Sam Altmans New AI Device Unleashes AI Agents (New Worldcoin Orb)

Sam Altman presented Worldcoin's next-generation Orb, a device that verifies you're human before AI agents become widespread. The new hardware is faster and simpler, with verification data split across multiple institutions using cryptography so no single entity sees your face. Worldcoin plans retail locations, delivery via Rappi, and preorders starting spring 2025.
Questions this video answers
Three times faster.
30% fewer parts.
Assembly can be done at twice the speed initially.
Five times the AI performance over the previous generation.
Capable of 100 trillion operations per second.
Chapters
It was going to be the most powerful tool that humans had yet created. It would change the economy, it would change society in all sorts of ways.
So Sam Altman recently spoke at an event where he is the co founder of the product which is called worldcoin. Now I personally do believe that this is largely going to become increasingly one of the most important products for the future years of our existence in considering the fact that AI agents are about to be extremely rampant. I'm going to pass it over to Sam Altman to show you guys exactly what was announced at today's event. But for those of you who don't know what this is, this is essentially worldcoin. Now worldcoin is essentially a product that is needed if AI agents are to ever come to fruition. Most of you guys know what AI agents are. They're essentially autonomous AIs that can go off out into the world and do different things, mainly the digital world. And if that does happen, it'll be increasingly hard to verify humans id. So worldcoin is basically a digital identification platform co founded by Sam Altman, Alex Blainier and Eventstone. And it aims basically to create a global identity network by providing a digital ID known as the World id, which actually verifies that users are humans rather than bots or AI entities. Now let's actually see some of the things that was announced today because I think like I said before, this is something that definitely needs to get off the ground right before we enter the AI agent era in late 2025 and of course in 2026.
Great, nice to see all of you. So we started this a long time ago with a belief that at the time seemed very unusual, which was that AI progress was about to go very rapid. And we thought that if AI happened, if, if AI fulfilled the dream, and we still have a long way to go, but we've made a lot of progress since then. It was going to be the most powerful tool that humans had yet created. It would change the economy, it would change society in all sorts of ways. And the thing that we hoped for the most with AI and the things we wanted to build around AI was is that AI would really accelerate people. This idea that AI was a tool that people would use, not an entity on its own, and that people were going to be able to do things in the world that they just couldn't have done before. And we started thinking about the infrastructure that society was going to need if AI really started to happen, you know, what does it take to get to this great outcome for the world. We could imagine a world back then and even more now with lots of AI tools that let people do far more than they ever could before. Lots of AI agents off doing things. And one thing we thought for sure is that the world was going to need a new layer of infrastructure for this to be able to happen. That you wanted the ability to verify humanity, you wanted to know who the people were, you wanted to be able to provide great infrastructure for humans and agents to be sending resources back and forth and doing all sorts of of things and communicating with each other. And we thought that maybe if there were a way to build a new network, that it would be a powerful step forward towards doing this.
So with that we came up with a simple plan. And that plan didn't change much since then, and I don't think it will change much going forward. And it had four big steps. That's it. The first one was about building a proof of human system that can be deployed globally that is fully anonymous. Then second, use that infrastructure to launch a digital token by giving ownership in it to eventually every human and use that to get the network to critical scale. And third, continuously scale and decentralize the network to then eventually, among many other great things that we hope this network will do, use it to make the benefits of AI accessible to everyone. And so we did one, we developed World id, which we think is that proof of human on Internet scale. We launched the token last July, which brought us to step number two. And so currently we are at three. And that means that a big, big focus of what we do is to continuous scale and decentralization of the network. And Sam, one of the big things you brought to the company very early on was this conviction about scale. And to me and a couple other team members back then, this felt very scary because it's like very contrary to usual startup advice, that's not what you usually do. But you had a lot of conviction about that and I gained it over time. But you want to talk about why that is.
So this is where Sam Altman actually talks about the scale of things. Of course, if you remember, we've always known that models are getting bigger and scale is all you need is a very interesting, you know, phrase that has echoed around the AI community for quite some time. So they're basically trying to scale up this so that it is actually effective. Because if you're going to be doing this, you know, across the Internet, you're going to have to make sure that you onboard as many people as possible.
