Seems very sinophobic. Deepseek and Manus have shown that China is legitimately an innovation powerhouse in AI but this article makes it sound like they will just keep falling behind without stealing.

That whole section seems to be pretty directly based on DeepSeek's "very impressive work" with R1 being simultaneously very impressive, and several months behind OpenAI. (They more or less say as much in footnote 36.) They blame this on US chip controls just barely holding China back from the cutting edge by a few months. I wouldn't call that a knock on Chinese innovation.

Don’t assume that because the article depicts this competition between the US and China, that the authors actually want China to fail. Consider the authors and the audience.

The work is written by western AI safety proponents, who often need to argue with important people who say we need to accelerate AI to “win against China” and don’t want us to be slowed down by worrying about safety.

From that perspective, there is value in exploring the scenario: ok, if we accept that we need to compete with China, what would that look like? Is accelerating always the right move? The article, by telling a narrative where slowing down to be careful with alignment helps the US win, tries to convince that crowd to care about alignment.

Perhaps, people in China can make the same case about how alignment will help China win against US.

Yes, it's extremely sinophobic and entirely too dismissive of China. It's pretty clear what the author's political leanings are, by what they mention and by what they do not.

Stealing model weights isn't even particularly useful long-term, it's the training + data generation recipes that have value.

Don't confuse innovation with optimisation.

How so? Spoiler: US dooms mankind, China is the saviour in the two endings.

In both endings it's saying that because compute becomes the bottleneck, and US has far more chips. Isn't it?