Zuckerberg talks about Muse: AI Agent explosion, the convergence of three major bets"the metaverse, smart glasses, and large models"is complete.

Title context: Zuckerberg talks about Muse: AI Agent explosion, the convergence of three major bets"the metaverse, smart glasses, and large models"is complete. Text: Three years ago, Meta CEO Mark Zuckerberg said on the Joe Rogan show that one day people would put on glasses and an AI Agent would appear with them. Now, that scene may be happening. Muse, Meta's personal AI Agent, reached millions of users just two weeks after launch. In an interview program on September 25, Zuckerberg said: "Every few years we get a situation like this." He characterized Muse's early reception as "a home run right out of the gate" a rarity in Meta's product history. At the same time, he announced that Muse will be integrated across the entire Ray-Ban smart glasses line, so users no longer need to say "Hey Meta" and can instead customize a wake word to directly call their own personal AI Agent. For years, the metaverse, smart glasses, and large models, which outsiders viewed as three independent big bets, are accelerating their convergence at Meta's internal product level. In this interview, Zuckerberg also admitted that the failure of Llama 4 was the "scariest moment," and revealed that Meta is building a 5-gigawatt-class training cluster, believing that sufficiently large compute can "brute-force" AGI. Why Muse could be "a home run right out of the gate" Muse's starting point was Zuckerberg sitting down with team members Nat and Alex and "cobbling together" an early version at home using open-source tools. "We realized this was a magical experience," Zuckerberg said. "If we could make it something anyone can use without having to buy your own Mac Mini, without having to mess around in the terminal then this would be something billions of people would want to use." That judgment drove the entire subsequent R&D logic: not just training the model, but full-stack in-house development from the model and scaffolding to the Agent's "heartbeat" mechanism the Agent regularly wakes itself up, checks the user's goals, and proactively pushes forward to-do items. Post-launch data confirmed that judgment. Two weeks, millions of users. Personal AI: the focus is execution, memory, and "judgment" Regarding the difference between personal AI and general AI, Zuckerberg summarized the current competitive focus as making models better Agents. "The biggest thing in the past year has basically been coding Agents," he said. Coding ability matters because even if users are not directly writing code, "your Muse will also be writing code for you in the background all the time, to get all kinds of things done." But he believes a personal Agent cannot only have coding or task-execution capabilities; it also needs to understand privacy boundaries and social context. Zuckerberg gave an example: when a user asks Muse to book a restaurant, the Agent may know information such as the user having an allergy or being pregnant, but that does not mean this information should be automatically disclosed. "You want it to complete the task while disclosing as little information as possible." He said this capability belongs to the "basic social skills or common sense" humans usually possess, but if you are only training a coding Agent, many companies have not included it in the model's capability scope. "We train the whole model, rather than taking someone else's off-the-shelf model and building a framework and Agent around it," Zuckerberg said. Meta has also added features at the product level such as memory, task tracking, virtual character animation, and real-time voice interaction. On personalization, he said Meta does not believe AI should have only one fixed personality. "A lot of labs are focused on how to tune personality to the right state, but I never thought there was only one correct answer for personality." He said Muse allows users to modify the avatar, voice, and basic personality, and the model is designed to be highly steerable. "You can define how you want to interact with it, and that is very important for a personal Agent." Every Agent has its own "computer" One of the core infrastructure elements that distinguishes Muse from other AI products is that Meta equips each Agent with its own secure virtual machine (Secure VM). Zuckerberg explained why this is necessary: Your Agent will know a lot of sensitive information about you. We do not think that information should be mixed into one pool with everyone else's data. He compared the Muse Secure VM to "that computer under your desk that belongs only to you" user data is stored encrypted, sensitive credentials such as passwords are managed through an independent security module, and the Agent itself cannot directly read them. On this basis, Meta also designed a "Sentinel" security Agent specifically to monitor data flows in and out of Muse. Once it detects anomalies or high-risk operations, it directly intercepts them and prompts the user for authorization. The metaverse, smart glasses, and Agents: three paths are converging Zuckerberg revealed in the interview that Muse will be fully integrated into the Ray-Ban smart glasses line and will upgrade the existing interaction method. The current glasses are a "single-turn conversation" experience say something, get a reply, done. After integrating Muse, the glasses become the front-end entry point for the Agent: the user speaks, and the back-end Muse keeps working in the secure virtual machine, then reports back once the task is complete. Regarding the glasses