Not always but surprisingly often I have found that scale is this very important emergent property across all sorts of things. You want to scale neural networks, you want to scale, like companies, you want to scale crypto networks, you want to scale all sorts of things. And I, human intuition, for whatever reason, seems to consistently undervalue the emergent properties of scale across all of these things. So one joke that I, I, I use sometimes is when in doubt, scale it up. It doesn't work for everything, but I think it's an interesting default. And one of the things that we've talked about with worldcoin is we want to see what happens when we do this at mass scale. If we're going to build infrastructure for all people, it's great that we've signed up so many, but there's like 99.9% left to go. Alex will talk about progress there, but, but scale has a really important quality across all of these things and it has been a pretty principle that we've been driving towards.
So this is of course where Sam Altman, well, not actually Sam Altman, the designer of the new Orb, actually talks about how it works. And essentially, if you're wondering what on earth is this orb? Well, this is how the orb scans your eye in order to ensure each user is a unique individual. So this is where you scan your iris and receive your World id and it generates a unique identifier stored on the decentralized blockchain, which, which cannot be duplicated. This is a new design. So this is what it is.
And today we will talk about the next generation of the Orb, something we worked on really hard for quite a while, and I'm really excited about. And so with that, we're going to hear first from Thomas, the designer behind the actual Orb, and then from Rich, who actually is manufacturing and building and engineering these devices.
Today, I'm incredibly excited to announce our new Orb and the spaces it will live in. The Orb is a very important part of our experience that we deliver to our users. It's where the user first meets us. So it's a very special moment, a moment where you prove that you are you, that you are human. The spaces we introduce today are the places where this orb will live. It's a bit of a temple, a place to visit, to get verified, but it's also designed with the values of a public space. It's inviting, simple, peaceful. It's focusing on the user, on the orb and the verification.
So besides being just incredibly beautiful, it was designed from the ground up with scale in mind. The entire software stack was completely rewritten for all orbs to enable verifications three times faster than before. The new orb was designed with 30% fewer parts and it enables us to assemble them at twice the speed initially. We're tripling the production capacity of the orb with plans to increase that even further. With new assembly partners and locations around the world, we're making progress towards the ultimate goal of decentralized orb manufacturing. The latest orb is built with the Nvidia Jetson architecture, which has five times the AI performance over the previous generation. It's capable of 100 trillion operations per second. So this enables the new orb to execute sophisticated AI models that prove humanness and operate entirely on device. And even faster than before. The new orb was built with more transparency and the ability to audit the code that's running on it. It has a fully removable SD card that contains all the operating instructions, so anyone can compare those instructions on the SD card versus our published source code. But to provide access to every human, we need more orbs. Lots more orbs, probably on the order of 1000 times more orbs than we have today. Not only more orbs, but more orbs in more places. And I want to talk about a few ways we're going to achieve that today.
So for those of you that are wondering how exactly this works in person, this is where the designer actually talks about how this is going to work as you manage to verify your user ID with your iris.
I'm excited to share the flagship locations with you today. Also, these are premium verification experiences. We have staff there available to answer any questions people have. We've opened our first two flagship locations in Buenos Aires and Mexico City just a couple weeks ago. This is our Mexico City location that opened just a few weeks back. I think they really look amazing. Next, we're expanding access to orbs through self serve verifications. Think of these as orbs inside existing retail locations. So your neighborhood convenience store or local coffee shop like you see behind me, these enable people to both discover the project through seeing an orb in a retail environment or specifically going to that retail store with the intent to verify with an orb. These partnerships are key to scaling the orb locations and getting more distributed in a region. They're providing access to even more people than we do today. And if you don't want to go find an orb, that's okay. So starting today, we're announcing a new feature called Orb on Demand. This will enable anyone with the touch of a button to request an orb come to them, much like a Pizza you would have delivered to your apartment. You can push a button, it shows up at your door and that driver then leaves the food behind. And I'm incredibly excited to announce a brand new partnership with Rappi, the number one delivery service in Latin America. They're integrating with our app and they'll dispatch their existing delivery fleet to bring orbs to anyone who wants to verify whenever they want to. And lastly, I wanna talk about a new type of operator we're starting today called community operators. So anyone can rent or buy the new orb to verify anyone in their community. We'll have a website up, you can place your pre order and starting in the spring of 2025, it'll be delivered to you. In fact, I want all of you guys here today to be our first community operators. Everyone here will get an orb.
