DeepSeek - a revolution in the world of AI from China

DeepSeek has become one of the hottest topics in the technology world at the very beginning of 2025, causing both delight among users and concern among technology investors. This Chinese model of artificial intelligence, has attracted attention not only by its performance, but also because of controversies over data security and training methods.

What is DeepSeek and where did it come from?

DeepSeek is an advanced artificial intelligence system from China that has rapidly gained immense popularity and begun to compete with the most powerful AI models in the world. The creator and current owner of DeepSeek is a Hangzhou, China-based company, DeepSeek Artificial Intelligence, founded back in 2023 by Liang Wenfeng.

In specialized terminology, DeepSeek is classified as aLarge Language Model (LLM), much like ChatGPT or Google Gemini. However, what sets DeepSeek apart from its competitors is primarily its architecture. The model uses an innovative approach called Mixture-of-Experts (MoE), which can be compared to a team of specialized experts, where only the most suitable ones are engaged for each task.

The second major distinguishing feature of DeepSeek is its open-source approach, which means nothing less than that it is available for free and can be modified by the community. This, in the context of the controversy we're about to get into, is particularly valuable for those wanting to understand exactly how DeepSeek works. What's more, the model can be installed locally and its underlying technology can be customized or integrated into one's own systems.

We mentioned at the beginning the anxiety of technology investors. It was caused by the surprise of the cost of implementing DeepSeek, which turned out to be significantly lower than that of ChatGPT. DeepSeek was trained on 14.8 trillion diverse and high-quality tokens, and the cost of the entire process was estimated at around $6 million. This is significantly less than the $100 million that OpenAI was expected to spend on GPT-4 training in 2023.

DeepSeek-R1 and DeepSeek-V3 - are they the same?

DeepSeek-R1 and DeepSeek-V3 are two advanced language models that differ fundamentally in their architecture, applications and mode of operation.DeepSeek-R1 is a specialized model focused on reasoning and solving complex problems, using learning with an emphasis on generating accurate or thoughtful answers. DeepSeek-V3, on the other hand, is a versatile general-purpose model designed as a competitor to models such as GPT-4o, excelling in content creation, but with limited reasoning capabilities.

DeepSeek - controversy and concern, or from admiration to anxiety

Almost immediately after DeepSeek's debut, an area of concern related to it became the issue of protecting users' personal information. DeepSeek has been logging user messages and bot responses to improve the model, although the company assures that the information is anonymous.

However, the fact that the data is stored on Chinese servers raises concerns in terms of state control and the ability of Chinese authorities to access the data. For this reason, several countries, including the United States, Australia, South Korea, Italy and Taiwan, have officially banned the use of DeepSeek on government devices.

These concerns have been compounded by controversy over the censorship of DeepSeek-generated responses on topics sensitive to China, primarily political content.

While other artificial intelligence models are sometimes characterized by a hallucinatory effect, or to put it bluntly, making up fictitious information if they don't have the right information in their training data, DeepSeek avoids and even refuses to give sensitive answers to the Chinese government.

Examples of Chinese censorship implemented in DeepSeek include the chatbot's refusal to answer questions about the 1989 Tiananmen Square massacre, the status of Taiwan or the persecution of Uighurs.

DeepSeek vs ChatGPT - a comparison of capabilities

A comparison of DeepSeek vs ChatGPT reveals significant differences in their approaches to similar problems. Both models represent state-of-the-art AI technologies, but each has its own unique strengths and limitations.

Criterion	DeepSeek	ChatGPT
Costs	Free, open-source	Subscription for premium features
Specialization	Technical tasks, mathematics	Creativity, conversations
Performance	Faster responses in coding	Better understanding of the context
Security	Risks related to China's jurisdiction	Stronger data security
Availability	Possibility of local installation	Only in the cloud

While DeepSeek uses the aforementioned Mixture-of-Experts architecture, ChatGPT commits all of its resources to each task (as if all the experts were working on each problem), DeepSeek activates only the parts of the model that are most appropriate for a given query.

DeepSeek V3, the latest version of the model, has an impressive 671 billion parameters, of which only 37 billion are activated for each token, which significantly increases the model's performance, allowing it to dynamically adapt to different types of queries while optimizing resources. For this reason, in performance tests DeepSeek shows impressive capabilities in technical tasks, especially in mathematics, where it achieves 90% accuracy - significantly outperforming many competitors. This makes it a particularly valuable tool for solving technical problems. ChatGPT, on the other hand, has traditionally demonstrated stronger capabilities in understanding context and providing more nuanced answers across a wider range of topics.

A good example of this would be the test conducted by the BBC, in which in one of the tasks DeepSeek produced a full story faster, while ChatGPT focused on generating ideas.

What else is important in terms of comparing these models, DeepSeek offers more extensive customization options, which is beneficial for those with some technical knowledge. ChatGPT, on the other hand, focuses on user-friendliness, offering a more refined experience that is accessible even to those new to AI.

DeepSeek - a revolution in the shadow of controversy

DeepSeek has already managed to revolutionize the AI market, which is still fresh after all, proving that advanced models can be built with minimal investment. On the other hand, while its efficiency and low cost are attracting developers and startups, controversies over data security and censorship are limiting applications in sensitive sectors.

Nevertheless, DeepSeek is a fascinating example of how a company with limited resources can challenge the biggest technology giants. The DeepSeek R1, with its innovative Mixture-of-Experts architecture and open-source approach, has won praise from users around the world.

Perhaps the aforementioned controversies over censorship, data security and training methods cast a shadow over this success, but at the same time they remind us that the development of AI technology is not taking place in a vacuum - it is deeply embedded in a broader social, political and geopolitical context. These issues will certainly shape the discussion around artificial intelligence in the coming years.