In an interview with Nikkei Asia on April 21, Microsoft Vice Chairman and President Brad Smith highlighted that a Chinese institution is poised to be among the frontrunners driving the generative AI revolution.
Smith identified three entities at the forefront of the innovation: "One is OpenAI working with Microsoft, the second is DeepMind, which is part of Google, and the third is the Beijing Academy of Artificial Intelligence (BAAI)."
"Who's ahead and who's behind can change a bit from one part of the year to another, but one thing has been absolutely constant: the gap is almost always measured in months, not years. And so there's this enormously competitive race to innovate," Smith said.
Here's what you need to know about BAAI:
BAAI, also known as Zhiyuan AI Institute, was established in 2018 with support from the Ministry of Science and Technology, the Beijing government, technology companies (including Xiaomi, Baidu, Megvii, Meituan, etc.), and leading universities to develop large-scale pre-trained models based primarily on Chinese. Currently, English development dominates large-scale models internationally.
Bringing together more than 100 AI experts and scholars from China and around the world, the organization has been working hard to build its own pre-trained model and has made notable strides.
In March 2021, it released Wudao 1.0, China's first homegrown super-scale intelligent model system. Three months later, Wudao 2.0, China's first trillion-scale model and the world's largest artificial intelligence modeling system at the time, came out.
Boasting 1.75 trillion parameters, Wudao 2.0 surpassed the previous record of 1.6 trillion parameters held by the Google Switch Transformer. Both models are recognized in the AI industry as "super-large-scale pre-trained models."
According to reports, Wudao 2.0 is close to surpassing the Turing test in multiple tasks such as poetry and couplet creation, text summarization, question answering, and painting.
Additionally, this language model either reached or surpassed state-of-the-art (SOTA) levels on nine benchmarks, as reported by BAAI.
The AI arms race, ignited by OpenAI's ChatGPT-3.5, pushes BAAI to go further as well. Shouldering the mission to drive China's AI for more "phenomenal" applications, it rolled out FlagOpen, an open-source ecosystem that includes large model algorithms and acceleration tools, data processing tools, large model evaluation, etc., in March this year.
According to BAAI, FlagOpen has already joined forces with more than 30 research institutes and universities. It has also formed alliances with the country's leading tech companies to improve the ecosystem.
When Wudao 2.0 is released in 2021, we wrote a story about BAAI, check it out.