SEPTEMBER 28, ACCORDING TO THE INTERFACE NEWS, SEPTEMBER 27, 2010 CEO Kai-Fu LeeON THE OCCASION OF THE 20TH ANNIVERSARY OF THE YANGTZE CEO'S CELEBRATION OF THE SCHOOL DAY, THE CHINESE GOVERNMENT ANNOUNCED THAT IT WOULD NOT BE ABLE TO FIND A SOLUTION TO THE PROBLEMDeepSeek To China AI The core contribution to development lies in the promotion ofOpen SourceEcological formation. “If, after 10 years, we recall how DeepSeek left China not behind the United States, the answer is not its technological capability per se, but it brought about the opening-up era of China.”
Li recalled that, since DeepSeek's opening, a number of domestic enterprises have developed large-scale open-source models that create a “open-source and faster-than-mix” virtuous competition. In his view, the open source model is highly compatible with the learning characteristics of Chinese enterprisesIT IS HOPED TO HELP CHINA CLOSE THE GAP WITH THE UNITED STATES IN AI.
1AI NOTED THAT LEE HAD BEEN ABLE TO SUPPORT THE MODEL ON SEVERAL PREVIOUS OCCASIONS. IN MARCH OF THIS YEAR, HE SAID ON THE DAY OF ARTIFICIAL INTELLIGENCE OF THE NAKAWAN FORUMDeepSeek's success proves that closing the source is a dead end, and opening the source will be much more advanced.
In his view, as the Scaling Law law of AI moved from the pre-training to the reasoning end, the AI model and AI application would usher in a new acceleration this year, and an industry slogan this year would be “Make AI Work” (to make AI useful). He stated that the direct commercial value of super-pre-training models was decreasing with the slowdown of Scaling Law。
1AI WAS INFORMED THAT THE FOUR REASONS GIVEN BY LEE KAI-CHUL WERE:The lack of data leads to the end of traditional pre-training; hyper-governed GPUs are less efficient, fault-tolerant, etc., leading to lower marginal effects; super-moderate pre-training models are expensive and slow; new reasoning Scaling Law will bring higher returns. The value of mega-pre-training models is shifting and will be further reflected in the role of “teacher models” and its essence will move further into the infrastructure of the larger model age。THE OLD LEARNING PARADIGM IS HUMAN EDUCATION AI, THE NEW LEARNING PARADIGM IS AI TEACHING AI.