DeepSeek is an AI firm from China working on large language models, and their new chatbot DeepSeek is drawing a lot of attention for its products. This, however, has triggered a drastic selloff in the technology sector due to concerns about censorship and data security breaches.
Background
Located in Hangzhou, DeepSeek is an AI enterprise that was started as a registered company in July 2023. Its founder is Liang Wenfeng. In January 2025 company released the DeepSeek-R1 model along with the Deepseek chatbot which brought them to notoriety.
Technical details:
- DeepSeek AI combines machine learning along with natural language processing and deep neural networks.
- The user does not need to follow any prompt template as instructions can be given in any form of natural language.
- Employs a Mixture of Experts (MoE) model. They’re only turning on relevant sub-networks needed for the queries. This improves performance without the need for extra hardware.
- DeepSeek-R1 includes a reinforcement learning reasoning iteration and feedback pipeline.
- It is capable of receiving a single request containing 128,000 tokens and simultaneously generating 32,000 tokens.
DeepSeek makes use of a sparse mixture-of-experts (MoE) architecture, which switches on fewer neurons as a response to stimuli which improves productivity.
Cost efficiency
Compared to its competitors, DeepSeek AI claims that its cost-efficiency is significantly higher. The company states that its V3 model was trained for $6 million. In comparison, OpenAI’s GPT-4 claimed to have cost $100 million in 2023. This stark difference in training costs has raised concerns and questions about the potential use of distillation techniques or other methods that could involve copying proprietary technology.
Some industry experts have even raised suspicions that DeepSeek might be copying ChatGPT technology, fueling the ongoing debate about intellectual property and technology theft in the AI sector.
DeepSeek’s inference costs are said to be approximately 2% of OpenAI’s, which is laughably low.
Censorship concerns
When asked questions that have to do with sensitive topics, DeepSeek’s chatbot responses showcase China’s rigid information control. Isaac Stone Fish, a China expert, cautions that this makes DeepSeek a danger to global free speech because it over-suppresses expected open thought concerning China.
Dangers to data security
DeepSeek’s ability to harvest user information has raised concerns among specialists regarding its possible data transfer capabilities to the Chinese government.
Feroot Security’s Ivan Tsarynny claims that the links to China-based merchants’ servers raise security issues since the Chinese government owns them.
Practical uses
DeepSeek AI has a few real-life applications:
- Healthcare. Chinese hospitals leverage its medical imaging analyses to detect diseases at an early stage.
- E-сommerce. Online platforms employ it for behavioral analysis that assists in formulating targeted product advertisements.
- Cybersecurity. Information technology companies utilize it for network surveillance aimed at identifying security breaches.
A strategy involving no restrictions
The DeepSeek-R1 model is exposed to the public under the MIT license which permits its use, alteration, and commercialization without restrictions.
From the University of Edinburgh, Dr. Luo Mai claimed that DeepSeek is the first AI company to use the full open-source option together with differentiating pricing that is lower than the competition, thus allowing students and researchers from all over the globe to engage in AI development.
Reaction around the world
The launch of DeepSeek has raised questions on how to ensure effective control of innovative technologies. Decision makers are seeking to establish policies that enable further development of such technologies while safeguarding the general public.