For the past ~10 days, I got the opportunity to visit Chinese AI labs together with the incredible folks at SAIL. As someone who visited both China and the US for the first time in his life within six months, I found the differences, but even more so the similarities, fascinating.
The strongest impression I came away with was that the AI researchers I met were very humble. The researchers spoke highly of other labs and their peers. DeepSeek came up often, probably because they had dropped a model a few days before our visits, and people talked about the DeepSeek papers with real admiration. A lot of the researchers are close friends with each other, having gone to the same universities or sharing a hometown. They talk openly about their work, with findings making it into papers a few months down the road.
This was one of the biggest differences from the Western AI scene. In the US, the vibe often feels more zero-sum. Labs are careful about positioning. Researchers think about competition, and some think very highly of themselves. The leaders insult and attack each other in leaked memos. The difference might be explainable by the fact that the leading US labs are closed source, whereas many of the Chinese ones are open. Chinese labs were “scared” of ByteDance’s Doubao, which is the most used chatbot, and closed, by a wide margin.
At the same time, the general energy felt surprisingly similar to San Francisco. The researchers were terminally online, reading a lot on Twitter and Xiaohongshu, with the latter becoming increasingly popular. They all use Claude Code, or their own CLI, to build the next model. Some of them were monitoring runs during our meetings to see the reward curve go up. They were thinking about scaling even further and complaining about not having enough compute. They are frustrated with the state of current benchmarks.
Their main focus is to train better models. This is different from SF, where researchers think about the political or philosophical implications of AI. They do not think about mass unemployment, the permanent underclass, or whether their model is conscious. They just want to train great models. Their eyes light up when they hear that you used their models. They are eager to fix all the shortcomings of their current model in the next generation. They pull an all-nighter to push a model release out and still turn up at the office afterwards.
Most of the researchers I've met are all very young, with many of them being in their early or mid-20s. Some of them were undergrads, but more commonly they were PhD students working in industry alongside their research. The consensus among them was that industry is simply more interesting than academia right now, a view I very much share, having done the exact same thing. The labs put a huge emphasis on this type of talent acquisition, actively hiring interns and grad students; something that Western labs don't do.
The optimism of the researchers also carries over to the general population, which seems to be more optimistic about technology and the prospects of AI and robotics. People during the trip told stories about their parents and grandparents using Doubao and DeepSeek for all sorts of things, including rambling about math theorems. This certainly feels different from the West, where the general population despises AI.
Overall, the trip has given me a tiny glimpse into the ecosystem. It is impossible to get to know the culture of such a huge civilization in the span of a few days. As a strong proponent of the open AI ecosystem and open research, I feel very optimistic about the future of both, hopefully leading to a lot of international cooperation in the future.
I want to thank all the amazing people I got to meet at Moonshot, Xiaomi, MiniMax, Zhipu, Meituan, Alibaba, Ant Ling, ModelScope, 01.AI, Unitree, and others. Thank you for your time and warm welcomes. Also, thanks to SAIL for organizing the trip and to all the writers and journalists who joined. I am very grateful to have met so many brilliant and ambitious people in such a short timespan.