Deepseek: Really revolutionary AI?

► In short

  • Deepseek is AI launched by Chinese society
  • It seems to compete with the strongest models
  • His arrival of the card does not match

AI has been accelerating for more than 2 years. In 2022, the world discovered the Chatgpt, a conversational robot with stunning, but also fun answers, because of its mistakes and other hallucinations. But 2 years later they are far from the smile before the wandering Openai AI.

Because in the meantime, the industry has evolved at high speed. And today it is clear that AI, except for saving on certain sides of our daily life, became a problem at the core of the news. An example with Deepseek, a solution developed by a Chinese company that has been a monopolizing conversation for several days.

Deepseek, a technological demonstration

Because Deepseek, it is primarily a model made for bluffing. Developed in a few months, for a budget of EUR 6 million, launched in an open source, with a performance almost similar to the last versions of the cottage, shows above all how accessible IA has become.

Because the project, which actually existed since 2023 in more rustic forms, has just completed the course. The Deepseek-R1 version launched in 2025 will therefore be offered a complete AI based on a chain of reasoning in logical stages and more than 671 billion parameters, can only impress. All the more impressive when we know that everything was done despite the American embargo on the NVIDIA GPU, usually essential for AI.

AI who likes to be silent

But this is for technological performance. Because once behind the keyboard we discover a strongly politicized AI. It is true that we could claim that Optai or Gemini offer already biased models, but they could still be forced to discuss any topic with the right challenge.

In the case of Deepseek, we will discover the new Cran de Franchi with pre -pricing that seems deeply anchored in the AI ​​parameters. So do not tell him about Tian’anmen or the possible abuse of the Chinese government: You will not have any answers. On the other hand, run it on the failure of various Western nations and Deepseek will be the most speaking there. And it is no wonder that if he asks if he treats this information differently, it is honest: “I am here to provide you with information with the official positions of the Chinese government”. Note that the restriction of the conversation agent seems to disappear on your machine locally.

Complex ecosystem

Despite its limitations, however, Deepseek never ceases to impress users around the world because of its many performances. It seems that the development, development of code, reflection, and at low prices succeeded in its bet. What to say that the rest of the world is exceeded? No, because the AI ​​domain does not stop on the main public models. Take an example of Google Deepmind; It is true that Gemini’s solution is trying to persuade, especially in the face of Chatgpt, and is now far looking for much more efficient. However, Mountain View does not attach the main importance! Because for her, the core of IA is scientific research models. And in this area, the California company always has impressive progress and regularly launches a solution with a very high innovation potential.

So you have to continue to look at the image as a whole. On every new powerful public model that will be launched, we will inevitably have the impression that the new actor is pushing for. And yet we must continue to maintain a slight look: AI is a complex sector, still in motion, which cannot be summarized by announcing the notification.

Leave a Comment