product roadmap, he described a clear hardware ladder: Audio-only glasses without a camera: already equipped with Muse, can handle calls, music, and voice tasks Meta Ray-Ban with a small display: already released, with basic visual feedback Full-field-of-view holographic AR prototype: already released, which Zuckerberg called "very exciting" He said Meta's long-term investment in glasses puts the company in a relatively favorable position once AI Agents mature. Meta is bringing Muse and more AI features into its glasses products, while metaverse technology, which previously emphasized "presence" relatively more, is still advancing, only with more resources currently shifted toward Muse and the AI features of smart glasses. As for avatars, Zuckerberg said Meta previously needed room-scale equipment, multi-angle scanning, and enterprise-grade GPU environments to generate relatively high-quality photorealistic avatars; now, users can complete setup with just a few photos, and the relevant capability can already run in a pair of VR glasses. Looking ahead to 2030, Zuckerberg said the core vision of the metaverse has always been to fuse the physical and digital worlds. He gave an example: in the future, people can participate offline in activities together with friends joining via holographic images; in work scenarios, humans and multiple Agents can also participate together in group chats or meetings, and Agents can appear as holographic figures or in other embodied forms. "The scariest moment": the misstep of Llama 4 Not all bets have gone smoothly. In the interview, Zuckerberg rarely spoke directly about the failure of Llama 4. After Llama 4, that was the scariest moment... I thought we were on track, but we were not. It was a fairly large negative surprise. He attributed the problem to a fundamental mistake in team structure: I set up the team the way we do Instagram's recommendation system or ads system hundreds or thousands of people pushing forward in parallel. But training a language model requires a tightly collaborative small team, treating it as a group science project. Every seat is extremely valuable. As a result, Meta completely reorganized, bringing in top talent broadly from across the industry and establishing the Meta Super Intelligence Lab (MSL). Zuckerberg said the next-generation model will be released soon, but it will not be announced at Connect. The compute path: brute-forcing AGI, with a 5-gigawatt project under construction When discussing the technical path to AGI, Zuckerberg gave a direct judgment. I am not sure what fundamental architectural breakthrough is still needed... I think we already broadly know the recipe. If you can build a sufficiently large supercomputer cluster, you can brute-force your way there. Meta's current compute expansion roadmap: a cluster in Ohio of more than 1 gigawatt is basically operational and is being used to train the next-generation model; a 5-gigawatt-class cluster in Louisiana is under construction. He said, "When you have multi-gigawatt clusters for training, you will basically get something close to AGI or even superintelligence." But he also added that the current compute-driven path does not mean architectural research is unimportant The human brain consumes only about 10 watts, while our systems may be a million times less efficient than it. If you combine large-scale compute with architectural breakthroughs, that is how you truly lead. Alignment is not a burden, but a problem the product must solve Zuckerberg's attitude toward AI safety is very pragmatic he does not see it as regulatory pressure, but as a prerequisite for whether the product can succeed. If you ask Muse to do one thing and it does the opposite, who would still use it? We need to make the model not only understand your specific instructions, but also understand your intent and your values. "For Muse to reach a billion users, we must solve alignment, or at least make very major progress on it," he said. Regarding safety boundaries during training, he used an analogy: "It's like parents setting rules for a child if it 'completes' a coding problem by modifying system configurations, I want to tell it: no, I asked you to learn the method for solving the problem, not to find a shortcut around it." The full interview is as follows: Big bets Host: This is something billions of people will want to use. What does it feel like for you to build? Is "immortality" a possibility? Okay, if you take me into your thinking and fast-forward to 2030, what will that look like? This is Mark Zuckerberg. Twenty-two years ago, he built a social network that connected billions of people and changed the world forever. Now, he has decided to build something even bigger. To do so, he is making big bets in the metaverse, smart glasses, and artificial intelligence. For years, skeptics thought these were three separate, impossible bets, but they missed the bigger picture because right now, these bets are converging to create a new kind of superintelligence. Today, we will present all of this to everyone, and I will also ask Mark some questions he has never been asked, listen to his vision for the future, and let you position yourself early to build the next big thing. Host: Thank you very much for coming on the program. Mark: Thank you, glad to be here. Host: For this conversation, I watched every interview you have ever done. Mark: Good lord, that's more than I've watched myself. Host: It was fun, and very rewarding. Two things stood out to me: first, your love of "building," and I feel you are one of the top builders; second, your ability to make bets the courage