智源发布“悟界”系列大模型，宣布围绕物理AGI进行布局

Group 1 - The core viewpoint of the news is the launch of the "Wujie" large model series by the Beijing Zhiyuan Artificial Intelligence Research Institute, focusing on advancements in physical AGI and breaking the boundaries between virtual and real worlds [1] - The "Wujie" series includes four models: Emu3, Brainμ, RoboBrain 2.0, and OpenComplex2, each targeting different aspects of multi-modal learning and applications in neuroscience [1] - Emu3, set to be released in October 2024, utilizes a novel token prediction paradigm for unified multi-modal learning, allowing for the encoding of images/videos into discrete symbol sequences that are isomorphic to text [1] Group 2 - Brainμ is built on the Emu3 architecture and tokenizes brain signals from various neuroscience modalities, enabling multi-directional mapping between brain signals and other modalities like text and images [2] - The model has been pre-trained on over 1 million units of neural signals and aims to support a wide range of applications in neuroscience, from basic research to clinical studies and brain-computer interfaces [2] - Collaborations with leading neuroscience laboratories and research teams in China, including institutions like Tsinghua University and Peking University, are being established to expand the scientific and industrial applications of Brainμ [2]