Pod Insights a16z Podcast Leveling Up with Roblox's David Baszucki About this episode Roblox cofounder and CEO David Baszucki and a16z's Jonathan Lai discuss the ways AI is changing games and virtual worlds and how we create them.
[0:00] Generative AI in games
[4:37] Fully personalized playthroughs
[7:30] Roblox's generative model
[11:08] Future of interfaces
[13:05] impact on creators
[16:16] NPCs as copilots
[18:29] VR & spatial computing
This conversation is part of our AI Revolution series, recorded August 2023 at a live event in San Francisco. The series features some of the most impactful builders in the field of AI discussing and debating where we are, where we’re going, and the big open questions in AI. Find more content from our AI Revolution series on www.a16z.com/AIRevolution.
Transcript 0:00 Speaker 0 Okay, I'm making the super dungeons and dragons thing, but as it watches you play, and maybe we know your history, you're literally you'll be playing a 3D experience that Snow has ever seen before.
0:39 Speaker 1 So let's jump into it. We've we just had a session talking about how um generative AI is potentially transforming all types of media. Tell us a bit about how AI is transforming the way folks are building games on Roblox.
0:52 Speaker 0 Yeah. We expansively go beyond gaming even and we think about the ultimate form of the way humans communicate at a distance. And what we like what we see in a lot on our platform in addition to playing is just how do uh kids connect with their friends in the middle of COVID. Uh how do people go to a concert with their friends when they can't go together in person? How do we in our own office simulate serendipity when some of us are remote? So we have a simulation of the Roblox office that we're playing around with inside of Roblox. So if you start, if we if we take a step back and say immersive 3D, it will with market forces follow the phone system, video communication, SMS, then the influencers are just massive.
1:38 Speaker 0 And I think there's three categories. I I like some of the earlier discussion. There is one category where people on our platform don't even think of it as AI, even though it's been going on for two or three or four years. Um quality of personalized discovery, uh quality of safety, civility, voice and text, uh monitoring, asset monitoring, quality of real-time natural translation, how good is our translation versus others. So that's that's the one that people don't notice. The next one is I think the one that's really exciting right now, which is generative, either code generative, 3D object generative, avatar generative, game generative, which is very interesting. And then and then the future one, which is really exciting is how far do we get to a virtual doppelganger or a general intelligence agent inside of a virtual environment that's very easy to create by a user?
2:34 Speaker 0 You want George Washington in your 12-year-old school project, how good is George Washington, or I want um, I'm a knot on Tinder, but if someday Tinder has a Roblox app, can I send my virtual doppelganger for the first 3D meeting kind of thing? So I think I think going all the way from the things we don't notice to the things that are exciting around generative to future, then general intelligence, these are all gonna change the way this works.
3:03 Speaker 1 That's very exciting. And when you think about the parts that go into building the game today, like there's just so many, so many, so many pieces, right? There's the um there's the concepting, the storyboarding, there's the writing, there's the creation of the 2D images, the 3D assets, then there's the code and the physics engine, and so Roblox has built many of these pieces into its own studio and its platform. Um what parts do you think will be most affected by this new generation of generative models that you just spoke about?
3:34 Speaker 0 Yeah, it's almost worth saying the anti-thesis, what will not be affected because ultimately there will be acceleration on all of these. We have a bit of an optimistic viewpoint right now because of the say 65 million people on Roblox. Most of them are not creating at the level they want to. And and we for a long time imagined uh simulation of Project Runway, where the early days of Roblox, we imagine Project Runway is just pretty skew morphic. You have sewing machines and fabrics, and it's all 3D simulated, and that's how you would do it. But when we think about it, even that's kind of complex for most of us. And I think now when Project Runway shows up on Roblox, it will be text prompt, image prompt, voice prompt, whatever you want.
4:19 Speaker 0 As if you're you're sitting there, and if I was helping you make that, I'd say I want kind of a blue denim shirt. I want some cool things, I want some buttons, make it a little more trim, fit it. What we'll see those kind of creations. So I I actually think we're gonna see an acceleration of creation.
4:37 Speaker 1 I think one of the things um uh you might have mentioned earlier in a podcast is the idea that people can create experiences inside of other experiences. And so even when they're inside a game, to your point, they have a fully postalized avatar or cosmetics. Yep.
