The video discusses DeepSeek’s parent company, Highflyer Quantitative Fund, founded by Leang win Fang, who turned down an offer from DJI’s founder to pursue his dream of building AI technology. Highflyer initially aimed to use AI to build trading models to generate profit for AI research. Although Highflyer faced profitability challenges, it developed significant technology in the AI large models field.
The speaker explains that the current bottleneck in generative AI development is not computing power but the depletion of training data. They argue that AI large models are still programs that cannot achieve AGI (Artificial General Intelligence) or self-improve. The speaker notes that while AI can replace humans in tasks like objective information retrieval and basic programming, it is unlikely to replace them in complex fields requiring on-the-spot decision-making, such as medical surgeries or artistic performances.
The video highlights DeepSeek’s ultimate goal, which is not to profit from API calls but to optimize the underlying computing architecture and provide cost-effective computing interfaces. They aim to offer free DeepSeek foundational large models, allowing more companies to train their own models on Highflyer’s computing infrastructure, thus positioning themselves as computing infrastructure providers. This approach is seen as a way to address the cost concerns of smaller companies and stimulate the AI ecosystem.
The speaker acknowledges DeepSeek’s work as a significant innovation that has broken the technological monopoly of U.S. high-tech companies in the short term, providing Chinese small and medium-sized enterprises with affordable and usable AI large models. This is viewed as a development that could widen China’s advantage over the U.S. in the hardware field and accelerate the U.S.'s failure in the tech war.