pov: you're an agi and this life is your test to be let out of the box

ive been rapidly updating the probability that this has in fact occurred, that the true marginal productivity in the US is very negative at almost every human capital margin. iow actual GDP is much less than what our capital endowment would return in pure maintenance mode

i'm sad to say that the golden age of inner beauty appreciation has gotten off to an inauspicious start, perhaps due to supply chain issues

"leech" and "leach" are an interesting pair of verbs in that they both have a sense of something being sucked out of a substrate but one is active and the other passive

i dont know if i believe the full implications of this. but its clear to me that the consequences of the Big Sorts that have shaken up every society on earth are only starting to make their upheavals felt https://t.co/Ka3zM8I6qk

you are selecting for those who don't value (on a relative basis) the roots that are the major part of their inheritance. you are selecting for those who will chase status to the ends of the earth. and who's to say what happens when that cohort leaves your country

when that funnel is a brain drain, you are also selecting for individuals who are most pessimistic about the home country's future. you are selecting for the ones who perceive the biggest gap between their talents and the reward they feel entitled to receive from society

if you build a society sized funnel to select for the "smartest" or most "hardworking", you will invariably select for other things. the individuals you select will also possess traits that are orthogonal to your goal but tightly coupled to the nature of the funnel

it's tempting here to think, imagine how it would be even better if all that human capital had stayed put in the country. but i wonder if the lessons hiding beneath the surface is something much more revelatory and discomfiting

within the diaspora, that china (esp. under the CCP) could reach parity with US was seen as preposterous. the more dismissive variants of this opinion were certainly driven by the remembrance of the brain drain, the sense that the best and brightest had already left the country

china's success in catch up growth is all the more striking when you consider the cream-skimming brain drain that was going on at an industrial scale behind the scenes for the large part of three decades

and even within this group, the selection effect was very strong. 70% of tsinghua going to the just the US sounds crazy, but it probably was directionally true. it was only 11% in 2018 and 7% in 2022 https://t.co/TQ7VrTn3u6

even one more wrinkle is that emigrants as a whole are a quite selective group. the further back you go the more true this is. hard to get reliable figures, but i have heard in the 90s, >30% of STEM undergrads at top 20 colleges would pursue international graduate studies

there's so much latent psychological energy here. china is so vast, individual life histories are fractally vast, even the counterpoints ("any price to avoid gaokao", "ackshually china is on the brink of economic collapse", freedom > security, etc.) are mutually contradictory

if you're even the tiniest bit curious about this, read this thread/replies. runs the gamut between QingBei emigrants comparing themselves to billionaires and party bigwigs and under-the-table restaurant workers who see the middle class path not taken https://t.co/C2jY5dvISK

the "pots not people" meme was used to argue that step changes in the archeological record did not necessarily entail genocidal warfare. turns out maybe it doesnt, maybe sometimes populations dwindle away because they just choose for no particular reason to have a tfr of 0.7

i remember spirited discussion circa 2018 about what Gen Alpha's social media would turn out to be (i.e. Facebook for millennials and Insta for zoomers), this was not an option anyone considered

if you wanted to be harsh, you might actually argue that in many academic disciplines, the selection has actually been for obtuseness, arbitrariness, etc to give more power to gatekeepers. for many domains you might have to pretty much start from scratch https://t.co/jDcSvBfrfc

more importantly, the specialized vocabulary and grammar for most domains has not undergone any selection process to be tractable in this sense. considering the aggregate mind-hours that went into math's "evolution," the creation of any robust/tractable DSL is a huge undertaking

so you wouldn't expect for example humor to work the same way. or making a movie. there is no "domain specific language" for these tasks. at best there is rudimentary jargon, but it has not been consciously designed to be tractable for llm-scaling

this is what allows for self-play and scaled searches to yield such striking gains in capabilities. so it's not an accident that that these are the areas where llms seem to obey scaling laws the best, they are in a sense the "perfect target" to shoot toward