4:53 Speaker 0 I I think that you know, the classic uh 3D creation is use these heavyweight Blender, 3D studio Maya, Roblox Studio, use these things. And and for a while, we at Roblox were thinking, how do we make Roblox Studio easier to use? Two million people use it. But there is another level above that. We saw early signs of this as our users tried to build, you know, tycoon builder, roller coaster builder, those kind of things. The more access they have to the 3D objects and the more persistence all of a sudden you get better creation as AI starts being a service available within any experience, rather than you creating, you know, we would like Project Runway to make the ultimate project runway experience, dynamic, persistent 3D clothing, not in Maya or 3D studio, but in the experience itself, accelerated my cloud AI tools to help that.
5:45 Speaker 0 And then there's so much leverage. We're not having to make Roblox Studio better. We're providing a lot of possibilities for then that very vertical creation experience, and we would see the same for cars and for automobiles and for all kinds of other things.
6:02 Speaker 1 Yeah. But I think one of the things you're getting at here is just the power of um the concept of a personalized playthrough. Yeah. And so the idea is that maybe every user in Roblox in the future has a fully unique experience that's tailored there, their preferences and the choices they made and so on and so forth.
6:19 Speaker 0 Yeah, I think you're touching on an even further extension with takes probably a lot of compute horsepower, which is completely personalized generation in real time backed by massive inference stuff. So you you could imagine, okay, I'm making the super dungeons and dragons thing, but as it watches you play, and maybe we know your history, you're literally you'll be playing a 3D experience that Snell has ever seen before. So I think we will see in certain types of play real-time procedural at some point that obviously you need enough inference in real time to generate what you would like. Right. Um one of the good things we've done, and I can see some Roblox people in the audience here, is we we for a long time really focused on building our own infrastructure.
7:08 Speaker 0 We have hundreds of thousands of servers, you know, many, many edge data centers, terabits of connectivity that we've traditionally used for 3D simulation, that the more we can run inference jobs on these, we can run super high volume inference at high quality at low cost and make this, you know, just freely available so the creators
7:30 Speaker 1 don't worry about it. Yeah. There makes a lot of sense. And actually, just in the topic of infrastructure, in your latest showholder letter, you had actually mentioned that Roblox is working on your own multimodal generative model, right? That will encompass many of the things that we just chatted about. Can you tell us a bit more about sort of what you're building there? And
7:48 Speaker 0 yeah, I think that I mean, the there's probably, you know, in any company like a Roblox, there's probably 20 or 30 end user vertical applications that are probably very bespoke, you know, natural language filtering very different than generative 3D. And at the end user point, we want all of those running. We want to use all of the data um in an opt-in fashion to help make these better, tune these better. But as we go down, down, down, there's there's probably a natural two or three clustering of general bigger, fatter type models in a company like ours. There's definitely one around safety, civility, natural language processing, natural language translation, generally a one more multimodal thing around 3D creation.
8:37 Speaker 0 How would how would you um just like in Project Runway, say um some combination of text, image, whatever, generate a great avatar? And then there's probably a third area which gets into the virtual human area, which is how would we take the five billion hours of human opted-in data, what we're saying, how we're moving, where we go together, how we work in a 3D environment, and could we use that to maybe inform a better 3D simulation of a human? So I would say, yes, looking at large models in those three areas, and I and I think you know, we're the market as we see it, there's there's gonna be these super big whatever God model massive LLM type companies.
9:22 Speaker 0 I think we are probably a layer below that. We're very fine-tuned for the disciplines we want, but the ability to train and run massive
9:31 Speaker 1 inference for this. But it's a massive um proprietary data set that you guys are working off of of all of the 3D assets that have been created in Roblox over that over the years and so on that you can train off of.
9:44 Speaker 0 There's a lot of data in in any kind of 3D immersive environment. Um the communication from people which can inform safety and civility, uh, 3D motion eye tracking, uh-huh how do humans interact. I think. I think what one area we're really watching, that's a very difficult issue problem right now, is true high quality 3D gener generation as opposed to 2D generation. There's a there's lots of wonderful 2D generation stuff out there. We're really double down on 3D generation.
10:17 Speaker 1 How do you think about what parts of the tech stack you want to partner with from external partners, for example, versus you know building yourself in house with all of that, the data and the compute resources that you have available?
10:31 Speaker 0 I think um areas that we're interested in are um there's a lot there's companies exploring what's the future of training cheaply at mega volume. What's the future of running inference cheaply at mega volume? What types of technology abstracts away different hardware devices? Um how can you run a mixed CPU GPU environment um over time? We're very interested in that. Um so I think we're watching those types
11:06 Speaker 1 of tech stacks a lot. Do you think you'll need to have a new like user interface or discovery mechanism just as the sheer quantity of games and experiences, you know, if we're moving towards a metaverse.