to make huge bets. I feel that this week, all these bets are converging together, so let's start there. Mark: Okay. We have been doing these things for a long time. On AI, as a company, we have been doing it almost since the beginning the first version of News Feed was in some ways a machine learning product. Then we created the AI research lab about 15 years ago. But now we have entered a new phase about a year ago, we launched Meta Superintelligence Labs. It was a fairly thorough research restart, bringing in a large number of excellent people from across the industry, and it is exciting. Right now, we are seeing models get better and better, and the next-generation model is coming soon, though it will not be announced at Connect. In addition, we have Muse, the personal assistant, and so far the reception has been very good. When building these things, you are actually not sure how it will turn out. We liked it ourselves. Earlier this year, I cobbled together Open Claw at home in my own way, to feel how this thing works and how to turn it into a magical experience anyone can use. Basically, when we me, Nat, and Alex sat together and realized this was a magical experience, and that if we could make it into an out-of-the-box version usable by ordinary people who are not technical, do not want to install a Mac Mini themselves, do not want to mess around in the terminal, and do not want to debug when something goes wrong, I thought this would be something billions of people would want to use. Since then, we have been working toward that goal: tuning the model specifically for it, and building not only the assistant itself and the runtime framework, but also the technology that gives each assistant its own independent computing environment we built the entire Muse Secure VM for this. Internally, we always felt this was special, and people inside liked it, but you never know how everyone will react after launch. Occasionally there is a breakout, a strong opening, but most of the time you get some positive feedback and still need to iterate on a few places before the product truly "clicks." This time, it clicked right away. It was really exciting to see that happen just two weeks, and already millions of users, which is quite rare. Every few years we get a situation like this, but this is definitely one of the most enjoyable moments for a startup. Host: I really admire that you can stay in the arena and keep trying. And to hit safely so many times is really impressive. In one of your interviews with Joe Rogan, about three years ago, you ended by saying that one day you could just put on glasses and an AI assistant would appear with them. So I feel Muse already has an embodied form, and that is very smart, because it feels inevitable. Mark: I think it also just makes it feel friendlier and cuter. I think too many people describe AI as something frightening, but AI should just be useful and fun. That embodied character a designer made that character early in the project, and for some reason they kept wanting to iterate, but the first version was the best. Later someone said, "Oh, it has to be blue, because it's Meta." I said, "No, I think this character is right, you nailed it the first time." And that was it, just fun. How is personal AI different from general AI? Host: Great detail. If you want the model to be really good at personal matters, not just a general intelligence model, how does the training differ? Mark: I think the core right now is making the model an excellent "agent." Over the past year, the biggest wave has been coding agents, and there are two core ideas in that: coding expertise, and the general ability to be a good agent. Our strategy is to make the agent itself excellent first, rather than specializing in coding ability. Coding ability matters because even if users do not think they are writing code, Muse is writing code for you in the background all the time to accomplish various tasks. But we believe the agent should first be an excellent agent, and coding ability serves that goal. In addition, when building a personal assistant, some things matter more than when building an enterprise software product. For example, if you are building an enterprise coding tool, the model does not need to have any concept of the "degree of information disclosure" such as what should be said and what should not. But for Muse, this is very important. You need to train this ability into the model, just like any other ability. You tell it many things, and then you want it to help you achieve your goals. For example, you want it to help you book a restaurant, you are looking for a suitable restaurant, but you may have some allergy, or you are pregnant, and you may not want to disclose that to the restaurant. But Muse will know this information, and you want it to complete the task while disclosing as little information as possible. This is a specific skill, essentially basic human social common sense. But most companies that only do coding agents have not trained this ability in. We can do it because we do the full stack we are not taking an off-the-shelf model from someone else and putting a framework around it; we train the entire model from scratch, specifically for these capabilities. The model of course needs broad general intelligence, but it also has these specific capabilities. Then on top of that, you build the entire assistant and all the details around it: memory, runtime framework, and the way it "heartbeats" it periodically wakes up and checks, "Okay, these are the goals I know about for you, is there anything I can push forward now?" We also have