Fei-Fei Li, recognized as the 'godmother of AI,' has launched a new billion-dollar startup called World Labs, focusing on enhancing AI's understanding of three-dimensional environments.

The startup has successfully secured $230 million in funding, achieving a valuation of approximately $1 billion, despite some skepticism within the AI community regarding the pace of advancements.

World Labs aims to elevate AI capabilities from 2D images to immersive 3D worlds, emphasizing the development of a foundation model that facilitates spatial reasoning.

Li and her team, which includes experts from Amazon, Meta's Reality Labs, and Google, believe the next decade will focus on generating content that integrates AI into real-world spaces.

As the field of AI continues to evolve, the advancements from World Labs are expected to have a profound impact, comparable to the innovations seen with language models.

Li has pointed out that current generative AI models lack a comprehensive understanding of 3D world structures, which is crucial for producing accurate AI outputs.

The models being developed will use a transformer-based architecture similar to OpenAI's ChatGPT, but will incorporate additional elements to better understand spatial dynamics.

Successful development of these large world models could significantly impact technologies such as autonomous vehicles, automated factories, and humanoid robots.

World Labs is based in Stanford, California, and its founding team includes notable experts like Justin Johnson, Ben Mildenhall, and Christoph Lassner, who have made significant contributions in AI and graphics.

The funding round for World Labs was co-led by prominent investors including Andreessen Horowitz, NEA, and Radical Ventures, with additional backing from notable figures such as Marc Benioff and Reid Hoffman.

Building a comprehensive world model that integrates spatial intelligence presents significant challenges in data, graphics, and AI, but the potential rewards are substantial.

The startup's mission includes creating 'large world models' that utilize both synthetic and real-world data for training, aiming to enhance applications in augmented reality and robotics.