Apr 17, 2024 · 50m · mad
AI at Roblox: Revolutionizing Game Creation | Morgan McGuire, Chief Scientist of Roblox
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the MAD Podcast, host Matt Turck interviews Morgan McGuire, Chief Scientist at Roblox, to discuss how the company leverages custom infrastructure, machine learning moderation, open-source AI research, and controllable generative tools to empower millions of 3D content creators worldwide.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 8.9% of the talking time here. How this is scored →
Matt as informed peer 1.5 Guest teaching 2.6 Guest disagreement 0.2 Matt pushing back 0.0
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 33:15 Rejecting prompt engineering as true Gen AI
Morgan lightly critiques prompt engineering practice, calling it an undocumented, strange programming language rather than true generative AI.
Hardest push from Matt ▶ 30:31 Challenging complete automated generationMatt pushes back on hype around full automation, asking if Gen AI only produces a draft that requires human intervention for the last mile.
Biggest teaching moment ▶ 14:40 Explaining fundamental shift to data-driven codeMorgan breaks down technical fundamentals, educating the host on how machine learning replaces explicit developer code with data-controlled execution.
Matt holds his own ▶ 30:31 Identifying draft-versus-execution limitationMatt demonstrates sharp domain knowledge by pointing out that Generative AI tools rarely achieve 100% precision in specialized design pipelines.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Overview of Roblox as a 3D Social Platform | 1 | 2 | 0 | 0 | Matt opens with standard background questions introducing Roblox as a 3D social platform rather than a game. Morgan explains the basic user metrics and community creation model. | |
| Core Infrastructure, Server Scale, and Engineering Teams | 1 | 2 | 0 | 0 | Matt prompts for behind-the-scenes details on scaling infrastructure and team size. Morgan outlines their 100k servers, edge data centers, and engineering distribution. | |
| Roblox Studio and the Developer Creator Ecosystem | 1 | 3 | 0 | 0 | Matt asks about pre-AI creation tools. Morgan provides an extensive breakdown of Roblox Studio, free instant publishing, Luau scripting, and prosumer integrations. | |
| Generative AI Strategy, Open Culture, and Weekly Releases | 2 | 2 | 0 | 0 | Matt compliments Roblox's rapid pace of generative AI releases. Morgan details their open roadmap, public iteration philosophy, and unique weekly deployment cadence. | |
| Evolution of Machine Learning in Safety and Moderation | 1 | 4 | 1 | 0 | Morgan reframes the conversation around AI definitions, contrasting data-driven ML with traditional code and detailing early transformer model adoption for real-time safety. | |
| AI Code Assist and Open Source AI Models | 2 | 3 | 0 | 0 | Matt asks if AI Code Assist is homegrown or third-party. Morgan explains their hybrid architecture, open-source StarCoder research, and domain transfer techniques. | |
| Generative AI for 3D Materials and Textures | 1 | 3 | 0 | 0 | Matt introduces 3D image/material generation. Morgan educates on the difference between 2D clip art and physically based rendering (PBR) materials applied to 3D meshes. | |
| Controlling Generative AI via ControlNet and AdaptNet | 3 | 3 | 1 | 0 | Matt asks a sharp question on whether Gen AI output is just a draft needing human last-mile edit. Morgan agrees and explains ControlNet research aimed at controlling neural networks. | |
| StarCoder Research and Open-Source Science | 2 | 3 | 0 | 0 | Matt asks about StarCoder research learnings. Morgan highlights the shift from model creation to data curation, normalization, and hallucination prevention in AI engineering. | |
| AI Team Structure and Morgan McGuire's Career Journey | 1 | 1 | 0 | 0 | Matt asks about team structure and career journey. Morgan shares his dual background in academia and industry, explaining why Roblox is ideal for 3D Gen AI R&D. |
Statements from this episode (16)
McGuire: Roblox has around 70 million daily active users
“We have about seventy million daily active users”
McGuire: Roblox's fastest growing segment is 17-to-24-year-olds
“And our fastest growing segment is actually 17 to 24 year olds,”
McGuire: Roblox operates roughly 100,000 servers on custom infrastructure
“We have about a 100,000 servers that we maintain with our own infrastructure running our own software stack on those.”
McGuire: Roblox operates 17 edge data centers globally
“We have edge data centers around the world. I think we're up to 17 of those”
McGuire: Roblox employs around 3,000 total workforce members
“So the company as a whole is about 3000 people at this point.”
McGuire: Almost all Roblox employees are engineers or engineering-adjacent
“And it's heavily engineering dominated. So almost all of that is engineers or engineer associated professions.”
McGuire: Roblox is the only platform allowing instant, unapproved 3D publishing
“And so it's the only platform where you can do for three D what today you can do for micro blogging, like text or images, which is one person without some sort of app store approval, without becoming a publisher, or that's, you can go and create something that…”
Roblox releases new client software to all players every Thursday
“Every single Thursday, we ship a new client to every single player around the world, releasing new features new access, new security mechanisms.”
McGuire: AI will take decades before its full impact is felt
“The core technology has decades before the full impact is felt.”
McGuire: Every human interaction on Roblox is monitored by technology
“Every communication between humans is monitored, monitored by technology.”
McGuire: Roblox releases a new AI feature every couple of weeks
“Every, every couple of weeks we're releasing a new AI powered feature.”
McGuire: Roblox AI Code Assist has suggested 300M characters of code
“Three hundred million characters of code have been suggested by this tool.”
McGuire: Roblox released its code-generation LLM as fully open source
“And so we shared our LLM with the world that we had built as fully open source that anybody could build on it.”
McGuire: Controlling generative AI will be AI's main challenge for 2–3 years
“I think the big challenge that the field is facing probably in the next two or three years that we've been focusing on is how do you control generative AI?”
McGuire: AI advances stem from data preparation, not new models
“A lot of the learnings in AI are not on the technology side. It's not about creating a new AI model from scratch. It's about learning how to prepare data, how to augment the data so that you don't need quite as much of it, learning how to normalize or regulari…”