DeepSeek Explained: A Few Words Why Is It Disrupting the AI Sector
DeepSeek is a Chinese artificial intelligence startup that has recently gained significant attention in the global tech industry. Founded in July 2023 by Liang Wenfeng, a graduate of Zhejiang University, DeepSeek is based in Hangzhou, China. The company has made waves with its AI chatbot app, which quickly rose to the top of Apple’s App Store charts after its US launch on January 10, 2025.
Key Features and Achievements
Low-cost development: DeepSeek claims to have built its AI models for just $6 million, significantly less than its US counterparts.
Competitive performance: The company asserts that its latest AI models match the performance of leading US models like ChatGPT.
Efficient chip usage: DeepSeek developed its models using a combination of high-end Nvidia A100 chips and less expensive, lower-end alternatives.
Open-source approach: DeepSeek’s AI models are largely open-source, allowing developers to examine and build upon their inner workings.
Reasoning capabilities: The DeepSeek R1 AI assistant provides detailed reasoning for its answers, which has excited developers.
The DeepSeek R1 AI assistant provides detailed reasoning for its answers, which has excited developers.
Impact on the Tech Industry
DeepSeek’s sudden rise has caused significant disruption in the global tech market:
- Stock market impact: The company’s emergence led to a sharp decline in shares of AI-related companies like Nvidia and ASML.
- Reassessment of AI development costs: DeepSeek’s low-cost approach has prompted a reevaluation of the massive investments made by US tech giants in AI development.
- Geopolitical implications: The success of DeepSeek has raised questions about the effectiveness of US export controls on advanced chips to China.
- Competitive landscape: DeepSeek’s rapid advancement has been described as “AI’s Sputnik moment,” challenging the perceived American superiority in AI technology.
- As DeepSeek continues to develop and expand, it is likely to remain a significant player in the global AI race, potentially reshaping the industry’s dynamics and challenging established tech giants.
Performance Comparison
- Technical Tasks: DeepSeek outperforms ChatGPT in technical applications, particularly in coding, solving complex equations, and logical reasoning.
- Conversational Abilities: ChatGPT remains superior in tasks requiring conversational or creative responses, as well as delivering news and current events information.
- Multilingual Capabilities: DeepSeek demonstrates exceptional performance in multilingual tasks.
- Problem-Solving: DeepSeek’s R1 model showcases advanced self-evolving reasoning capabilities, allowing for more autonomous problem-solving.
- General Performance: For most general inquiries, both models deliver comparable results.
DeepSeek V3 operates with 600 billion parameters, while ChatGPT-4 uses 175 billion.
Key Differences
- Parameters: DeepSeek V3 operates with 600 billion parameters, while ChatGPT-4 uses 175 billion.
- Training Data: DeepSeek V3 was trained on 14.8 trillion tokens, enabling it to handle highly complex tasks.
- Learning Approach: DeepSeek heavily relies on Reinforcement Learning (RL) for developing self-improving reasoning capabilities, while ChatGPT uses a hybrid approach combining Supervised Learning and RL.
Cost and Accessibility
DeepSeek offers its advanced features for free, including web-search capabilities and file uploads, while ChatGPT requires a premium subscription for similar functionalities25. This cost difference makes DeepSeek an attractive option for developers and businesses, with significantly lower API pricing compared to OpenAI.
In conclusion, while both models are highly capable, DeepSeek appears to have an edge in technical and specialized tasks, whereas ChatGPT maintains its strength in general-purpose and creative applications. The choice between the two depends on the specific use case and user requirements.
Final note
DeepSeek managed to acquire a significant stockpile of Nvidia A100 chips before the U.S. export restrictions on advanced GPUs to China took effect. Liang Wenfeng, DeepSeek’s founder, reportedly accumulated over 10,000 Nvidia A100 GPUs during this period. These chips became a foundational resource for training their AI models, enabling the company to develop its competitive AI systems despite subsequent restrictions on high-end chip exports to China.
Links
- DeepSeek, unravel the mystery of AGI with curiosity. Answer the essential question with long-termism. – DeepSeek
- Sell pre-owned AI device online – iGotOffer
DeepSeek Exposed: How Good Is It Really? (Tutorial for Beginners) [Video]
Video uploaded by Kevin Stratvert on January 30, 2025.
Facebook
Twitter
RSS