ATRIUMsearch → argument graph
Audio · 2026-02-05 · 44m · 6 moments

Introducing 4D Creation Open Beta: NPCs, 4D Worlds, and the Future of Gaming with Roblox CEO Dave Baszucki

From “virtual doppelgängers” to “real-time dreaming,” online gaming platform Roblox is using AI technology to build the “Holodeck” envisioned in science fiction decades ago. Sarah Guo and Elad Gil sit down with Roblox CEO Dave Baszucki at Roblox headquarters to explore the intersection of AI, physics simulation, and the future of human connection. Dave discusses the evolution of the 4D creation tool in Roblox, a high-fidelity simulation that enables thousands of people to interact in real-time w ✦ AI generated

timeline · colored by role

01
Context

The ultimate vision for Roblox, held for 20 years, is a high-fidelity physics-simulated space where people can co-experience — build a car, drive it, go to a birthday party, chop trees — and AI's role is to superpower the creation of that holodeck and make it photorealistic, scalable to 10,000 people, and acoustically simulated.

Dave frames Roblox's 20-year vision as building the 'holodeck' — a high-fidelity simulation backed by real physics where people co-experience. AI exists to superpower that vision, not replace it.

transcript

Dave Baszucki: And what we had always imagined is if this high-fidelity space was backed by physics simulation and reality simulation, you'd be able to do stuff. You'd be able to build a car, put wheels on it, drive it around. You'd be able to go to a birthday party and blow out the candles. You'd be able to chop trees down and make a house of it. So that was literally the genesis of Roblox. And here we are 20 years later, there's so much more to do. But one vision of AI is how do you superpower the creation of that holodeck. And how do you get it photorealistic as soon as possible? How do you get 10,000 people in it instead of 100? How do you do acoustic simulation? That sounds realistic with 10,000 people.

02
Definition

There are two extremes in the future of immersive 3D: one is a high-fidelity communication platform with people you know, and the other is 'real-time dreaming' — a single person alone in a world of NPCs, like in Vanilla Sky — and everything in between is possible.

Dave contrasts two poles: the 'holodeck' as a multiplayer communication platform with real people, and 'real-time dreaming' — a single-user dream world of NPCs, like Vanilla Sky. He sees all hybrid products in between as possible.

transcript

Dave Baszucki: There is another interesting vision, and I think it's useful comparing these product things that's much more, what's the future of just a single person by themselves dreaming what they want to dream? And we sometimes call that real-time dreaming category. It's a category that we can see a bit of the trappings of with short-form video. There's a little bit of that in short-form video. You know, someone's 2 A.m. in their bed doom scrolling, it's reacting to dwell time, it's reacting to what you favorite, and you're kind of getting a little bit of a pre-built dream. That, I think, we sometimes imagine could go all the way to the famous Tom Cruise movie, Vanilla Sky, where he literally real-time dreamed for several years in an imaginary universe where everyone around him was primarily an NPC, and he didn't even know he was doing it. So that's more world-building as content. That is exactly right. And so if you put those at 2 extremes. One is communication platform, high fidelity with people you know. The other is real-time dreaming where everyone's an NPC. Everything in between is possible.

extends · 1provides context · 1

03
Prediction

4D immersive simulation will become a superset of video communication — once it's photorealistic, video will be like 'legacy analog mode' — and will solve problems video can't, like natural spatial audio and acoustics at scale.

Dave argues that as 4D simulation becomes photorealistic, it will be a superset of video — video becomes the 'downsampled' version. He highlights spatial audio: in a 3D simulation with 1000 people, you hear someone based on distance, which is more natural than Zoom's mixing.

transcript

Dave Baszucki: I think as the technology for multiplayer 4D simulation gets better and more photorealistic, it's almost going to be like video is the downsampling. And it's like if we were having this 4D immersive communication, we'd say go to legacy analog mode and make it look like Zoom, basically. And we could always do that, but we'd be able to do things we could never do on Zoom. You know, you'd say, hey, let's pop up and go walk around my office. And I want to show you something, come and follow me kind of thing. So I do think the 4D simulation, as it gets photorealistic, will be a superset of video. The other thing for business is there's a few things that 4D, and I use 4D because it's not just 3D shape, but it's function in a way, and will provide that video does not provide is acoustical things. And there's some real tricky physics problems when you're having a company meeting of 1000 people. If we try to do it on Zoom, we have like a bunch of squares, who's live, who's not live, how do we mix the sound, which is very tricky. If we do it in a three-dimensional simulation, I can hear everyone with attenuation based on how far they are, and it can actually be a much more natural experience.

explains mechanism · 1extends · 2provides context · 1

04
Mechanism

The ultimate technical challenge for large-scale multiplayer is not just physics simulation but finding the most efficient representation for synchronizing the state and memory of 10,000 people in real time — whether that lives in video latents, native 3D format, or a new hybrid 3D-video latent space.

