Beijing Academy Unveils WuDao 2.0, World's Largest AI Model
The Beijing Academy of Artificial Intelligence introduces WuDao 2.0, a 1.75 trillion parameter AI model that surpasses Google's previous record. The Chinese-language system shows near human-level cognitive abilities in creative tasks.
The Beijing Academy of Artificial Intelligence (BAAI) unveils WuDao 2.0, a massive artificial intelligence model containing 1.75 trillion parameters. This new system breaks the previous world record of 1.6 trillion parameters set by Google's Switch Transformer, making it the largest AI pretraining model globally and the first trillion-scale model developed in China.
WuDao 2.0 achieves impressive results across nine benchmark tasks in the pre-trained model field. The system shows near human-level performance in creative and cognitive tasks such as writing poetry, generating text summaries, answering questions, and painting, pushing the boundaries of what AI accomplishes beyond the traditional Turing test.
Unlike most massive international models that focus on English, WuDao 2.0 builds on a Chinese language development environment. Backed by the Beijing Municipal Science and Technology Commission and the Haidian district government, this homegrown system represents a major step for China in the global race to develop super-large-scale intelligent models.