Last Monday, a quiet earthquake shook the AI world. The day before Donald Trump’s $500 billion AI infrastructure announcement, a small Chinese company, DeepSeek, unveiled a Large Language Model (LLM) that rivals the best from OpenAI and other major players.
DeepSeek say they built this model in just 55 days for $5.6 million. OpenAI’s equivalent models are estimated to have cost anywhere from $100 million to $500 million each using far more advanced hardware. Proving it to be no idle boast, DeepSeek shared a detailed paper explaining how they did it and made it open source for anyone to explore and adapt.
A lesson from history
After the devastation of World War II, Japan’s ambition to become a manfacturing powerhouse seemed laughable. “Made in Japan” was synonymous with cheap, low-quality products. However, Japanese companies adopted ideas from American quality pioneers like W. Edwards Deming and Joseph M. Juran and gave birth to “Kaizen”.
Despite having far fewer resources than its American rivals, the Toyota car company rethought the car manufacturing process entirely. The result was Lean Production—an efficient system that disrupted global manufacturing.
Similarly, DeepSeek, denied top-end computing technology, has upended the assumption that training world-class LLMs requires vast computational resources. While Microsoft considers the creation of its own nuclear power station to power their models, DeepSeek has shown that lean, resource-efficient processes can achieve comparable outcomes at a fraction of the cost. Their models are cheaper to train and far less expensive to run.
What are the implications?
1. A level playing field: If DeepSeek’s approach holds up, the barriers to entry in generative AI are dramatically lowered. Any organization with modest resources can now access and adapt cutting-edge AI.
2. China’s competitive edge: DeepSeek demonstrates that China is either catching up or has already closed the gap in AI innovation, particularly since this model was developed by a hedge fund’s “side hustle.”
3. Open ecosystems foster innovation: By open-sourcing their work, DeepSeek has created a platform for domain-specific AI development. Just as Microsoft’s open operating system fueled a thriving ecosystem of software innovation, DeepSeek’s openness could spark a wave of new applications.
4. A rethink for incumbents: OpenAI and other major players are pursuing ambitious goals like Artificial General Intelligence (AGI). But DeepSeek’s lean approach challenges their resource-heavy strategies. Their open, adaptable model might just prove to be more practical and disruptive in the near term.
Put DeepSeek on your radar. They might be the Toyota of AI.
コメント