2024 年

This report discusses the emerging field of Agent AI, which aims to create artificial intelligence systems that can perceive and act in various domains and applications. Agent AI training has demonstrated the ability to understand multi-modal information in the physical world and can leverage generative AI and cross-reality data to apply to both physical and virtual worlds. The report presents an overview of an Agent AI system that can serve as a route towards Artificial General Intelligence using an agent paradigm. The authors suggest that multi-modal AI systems will become ubiquitous in our daily lives and that embodied agents within physical and virtual environments can facilitate the ability of models to process and interpret information more effectively.