A look at the Beijing Academy of Artificial Intelligence, a government-sponsored research lab, and its work on a new language model similar to GPT-3
Will Knight / Wired : Tweets: @lsenews , @willknight , and @wired Tweets: LSE / @lsenews : “Innovation by its very nature is sort of uncertain, and perhaps nowhere more so than in AI.” @LSEManagement's Professor Noam Yuchtman speaks to @wired's @willknight. https://www.wired.com/... Will Knight / @willknight : The Chinese government is funding a competitor to GPT-3, which it plans to use as a virtual administrative assistant. “This is a big project,” says Ji-Rong Wen, the scientist leading the effort. “It takes a lot of computing infrastructure and money.” https://www.wired.com/... @wired : China produces as many artificial intelligence researchers as the US, but it lags in key fields like machine learning. The government hopes to make up ground. https://www.wired.com/...
Context & Ripple Effects
The piece lands mid-arc in China's push to close its AI gap with the US. Six months after Beijing unveiled its plan to lead the world in AI, experts flagged that Washington had done little to respond (that 2018 assessment); BAAI's GPT-3-style project is the kind of state-funded answer that plan called for. Ji-Rong Wen frames it plainly: a big project needing heavy computing infrastructure and money — resources only a government sponsor can guarantee at that scale.
The choice of a virtual administrative assistant as the target application matters too: rather than chasing OpenAI's consumer buzz, the lab is aiming the model at government workflows, making it an early instance of what would become a broader pattern of non-English general-purpose language tools built outside the US.
First-order effects
- The Chinese government gains a domestically funded path to frontier-scale language modeling, with BAAI's compute and budget underwritten by the state rather than venture capital.
- The planned administrative-assistant deployment puts a GPT-3-class model directly inside government workflows, tying the lab's research agenda to state operations.
Second-order effects
- Startups in China, South Korea, Israel, and Germany follow with general-purpose language tools for their own languages, fragmenting what had been an English-centric field around OpenAI's GPT-3.
- Beijing's municipal government later formalizes the pipeline, backing ChatGPT-like startups and concentrating nearly a third of China's core AI companies in the capital.
Third-order effects
- State-sponsored labs become a standing feature of frontier AI: by 2025 China is running explicit industrial policy, including a pledged $8.5B for young AI startups, to help its companies close the gap with US rivals.
- The pattern also exposes its limits — insiders report Chinese firms still trailing US generative AI by roughly a year and leaning on open US models like Meta's LLaMA — suggesting state funding alone does not guarantee parity.
The trend: Frontier language models are becoming instruments of national industrial policy, with governments funding sovereign-scale labs whose output feeds both state administration and domestic startup ecosystems.