A Western scholar's visit record to a Chinese AI laboratory: humility, openness, no discussion of philosophy, only aiming to train better models

Author: Florian Brand

Translation: Deep Tide TechFlow

Deep Tide Introduction: The background of this article is that SAIL (a media alliance that brings together top AI writers on Substack, including Nathan Lambert, Sebastian Raschka, ChinaTalk, and others) organized a visit group to Chinese AI laboratories. The author Florian joined the tour to the Dark Side of the Moon, Xiaomi, MiniMax, Zhipu, Meituan, Alibaba, Ant, Modao, Zero One All Things, Yushu, and more than a dozen companies, and wrote this impression.

Florian Brand is a doctoral student at Trier University in Germany and the German Research Center for Artificial Intelligence (DFKI). His research focuses on the application and evaluation of large language models.

Not exactly “famous,” but has some visibility in the open-source AI community. It’s also quite interesting to view China’s AI ecosystem from a foreign AI practitioner’s first-person perspective.

Main text:

Over the past roughly 10 days, I had the privilege of visiting Chinese AI laboratories with the SAIL team. As someone who visited China and the US for the first time in six months, I found the differences between the two places fascinating, but even more intriguing were the similarities.

What left the deepest impression on me was that all the AI researchers I met were very humble.

They highly praised other labs and colleagues. DeepSeek was mentioned frequently, probably because a few days before our visit they had just released a model, and people spoke about DeepSeek’s papers with genuine admiration.

Many researchers are close friends with each other, coming from the same university or sharing the same hometown. They openly discuss their work, and research results are published as papers months later.

This is one of the biggest differences from the Western AI scene. In the US, the atmosphere is often more zero-sum. Labs are cautious about positioning. Researchers think competitively, and some hold themselves in high regard. Leaders insult and attack each other in leaked memos. This difference can perhaps be explained by facts: leading US labs are closed-source, while many Chinese labs are open-source. Chinese labs are somewhat wary of ByteDance’s Doubao, which is the most used chatbot and is closed-source, giving it a significant advantage.

Meanwhile, the overall atmosphere is surprisingly similar to San Francisco. Researchers are extremely online, reading extensively on Twitter and Xiaohongshu, the latter becoming increasingly popular. They all use Claude Code or their own CLI to build the next model. Some monitor training runs during meetings, observing reward curves rising. They are thinking about scaling further and complain about insufficient computing power. They are frustrated with the current benchmark testing status.

Their main focus is on training better models. This is different from San Francisco, where researchers consider the political or philosophical impacts of AI. They don’t think about mass unemployment, permanent underclass, or whether their models have consciousness. They only want to train excellent models.

When they hear you used their models, their eyes light up. They are eager to fix all the flaws in the next generation of models. They work overnight to push model releases, and still show up at the office afterward.

Most of the researchers I met are very young, many in their early 20s or around 25. Some are undergraduates, but more commonly they are PhD students working in industry simultaneously. Their consensus is that industry is more interesting than academia right now, a view I strongly agree with because I have done the same. Labs place great importance on recruiting such talent, actively hiring interns and graduate students; this is not common in Western labs.

The researchers’ optimism extends to the general public, who seem more optimistic about technology and the prospects of AI and robots. During the trip, someone shared stories about their parents and grandparents using Doubao and DeepSeek for various tasks, including discussing mathematical theorems. This is quite different from the West, where ordinary people generally dislike AI.

Overall, this trip gave me a little understanding of this ecosystem. It’s impossible to grasp the culture of such a vast civilization in just a few days. As a staunch supporter of open AI ecosystems and open research, I am very optimistic about both futures and hope for extensive international collaboration in the future.

I want to thank all the incredible people I met at the Dark Side of the Moon, Xiaomi, MiniMax, Zhipu, Meituan, Alibaba, Ant Lingxi, Modao, Zero One All Things, Yushu, and elsewhere. Thank you for your time and warm hospitality. Also, thanks to SAIL for organizing this trip, and to all the writers and journalists involved. I am very grateful to have met so many outstanding and ambitious people in such a short time.