Almost 2 trillion dollars of technology companies have disappeared in just a few hours ago in the stock market, the reason is the new version of Chinese artificial intelligence called Deepseek R1, which is capable of surpassing all artificial intelligences with infinitely less use of resources. than its competitors.
It is clear that there is a race of technological superpowers but a Chinese startup has emerged as an unexpected contender Deepseek developed an Artificial Intelligence superior to those of the West, but also challenges the established norms of how these models are built and how they are financed with an approach in efficiency and innovation.
Deepseek has shown that it is possible to compete with the elite of Artificial Intelligence, without depending on the unlimited resources that have characterized the Western Giants until today.
On January 20, the Chinese technology company launched its Deepseek R1 model, a partial open source system that quickly became the center of attention in Silicon Valley. The interesting thing is that Deepseek R1 is not just an Artificial Intelligence model.
Deepseek R1 outperforms leading models such as OpenAi in various math and reasoning tests, this achievement is not only impressive for its technical prowess but also for its effectiveness, cost efficiency and open source approach, which which makes it a very attractive alternative for researchers and developers.
Deepseek R1 not only matches the performance of Western models, but it does so using a fraction of the computational resources and this is the key, for example, the Deepseek R1 model required a tenth of the computing power used to train for example the model called 3.1 meta.
And this efficiency is the result of a combination of technical innovations, such as the multi-head latent attention that it calls MLA, and the expert mixing approach that allows the model to optimize the use of resources without sacrificing performance, the program just like its counterpart American processes requests using something called thought chains that mimic human reasoning processes with almost eerie precision.
This system was not born out of nowhere, it was trained from the operation of chatgpt v3 also developed by Deepseek using reinforcement learning techniques that reward not only the correct answer, but also the system's ability to explain its own thinking and it does so during the problem solving, but that's not all, the researchers took this model a step further by implementing something called "expert mixing architecture."
The “expert mix architecture” is an Innovation that allows the model to decide with an almost warlike Maqui coldness which processing networks should be activated for each specific task, the result is a machine that not only thinks, but does so in a manner strategic and efficient, in addition, it is capable of emulating human reasoning quite effectively using “chains of thought” that imitate human cognitive processes
Deepseek R1 costs one-thirtieth of what it means to use Open Ai 01, which not only democratizes the use of Artificial Intelligence, but also raises a very uncomfortable question for Western giants: how is it possible for a Chinese model developed in a context of restrictions and sanctions has achieved such a feat.
The cost of training this technological prodigy is around 6 million dollars,
a figure that in comparison with the more than 60 million meta invested in llama seems almost laughable or the billions that the rest are using, this monumental saving in computing resources is not only a technical achievement, it is a kick to the table that shakes the foundations of the industry.
The cost of training this technological prodigy is around 6 million dollars,
a figure that in comparison with the more than 60 million meta invested in llama seems almost laughable or the billions that the rest are using, this monumental saving in computing resources is not only a technical achievement, it is a kick to the table that shakes the foundations of the industry.
We are in a moment in which efficiency and innovation have proven to be more powerful than unlimited resources and financial power, the one that comes from Silicon Valley,
The good thing will come in the coming months, at the level of artificial intelligence and politics, the year did not start boring at all.
