bunch of crazy things that we never even thought possible.

These systems are already human level in specific tasks and of course in a lot of tasks they're not. If you look at the trajectory of improvement, systems like GPT3 were maybe, let's say, you know, toddler level intelligence. And then systems like GPT4 are more like smart high schooler intelligence. And then in the next couple of years we're looking at PhD level intelligence for specific tasks.

So two things from that clip. Number one, we also got the PhD to high schooler kind of analogy again. And this is kind of remarkable considering the fact that this is the second time that they basically said that, you know, GPT4 is like a high schooler. And then of course this one is like, you know, a PhD level. So I'm guessing that potentially GPT5 or GPT6 is going to be at the level of a PhD researcher. Now I do want to say that what's kind of interesting about Mira Marathi is that some of her statements have been not quite in line with what some of the other OpenAI employees state. That doesn't mean that they're not true. It just is a little bit confusing sometimes because I do wonder if everyone at OpenAI is of course on the same kind of wavelength, if they all know exactly what's going on or if there's different timelines being floated around from different departments or based on what they're allowed to say. So I'm guessing that maybe she kind of gave us an early timeline here because we didn't say a year from now, she said a year and a half, let's say. So either it means that we're getting GPT5 in a year and a half, which is going to put it at that's actually 20, 26, or potentially they've got some, you know, the next level of model baked in. They've got like GPT6 mapped out to the point where with scaling laws they're like, okay, we know what GPT6 is going to roughly be in terms of the capability. So let's say a year and a half from now, when GPT6 is around, we can say that that is going to be at the level of PhD, which I mean, if you're trying to do some planning, if you're trying to understand exactly what's going on, I think this is something you're probably going to want to write down. So either way, as she said, things are changing rapidly. And I'm trying to reiterate this point because a lot of people right now they're stating that OpenAI is not shipping, OpenAI isn't developing good products, yada yada yada. But I would just say, just remember that, remember they had GPT4 for quite some time and it was so advanced. They have had so much more time than anyone with all of the leading researchers to be able to develop next generation models. And I think with the amount of compute that they're building up with Microsoft, they're truly going to be another step ahead of the game when they reveal their next generation model, GPT5. So I would say stay tuned for that because that is truly going to change the economy like we've never seen.

Before, transcendent in its importance. For the reason that I said that I just described it has this recursive aspect where it's able to improve itself, and that differentiates it from other technologies. Any effort to stop this were to just slow it down for the sake of slowing it down without also thinking about its positive development, would be to cede the field to the irresponsible, would be to cede the field to potential adversaries of the United States would be to cede the field to those whose vision of AI might be as a tool of totalitarianism rather than as a tool of human emancipation.

So we can see here that one of the key board members, Larry Summers, who is currently at OpenAI serving as one of the board members, is actually as what many would regard to a accelerationist, which just essentially means that he doesn't want to slow this down. Now, he's not the accelerationist in the sense that, you know, he just wants everyone to go faster for the sake of it. I think the point that he makes here is one that's rather important if you slow down, unfortunately, and this is rather unfortunate, because it doesn't give us as much time as we'd like to work on safety. It gives the adversaries of the United States, slash of the free world, the ability to catch up and use that as a tool, as he stated, you know, of human, not enslavement, but like totalism and dictatorship, which is a real, real problem. If you've read any books on the future and about how technology can be used to strip citizens of their basic rights, something that I would say is very important to pay attention to. So with the free world on the brink of the AGI asi, I think it's important to keep advancing with this technology to ensure that you can use it to safeguard the rest of the world. Because if you do slow down and they do get to it before us, it's not going to look good for anyone that doesn't have that technology. So it's actually pretty good to hear this from Larry Summers. Now, this is a tool that isn't available yet. I probably should reach out to Magnific to see if I can actually get my hands on this demo. But essentially they have something really cool here. This is going to be used a lot, and I can't wait to get my hands on this. So basically what this is, this is called relight. It essentially allows you to relight a product in any kind of lighting you can. So you can see it can be under the water, can. It can be on a bed. It can be on Saturn, Jupiter. Jupiter Whatever, whatever. I don't know what planet it was. But the point here is that this is actually really cool. And I can see so many different use cases for this. So if you're someone that works in media, in the creative industry, this is definitely going to help you out. And I know you might be thinking, ah, it's just changing the background, trust me guys, changing the lighting. Like this is something that is so painstakingly long that people have to do. And I would say that this is, you know, one of the great things about generative AI that actually makes certain creative workflows a lot better.

So what is asi? That's something that I would like to discuss here. Asi, you may heard the word of AGI because I used it last year at agm. AGI is artificial general intelligence. That's a kind of a bigger of AI. But AI is a kind of a functioning some part of it. And if you the way of thinking or the intelligence wisdom, if you exceed humanity, then that becomes AGI. AI becomes AGI. So about 1 times to 10 times intelligence, more intelligence and humanity is AGI. So we were aiming at. AGI is a kind of AI experts target and objective now, common objective now, but targeting AGI, aiming AGI what we should do with it. Because AGI is almost the same or equivalent as humanity. Maybe same or equivalent to smartest people in humanity. Genius people, genius brain, about one to ten times of humanity. One time is about the general person. Ten times genius. I would say so within humanities skills or capability. That's AGI. So I said AGI is coming in 10 years. That's something that I said in the public speech, but in reality the that was a bit conservative. My real thinking is that that will come much earlier, within less than 10 years. I believe that all the AGI is going to be exceeding humanity. Human brain, that doesn't. If that's going to be about the same as human or smarter human, then that we don't need to change our way of living. We don't need to change the structures of human lifestyles. Because one to ten times it doesn't make much changes anyway. But when it comes to asi, it's a totally different story. Asi, you see a bigger movement. Those are the AGI moving around. So AGI creating here and there. And that's going to be connecting each other. And it's like a neuron, a brain and connected and that becomes 10,000 smarter than human. It's not 10,000 people, but 10,000 times smarter than any genius. That's the kind of intelligence level that we talking about. Si artificial superintelligence.

That's a side A very interesting discussion by SoftBank CEO. So this is the Softbank chief talking about artificial superintelligence and artificial general intelligence, basically saying that, you know, it's going to be here in 10 years. And I'm not going to lie to you guys, I agree with this because after looking at pretty much every single prediction by every major AI researcher, I think the conservative estimate is around 15 years on the far end. But on the close estimate, the closest estimate that I've seen that's actually reasonable is 2027. So we can say anywhere between three to seven years for something that's dramatically going to change society in a way that we've never seen before. I mean that's not a long time, guys, when you think about actual long periods of time, like if you go back in time, for example, 20 years ago, if it is 2024 right now, we can see that the early 2000s, when the Internet was just a thing, the Internet has truly changed.