Context: The Chinese start-up DeepSeek, has created a buzz with the launch of its cutting-edge AI models ‘DeepSeek-R1’ & ‘DeepSeek-V3’, claiming they nearly match the capabilities of top AI models in the U.S., while being far more affordable.
Relevance of the Topic: Prelims: Basic understanding of terms like Large Language Models, DeepSeek AI.
What is DeepSeek?
- DeepSeek is a Hangzhou-based Chinese startup that has recently launched artificial intelligence (AI) chatbot built on a low-cost Large Language Model (LLM) infrastructure.
- The AI is optimised for tasks like maths and coding, making it a strong competitor in the AI space.
DeepSeek vs global LLMs:
- Low training cost: DeepSeek reportedly trained its model for just $6 million, significantly lower than the estimated $100 million expenditure behind OpenAI's GPT-4.
- DeepSeek was able to dramatically reduce the cost of building its AI models by using NVIDIA’s H800 chips, an older generation of GPUs in the US.
- High efficiency & Low cost: The model is being praised for its efficiency, as it uses advanced and lower-grade chips to deliver high performance at a lower cost.
- E.g., Reportedly, DeepSeek-R1 is 20 to 50 times cheaper to use than OpenAI o1 model (depending on the task).
- DeepSeek’s R1 may not be quite as advanced as OpenAI’s o3, it is almost on par with OpenAI o1 on several metrics.
- Innovative & Adaptable:
- DeepSeek-R1 uses reinforcement learning to naturally (autonomously) evolve its reasoning capabilities. The model self-improves through feedback loops during training, without needing massive labeled datasets.
- DeepSeek-R1 can transfer (distill) reasoning capabilities into smaller models (SLMs), which are faster and more resource-efficient. Thus, DeepSeek-R1’s reasoning capabilities are scalable across different model sizes, making it highly adaptable.
- Affordable for users: DeepSeek's paid subscription comes at $0.50 a month, while ChatGPT costs $20.
Concerns:
- Censorship on digital content & bias:
- Unlike many Western models, DeepSeek follows China's strict censorship rules. When asked about sensitive topics, it avoids direct answers, reflecting government control over digital content.
- Furthermore, the chatbot is expected to have a pro-China bias.
- Potential security risks: Experts have warned about potential security risks associated with the DeepSeek AI app, pointing out the need for scrutiny in data privacy and AI ethics.
Global Impact:
- Sputnik moment: Much like the Sputnik in the 1950s, DeepSeek brings a new technological frontier into the great power competition.
- Market disruption: The launch of the DeepSeek AI resulted in a historic $600 billion market value drop for Nvidia, a key player in AI chip production.
- Policy implications: DeepSeek heralds an escalation of the geopolitical rivalry between the US and China. It risks escalation of the US government's restrictions on advanced chip exports to China.
Read more: What are Small Language Models?