Dwarkesh Patel: evaluation

Transcript context

…You have to stack every layer, right? The beautiful thing about AI is that because it's growing so fast, every layer is being stressed to an incredible degree. Of course, China has been hacking ASML for over five years and ASML is kind of like, "Oh, it's fine." The Dutch government's really pissed off, but it's fine. They already have those files in my view. It's just a very difficult thing to build. The same applies for fab recipes. They can poach Taiwanese nationals. It’s not that difficult because TSMC employees do not make absurd amounts of money. You can just poach them and give them a much better life and they have. A lot of SMIC's employees are TSMC, Taiwanese nationals, especially a lot of the really good ones high up. You go up the next layers of the stack. Of course, there are tons of model secrets. But how many of those model secrets do you not already have and just haven't deployed or implemented or organized? That's the one thing I would say. China just clearly is still not scale-pilled in my view. If you could hire these people, it would probably be worth a lot to you because you're building a fab that's worth tens of billions of dollars. This talent knows a lot. How often do they get poached? Do they get poached by foreign adversaries or do they just get poached by other companies within the same industry but in the same country? Why doesn't that drive up their wages? It's because it's very compartmentalized. Back in the 2000s, before SMIC got big, it was actually much more open and more flat. After that, after Liang Mong Song and after all the Samsung issues and after SMIC's rise, you literally saw—…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.