At the Beijing Zhiyuan Conference 2025, researchers unveiled a suite of advanced AI models that could reshape our approach to artificial general intelligence. The newly released "WuJie" series represents a significant leap in developing AI systems that interact seamlessly with the physical world.

The comprehensive system comprises four specialized models, each targeting different aspects of physical AGI development. The Emu3 multimodal model stands out with its ability to process and generate text, images, and video through a unified architecture. Unlike conventional systems that rely on diffusion models, Emu3 uses self-regressive techniques to create homogeneous token sequences across modalities.

Neuroscience takes center stage with JianWei Brainμ, which has achieved the unprecedented feat of tokenizing neural signals like fMRI and EEG data. With pre-training on over one million neural signal units, this model shows particular promise for brain-computer interfaces and medical diagnostics. Researchers compare its potential impact to AlphaFold's revolution in protein folding.

For robotics applications, RoboBrain 2.0 serves as an embodied intelligence platform, while OpenComplex2 models microscopic biological systems. Together, these components form a complete chain from perception to action in physical environments.

The institute has established partnerships with leading academic institutions including Peking University and Tsinghua University, along with neurotechnology company BrainCo. These collaborations aim to translate laboratory breakthroughs into real-world applications across scientific research and industry sectors.

Key Points

1. The WuJie series introduces four specialized AI models targeting physical AGI development
2. Emu3 unifies text, image, and video processing without diffusion architectures
3. JianWei Brainμ represents a major advance in neuroscience AI applications
4. Collaboration network includes top Chinese universities and tech firms
5. System covers full spectrum from perception to action in physical environments