Roblox launches AI tech that generates functioning models with natural language

By Zaheer Kachwala

Feb 4 (Reuters) – Roblox launched an artificial intelligence technology on Wednesday that can generate fully functioning in-game models using natural language prompts, as the company aggressively pursues machine ‍learning to attract more developers to its videogame platform.

The technology, dubbed 4D creation, released into beta, and builds upon the company’s earlier model which could generate static three dimensional objects. The new model allows users to generate a vehicle, interact with its doors and drive it with accurate physics ‌mechanics.

The move serves to lower the entry barrier ‌to Roblox’s developer community, which is user-led and responsible for the games played on the platform, as the firm looks to spur the creation of new titles and boost its already sprawling user base. The company ​had more than 150 million average daily active users at the end of the third quarter.

“You want to create an object, that ‍is one path, and for the ​artist, the visuals might be easy, the coding might ​be hard. The other is, you’re a game developer who finds the ‍visuals harder, coding easier. So we’re trying to bring all of them together, and our highest goal would be that a player can create inside a game,” Roblox’s senior vice president of engineering for foundation AI and infrastructure, Anupam Singh told Reuters.

Roblox has been ‍investing heavily in expanding server capacity to accommodate the sharp growth in its player base, while also putting funds towards developing AI models and ‍safety – an issue that ‍has come under heavy scrutiny from American state ​and international governments.

4D creation is a part of ​Roblox’s larger ⁠push to develop AI world models, which can ‌understand the rules and dynamics of an environment so that it can predict and generate future gameplay.

The announcement comes a week after Alphabet’s Google launched an AI model allowing users to simulate and generate a real-world environment through prompts or uploaded images.

(Reporting by Zaheer Kachwala in Bengaluru; Editing by ⁠Krishna Chandra Eluri)