2025-09-20
the end of eternity
There’s a piece by Scott Kennedy in CSIS which describes how the Chinese government is attempting to deal with involution in manufacturing, which has many interesting implications: for example, now that solar panel costs have fallen so dramatically beyond the expectations of most experts, most people now make projections assuming that it will continue to do so. Anyway, David Fishman has some interesting commentary on this, defining involution as excessive competition, and noting what this means in terms of various causes or outcomes depends on the specific sector and circumstances. In the case of manufacturing, the obvious solution is to centralize production under cartels, but the government does not do this because they understand, from the historical underperformance of state-owned enterprises, that this cure would be worse than the disease. Related, there’s an interesting article by Snowden Todd on wages in Taiwan and Korea, which notes that excessive competition in one area often is the result of insufficient competition in adjacent ones.
Epoch AI has a piece on the implications of increasing LLM inference context windows. Particularly interesting is the last section, which notes that increasing the amount of time it takes to fully saturate the context, also means that experiments take longer to run, which extends the length of the feedback loop. Related, there is an interesting tweet by Dustin Tran which notes that what AI researchers really want in our current compute constrained environment, is for their company to increase compute per capita, which means more experiments they themselves can personally run. In terms of researcher incentives, moving the bulk of time spent on experiments from training to inference might be a good thing, in that it allows more researchers per compute, which maintains a more talent-constrained environment.
Byrne Hobart review of IABIED, and well as a Tyler Cowen review roundup.
Awais Aftab reviews Elusive Cures, on the lack of success in the field of molecular neuroscience and potential avenues for rectifying them.
Lilian Wang Selonick reviews Vanishing World in the Metropolitan Review. I felt it was quite hard to suspend disbelief that a culture driven by declining birth rates would stigmatize spousal intimacy, but then I remembered the recent drama involving Orchid Health where the CEO brought up the idea that natural conception would be soon regarded as irresponsible1. Somewhat related, it’s both interesting and unfortunate that for the most part the state of the art in public discourse, like this piece by Brett Devereaux, is still just people realizing that The Population Bomb is not what is actually happening.
Nathan Goldwag on Thaddeus Stevens and the ebb and flow of social and political movements.
