Ep36:From Oligopoly to Inclusivity, the DeepSeek‘s AIEconomy

This episode discusses DeepSeek as a representative of Chinese AI innovation, which is changing AI economics and the scaling law, and will profoundly impact the global AI industry. DeepSeek's Rapid Growth: DeepSeek-R1, a chatbot launched in January 2025, rapidly gained users, reaching 125 million within two weeks, surpassing ChatGPT's growth rate. By early February 2025, DeepSeek's daily active users (DAU) exceeded 40 million, approaching ChatGPT's DAU. Impact on the AI Industry: DeepSeek's rise has had a notable impact. On January 27, 2025, it topped the U.S. App download chart, coinciding with a significant drop in NVIDIA's stock value. AI Laws: Sam Altman of OpenAI has stated three AI laws, in which AI model intelligence is proportional to the logarithm of resource input, AI usage costs are rapidly declining alongside exponentially growing demand, and the linear growth of AI leads to an exponential increase in socio-economic value. DeepSeek's Advantages: DeepSeek brings clearer technology routes, significantly reduced development and application costs, and a more open-source sharing model, reshaping the AI industry. This shift transforms the scaling law from an oligopoly-centered model to an inclusive one. Performance: DeepSeek has reached a level comparable to the most advanced U.S. models in both pre-trained foundational models and post-trained inference models. DeepSeek-v3 excels in Chinese question answering accuracy, long text processing, mathematical abilities, and coding, outperforming GPT-4o and Claude-3.5-Sonnet. The cost efficiency is notable, with DeepSeek-v3's training cost approximately one-tenth of that of leading European and American models. Innovation: DeepSeek focuses on engineering and low-cost innovation within existing frameworks. For pre-trained models, DeepSeek optimizes memory usage, workload distribution, and hardware interaction. For inference models, it uses a simpler and more efficient path, drawing inspiration from AlphaZero's training methods. Engineering Innovation: The success of DeepSeek highlights the value of engineering innovation in the AI industry. Unlike the "AI Seven Sisters" in the U.S., DeepSeek's origins lie in a Chinese quantitative hedge fund research team. Changing AI Economics: DeepSeek's advancements are changing AI economics and the scaling law. The original oligopoly-driven scaling law, dominated by U.S. tech giants, is shifting towards a more inclusive model due to reduced innovation costs and open-source approaches. This shift broadens participation in the AI industry beyond the dominant U.S. tech companies. Cost Reduction: DeepSeek initiated a price war in China's large-scale model API market in May 2024, forcing other companies to follow suit. This has led to significantly lower AI usage prices in China compared to the U.S.

There are no comments yet.

Please log in to write the first comment.