So now I know some of you might be skeptical of the idea for a company connecting your data, collecting your data to some kind of digital id, but I do think it is rather fascinating to see how they talk about the anonymous layer for humans and how everything is decentralized so your data remains anonymous.
But even that is a pretty big, pretty detailed statement to make. And if I had said a year ago to a crowd of people, all of you, all of you very security minded, very concerned about privacy, you would have said, it's just not possible. And I think you probably would have been right at the time it wasn't possible. How we would build something that would scale to 7 million users, much less the billion that we're targeting over the next several years, is just not possible a year ago. And yet when we went and talked to the team of cryptographers at worldcoin foundation, they said, that's interesting, it might be possible. And they began to think about it. Remco Philip, how do we use existing technologies that don't quite scale, that don't quite operate at the level that we want them to work and build something that becomes part of the protocol. And they started experimenting with something called Secure Multi Party compute earlier this year. We talked about the fact that we were beginning to see the path towards using secure Multi Party compute to build something that would allow data to be decentralized across several different parties. So there wouldn't be one single entity or, or even two entities or any individual entity that could see that data. It wouldn't be held by the worldcoin foundation, it wouldn't be held by any one company. And as we began going through that experimentation, began to publish research and began to understand what was possible, we developed something that we're now calling the anonymized multi party computation. The key attribute that it provides is the ability to take a piece of data, know whether it is unique, but know nothing else about that piece of data on an individual basis. Multiparty is a really important part of this because the data is broken into pieces that can't be brought back together and shared with different parties. And then through a lot of math that Remco will talk about a little bit later today, gives us that key attribute of it being anonymous. Multiparty is a critical part of multi party computation, as you might expect.
Right.
And it's important that none of the parties that are going to be holding the data that's necessary for the system to work are going to be centralized. Over the last year, we've worked to find parties that we think provide certain key attributes. They have to be trustworthy, they have to be globally distributed, they have to be independent from one another. At this time, we're announcing three partners who are already beginning the process of standing up infrastructure and onboarding the data that's necessary. Those are Berkeley center for Responsible Decentralized Intelligence, based here in California. I'm actually a Cal alum myself. FAU based in Germany, and nethermind, a global blockchain and crypto security company that has incredible amount of respect in that community. In addition to that, we're working with the University of Zurich to finalize terms to begin onboarding them as well. By November of this year, we expect that all of the data that is behind World ID will be held by those parties broken into pieces in such a way that no one of them is able to tell anything about that data other than answering the question, is this unique the core to our uniqueness service?
Now.
I build infrastructure for a living. I do security for a living. I do privacy for a living. I could spend. In fact, I'm obligated to spend all day, every day thinking about those problems. But what's more exciting than thinking about those is thinking about how you can begin using that uniqueness attribute, anonymized and at Internet scale, to enable new use cases. How many of you have a driver's license in your pocket? How many of you borrowed a passport off of one of your friends earlier today? Thank you, Ajay. I promise I'll give it back. I can't actually use anything about it. We'll go into the details of that in just a minute. One of the important attributes to understanding whether or not a government credential is real is whether it's the one instance of that government credential. Uniqueness at scale is one of the challenges in being able to use credentials issued by government authorities in any meaningful way on the Internet. There are other challenges as well. That driver's license. How many of you have tried to buy beer with your driver's license recently? I know you don't all look as young as I look. Did the person that was selling you the beer, say, 155 pounds, maybe you don't need this beer? Did they say, oh, I know somebody that lives on your block because you've got your address on your driver's license? Currently, the way that we use credentials on the Internet is we disclose all of the information that those credentials use in order to answer one simple question. Are you over 18 and should you have access to this content? Or if you're trying to participate in some process provided by your government? Are you a citizen of your government? You don't need to answer anything more than that. In fact, there are reasons why anonymity is critical for those kinds of interactions and so important. So uniqueness gives us the ability to determine that the document is real. Anonymity gives you the willingness to actually use the information contained in that document. So today we're beginning a beta of world ID credentials, and there's two characteristics that I think are critically important here that I want to point out. This is only possible because of the scale and anonymity that AMPC provides. So we think we're unique in the ability to make this possible. But of course, it's going into the protocol. So anybody can take advantage of it, anybody can use it, and we would encourage anybody and everybody to do that. The second attribute, because we are so committed to privacy, that passport information is only ever going to be on your phone. We're not copying it up into the cloud, we're not making an instance of it and putting it into the back end. You will hold the information about your passport on your device and AMPC will be used to do uniqueness check, an incredible way to provide private access and the ability to interact in an anonymous way, while providing the ability to verify certain attributes that are critical to access online services. Truly adding to that experience, we're rolling out in a number of countries today, United States passports work today. Passports from a number of other countries are also in development, and we're rolling them out over the next couple of weeks. Now, this isn't just about being able to tell more about yourself and interactions online. It's also about Providing access to the network to people who otherwise can't get to it. Now, Rich's team has built a lot of orbs. I'm not going to give away any secret numbers, but it's also the case that there just aren't enough of them out there right now, and there aren't enough users able to get to them as quickly as we would like. And so we think the ability to onboard to the network by using a unique ID and a unique document will allow us to accelerate the growth of the number of people that are on the network. So that's a critical capability as well, of course.
