DeepSeek is a Chinese artificial intelligence (AI) startup that has recently garnered significant attention for its rapid advancements in AI technology. Founded in Hangzhou, DeepSeek has developed a series of open-source large language models (LLMs) that have demonstrated impressive performance in various benchmarks.
Key Developments:
DeepSeek-R1: Launched in November 2024, this model focuses on logical inference, mathematical reasoning, and real-time problem-solving. It has been reported to outperform industry leaders in specific benchmarks, achieving high accuracy in mathematical problem-solving and programming tasks.
DeepSeek-V3: Released in December 2024, DeepSeek-V3 is a mixture-of-experts model with 671 billion parameters. It was trained over approximately 55 days at a cost of around $5.58 million, significantly lower than the expenditures of competitors like OpenAI. Benchmark tests indicate that DeepSeek-V3 outperforms models such as Llama 3.1 and Qwen 2.5 while matching the performance of GPT-4o and Claude 3.5 Sonnet.
Market Impact:
The emergence of DeepSeek has had a notable impact on global tech markets. On January 27, 2025, major U.S. tech stocks, including Nvidia, Microsoft, and Tesla, experienced significant declines, potentially losing up to $1 trillion in value.
This downturn was attributed to DeepSeek's development of a competitive large-language model at a fraction of the usual cost and without reliance on advanced Nvidia chips. Following the news, Nvidia's stock fell by nearly 13%.
Technological Approach:
DeepSeek's models are notable for their cost-effective development and efficiency. The company has utilized approximately 2,000 Nvidia H800 chips, which comply with U.S. export controls, to create its models efficiently. This approach challenges the prevailing high-cost AI development models and suggests a potential paradigm shift in the industry.
Controversies:
Despite its technological achievements, DeepSeek has faced criticism for producing responses that align closely with Chinese state-sanctioned views, particularly on sensitive political topics such as human rights and Taiwan. This has raised concerns about censorship and the influence of state narratives in AI outputs.
