Table of Contents 64q43
Last weekend, an artificial intelligence called DeepSeek, developed in China, caused a stir in the technology world and caused the value of shares of American giants such as NVIDIA, Google, Meta and Microsoft to plummet by more than 1 trillion dollars. Understand how it works and how it can impact the world of technology, but specifically OpenAI's ChatGPT. x3w3z
What is DeepSeek 2j1s3a
DeepSeek is a Chinese artificial intelligence company, established in 2023 by Liang Wenfeng, Chinese entrepreneur and co-founder of the quantitative hedge fund High Flyer. This fund, famous for using machine learning techniques in stock investing, is also the current financier of the company's AI research.
Since 2023, the company has been attracting global attention due to the rapid development of its language models or LLMs.
Low cost development 3yq54
The main advantage of the models is their development cost. In a recently released article, the company explained that its language model DeepSeek-V3 used 2 NVIDIA H800 cards, which are already relatively old, for training at a total cost of $5,5 million.
For comparison, this is just a fraction of the cost of training a model like Meta's Llama 3.3, which this year alone is expected to cost more than 65% billion dollars in its improvement. In this comparison, the costs of DeeppSeek end up being below even the annual salary of some executives at Meta's company. According to analysts, the cost reductions are impressive 95%.
This feat puts an end to the paradigm that high investments in data center infrastructure and electrical energy would be necessary to train AIs, making DeepSeek an AI that is not only cheaper, but also more ecologically sustainable.
In theory, instead of using specialized data centers with expensive GPUs designed for high AI demands, DeepSeek's models could be trained even on home computers, with traditional GPUs, which causes both astonishment and fear in companies like NVIDIA, which has been so leveraged by the recent rush to build data centers by companies seeking a share of this market.
For this reason, in what has become known as the “DeepSeek Panic Day” ou “DeepSeek panic day”, In free translation, January 27th was marked by billion-dollar falls in shares on NASDAQ, the stock exchange for American high-tech companies, reaching 3,5% of the total traded. ChatGPT's competitor has everything to bomb.
DeepSeek-V3 and DeepSeek-R1 5s3s3r
The company's models are also an example of efficiency. DeepSeek-V3, for example, boasts impressive performance across multiple benchmarks while requiring significantly fewer resources than competitors like Meta's Llama 3.3. And the DeepSeek-R1, launched in January 2025, focuses on reasoning tasks and challenges OpenAI's o1 model with its advanced capabilities.
In essence, both Chinese AI model options learn by interacting with their environment and receiving on their actions, similar to how humans learn through experience. This allows them to develop more sophisticated reasoning skills and adapt to new situations more effectively. This approach is also cheaper than the “human-supervised learning” used by competitors.
Another advantage of this Chinese AI is in resource saving. Instead of using all the model parameters to answer a query, DeepSeek models select only a few “specific parameters” or “experts” for the topic. This selective activation significantly reduces computational costs and increases model efficiency.
Explaining it in a simpler way, it would be like having a large team of specialists available for all types of tasks, but calling only those most suited to carry it out in a specific case, ensuring efficient use and expertise in service.
O DeepSeek-V3 also employs distillation techniques, meaning that it transfers knowledge and capabilities from larger models to smaller, more efficient models. It is like a teacher transferring his or her knowledge to a student, enabling the student to perform tasks with similar proficiency but with less experience or resources.
The result is an AI that takes up less space and resources, with high efficiency to work on smaller devices, such as smartphones and notebooks. To make the application available for smartphones and also in the web version, the company used DeepSeek R1 as a base and transformed it into an app.
Open source model 3d415w
In addition to combining efficiency and performance techniques, the models also take advantage of the collaboration inherent in design projects. open code (or open source, in English).
Here, unlike a company like OpenAI, which works privately on AI development, DeepSeek's models have their code made available on the web, and can be accessed, tested and even corrected by the global community of developers, which also reflects in more transparency and trust in the results presented.
In fact, the company wants to encourage anyone to test its models and is releasing licenses to its professionals to make it easier for other companies to use them.
The strategy also strengthens competitiveness by attracting talent and highlighting DeepSeek's leadership position, in a kind of free marketing. The company does not benefit directly from this, but it will benefit from the mass adoption of the model, being able to monetize through complementary services, such as technical or versions, while fostering a collaborative and innovative ecosystem.
Number 1 in the app stores 2l2
Curiosity about DeepSeek wasn't just among CEOs and tech engineers.
Launched on January 20th, the R1, DeepSeek app for iPhones and Android smartphones, it seems to be following the success path of its competitor ChatGPT. It is already the most ed application in the American Apple App Store and the second in Brazil, behind only the OpenAI competitor.
Na Play Store, Android store, the success is also undeniable: worldwide, the Chinese chatbot has already been installed more than a million times.
Sam Altman, founder and current CEO of OpenAI, spoke about DeepSeek's success. He said he's keeping an eye on what Chinese AI can do and that the first model is impressive. See the tweets and translations below.
The Deepseek R1 is an impressive model, especially considering what they can deliver for the price. Obviously, we will deliver much better models and it is also genuinely refreshing to have a new competitor! We will be doing some launches.
But most importantly, we are excited to continue executing on our goals and believe that more computing is more important now than ever to succeed in our mission.
The world is going to want to use A LOT of AI and will truly be amazed at the next-gen models that are coming. We can’t wait to bring you all of AGI and beyond.
Sam Altman on the success of Chinese AI DeepSeek.
Implications and political bias 4m2pp
But with success comes suspicion. Experts question whether the total number of GPUs used to train DeepSeek's models was underreported, with the intention of hiding the use of NVIDIA hardware that was recently banned from sale in China due to the trade war with the United States. But the company denies this, saying that it was able to train its model even with low-power GPUs.
There are also signs of censorship in the tool, in line with the precepts or political bias of the Chinese Communist Party. For example, when asked about the “1989 Tiananmen Square massacre in Beijing”, where hundreds of activists were killed or injured by the Chinese army, the AI refuses to answer the question.
The same thing happens when the AI is asked about the conflict between China and Taiwan. DeepSeek cites that Taiwan is part of China and that “The Chinese government is committed to the one-China principle and seeks the peaceful reunification of the country.
What comes next? 4n6s2p
The possibility of having such efficient AI models, at significantly lower infrastructure costs, is making investors rethink buying or even selling their shares in technology companies focused on this sector.
One of the legends of venture capital American, Marc andreessen, sees the emergence of DeepSeek as the “AI’s Sputnik moment”, in reference to the historic launch of the first satellite into orbit in 1957, which marked the moment when the Soviet Union temporarily sured the United States in the field of space technology.
The fact is that American superiority in the field of artificial intelligence has been put in check. And for this reason, it is likely that retaliation from the American government, perhaps through more sanctions on the commercialization of American technologies and chips to Chinese companies, will be announced.
In response to the case, new President Donald Trump called the rise of Chinese company DeepSeek “a wake-up call” for the US technology industry, a sign that a reaction is coming soon.
We will follow the topic closely. In the meantime, let us know Comment: Do you think DeepSeek can take out the ChatGPT leadership?
See the video 81n20
See also other features
With information: Yahoo l Groundd News l InfoMoney l Forbes (1 e 2)
