After the explosion of Deep Seek, is AI Arms Race 2.0 coming?

Wen| Internet things

During the Spring Festival, the only player who can compete with the Cyberyangko Team for popularity may be Deep Seek.

After all, the former brings us physiological shock from silicon-based life; the latter brings us intellectual shock from silicon-based life.

AI Arms Race 2.0 is coming

As an artificial intelligence start-up, the in-depth search from Hangzhou can be said to have made the most of the limelight during the Spring Festival.

On January 27, the company’s application Deep Seek rushed to the first place in the free app rankings of Apple in China and the United States. While pressing Chat GPT to become the number one AI brother in the minds of users everywhere, it brought a huge industry shock to U.S. technology stocks.

At the data level, chip giant Nvidia’s share price plunged nearly 17% that day, semiconductor company AMD’s share price fell 6%, Broadcom’s share price fell 17%. Well-known names such as Microsoft, Meta, and Google’s parent company Alphabet were also involved in this incident. There have been varying degrees of stock price fluctuations.

The reason is that DeepSeek-V3 offers extremely low training costs for performance comparable to top models in the industry such as Claude Sonnet 3.5 and GPT-4o. Using the products in front of you to make AI companies and technology companies around the world realize that blindly accumulating computing power is essentially a powerful behavior, and it is a blind consumption of company capital reserves and investors ‘capital injections.

After all, in addition to the DeepSeek-V3, the DeepSeek-R1, which has developed less than US$6 million in basic computing capacity in two months, can also compete with models that have spent hundreds of millions or billions of dollars to develop. Once again, it has been proved with strength that a sufficiently sophisticated algorithm can achieve a balance between performance and cost, so that the development cost of artificial intelligence models is no longer measured in hundreds of millions.

An article published by the Jiangsu Province Association for Science and Technology on February 6 pointed out that at present, three established technology giants, Nvidia, Microsoft, and Amazon, have announced that they have reached a cooperation with Deep Seek and formally decided to connect to DeepSeek-R1 to enhance their product competitiveness in the AI era.

Also in the domestic market.

An article published by People’s Post and Telecommunications on February 7 also pointed out that the three major operators (China Mobile, China Unicom, and China Telecom) have announced full access to Deep Seek and hope to integrate it with their own platforms and resources. In this way, we will open up Deep Seek’s multi-scenario and multi-product application ecosystem and accelerate the realization of the vision of AI inclusiveness.

Deep Seek salary exposure

During this period, what attracted netizens ‘attention was not only the interactive capabilities of the Deep Seek series of models, but also the salaries and benefits of Deep Seek team members.

Through a survey on the quantitative magic squares of Deep Seek’s parent company, some media found that its current recruitment focus is data R & D engineers, deep learning researchers, and core system R & D engineers, and most of its work locations are in Hangzhou or Beijing.

In terms of salary and benefits, in addition to adopting the 14-salary system, the company also provides a starting salary of 20,000 +/month for most positions. As for more competitive positions such as Deep Learning Researcher (AGI), their monthly salary has been pushed to 80,000 – 100,000 per month.

In addition, the relatively high-paid positions in the Deep Seek team include data R & D engineers and core system R & D engineers, with salary ranges of 40,000 – 70,000 per month and 70,000 – 80,000 per month respectively. Even the salary of interns in AGI positions is as high as 500- 1,000 per day, which can easily achieve the life goal of a monthly salary of tens of thousands.

It is worth mentioning that Luo Fuli, the AI genius girl recruited by Lei Jun at the expense of tens of millions of annual salary, is also a former member of the Deep Seek team (she once participated in the DeepSeek development project of DeepSeek V2 as a deep learning researcher). Lei Jun’s decision to hire Luo Fuli to serve as the leader of Xiaomi’s AI model team also shows us to a certain extent how recognized Deep Seek members are by various technology companies.

The “2024 Talent Migration Report” pointed out that among the TOP 20 hot recruitment positions in 2024, 5 are related to AI-related scenarios. Among them, positions such as algorithm engineer, large model algorithm, natural language processing, and artificial intelligence engineer are all popular positions.

In addition, as of December 2024, the number of users of generative artificial intelligence products in China has reached 249 million, accounting for approximately 17.7% of the total population. Faced with huge user needs and user scale, the current job market model and artificial intelligence-related talent reserve are also slightly insufficient. As a result, some industry reports predict that in 2030, China’s AI talent gap will reach 4 million.

All of the above information also makes us realize that for a long time in the future, not only will the artificial intelligence major become the T1-level choice for students, but the treatment of relevant track employees will also be further amplified with the scale and needs of users. Become an object of envy.

AI can’t escape the popularity of people

As the saying goes, people are popular, and AI is no exception.

The popular Deep Seek has not only attracted the attention of investors and regular users, but also attracted a small number of people with ill intentions.

On February 6, Qianxin released a report stating that during the period from December 1, 2024 to February 3, 2025, a total of 2650 websites that counterfeit Deep Seek appeared. These people hiding behind the copycat website either hope to steal personal information in the name of Deep Seek, or they want to spread malware in the guise of Deep Seek to defraud subscription fees.

In addition to registering phishing websites, there are currently a large number of paid courses on the market under the banner of Deep Seek taking you to make money while lying down and how to make 1 million yuan with Deep Seek Even communities where AI enthusiasts communicate on weekdays have taken the opportunity to launch Deep Seek’s paid exchange group to accumulate wealth in the name of paying membership fees and interacting with industry leaders.

The emergence of these situations also reminds us that for ordinary people, the first and most important step in playing Deep Seek well is to learn to identify information on the Internet and help themselves avoid those copycat websites and cutting leeks. Paid tutorial written on their faces with three words.

As Bao Ran, vice chairman of the Interactive Media Standards Promotion Committee of the China Communications Standardization Association, said, ordinary users do not need to purchase so-called tutorials at all, because using Deep Seek no longer requires polishing the prompt words as before. Now all major mainstream AI can communicate directly with it in natural language.

In the past, the core concept behind the rise of paid AI tutorials was the existence of information gaps.

Due to the relative immaturity of the technology, models at the time often required users to enter a specific instruction set to get the feedback they deserve. That’s why people summarize their experience in conversations with AI and compile it into specific speaking techniques and writing formats to help Xiaobai users improve their experience.

However, with the full development of model distillation and data training processes, Deep Seek’s dialogue capabilities have been fully upgraded today. So-called structured dialogue tutorials have become a thing of the past. Users only need to provide simple spoken and daily instructions to get the feedback they want.

Not to mention that there are many AI training courses on the market that just piece together public information on the Internet. Some sellers don’t care about the quality of the content at all. They just want to make a fortune when the gold rush comes.

reference：

Jiangsu Province Association for Science and Technology: Three companies announced on the same day that they would access DeepSeek, and competition in the AI field entered a new stage

People’s Post and Telecommunications: The three major operators fully connect to DeepSeek to accelerate the development of AI inclusive

Yangpu, Shanghai: DeepSeek recruits with high salaries: interns earn tens of thousands a month, and researchers earn millions a year! There is a shortage of 4 million such talents→

Qilu One Point: More than 2000 copycat DeepSeek websites have appeared, and DeepSeek has responded!

Extreme News: The first wave of people to use DeepSeek to make money has emerged