Dave identifies the deep infrastructure problem: synchronizing not just real-time state but also 5 hours of memory for 10,000 people in 3D space. He compares world model approaches (storing memory in video latents) with Roblox's approach of storing everything in vector format, enabling infinite playback from any angle.

transcript

Dave Baszucki: The technology for this, I think we're going to start unraveling it over the next 5 to 10 years is what's the most efficient way to synchronize the state of 10,000 people. And when we synchronize the state and the memory over the last five hours of 10,000 people. Where have they been? Where have they been in 3D space? What have they been doing? The world model thing is very exciting because they're storing memory in literally video latents. A minute of frames, it's showing super early promise. I think the technical revolution is going to be what's the ultimate synchronized state for 10,000 people? Is it in video latents? Is it a much more native 3D format? Is it a hybrid new discovery of some 3D video latent space to be determined? I will share that the format we're using, part of the vision for us is to ultimately store the history of everything on Roblox. and to store it, not raster like video, but store it in vector, which means the ability to play back anything that's ever happened on Roblox, that's 13 billion hours a month.

explains mechanism · 3provides context · 1

05
Mechanism

Roblox is building NPCs through three levels: first, NPCs that can competently play any Roblox game by training on the platform's 13 billion hours of monthly 3D vector data; second, personalized virtual doppelgangers that learn an individual's gestures and behaviors; third, agentic doppelgangers that can be sent out to act on your behalf.

Dave outlines a three-tier NPC roadmap: (1) NPCs trained on Roblox's 13B hours/month of native 3D data that can play any game, (2) privacy-compliant virtual doppelgangers that learn your personal behaviors, and (3) agentic doppelgangers you can send out to act for you — like playing with your child when you're busy.

transcript

Dave Baszucki: What we are starting to train is NPCs based on a combination of all of the data we have in a privacy compliant way that are starting to navigate and starting to play games. The challenge, I think, would be challenge level one. Everyone has access to an NPC that can get pretty good at playing any Roblox game. Super cool. Challenge level 2 is by watching, if I so opt in a privacy compliant way, my behaviors, the way I act, literally my gestures, the way I look at things, the way I talk, would we allow you to have your own virtual doppelganger if you wanted. Wow, that's fascinating. And then number three would be is if I have a good virtual doppelganger, just as agentic is all the buzz in other areas, is there a simple user interface for agentic virtual doppelgangers? And that starts to get interesting because one could imagine sending their virtual self out to go do something. It could be, my son or my daughter wants to play with me. I'm having to do some work. Could my virtual doppelganger fit in for 15 minutes? Sure, that's great. That could be really interesting.

explains mechanism · 1extends · 1provides context · 2supports · 1

06
Prediction

The role of game designers will be more leveraged and creative — not eliminated — as AI raises quality expectations: a team of five will produce astounding work, creators will spin up agentic testing and iteration overnight, and the human touch has a long runway against AI slop.

Dave is optimistic about the future of game creation: he believes human creators will be more leveraged, not replaced. AI will enable creators to spin up agents that iterate and test their experiences overnight, sending NPCs into games on emulators to tune them. Quality will rise dramatically, but human touch remains essential.

transcript

Dave Baszucki: I think I'm optimistic that typically industrial revolution or whatever revolution, I think we've chatted about stuff like this before, there's a thought that there won't be any jobs after this. I think I'm a little bit more optimistic that a new class of things will show up. And I think there's a huge opportunity for more creators. I think I'm not to the point where I believe what we call AI slop today is going to take over. I think the human touch has got a long runway on it. So you could imagine more diversity or quality of experiences supplemented by AI. But I do think the role will be more leveraged and the quality coming out of five people will be astounding, but we'll come to expect that quality. So quality will go up. I think iteration on testing will go up. I think where we see platforms like Roblox going is just as people are starting to spin up agents to develop code. I think people, hopefully on the Roblox platform against the Roblox cloud, you'll be able to go away for 24 hours and your agents are going to be like tweaking and testing, spinning up a Roblox experience, sending 20 NPCs into them on various client emulators.

extends · 2provides context · 3supports · 1

Highlight slides
Roblox's 20-Year Vision: The Holodeck✦ from: The ultimate vision for Roblox, held for 20 years, is a high-fidelity physics-simulated space where people can co-experience — build a car, drive it, go to a birthday party, chop trees — and AI's role is to superpower the creation of that holodeck and make it photorealistic, scalable to 10,000 people, and acoustically simulated.AI Superpowers the Holodeck Across Three Dimensions✦ from: The ultimate vision for Roblox, held for 20 years, is a high-fidelity physics-simulated space where people can co-experience — build a car, drive it, go to a birthday party, chop trees — and AI's role is to superpower the creation of that holodeck and make it photorealistic, scalable to 10,000 people, and acoustically simulated.Three-tier NPC roadmap✦ from: Roblox is building NPCs through three levels: first, NPCs that can competently play any Roblox game by training on the platform's 13 billion hours of monthly 3D vector data; second, personalized virtual doppelgangers that learn an individual's gestures and behaviors; third, agentic doppelgangers that can be sent out to act on your behalf.Training data scale✦ from: Roblox is building NPCs through three levels: first, NPCs that can competently play any Roblox game by training on the platform's 13 billion hours of monthly 3D vector data; second, personalized virtual doppelgangers that learn an individual's gestures and behaviors; third, agentic doppelgangers that can be sent out to act on your behalf.Use case: agentic doppelganger✦ from: Roblox is building NPCs through three levels: first, NPCs that can competently play any Roblox game by training on the platform's 13 billion hours of monthly 3D vector data; second, personalized virtual doppelgangers that learn an individual's gestures and behaviors; third, agentic doppelgangers that can be sent out to act on your behalf.
Related episodes