DeepSeek
Stay updated on DeepSeek developments!
Get the latest updates and in-depth analysis of DeepSeek’s impact on AI, security, geopolitics, and more.
DeepSeek is more than just an AI platform—it’s a paradigm shift.
About DeepSeek
What is Deep Seek?
DeepSeek is a Chinese startup specializing in the development of open-source AI models and applications. Currently, there are two primary applications under the DeepSeek umbrella:
-
DeepSeek-V3: A large language model that forms the foundation for DeepSeek’s AI capabilities.
-
DeepSeek-R1: A reasoning model built on top of DeepSeek-V3, designed to enhance logical and analytical tasks.
Together, DeepSeek-V3 (the large language model) and DeepSeek-R1 (the reasoning model) represent the core of DeepSeek’s innovative AI solutions.
What does open-source mean for DeepSeek R1?
DeepSeek R1’s weights are freely accessible, allowing developers worldwide to use and modify them. Hosting the model on private servers ensures data sovereignty, preventing transmission to China if operated outside Chinese jurisdiction.
Does DeepSeek require supervised fine-tuning?
Unlike traditional AI models, DeepSeek R1 challenges the need for supervised fine-tuning, developing R1-Zero, which eliminates this stage, questioning its necessity.
Is pre-training essential for AI?
DeepSeek questions conventional pre-training, which demands vast datasets and computational resources. As experts predict a future scarcity of quality training data, alternative methods become crucial.
What were the training costs for DeepSeek?
Training required 2,788 thousand H800 GPU hours, costing approximately $5.576 million ($2/GPU hour). DeepSeek, however, notes that the costs only include the costs of the official training of V3, excluding the costs associated with prior research and experiments of architecture, algorithms and data (DeepSeek-V3, p.5).
What makes DeepSeek unique?
- Uses a mixture of expert models instead of a single massive model, reducing parameters from trillions to billions.
- Implements quantization, cutting weights from 32-bit to 8-bit, enhancing efficiency.
- Tokenizes phrases instead of words, improving processing speed.
Open Source
Why did DeepSeek choose open-source?
CEO Liang Wenfeng emphasised that open-source attracts top talent. He stated:
“In the face of disruptive technologies, closed-source moats are temporary. Even OpenAI’s closed-source model cannot prevent competition. Our strength lies in our team’s expertise and culture of innovation.”
What is 'open' in DeepSeek compared to other LLMs and AI models?
| AI aspects | LLaMA 2 | Mistral | BLOOM | DeepSeek |
|---|---|---|---|---|
| Algorithms | Yes | Yes | Yes | Yes |
| Pre-trained models | Yes | Yes | Yes | Yes |
| Weights | Yes | Yes | Yes | Yes |
| Datasets | No | No | Yes | No |
| Frameworks & Libraries | Yes | Yes | Yes | Yes |
| Deployment & Serving tools | Yes | Yes | Yes | Yes |
| Model training pipelines | Yes | Yes | Yes | Yes |
| Evaluation & Benchmarking | Yes | Yes | Yes | Yes |
| Data annotation & Labeling | No | No | Yes | No |
| Monitoring & Observability | No | No | No | No |
Does open-source AI pose security risks?
Yes. Open-source software can introduce vulnerabilities like Trojan backdoors. However, rigorous audits and independent verification can mitigate risks.
Is DeepSeek a copy of OpenAI?
No. DeepSeek integrates existing AI knowledge while innovating in reinforcement learning, achieving greater efficiency.
🔍 Explore more on open-source AI trends at Digital Watch.
Hardware and semiconductors
Will DeepSeek disrupt the GPU market?
DeepSeek optimizes models, reducing GPU requirements. However, major cloud providers continue investing in AI hardware infrastructure.
Did DeepSeek violate GPU sanctions?
No. DeepSeek was trained on NVIDIA H800 GPUs, which are not subject to current U.S. sanctions.
What is Huawei’s role in AI chips?
Huawei’s Ascend AI chips are positioned as a competitor to NVIDIA’s dominance.
🔎 Follow the latest AI geopolitics at Digital Watch.
Geopolitical impact
Is DeepSeek a data security risk?
Possibly. Chinese regulations could impose data access requirements, but self-hosting on private servers ensures data remains secure.
Does DeepSeek pose a geopolitical risk for the West?
Not directly. Its open-source nature allows independent adoption, demonstrating China’s growing AI expertise.
🛡️ For geopolitical aspects of DeepSeek moment, check here.
Content policy
Why does DeepSeek censor certain topics?
Chinese law restricts discussions on sensitive topics (e.g., Falun Gong, Taiwan).
Can this censorship be bypassed?
Yes. Self-hosting DeepSeek allows users to control content moderation outside China’s legal framework.
⚖️ Learn more about AI and content policy at Diplomacy.edu.
Economic impact
How is DeepSeek reshaping AI business models?
DeepSeek accelerates the shift from expensive proprietary AI to accessible open models, shifting economic value to AI applications and services.
Does DeepSeek enhance AI competition?
Yes. It challenges AI monopolies, encouraging broader innovation across the industry. Here is a comparison of the costs of OpenAI and DiploSeek:
|
OpenAI-01 |
DeepSeek-RJ |
|
|
1M input tokens |
$ 15.00 |
$ 0.14 |
|
1M cashed input tokens |
$ 7.50 |
$ 0.55 |
|
1M output tokens |
$ 60.00 |
$ 2.19 |
|
Month pro subscription |
$ 200.00 |
Free |
💰 For AI and the economy, visit Digital Watch.
Law and regulation
What is DeepSeek’s legal basis for open-source?
DeepSeek follows the MIT license, allowing free use, modification, and commercialization.
How does the EU AI Act apply?
The impact of open-source AI on EU regulations remains under review, as policymakers consider new governance frameworks.
📜 For AI legal developments, check Diplomacy.edu.
Winners & losers
📉 Potential Losers:
- Semiconductor manufacturers (reduced high-end GPU demand)
- Meta (losing open-source AI dominance)
- Energy companies (lower AI energy consumption)
📈 Potential Winners:
- Software developers (higher ROI from AI applications)
- Apple (gains in edge inference efficiency)
⚖️ Mixed Impact:
- Microsoft (loses as OpenAI backer, but gains from AI-driven services)