DeepSeek, a Chinese AI start-up founded in 2023, has made significant strides in the global AI race with its resource-efficient and open-source models. The company is based in Hangzhou, a tech hub that is also home to Alibaba.
Key Points about DeepSeek:
- Resource Efficiency: DeepSeek has developed competitive AI systems like the DeepSeek R1, rivaling industry leaders such as OpenAI, while using fewer resources. Its V3 base model was developed in just two months with a budget of under US$6 million.
- Open-Source Approach: DeepSeek has embraced open-source methods, promoting collaborative innovation1.
- Technological Innovation: DeepSeek uses innovative methods such as test-time scaling to enhance performance during model deployment. They have also made progress on Multi-head Latent Attention (MLA) and Mixture-of-Experts, which make DeepSeek models more cost-effective by requiring fewer computing resources to train.
- Government Support: China’s government policies, funding, and AI graduates have helped Chinese firms like DeepSeek to create advanced LLMs. Zhejiang province, where DeepSeek is based, aims to establish itself as a leading hub of innovation in AI.
- Supercomputing Integration: China’s national supercomputing network has integrated DeepSeek’s open-source large language model into its platform to enhance the accessibility and utilization of advanced AI technologies for local and global users. The national supercomputing network platform is linked to over 20 supercomputing and intelligent computing centers across 14 provinces in China.
- US Export Restrictions: DeepSeek has achieved its success despite US export restrictions on critical hardware. The company had to come up with more efficient methods to train its models. They optimized their model architecture using custom communication schemes between chips, reducing the size of fields to save memory, and innovative use of the mix-of-models approach.
DeepSeek’s rise has altered investor perceptions regarding China, shifting attention away from typical macroeconomic issues towards this domestic success story.