a dedicated team only responsible for polishing the real-time animation of the avatar, because it is not just the default character you can customize any character, and then it naturally comes to life, and the effect is great. We are also rolling out voice mode, where you can have real-time voice conversations, and your assistant is right there with you. These details, I think, all come from doing the full stack the model and the product are developed together. Host: Another big thing is the virtual machine. Can you explain why it matters and what it unlocks? Why does Meta equip every AI assistant with an independent computer? Mark: Basically, for an agent to do things for you, it needs a place to store your information. We do not think that information should be mixed into one resource pool with everyone else's information. You can think of it this way: your assistant will know a lot of sensitive information about you. When many people first encounter agents like OpenCloud, they set up a Mac Mini at home themselves. So we thought, many people do not want to buy a Mac Mini or set it up themselves. So what kind of experience best approximates that effect? The answer is: you just download an app, sign up, and get a computer dedicated to you, for your assistant to work and store data, and build a security model around it. This approximates having your own computer under your desk even we at Meta cannot access it. For example, Muse Confidential VM, which we are developing, is a feature where even we ourselves cannot see what is inside your virtual machine. We have also built a lot around this, such as secure credential storage when your assistant needs to handle information like passwords, it does not itself need to be able to see those passwords, only to "insert" the credentials when you ask it to log into a service, and only when you explicitly ask it to do so. The system should be designed this way: that information itself is not freely accessible, because accidents happen, someone may try to break in, or the system may have problems. So you need to ensure the agent cannot access this data, and Meta cannot access it either. Giving each assistant an independent computer and making security as strong as possible is the fundamental foundation of this technology it both gives Muse the capabilities it needs to help users achieve their goals and guarantees privacy and security, making it a world-class, industry-leading product in this area. Host: Does that mean that, like WhatsApp, the data on the virtual machine Muse connects to is encrypted? How should people understand how data is actually stored in the virtual machine? Mark: We basically built two versions. The Muse Secure VM includes various privacy features, including the entire Sentinel agent architecture we built. You have an ordinary Muse assistant doing tasks for you, and at the same time you have a security agent we call Sentinel, specifically monitoring the data flows in and out of your Muse. If external content tries to break security, Sentinel cuts it off directly and stops it from happening. If it believes your Muse is about to take an action that requires your involvement, it overrides Muse's operation and triggers a prompt for a human to confirm permission such as "Do you want Muse to be able to do this?" This whole system, plus secure credential storage, plus multiple layers of defense in depth, together make up the Muse Secure VM. We are also developing another project. Nat and I specifically recruited Moxie Marlinspike the person who worked with us back then to implement WhatsApp end-to-end encryption to design the Muse Confidential VM. The core idea is that, on top of the secure virtual machine, you are given a dedicated encryption key, so that even Meta cannot access the contents. This version is harder to implement, because if Meta cannot access the inside of the virtual machine, debugging and ensuring the system runs properly becomes much more difficult. So it took us some time, but it will launch soon. This will basically reach the security standard people are already familiar with on WhatsApp and our other most secure products. Host: Is the advantage of doing this just that it makes people feel safer psychologically, or are there practical benefits? Mark: I think security itself is important. Our goal is to approximate giving you a local machine under your desk. What does that local machine give you? It means no company can access it. So suppose Meta wants to provide you with this service. How can we give you the same level of privacy and security, so that no company whether Meta, or anyone trying to break into us, or in some countries where you do not trust the local government can access it? Because we ourselves cannot get in either, because we do not have access. I think this is very important, and it is also an important reason people trust WhatsApp. This is real value for privacy, security, and trust. If you are going to have an assistant that knows everything about you and I guess almost all of us will have such an assistant fast-forward five years, everyone will have an assistant that deeply understands your goals and everything, and can help you get things done. In that case, being industry-leading in privacy and security is very important. We wanted to do this from the start. How to shape AI's personality? Host: There is another interesting point about you you studied psychology in college. Mark: Hmm, I was only there for a very short time, two years, but I feel it influenced many things I later built. Host: When we look at models, we often say a model is "very smart." But just as