호랭교관
10.9K posts



The SEC will host a roundtable on September 17 to discuss moving towards 24-hour trading in the U.S. equity markets. Agenda, panelists, and registration info will be available soon. See additional info below ⬇️ sec.gov/newsroom/press…

Only 2.2% of U.S. households have a paid AI subscription We are still so early









Will do. 1. China doesn't have the access to the latest or as many GPUs. Nor do they have access to lithography tools and other precursors to make advanced ASICs. 2. With a fraction of the GPUs of American Labs, Chinese Labs are able to always be right behind American Labs. 3. One read of this is that Chinese Labs have incredible efficiency in training models. This would imply demand for AI buildout would be greatly reduced. 4. The problem with step 3 is that Chinese Models have no additional efficiencies in inference or when you query the LLM. Training and inference efficiencies come hand in hand. 5. Thus from step 4, it's not likely the Chinese Labs have incredible training efficiencies and no inference efficiencies to match. 6. Thus the culprit is that Chinese Lab's "training efficiencies" is basically distilling from American models aka using American models to train Chinese Models. 7. American models will still continue to push AI forward and buy alot of GPUs for training. 8. Whether you query (inference) a American Frontier Model or Chinese Open Source Model - you still need around the same amount of GPUs. The demand for inference GPUs is growing faster than training GPUs. Then there are those who say Chinese ASICs like Huawei Ascend are doing great. This however, is contradicted by the fact that if you go to China, their own open source models are heavily nerfed due to lack of compute. Also Chinese developers use VPNs to get access to Claude code. Chinese models are in fact good at coding, yet in China they don't have enough compute so the Chinese models get nerfed when served in China.





One of the things I learned at ICML this year is that Korea is only pretending to take AI seriously. The hype surrounding Korean AI startups is absurdly inflated, and the same goes for university researchers. Compared with China, we are honestly nothing. Korea needs to copy China’s playbook. We should launch something similar to the Thousand Talents Plan to bring back brilliant Koreans working overseas and aggressively poach top foreign AI talent. Otherwise, Korea will simply become an AI-dependent country.





















