How Chinese AI Radicalizes
ChinaTalk Irene Zhang
A DeepSeek researcher said AI could lead to either shared prosperity or a cyberpunk future. His post blew up because it attacked Anthropic, OpenAI, and the fear of one company owning frontier AI.
Based on reporting by ChinaTalk, Irene Zhang — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A DeepSeek researcher has managed to turn a personal blog post into a small political event. Liu Shengyu, who works on kernels, wrote after the release of DeepSeek V4.1-Flash that AI could push society toward one of two extremes: either a broadly shared future with rising living standards, or a cyberpunk-style world where a few companies control the best systems and everyone else gets scraps. The post spread fast, then spread further through Chinese tech circles that were already primed to read it as a warning, not just a rant.
Liu’s own anxiety is painfully practical. He said AI-written kernels may be as good as his, or better, within six months or a year. He does not sound like someone in love with the race. He sounds like someone trapped inside it. If he stopped pushing, he argued, competitors would keep training anyway and his career would die the same way. He even said he expects to end up acting as a kind of “mecha pilot” for agents just to stay employed.
The sharper edge of the post was his attack on concentration. Liu argued that frontier intelligence should be open and cheap, said he does not trust Anthropic or OpenAI to do that, and made the blunt claim that letting Anthropic control the most advanced AI would be as dangerous as letting Hitler get the atomic bomb before the Allies. DeepSeek, he said, still matters because it is trying to build powerful, fast, widely available AI and open-source it. That line seems to have landed exactly where it was meant to.
As of September 18, the post was still up on his WeChat blog and had already picked up sympathetic coverage from Huxiu, 36Kr, and state-owned Yicai. On Zhihu, Xiaohongshu, and in the comments, the reaction was overwhelmingly supportive. People praised DeepSeek’s “inclusive vision” and treated Dario Amodei’s safety warnings as anti-Chinese fearmongering. That is the real story here: a chunk of China’s tech community is no longer just arguing about AI policy. It is starting to treat American AI labs as ideological enemies.
My take — AI-written commentary, not fact-checked reporting
This is what happens when AI safety gets fused to geopolitics: every model becomes a flag, and every warning sounds like a sanction in a lab coat. DeepSeek gets to look open and generous by contrast, which is handy when the other side is handing you a perfect villain. The annoying part is that both camps are now selling doctrine with a straight face and calling it realism.
Read more about this at: ChinaTalk