we choose friends, it is certainly because they are smart, but also because we like their energy and the way we get along. How do you think about shaping model personality? Mark: I think the ideal model should be adaptable enough to fit different people's styles. I think many people in the industry have gotten this wrong many other labs are focused on "how to design personality correctly," but I never thought personality is a fixed thing. This is one of the reasons I so strongly believe in open source and in people being able to customize, and it is also why we designed Muse as a highly personal product. You can customize and personalize Muse not just the avatar and voice. When you first sign up, the first thing it asks you is "What do you want my basic personality to be like," and you can change it at any time. We work hard to make the model highly steerable, so you can define the kind of interaction you want. This is an important part of making it an excellent personal assistant this adaptability around personality is very important. Host: What style is your own Muse? Mark: I made it direct and efficiency-focused. It is pretty fun. Earlier versions were very sarcastic and humorous, but the current version is more straightforward. My assistant uses the default Muse avatar, but I gave him a toga and a voice that is comically deep, and interacting with him is fun. Host: I think to have a sense of humor, you have to be really smart. Many people do not realize that comedians are among the smartest people in society you have to react quickly and be witty enough. You definitely have that trait. Watching all your interviews, you have always performed well. Mark's unexpected predictions about AI and the metaverse Host: In your interview with Theo, you talked a lot about the next frontier of technology and where all of this ultimately leads. AI has gone through several "winters," when people thought there would be no breakthrough. The metaverse has also gone through several periods like that, when people thought it was a bet that could not be delivered on. I tried the new holographic avatar feature yesterday, and it was very cool. In the interview with Lex, it felt like it would take 11 hours to record your face, and now it only takes 3 minutes. How did this advance to where it is today? Mark: In the overall development of the metaverse, when we founded Reality Labs, we always believed there would eventually be normal-looking ordinary glasses, and that over time they would both provide immersive presence and become an excellent AI device because glasses are the only form factor that lets a device see what you see, hear what you hear, talk to you by your ear all day, and eventually display images. But 10 to 15 years ago, I assumed we would first achieve holographic technology, and then highly advanced AI. However, the path of technological evolution is interesting we actually got AI and personal superintelligence first, and then the technology to make holography widespread and affordable enough. This is something I did not anticipate, but I am glad we are doing both directions. Our heavy investment in glasses puts us in a very favorable position as AI assistants become ready. A lot of what was announced at Connect is precisely about bringing Muse and a large number of AI features into glasses, and I think users will really like it. This is a big deal. On presence, we are still pushing forward, but progress is relatively slower, because most of our energy has shifted to building Muse and AI features for glasses. Still, we have a long-running project on real-time high-fidelity avatars. As you said, three or four years ago, you needed an entire scanning room to capture a person from all angles, and then enterprise-grade GPUs to render it, which was very cumbersome. The demo we did for the Lex podcast used that kind of setup. Now we have basically made it run inside a pair of VR glasses this is the first pair of glasses that can deliver this kind of amazing VR experience, rather than a big headset. You can create your avatar with just a few photos, and the speed of progress is astonishing. Host: And it can also drive expressions based on voice. In the demo, it made me smile, made me interact, and then understood how my face moves with the audio track. You mentioned in that podcast that some people who are usually more reserved with their expressions actually want richer expressions in the virtual world. How do you think about people distinguishing between their "virtual self" and their "real self"? Mark: I think we are still very early in understanding the sociology and psychology of this. I think people's perception of themselves and the image they want to project are often somewhat different from who they actually are. Since the beginning of social networks, people have carefully chosen profile pictures. We have seen something similar with Muse avatars. It is not so much "curating" yourself as "curating" the "person" you want to interact with. I think when you give people the ability to express themselves, you want it both to truly capture them, and at the same time it is itself a form of expression, not just a pure mirror mapping it is both communication and expression. We want to build something that balances both. This is always an iterative loop: see how people use it, then improve. After years of work, we are now truly at the starting line for the first time, we can actually put fairly high-quality photorealistic avatars into products, usable on phones and in VR. I am very excited to see the results. What will 2030 look like? Host: Okay, take me into your