So now one of the things that this actually does try and solve is the issue of fake live stream, slash webcam. So you can see on screen an open source tool that allows anyone to basically pose as anyone. And this is a really big issue because company directors have been scammed out of millions. People have lost their life savings in situations where you would believe that the person on the other end of the webcam video stream is the real person. So it is rather important that we do manage to get technology on the other side and that allows us to verify who we're talking to is a real person. And that's exactly what this next section focuses on.
I don't know about you, but at least for the last four or five years, I've spent far too much time in video conferencing. Is that true for anybody else? Maybe a few of you? Is there anybody that's currently video conferencing?
That's here.
I'm a security person. The news that I read is different from everybody else's. That's kind of the way of silos these days. One of the things that I've heard about four or five times over the last year is an instance of a member of a company working in the finance department receiving an urgent call or an urgent text says, get on Zoom. This is your CEO or this is your cfo. We need to have a conversation. Okay. Unexpected, but doesn't seem scammy because I'm gonna get on a Zoom with them and I'm gonna interact with them and I'm gonna learn what it is that's being asked for. And that member of the finance department then having an interaction where they're told we need to transfer money. And it's a. It's their CEO, they're talking to their CEO, it's their cfo, they're talking to them. It's their voice, it's their face, it's a Real interaction. Now, it probably hasn't happened to any of you. If it has, I want to know more details about it. But it's something that we're concerned about. It's certainly something that we're beginning to think about, and it has started to happen in the real world. World ID is in a unique position for a few reasons. The first is that it's the only system right now that at a global scale, is endeavoring to have a trusted interaction where an image of a person is being captured, signed, and then being made available in an anonymous, trustworthy way. That's actually a very, very unique technical attribute of the system. The other thing that we announced earlier this year, originally part of Face, originally part of World ID 2.0, currently available, was Face off, where we use the image that's contained in that personal custody package captured by the orb and made available and held on the user's device to do a comparison to make sure that the person who's interacting with the device is actually the same as the person who originally signed up. That authentication is really critical for being able to decide where they can interact on the Internet. And so as we were looking at this challenge of deepfakes, we began thinking, is this something that we can help solve? And it turns out we can. We can connect those two things together combined with a challenge being issued by the person who wants to make sure that the person they're interacting with in a given transaction is actually the same face as what it is that they're seeing online, or what it is that's been presented to them in the past. And so we call that World ID deepface. We're announcing it today. We're working on beta of it right now, which is gonna be coming out very shortly. The implementation is, of course, as an SDK. We wanna make it possible for anyone to incorporate it directly into their application, make it as seamless as possible, tie into your infrastructure within your enterprise, make it part of your Active Directory or whatever LDAP system it is that you're using to sort of bring it together and make it as seamless as possible. But in addition to that, we wanna make it available so people can begin playing with it and understanding how it works right now. So we're announcing a beta. And the beta is an integration with existing applications, right? You and the devices that you're using right now will be able to work with this beta, will be able to confirm that the person you're interacting with is the one that you think you're interacting with in an online conversation. And it will work with all of the applications that currently use video, so it's not dedicated to any one of them. Very excited to make that announcement.
Well, as.
As well.
So let me know what your thoughts are about this AI device allowing you to verify your identity. Do you think this kind of thing is needed? And I'd love to hear your thoughts about the future of AI agents and verification.