11:19 Speaker 0 You know, ultimately, whether it's Amazon or Netflix or TikTok or YouTube, I think that for us the difficult discovery problem is you're trying to optimize tenure enterprise value along with your stock price every year that you get there. And a lot of short-term discovery mechanisms do not optimize long-term value if some amazing creation gets suppressed where it could have added five percent to the platform. So I think um there's a lot of opportunities in discovery. I think super hard to not a solved problem, although many companies are solving it in really interesting ways right now. I think the user interface, there's ways that can change as well. I think you know, there's there's a lot of opportunity in addition to thinking of this just as content and thinking of this as your real-time social graph and thinking through what other people are doing in addition to the
12:17 Speaker 1 content. It's fascinating because I think one of the examples of um AI being used by big companies that Martin just talked about was Netflix and then I think TikTok as well, if there's sort of a personalized imaging recommendations and you could maybe imagine a feature where a user that onboards in the Roblox isn't doesn't actually see a library or a catalog of games, but it's just presented with like a feed, and it's almost like you're just going from one thing to the next.
12:44 Speaker 0 This is really right. I think we're we are constantly testing, you know, the new user experience. Um, should that be 2D? Should that be 3D? What's the weighting between creating your digital identity versus discovery? What's the weighting between connecting with your friends and optimizing all that? And and we may find that it's it has to be personalized.
13:06 Speaker 1 So what is ship briefly to the player perspective? Um now that you've had you have a couple of AI tools, like things like the code assist and the material generator out. Um what has been the reception from the creators, the players, and then counter disks, like any any feedback?
13:23 Speaker 0 Whenever we give more capabilities to the user or the player base, we we have historically seen a lot of acceleration, even if it's very um early. And so I think it is it's still super early. Um the things we have launched right now are are just a very light layer, you know, code generation, super exciting, uh material generation, super exciting, potentially um prompt driven avatar creation even more exciting, potentially, you know, general purpose 3D asset creation, even more exciting.
14:00 Speaker 1 Do you think the types of games that leverage these sorts of technologies might change themselves in terms of the gameplay mechanics, like the genres that you might see emerge?
14:10 Speaker 0 Yeah, I think we will see like I'm I'm optimistic. We will um the more people have control of inexperienced creation, for example, going back to the project runway or Carly Kloss type example experiences where uh creation amongst 65 or 70 million people is just part of the way it goes, have not been possible. But an experience where there's millions of people acting as fashion designers and voting and picking who's got the best stuff, and then possibly imagining some of that, you know, going off and being produced in real life, or some of them being plucked up by parsons and say, okay, the future designer. So you could imagine just like in um, you know, I think there's a race car driver that got picked because they were so good at a simulator.
15:01 Speaker 0 You know, you could you can imagine other genres like this fashion design, other types of creation where you actually create on platform and then get identified as
15:12 Speaker 1 a future star. I'm just really excited about the notion of taking the creativity that you find and kids doing imaginative play and then turbocharging that basically the AI tools where it's like, what if this would have happened? What if there was a natural disaster in the middle of the stream? And that's right. Right now, it's actually it's very difficult for those creators to make those experiences, but like with a text prompt, you could imagine there there's games that just materialize and just like that diversity and uh the wildness of those experiences can be.
15:41 Speaker 0 Yeah, I think having a text prompt that's generative, not at the developer level, you know, building the big heavy game, but having text or voice prompt as just something that's naturally part of any experience wherever you go. And it's almost like rather than the traditional ways with either voxels or this of building things in 3D experiences, it just becomes natural that you use a text prompt. Just like um in a traditional avatar editor, rather than sliders and radio buttons that will move to, I think a more interactive type text
16:16 Speaker 1 prompt thing. And something we've been excited about is just the notion of um uh MPCs as teachers and co-pilots with you as well, right? And so let's say the first time that you're onboarding the Roblox, you could have a guide that basically shows you what to go where to go next. And then if you're interested in building, you know, helps you lay your first few, you know, experiences together and so on. And that feels like it's potentially very powerful as the concept of like MPCs as co-pilots and sort of co-players with us as needed. Yeah, I think
16:46 Speaker 0 there's a lot of um interesting stuff going on in various ecosystems around this co-pilot notion. Um there is one co-pilot where we're all wearing like our little earbud all day long, and that co-pilot is talking to us. So that's more maybe more consumer real-time co-pilot. There's obviously many companies trying to build the co-pilot that you hook up to your email, your text, your Slack, your web browser, and whatever, and it starts acting for you. I'm really interested in the notion that co-pilots will either will talk to other co-pilots using like natural English, I think will be the universal interface of co-pilots and you you would imagine NPCs being created by prompts, you know.
