AGI市场社会主义：中美两种路径的对比分析

> You divide this 5% over all US households Tbh, I think it's still garbage. Incentivizes the creation of an all-important Guild with a ton of leverage, and minority shareholders would be pretty powerless. Nevertheless it does seem like the insistence of the Nation State to perpetuate itself leads to some form of AGI Market Socialism by default. The question is in details. I see two coherent attractors: 1) apparent American one. Private AI companies compete, one of them gets closer to "AGI", starves others of revenue (which is, past some point, not new value creation but redistribution from non-AGI businesses, because autonomy in extant domains just improves vastly faster than people come up with new use cases), starts to rent their compute, accelerating the flywheel, develops its own chips, secures the supply chain, buys out legacy companies for data (those which haven't already *paid it* to process their data yet, as Karp argues), bankrupts everyone else and devours most of the economy. For all the blather about empowerment, this seems to me like the explicit plan of both major "labs" and a few aspiring ones, the promise underpinning the value of their equity that so many high IQ doers and builders on this website assume to be their ticket out of the "permanent underclass". Call me Russian but I don't think the American state would suffer such an entity to exist. So this champion, should he get his wish, next gets some form of nationalization: either literal commandeering if tensions are high, or gradual acquisition (following the Intel case and proposed OpenAI 5% deal), or forced equity buyout and redistribution to the citizenry (again, either directly a la voucher auction or more likely via some kind of a fund), or just ad hoc over-the-top taxation to feed UBI. The staff will still make disproportionate wealth on it, but they won't form AGI Aristocracy. 2) My headcanon for how the Chinese state might approach this, based on Liang Wenfeng's ideas. A lab leading in research and training (or realistically a few, because they don't like monopolies, but they'd form a unified ecosystem at this rate) gets state funding, below market rate credit, provincial incentives, all the Industrial Policy package for semiconductors; and in return keeps most of its work open (either MIT grade open weights+code as it is now, or open-to-qualified-customers past some level), making modest margins on its own inference compute. Their artifacts are served at vastly larger scale by hyperscalers and locally by enterprises. Hyperscalers are a new kind of utility, with thin margins; possibly they even integrate with the state power grid. There's a thick layer of private data pipelines and customization services. There is, in short, no Guild or House, but a legitimate AGI Lab funded to boost overall societal efficiency and velocity, and massively decentralized value capture by smaller agents, who are taxed by the state to maintain the current status quo of power in a world of high automation. This is the steelman of the "China will do applications" vision, which can hold well into the AGI era.

I happen to believe that #2 is by default smarter, more dignified, more likely aligned with the broad interests of extant humanity and the long-term flourishing of our species, and less degenerate. On the other hand, there's a concern about the development velocity penalty from data and compute fragmentation, both in terms of capex reinvestment and generalization downstream (could be ameliorated with some data sharing agreement embedded into compute procurement? Just spitballing). It remains to be seen if China will choose this path, but it currently has the institutional and political prerequisites for that and its (ironically, somewhat less accurate) theory of AI makes it more palatable for the decisionmakers. The US… would need to change course, before it reaches the fundamental dialectical contradiction of its economic model. But perhaps that contradiction will give birth to the all-conquering synthesis. At least that's the plan.