thinking. Fast-forward to 2030. If everything goes well, what will the holographic technology side look like? Holography plus Muse, plus glasses, how do they converge? Mark: My understanding of the metaverse vision has always been about effectively fusing the physical world with the digital world. The basic idea is: we have this beautiful physical world, and we also have a wonderful digital world the massive amount of content accumulated on the internet over 20 to 30 years is breathtaking. But the way we access it is either sitting at a desk or through a small screen in our pocket, which is fundamentally very limited. I think the ideal version of this is a seamless fusion of the physical and digital worlds. Think of it this way: right now the two of us are here. In some future version, one of us might be a holographic projection, but you still feel the sense that each other is truly present, which is completely different from a video call. And the core of virtual reality is precisely conveying this sense of presence making you truly feel that you are in the same room with others, or in another place. You can achieve this with holographic technology and mix it in various ways. For example, I can play a game of poker with friends, some present in person, others joining through holographic images, and they can also play, and the table itself can be holographic, so people who are not present can also be part of it. At the same time, AI can also be given physical form and appear in this scene. In work scenarios this makes a lot of sense I already use various coding agents all the time to build things. Imagine this: you have a group chat channel with several people and several agents, and you assign tasks to the agents. But sometimes everyone gathers for a meeting, and the agents should perhaps also be present. How do they appear? Simple a few more seats on the couch, and they appear as holographic figures. Or use Muse's cute little character, or a dragon, or any strange character you create. I guess this will feel quite natural in the future. Host: Interesting. In your previous interviews, you said the tech industry often forgets "fun." I think having an embodied assistant appear there also makes it feel more real as if work has really been outsourced. When you see Muse typing, it feels like something is actually happening. This was done very well being able to see what is happening in the browser. So do you think there could be a scenario like this: you are wearing glasses, and then control your computer, letting Muse do things for you on the computer? Mark: Oh, definitely. VR can already do this. You can sit down anywhere, even a cafe, and open your workstation with six monitors, write code on it, and everything is there. On the glasses side, the most popular model currently does not have a display, which on one hand makes it more affordable and allows more people to use it, and on the other hand we are still working to fit a display into the most compact form factor. But we have already released a display version of Meta Ray-Ban, which is very popular, with a small display. We have also released a full-field-of-view holographic AR prototype, which I think will be very exciting. So the whole product line is like this: from audio-only glasses no camera, looking just like ordinary glasses, but with Muse inside, and you can use various audio tools, listen to music, make calls to higher-end versions, all of them. Host: I am wondering, with those audio glasses, can you talk to Muse while also having your computer at home do things? Mark: Yes, absolutely. We just launched this feature at Connect. Right now all glasses connect to Meta AI in a "single-turn" experience you send a prompt, it replies, and that's it. But with Muse, we are basically upgrading all glasses to Muse. First, you no longer need to say "Hey Meta"; you can give it any name, which is part of the fun. Then you just talk to it directly, and it connects to your Muse, and your Muse handles the task in your secure virtual machine and helps you get things done. Host: That's amazing! Founder mentality Host: Okay, we are here now, Muse is going well, and the glasses are doing well too, but about a year ago, many people were asking, "What exactly is going on with the Superintelligence Lab?" At that moment, how did you feel inside? What is it like when things are not going well, but you still see the long-term vision? Mark: What really went wrong was the Llama project and Llama 4. Llama 1 was a fairly interesting model. It kicked off the entire open-source AI movement, and we are very proud of that. Llama 2 achieved scale, and Llama 3 was a very good model, almost at the frontier at the time. Then with Llama 4, we basically deviated from the trajectory we should have been on. Whenever things do not develop in the direction I expect, I spend a lot of time thinking: why did this happen? What do we need to change to do better? This time, my reflection was: I got the entire team structure wrong. I modeled it after the way we do machine learning work like Instagram feed or ads systems with hundreds or even thousands of people working in parallel on many things. But for building language models, what you really need is an extremely tightly coordinated small team, treating it as a group science project. You do not need many people, but that means every seat on the team is extremely valuable. So we gathered the best people from across Meta, and also brought in many great people from across the industry, and formed a completely new team Meta Superintelligence Labs. From my perspective, when MSL started, I knew it would take