According to Business Insider, Roblox CEO David Baszucki recently took a sabbatical where he tried to catch up on the latest AI research. He found the experience “very humbling,” admitting it’s now extremely difficult to understand all the new research papers being published. Baszucki, who founded Roblox in 2005, recalled a time when he could read and comprehend all the research relevant to gaming, from physics to rendering. Now, he describes a “horizontal wave” of AI advancements—from transformers to diffusion models—that is “so massive and so fast.” His key takeaway is that AI for 3D and the physical world is still “very early,” as models are trained on human-made text and images rather than the raw 3D material of reality itself.
The Age of Information Overload
Here’s the thing: Baszucki is articulating a feeling a lot of us in tech have right now. It’s not just imposter syndrome. The sheer volume and velocity of AI research has created a genuine knowledge gap. When the founder of a multi-billion dollar platform that’s essentially a massive 3D creation tool says he can’t keep up, you know it’s serious. We’ve moved from a pace where a dedicated expert could theoretically stay on top of their niche, to a firehose of daily papers from corporate labs and academia alike. And as he pointed out, the research itself is becoming more private, with companies like Google pulling back on publishing. So not only is there more of it, but less of it is even accessible. How is anyone supposed to build a coherent mental model of the field?
Back to the Age of Research
Ilya Sutskever’s recent comment, which Baszucki referenced, really hits the nail on the head: “It’s back to the age of research again, just with big computers.” That’s a profound shift. For a while, the narrative was all about scaling—bigger models, more compute, more data. Now, the frontier seems to be moving back to fundamental breakthroughs. The companies that win might not be the ones with the most GPUs (though that helps), but the ones with the sharpest researchers asking the right questions. It’s a different kind of arms race. You can hear Baszucki’s full thoughts on the “Access” podcast if you want to dive deeper.
Why 3D Is the Next Frontier
Baszucki’s most interesting insight, to me, is about the “physically unnatural space” of current AI. He’s right. We’ve trained these brilliant models on a 2D shadow of the world—text and flat images. It’s like teaching someone about objects solely through their silhouettes. The “3D raw material of the world itself” is a completely different beast. Understanding physics, spatial relationships, and material properties from first principles is what’s needed for true simulation and creation. That’s why it feels early. Roblox’s entire business is built on user-generated 3D worlds, so he’s obviously looking for the tech that will blow that wide open. When that research matures, it won’t just change gaming. It’ll revolutionize everything from industrial design and architecture to robotics. Basically, we’re still playing with the easy data.
What It Means for Everyone Else
So what’s the takeaway for the rest of us who aren’t CEOs on sabbatical? First, give yourself a break. If a leader at that level feels humbled, it’s okay to feel overwhelmed. Second, it signals that genuine, deep expertise in a specific sub-domain of AI might be more valuable than a surface-level understanding of everything. The field is specializing. And finally, Baszucki’s focus on 3D is a great prediction of where the next wave of practical, world-changing applications will come from. The 2D text and image revolution is still unfolding, but the race to understand and generate the 3D world is just getting started. That’s where the real magic—and the real headaches for anyone trying to keep up—will be.
