Home / Technology / AI's Next Frontier: World Models Emerge
AI's Next Frontier: World Models Emerge
23 Apr
Summary
- World models understand 3D space and physics, unlike chatbots.
- Nvidia, Alibaba, and Tencent released world models recently.
- Pokemon Go data aids robot delivery and geospatial modeling.
- DoorDash and Instacart are collecting real-world data for AI.
Attention is shifting from large language models like ChatGPT to 'world models,' a frontier in artificial intelligence that imbues systems with an understanding of three-dimensional space and physics. This capability is crucial for advancements in robotics, autonomous vehicles, and augmented reality, areas where current chatbots fall short due to their lack of grounding in physical experience. In the past two weeks, leading technology firms including Nvidia, Alibaba, and Tencent have unveiled their own world models, indicating a significant push in this domain.
Unlike language models that mimic understanding based on text patterns, world models aim to replicate human-like comprehension of object permanence and physical interactions. Companies are leveraging diverse data sources to train these models. For instance, Niantic, the creator of Pokemon Go, is utilizing its extensive geospatial data to build a Large Geospatial Model, while DoorDash and Instacart are collecting real-world task data from gig workers to train robotics firms. This data collection is vital for developing AI that can interact meaningfully with the physical environment.
The development of world models represents a critical next step towards achieving more human-like artificial intelligence, pursued by major players like OpenAI, Anthropic, and Google. Startups like World Labs, founded by AI pioneer Fei-Fei Li, are also making significant strides, raising substantial funding to develop their virtual world generation models. The future of AI may heavily rely on synthetic data, especially for rich 3D content, potentially leading to new business models in gaming, VR, and robotics training.
China is poised to play a substantial role in the world of physical AI, given its manufacturing prowess and significant output of humanoid robots. As world models evolve with diverse approaches and potentially more open licensing compared to language models, companies in China could emerge as leaders in shaping the next decade of physical AI. The landscape is dynamic, with ongoing debates about commercialization strategies and the maturation of use cases for this powerful technology.
