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

Roblox’s foundational vision, laid out nearly 20 years ago, was to create a new category of “human co-experience”—the ultimate high-fidelity simulation, or “Holodeck,” backed by physics. This platform would enable users to interact with a simulated reality in meaningful ways, from building and driving a car to participating in a virtual birthday party. Today, AI is viewed as the key to supercharging this long-standing vision, accelerating the path to photorealistic, large-scale, and acoustically accurate immersive worlds.

The Future of Communication: 4D Simulation

The evolution of human communication has always trended towards higher fidelity, from smoke signals to video conferencing. The next logical step is a 4D immersive simulation, which is viewed not just as a 3D space but one that includes function and physics over time. This medium is a superset of video; one could always down-sample the immersive experience to a 2D video call, but not the other way around.

Key advantages of 4D simulation for communication include:

• Natural Audio: In a meeting with a thousand people, a 3D simulation allows for natural sound attenuation based on distance, creating a more intuitive experience than the grid of squares in a video call.
• Shared Interaction: Participants can move around and interact with a shared environment, such as walking through an office to look at a physical object together.

A significant technical hurdle is latency, particularly for synchronized activities like singing “Happy Birthday.” The proposed solution involves “time-forward extrapolation,” where each participant’s client predicts the others’ actions to remix the audio in real-time, creating a seamless experience despite the milliseconds of delay.

Training Native AI on Unprecedented Data

Roblox is leveraging its massive dataset—13 billion hours of user interaction per month—to train its own native AI models. A crucial distinction is that this data is stored in a vector format, not as rasterized video.

The data we have which is 13 billion hours a month is can be reproduced from any camera angle and can interact with the 3D space. So it’s very powerful data.

This vector format provides a rich, replayable history of every action and interaction within the platform. While respecting user privacy, this data can be used for practical applications like safety incident reviews, creating cinematic replays of personal moments, and most importantly, as a vast training ground for AI. This dataset of “hybrid video ASDW human interaction data” is invaluable for creating sophisticated NPCs that go beyond simple language models.

Evolving NPCs: From LLMs to Virtual Doppelgängers

The goal is to use this rich interaction data to create NPCs that are more than just conversational agents. This represents a shift from heuristic-based behaviors (like decision trees) to end-to-end learned models, similar to the evolution seen in self-driving technology.

The development of these NPCs is envisioned in three stages:

1. Game-Playing NPCs: Provide creators with general-purpose NPCs capable of competently playing any game on the Roblox platform.
2. Virtual Doppelgängers: Allow users to opt-in to have their behaviors, gestures, and speech patterns modeled to create a personalized virtual version of themselves.
3. Agentic NPCs: Enable these virtual doppelgängers to perform tasks on a user’s behalf, such as playing a game with a family member when the user is busy.

These NPCs would have a persistent memory embodied within the platform, able to access and learn from everything they have ever said or done in a lean, efficient format.

AI’s Role in Creation and the Game Industry

AI is set to fundamentally change the creator workflow and the economics of game development. While generative tools make assets cheaper to create, consumer expectations for quality rise in tandem, meaning development complexity doesn’t necessarily decrease.

Within Roblox Studio, AI will act as a powerful assistant, integrating with standard AI coding workflows while also providing unique capabilities for environmental generation. The vision is a seamless process:

• A creator uses a prompt (text, image, or even by walking around in a world model) to define an experience.
• AI iterates this into a primitive 3D skeleton for review.
• Finally, it’s fleshed out into a fully functional game.

Furthermore, AI can be used for “just-in-time” 3D upsampling. A creator can build a primitive experience and use a prompt like, “make it look medieval and more realistic,” and the platform’s cloud infrastructure will automatically upgrade all the assets to a higher fidelity on demand.

Strategy: Long-Term Vision and Unconventional Hiring

Roblox balances its 20-year vision with a culture of rapid weekly iteration. The long-term goal of building the Holodeck provides a stable, multi-decade specification, while individual teams work in fast cycles to make progress towards it. This is framed by a concrete intermediate goal: capturing 10% of the global gaming market.

To power this vision, Roblox has adopted a unique, data-driven hiring philosophy. By acquiring a company specializing in 3D assessment tools, Roblox developed its own testing to identify top-tier problem solvers, independent of their educational background.

We have found that not necessarily to be true and and we have found community college, you know, the small Midwestern engineering school, like because we’re assessing our own way, we we kind of we basically ignore the signal of where you went to university.

This meritocratic approach has allowed them to find and cultivate untapped talent, building a strong internal pipeline of future leaders.

• Tags:
• Feature
• Simulation
• World Models
• NPCs
• Game Development
• Virtual Worlds
• Roblox