DeepSeek is a Chinese AI company that has garnered significant attention for its cost-effective and high-performance large language models (LLMs), notably DeepSeek-R1 and DeepSeek-V3. Founded in July 2023 by Liang Wenfeng in Hangzhou, Zhejiang, DeepSeek is backed by the Chinese hedge fund High-Flyer
Key Developments
- DeepSeek-R1: Released in January 2025, DeepSeek-R1 is an open-source LLM that has demonstrated performance comparable to OpenAI’s GPT-4 and Anthropic’s Claude 3.5. Its efficient training, costing approximately $5.6 million, contrasts with the billions spent by Western counterparts
- DeepSeek-V3: An advanced version of R1, DeepSeek-V3 employs a Mixture-of-Experts (MoE) architecture with 671 billion total parameters, activating 37 billion per token. It was trained using 2.788 million H800 GPU hours, significantly fewer than models like Meta’s LLaMA-3 405B, which required 30.8 million GPU hours
Global Impact
- Open-Source Model: DeepSeek’s decision to release its models under the MIT License has democratized access to advanced AI, enabling researchers and startups to develop AI applications without relying on major tech firms
- Microsoft Collaboration: In response to DeepSeek’s success, Microsoft began offering DeepSeek’s R1 through its Azure AI Foundry platform, ensuring data security by not transmitting data to servers in China
- Healthcare Integration: DeepSeek’s AI models have been integrated into China’s tertiary hospitals, enhancing diagnostic accuracy and streamlining workflows, though this raises regulatory and ethical challenges .
Concerns and Criticisms
- Censorship and Bias: DeepSeek’s models have been found to comply with Chinese government censorship policies, limiting responses on sensitive topics. While uncensored versions exist, they may exhibit biases reflecting Chinese government viewpoints .Wikipedia
- Security and Privacy: Experts have raised concerns about DeepSeek’s data collection practices, with reports indicating that user data may be sent to Chinese companies, including ByteDance. This has led to regulatory scrutiny in countries like South Korea, which has suspended new downloads of the app .Wikipedia
- Data Breach: In January 2025, a significant data breach exposed over a million sensitive records due to improper cloud storage configurations, raising questions about DeepSeek’s data security practices .Wikipedia
🔗 Access and Resources
- Official Website: deepseek.com
- API Platform: platform.deepseek.com
- GitHub Repository: github.com/deepseek-ai/DeepSeek-V3