17:32 Speaker 0 Hey, I I'm building the historical constitutional thing. I want George Washington there, but I want George Washington to act at the highest level of civility and usher new users through the experience, tell them a little about constitutional history go away when they're done. Like I actually do think you will see those kind of assistance.
17:55 Speaker 1 That's an incredible example because what better way of learning history than actually talking to the folks that were part of that historical scene or time period or so on, right? And so you could actually see how this advances the educational.
18:08 Speaker 0 Yeah, we've been we've been talking a lot about the communication aspect of going to ancient Rome with your friends. The augmentation of you'll go to ancient Rome and most of the people you'll interact with will be historically as good as we can create representations to enrich in that experience. It'll be pretty fascinating. Yeah.
18:30 Speaker 1 Awesome. Um AI question that I wanted to tee up and then when you got into our closing questions. Um a couple of weeks ago, you were tweeted that um uh the Roblox app on uh MetaQuest had actually hit a million downloads in just the first five days and it's beta form, it was out on that's on the actual Oculus store. So what are your thoughts on VR, spatial computing? Yeah, so our our
18:59 Speaker 0 thesis has been that just as when the iPhone shipped and all of a sudden we had 2D HTML consumable on a small screen rather than a large screen with the pinch in Zoom, and now we take it for granted. I think my kids probably don't realize there was some cheesy mobile web thing years ago pre iPhone where browsers were large screen things. Um now we just assume 2D HTML is everywhere. I think 3D we feel is the same. It's um it's immersive multiplayer in the cloud, simulated 3D. And because of that, every device has better optimal for the device camera, optimal for the device user interaction, different levels of immersiveness.
19:45 Speaker 0 Your phone is not as immersive as um your VR headset, but your phone is more spontaneous. So I think we've we felt that. And we think the market ultimately figures out which device you consume this with. And and we're we're not trying to move the market, but we do want to be in every place so we can watch the market move with our players. I I would say one of the historical things we've done is all of our 3D engineers are working on crappy Android devices as their primary test device, which has been enabled that helped with Oculus because I'm not saying it's a crappy Android device, but I'm saying it's got two, um, it's rendering two eyes rather than one, which is technically challenging.
20:36 Speaker 0 And so having a foundation of more working at high performance rather than harvesting 16 cores helped us to do
20:45 Speaker 1 that. And I think it's exciting to think about um potentially different types of experiences being optimized for different devices, right? And so maybe you play a tycoon game on your PC where you have an awesome keyboard and then you play a first person shooter in the VR headset. That's how you're picking the right. That's exactly right. It's impossible to abstract all the way across hand controllers down to a phone. Right. Makes sense. Excellent. So quick um lightning round of fun questions to players. Cool. What is your favorite book? Infinite game. What is your favorite game? Chess. Chess.
21:21 Speaker 1 And um, what is the best advice that you have received as a young person from a mentor or a friend or so on?
21:29 Speaker 0 Wow, I can I kind of feel I didn't get much good advice when I was young, actually. Um the worst advice I got from my dad is you should go into accounting in business. Um so I don't know. I guess my advice would be I'm trying to think who gave me great advice. Uh my general advice to everyone would be I thought, you know, my first two jobs were a disaster and just like ah, you know, whatever, but life is long. And I think if if my general advice, if you find something you're really good at and you love at the same time, that's the that's obviously the optimal Venn diagram. You know, someone follow your passion, it's not follow what you're good at. Can you overlap those Venn diagrams, which is like the sweet spot?
22:13 Speaker 1 And what is one what is one wild prediction for how you think the world will evolve over the next five years?
22:22 Speaker 0 Five years. I have a touring test question for AI, and that would be if we took AI in 1633 and trained on all the available information at that time, would it predict the Earth or the sun as the center of the solar system? Even though 99.9% of the information is saying the earth is the center of the solar system. So I think five years is right at the fringe of if we were to run that AI touring test, it might say the sun.
22:52 Speaker 1 Interesting. Got it.
22:53 Speaker 0 Do you have a different answer if it was 10 years? 10 years, I think it'll
22:56 Speaker 1 say the sun. Awesome. Thank you, David. This has been a pleasure.
Previous Engineering Rapidly Reusable Rockets Next AI Revolution: Top Lessons from OpenAI, Anthropic, CharacterAI, & More Page updated August 25, 2026 · transcript and summary are machine-generated