I agree that US staff can probably move abroad (including to join foreign companies). My claim is that such a move can’t include IP, money, or sufficient numbers of chips. Given the importance of these three things, I still conclude that a move would probably destroy the lead of the AI company (or prevent the foreign company from catching up even with new staff).
I also think a clear mass exodus would be more politically actionable than the trickle of people to Mistral.
I think this is fair evidence and provides, to me, a weak update towards the administration being more permissive than I would otherwise think.
However, for the reasons I mention in the article, I think chip exports are somewhat disanalogous.
The current administration has been more lenient about chip exports, but that latitude is unlikely to extend to a frontier company relocating. The cases are meaningfully different:
Chip exports have US-based beneficiaries (NVIDIA, AMD, and other companies in the chip industry) whose profits, jobs, and tax revenue remain domestic. Allowing a frontier company to leave would mean ceding some profit to another jurisdiction for little economic benefit.
Lenient chip exports can be justified on the grounds of increasing diffusion and global reliance on US companies. Allowing a major company to move its headquarters overseas has no analogous justification.
Chip access can be restricted if AI becomes more strategically important, but it is difficult to force a company to return to the US once it has already left.
I also, more broadly, think that letting China use US-made chips is less obviously destructive to the US lead in AI than letting a US AI company leave. So, I think its pretty likely the president would block a company’s departure even if export controls aren’t supported by the president.
I agree that late in the singularity, AI workflows may be so different from humans’ that we learn very little from extrapolating from human returns to software R&D, but I expect that early in the takeoff, AIs may look significantly more like large-scale human labor (especially if they are still largely managed by humans). If existing returns are insufficient for a takeoff, that should update us against a software-only takeoff in general because it makes initial compounding less likely.
I also expect to observe relevant returns in the near future as AIs increasingly automate AI R&D (many of the points in the above post would include this). Early automation may give us some evidence on dynamics mid-takeoff.
I’m not sure how “being in the sphere of another power” will affect things going forward. For example, if a company moves to Canada and produces near-human AGI, the power dynamics between Canada and the US might shift substantially to make Canada more independent.
I think this is a reason the US wouldn’t be comfortable with companies going to allied nations. It also might mean that many nations are reasonable options for AI companies as long as those companies can get financing. Saudi Arabia might be one example.