Why is Deepseek being viewed as a “bug”?

Chinese Strartup Deepseek shook the US economy by launching a new artificial intelligence model (AI) that competes with the best American models. “And now?”

Over the past two years, AI models of companies like OpenAi, Google and the goal have become able to perform increasingly complex tasks. To achieve it, technological companies spent tens of thousands of millions of the AI ​​development.

But without (apparently) nothing to predict, on January 20, the Chinese startup DeepSeek Freely available its R1 model, published an article that demonstrates the impressive performance of AI e.

The Deepseek model performed comparable to OpenAi’s 01 model 01 in various math and coding tests. This would not be a shock, if I had not been able to be 20 to 50 times cheaper to train and use significantly fewer computing resources.

This invention could help more organizations and companies benefit from AI services. And the fact that it is open source and freely available could Forcing US competitors to lower their own prices.

Deepseek has been quite transparent in the publication of its results, in the reference tests. However, it is not yet known exactly to what extent the Deepseek model compares with major US models.

After all, this “bug” may not be as revolutionary as it seems.

As noticed, the model has certain Incorporated prejudices in their formation that are intended to comply with restrictions Chinese governmentfor example, to avoid discussion of the treatment of the Uigur ethnic minority and follow the Chinese Communist Party line about Taiwan’s political status.

Deepseek has access to Chinese and foreigner computing resources stored in the cloud – which are not subject to US export controls that have sought to limit the use of American chips in China. These resources probably helped the startup to discover more efficient techniques, generating First synthetic training data and then allowing a wide trial and error experimentation.

Despite these US export controls, the company also had access to a recent generation of AI chips. DeepSeek trained its AI models in Nvidia H800 chips that were designed to circumvent older and less effective US export controls, they pointed out.

However…

“The real test comes when these data centers need update or expansion – a process that will be easier for American companies, but challenging for Chinese companies under the US export controls“, consideraramm Lennart homeyes row corporation, e Sihao huangfrom the University of Oxford, point out, cited by New Scientist.

“O moment Sputnik da Ia”?

“If DeepSeek won’t completely disturb the US -based AI world, why are the US technology market and investors panicked? – Asks New Scientist.

The same magazine points out that, in recent years, An advertising bubble was created IA, which led to high assessments of technological companies; and compares the current scenario to “SPUTNIK MOMENT” where the Soviet Union surprised the US by launching the first artificial satellite for space.

In that case, The US was in fact aware of the USSR in terms of spatial technology and well aware of Soviet progress. Only the public was not aware, due to the secrecy of the US government.

There is a way for this moment to reflect the age of Sputnik: It may be a situation in which a sensationalist history provokes an exaggerated reaction from the public and, “hidden”, Encourage more US expenses with AI.