No Priors

How AI Will Transform Roblox Games into Photorealistic Worlds | CEO David 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 with photo-realistic graphics and acoustic physics. Dave reveals how Roblox is leveraging 13 billion hours of monthly user data to train native AI models that go beyond simple LLMs, enabling NPCs that can navigate and play games with human-like intuition. He also talks about how immersive communication will change video conferencing, how Roblox searches for unlikely talent outside of traditional elite universities, and how he balances rapid weekly iterations with keeping a “long view” on Roblox’s vision. Sign up for new podcasts every week. Email feedback to show@no-priors.com Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @DavidBaszucki | @Roblox Chapters: 00:00 – Cold Open 00:36 – Dave Baszucki Introduction 01:16 – Realizing Robolox’s 20-Year Vision 05:29 – Using 4D Immersive Simulations in Virtual Interactions 08:22 – Physics Engine vs. Photorealism 11:50 – Storing Roblox History as Vector Data 14:00 – Training NPCs - Moving Beyond LLMs 18:05 – The Future of the Game Designer 19:54 – Video Latent World Models 23:53 – Social Simulation - AI Companions and Virtual Relationships 27:26 – Why Asset Costs Haven’t Changed the Gaming Industry 29:52 – AI Coding in Roblox Studio 31:36 – The Roblox Creator Economy 33:57 – Long-Term Conviction vs. Weekly Iteration 37:50 – Dave’s Hiring Philosophy for Roblox 43:44 – Conclusion

EGElad GilhostDBDave BaszuckiguestSGSarah Guohost

Feb 5, 2026·43m·Watch on YouTube ↗

📖 CHAPTERS

1. 1
   0:00 – 0:36

   AI relationships, human-AI barrier, and why Roblox aims to last 40 years

   The episode opens with a provocative question about whether today’s children will form serious relationships with AI. Dave Baszucki argues we’re not yet crossing the human–AI barrier, but the question is useful for thinking about where immersive worlds and AI companions could go over decades.

2. 2
   0:36 – 1:16

   Roblox’s 20-year “Holodeck” thesis: co-experience, physics, and photoreal scale

   Baszucki describes a 20-year-old Roblox business plan that envisioned a new category centered on shared human co-experience. The core idea is a high-fidelity simulated world grounded in physics where people can do real things together—ultimately a Holodeck-like platform.

3. 3
   1:16 – 5:29

   Two futures: social co-presence vs. “real-time dreaming” worlds of NPCs

   Baszucki contrasts a communication-first metaverse with a solitary, AI-generated “real-time dreaming” experience optimized for personal immersion. He argues the future will include everything between these extremes, including new product categories we can’t yet name.

4. 4
   5:29 – 8:22

   4D immersive communication: why simulation can beat Zoom

   The conversation shifts to business and collaboration use cases, arguing that immersive simulation could become a superset of video calls. Baszucki emphasizes acoustics and spatial interaction as key reasons 4D environments can feel more natural than grids of faces.

5. 5
   8:22 – 11:50

   Physics engine versus creator-driven fun: realism isn’t the only winning strategy

   Roblox’s roots in physics simulation remain central, but Baszucki notes that creators often build viral experiences that aren’t realism-first. Platform capabilities like persistence and cloud features unlock diverse game mechanics beyond pure physical simulation.

6. 6
   11:50 – 14:00

   The scaling frontier: synchronizing 10,000 players and the “state of the world”

   Baszucki describes the core technical challenge as multiplayer state synchronization, not just better physics. The key question becomes how to represent and efficiently synchronize the world’s memory for thousands of participants over time.

7. 7
   14:00 – 18:05

   Storing Roblox history as vector data: replay, safety, and a training goldmine

   Baszucki outlines a vision to store the history of everything that happens on Roblox as vector (not raster video) data. This enables privacy-conscious replay for safety investigations, personal memories, and powerful training datasets that can be re-rendered from any camera angle.

8. 8
   18:05 – 19:54

   Training next-gen NPCs: beyond LLM chat to embodied, game-playing agents and doppelgängers

   Roblox is training NPCs using platform-native behavioral data to create agents that can navigate and play games, not just talk. Baszucki sketches a progression from general game-competent NPCs to opt-in personalized doppelgängers and eventually agentic virtual selves you can “send” to act for you.

9. 9
   19:54 – 23:53

   What creators build with richer NPCs: from GTA-style worlds to interactive education

   The hosts explore how better NPCs could change the kinds of experiences creators make. Baszucki points to both high-end emergent worlds and accessible educational or historical interactions as likely outcomes, enabled by cloud inference and platform services.

10. 10
    23:53 – 27:26

    The future of game designers: AI-leveraged teams, autonomous testing, and faster iteration

    Baszucki argues AI will amplify creators rather than replace them, increasing both output and expectations. He anticipates agent-driven development workflows: automated playtesting, tuning, and iteration running continuously in the cloud.

11. 11
    27:26 – 29:52

    Video latent world models and a hybrid rendering stack for multiplayer worlds

    The discussion turns to world-model research that generates interactive experiences directly from video without explicit physics. Baszucki expects a hybrid architecture combining synchronization engines, NPC services, 3D upsampling, and client-side 2D upsampling rather than a single monolithic model.

12. 12
    29:52 – 31:36

    Social simulation, AI companions, and memory with embodiment

    Baszucki reflects on AI companions and social simulation scenarios (including a Black Mirror-like thought experiment) while emphasizing Roblox’s non-dating stance. The key product opportunity is combining embodiment, persistence, and memory so NPCs can recall and evolve based on platform history.

13. 13
    31:36 – 33:57

    Why cheaper assets haven’t reshaped gaming yet: rising quality bars and cloud-native assets

    Asked why generative tools haven’t changed industry dynamics, Baszucki argues consumer expectations rise as fast as production gets cheaper. He outlines a cloud-first future where assets have dynamic LOD and may shift from files to prompts generated on demand.

14. 14
    33:57 – 37:50

    AI coding in Roblox Studio and end-to-end environment generation workflows

    Baszucki describes AI assistance in Studio via both industry-standard model integrations and Roblox-native assistants. The longer-term vision is multimodal environment generation that starts with prompting and iterates into a functional, testable game backed by Roblox’s cloud tooling.

15. 15
    37:50 – 43:44

    Creator economy, live ops cadence, and transparent discovery—plus how Roblox balances vision and weekly shipping

    Baszucki explains how successful Roblox development has shifted toward continuous live ops and frequent updates, supported by cloud deployment. He also discusses transparent discovery signals, long-term vision (Holodeck) paired with rapid iteration, and a measurable stepping-stone goal of reaching 10% of gaming.

16. 16
    43:44 – 43:44

    Hiring for problem solving: Roblox’s 3D assessments and rethinking university signals

    Baszucki shares Roblox’s systems-driven hiring approach, including acquiring Imbellus to build 3D assessments for creativity and problem solving. He argues this approach can be fairer and more predictive than relying on university prestige, surfacing strong candidates from a wider range of schools.