2025-09-12
nightside of eden
Afra Wang interviews a Chinese AI researcher on the Chinese AI landscape. Afterward, it’s interesting to listen to this interview by Steve Hsu with Kyle Chan on Chinese technological infrastructure and talent. The descriptions of the differences in “topology” between American and Chinese software seem almost like an inversion of the situation for manufacturing, where one side is highly interconnected and dynamic, and the other contains isolated islands of excellence within a framework of sluggish overplanning and shrouded with mystery and uncertainty. The other parallel I found particularly interesting is the analysis of weaknesses in the American ecosystem, with only “three companies, talent circulating within the same Silicon Valley bubble, HR departments practically operating as cartels”, which arises because American AI is talent-constrained, whereas China has a talent oversupply1 and therefore many companies to specialize to efficiently split the pie. It’s a compelling story that, as with every other sector where they surpass the US, if Chinese AI ever gets ahead, then it will continue to shoot far beyond. And since American AI has no moat and the Chinese ecosystem more or less knows what it needs to do, they clearly see the prize as ripe for the taking2.
Charles Yang also has some vignettes from the same China trip, which includes some salient examples of empty apartments, the corporate ecosystem, and the prevalence and focus on green energy. This is perhaps a good place to note that David Fishman has two recent posts, one defending involution in green energy production, and another arguing against the ghost-city narrative with the example of Ordos City. The criticisms he is arguing against ought not be taken too credulously, since they are made with selective information due to ignorance. On the other hand, it’s also probably not necessary to update too far in the other direction, since defenses such as these also often involve various levels of selectivity. What can you say, other than that there is an “invisible ecosystem of incentives and factors … difficult to grasp and understand at times”.
Scott Alexander reviews If Anyone Builds It, Everyone Dies3. There’s also an interview between Patio11 and Emmet Shear on what he sees as the solution to AI alignment, which involves identification with the greater system4. Coincidentally, there was a debate between Emmet and Nate at Manifest this year on the probability of doom. Anyway, it occurs to me that it would be ironic if China passes America in AI, and the noted disinterest in AGI and superintelligence in China results in a familiar end state where things are made in China but discussed in America.
Lionel Page on sports from a status perspective. Personally, I feel like watching sports should be lower status because it teaches you bad habits around status and group signalling, but playing sports should be higher status because it teaches agency and discipline. Unfortunately, it’s not clear to me how these two can be meaningfully combined.
Andy McKenzie on short and long-term goals in the field of brain preservation.
Nuño Sempere prediction markets linkthread.
Although it’s a question to what extent this talent limitation in America is actually due to their HR and hiring pipelines being terrible; likewise to what extent China is really oversupplied or only relatively so due to being hardware constrained is also unclear. Nevertheless, for now, the effect is that you can probably treat both statements as being true.
Actually, Michelle Ma has an interesting argument that China is already “winning” AI because they have a better ecosystem and faster diffusion through society. I feel like even if this is true, it still assumes that China can and will catch up in hardware (which is why Deepseek, as the frontier AI competitor, is using Huawei chips to their short-term detriment), and then to what extent the party harvesting the gains will end up hindering or constraining the path forward.