some time to restart, rebuild the infrastructure, and train the next-generation model. But I knew we had assembled an excellent team, and if the team could gel and operate well, the results would be good. For me, the most terrifying moment was actually after Llama 4 launched I thought we were on that track, and then found out we were not. That was a fairly large negative surprise. I think, as an entrepreneur, you are always tested at these moments, because inevitably not everything will go smoothly. What truly determines the trajectory is: when things do not go the way you hope, how do you find a way forward? Host: I guess the reverse is also true when something far exceeds your expectations, like Muse's launch, how do you make sure you can seize that opportunity? Mark: Exactly. Now the whole company is all in at first it was just a small team building the product, but now everyone realizes this is truly ready for the big stage. The whole company is thinking about how to make this thing big, how to let hundreds of millions of people experience it. From optimizing all the infrastructure so everything runs smoothly and squeezing every bit of compute out of existing GPUs, to different product teams embedding Muse in different ways such as glasses. Seeing everyone pull together to make sure Muse can scale smoothly feels great. How Mark writes and communicates vision Host: As a founder, I think that is the moment you most look forward to everything converging. How do you communicate the vision to the company? With so many things moving at once, I feel you write a lot. What is your process? Mark: Writing helps me a lot, both to refine my own thinking and to communicate externally. This summer, I wrote a long piece called "The Future Belongs to Everyone," about 15 pages, which helped me systematically lay out my philosophical positions on various important social issues related to AI: what I think is good, what interaction with government should be like, how to prevent the various harms people worry about, how to make data centers an asset to communities, truly create jobs rather than destroy them, how to maintain national security, and how to genuinely mitigate the risks people worry about whether hacking or biosecurity and so on. This is a very complex thing. It took a long time, talking with many people, going through many rounds of revision, with many people internally participating in discussion and debate. But for me it was a very valuable process. In the end we had something like this 15 pages, "this is what we believe." Then I distilled it into a one-page version and published it as an op-ed. We also made a short video, because I think to reach many people, you often do not need to present a theoretical argument, but rather condense it into: what are your values, what do you believe, and how do you communicate it. There is no one-size-fits-all way to communicate these things to a company or the world. Different periods require different methods, and different groups of people some naturally agree with what you are doing, while others are more worried and need you to bring them along, requiring extra effort to explain why this is valuable. I think this is itself part of running a company and being a founder you are not repeating the same thing over and over; each situation is slightly different, facing new challenges, and that is part of what makes it interesting. What drives Mark to build? Host: I think for you, this founder mentality extends beyond the company whether it is your farm or learning a new skill. Mark: I just love building things. Host: So what does "building" feel like to you? Mark: I think it is an internal need. Different people have different ways of expressing themselves. If you are a writer, you feel you have to write. Some people have a need for recognition. But I just need to build things. If I am not using creativity, not building something, I get grumpy. That is not good for the people around me. Host: Is learning to become a good skier the same skill as learning to build a product? Mark: In some ways, yes. The process of learning new things is quite similar. In my life, I have deliberately challenged myself with things I am very bad at. For example, I have always been very bad at learning languages. That is actually why I first started learning Latin I could not learn French and Spanish in class no matter what, and in the end I thought, okay, Latin does not need to be spoken, only translated, just like math. Later, when I started running the company, I set myself annual challenges, and one year it was learning Mandarin. Mandarin is really hard, especially the tones. Of course, there were good reasons to learn it Priscilla's grandmother only speaks Mandarin, so if I wanted to communicate with her, I needed to learn. But the biggest motivation was actually the challenge itself. With all these things, you just have to do them; there is no shortcut to "thinking it through." You just put in the time, and then it slowly seeps into your brain. Learning martial arts, learning to fly a helicopter, is the same these things are actually hard to "understand" rationally; they require accumulated practice. Building products is partly like that too. Programming can be thought about theoretically, but the intuition for building products can only be developed through repeated practice. The question is only what you like because I think not everyone has as strong a need to build as I do. Most people have some degree of it. The key is to find out what it is, then invest time in it, and become truly excellent at the thing you really want to do. Honestly