比Open Ai 最新的模型差一點也是了不起的成就，微軟、穀歌、Meta都公開稱讚DeepSeek，目前公開的模型是包含全部Weight以及代碼的真模型，也就是說任何人都可以下載下來在自己的機器上跑，也能表現的像ChatGPT一樣好，前提是你的電腦要十分強大，不需要再訓練，因為已經是訓練好的，隻是inference，6000B的參數可不是一般的電腦的能跑得動的。
1 DeepSeek model is still built base on Nvidia H800 GPU chip, Google transformer and OpenAI's previous ChatGPT model development, not all self-invented.
2 DeepSeek is very slow, 8-10s waiting time, comparing to American's leading model.
3 DeepSeek is a crippled model, good at math and coding, but not good at humanity, which is intrinsically due to Chinese gorvenment restriction. Any model from any Chinese local company will have this limitation, and there is no solution in any time soon As long as CCP still run the country.
4 Chasing for efficiency and low price has been build-in gene for all chinese companies throug last 30 years history in China. Otherwise they can't survive. Same thing in AI industry. It is not surprising that Deepseek started to find a cheaper way from the begining.
5 The success of DeepSeek couldn't draw a conclusion that it is a mistake for US to restrict advanced semiconductor chips getting into China, either that the restriction enhanced the innovation from Chinese companies. China has 1.4b people and many talents there, and they are the 2nd biggest economy, naturally there will be some talent people have the passion of innovation. On the other side, if they have more advanced GPU, they definitely will be able to develop even better model.
6 The success of DeepSeek doesn't mean the demand of GPU will be less, on the contrary, the demand will be more. AI demand more advanced chips other than low end chips. The semiconductor chips (GPU, CPU, NPU, TPU, ASIC and all other supporting chips, like HBM DRAM, power chips, connection chips etc.) will thrive in the coming decade for sure.
7 AI has a long industry chain,from hardware chips and connections, networks, to deep learning, model training algorithm to all kinds of applications. Overall it is obvious that US is leader now. China is at the 2nd position, but far ahead of any other rest nation.
8 The success of DeepSeek is good news to the industry and all consumers. It tells even small company with limited resource still has the opportunity to invent great technology and influence the world. AI is not only the game of the big Techs.
9 It encourages Chinese and chinese companies, so does the other nations. The hot wave of AI will continue to attract money and people to come-in all overall the world.
10 AI is still far in its early stage. It will have at least 10-20 years fast development time period. It will disrupt human society in all aspects, there is no limit for the 4th revolution in human history, the greatest ever. It is too early to say who is the biggest winner.
=====================================================
太宇講的沒錯，不但效率而且良率也非常高可以比肩台灣廠而且有時還超前。前段時間對此報道很多，因這的確有出乎人們的想象。美國方麵的做法是招募了大批的退伍軍人。當然這種事大外宣是不願讓人知道的“隻影”想知道可以自己上網去查（如果你真的不是在牆內）如果不想知道建議你洗洗睡吧。