news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

"When I graduate at the age of 24, my annual salary will be 500,000 yuan. By the age of 30, I can probably be promoted to P7 (note: job title), and then I can earn a million yuan a year."

After graduating with a master's degree from Shanghai Jiao Tong University, Zhao Hong, who was born in 2000, joined Tencent this year as an AI algorithm engineer, becoming one of the first young people to be employed in the AI ​​boom.

As artificial intelligence (AI) becomes increasingly popular, the "100 Model Wars" have begun fiercely, and AI talents have set off a recruitment boom. According to data from the job search and recruitment platform Liepin, in the first quarter of this year, AI-related positions increased by 321.7% year-on-year, and the number of talents applying for this field increased by 946.84% year-on-year. At present, the most in-demand large model algorithm positions have a talent supply-demand ratio of only 0.17, which is roughly equivalent to 6 positions competing for 1 talent.

In terms of salary, an annual salary of 500,000 to 700,000 is the general price, and an annual salary of over one million or even between 2 million and 3 million is not uncommon in the industry.

After leaving the grassroots stage, large models have gradually become a fiercely competitive red ocean. According to third-party data, as of the end of April this year, a total of 305 large models have been launched in China. In addition to the top manufacturers, there are alsoDark Side of the Moon、Baichuan Intelligence、Zero One Everything、MiniMax、Zhipu AIIn this competition, whether it is financial background, technical strength or talent reserves, it has become a battlefield for technology companies to fight fiercely.

Has a new round of "war for talent" begun? Many industry observers said that the Internet ten years ago was called "grabbing talent", while the current big model is "careful selection".

"In the Internet era, when salaries were raised to attract talent, there were cases where salaries were doubled or even 2.5 times higher than the original salary. But now in the field of big models, 'recruiting talent' is more pragmatic and rational," Maimai CEO Lin Fan told reporters. "Although salary increases have generally reached more than 50%, no company will spend money desperately to attract talent anymore."

Recruiting "few but elite": annual salary of top university recruits exceeds 2 million

"As a graduate student joining a large company, the lowest annual salary is about 400,000 yuan, and the highest is about 550,000 yuan." Zhao Hong revealed that his undergraduate major was electronic chips, and he switched to the AI ​​industry through an internship at Huawei. It is not uncommon to see colleagues around him who have transformed from traditional mathematics and engineering.

"700,000 yuan is generally considered a high salary for graduate students. For a PhD, the salary after graduation can reach one million yuan." Ni Yue, a headhunter who has deep experience in the AI ​​industry, revealed to reporters that it is common for large companies to offer annual salaries of more than 2 million yuan to PhD graduates from the domestic C4 (Tsinghua, Peking University, Fudan, and Jiaotong University) key laboratories.

However, such high salaries are only limited to core technical backbones. "90% of the domestic talents who work on base models come from Tsinghua University, and there are no more than 200 people who can really adjust and train models." Liu Chenghui, founder of Yizhe AI, said that due to the limited talent pool of the industry itself and the sudden popularity of large models, startups need to "lock" relevant resources in core laboratories in order to recruit the best technical talents.

"There are not many truly valuable AI jobs," Zhao Hong admitted. "The core code of large models mainly relies on a few great people. Many companies offer AI jobs that require installing and running open source code, adjusting parameters, and doing a lot of repetitive work." Compared with large Internet companies with tens of thousands of employees, AI unicorns generally have only a few hundred employees.

As one of the first people to catch the AI ​​boom, Zhao Hong was approached by top companies including Huawei, Alibaba, and Tencent during this graduation season. “Huawei spent three hours on the phone trying to retain me, hoping that I would stay in technology for a few more years so that I can go further in the future.” But the atmosphere at both Alibaba and Huawei was more scientific research-oriented. “I added the WeChat account of Alibaba’s supervisor and found that he was holding various international conferences around the world every day. I was very nervous at the time.” Because he preferred the application direction of large models, he ultimately rejected Alibaba and Huawei and chose Tencent.

"From the current market situation, the competition for AI talents among employers, especially high-tech companies and Internet companies, has reached an unprecedented level of intensity." The head of a recruitment company revealed to reporters, "In order to attract and retain AI talents, most employers have offered high salaries. These high salaries have undoubtedly intensified the competition in the AI ​​talent market. Even some large companies are no longer limited to the domestic market, but have expanded their search scope to the world to strengthen their research and application in the field of artificial intelligence."

Maimai data shows that the average monthly salary of large model R&D technical personnel has exceeded 54,000 yuan. Among the top ten high-paying jobs, the average monthly salary of new large model algorithm jobs ranks first, exceeding 67,000 yuan, followed by analog chip design and digital front-end engineers.

The talent war is surging, ByteDance has many positions and Huawei has high salaries

Yan Junjie, founder of unicorn MiniMax, once said in an interview that there will only be five large-scale model companies left in the world in the future. In the AI ​​era, the market share of large companies and start-ups may be even more miserable than that of Internet or mobile Internet companies. "The extreme ratio may reach 9:1."

According to the analysis, there have been 107 financing events in the global AI field this year, among which 20 domestic large-scale model companies have raised hundreds of millions of yuan. Five of the "Six Little Dragons of AI" (Zero One Everything, MiniMax, Baichuan Intelligence, Zhipu AI, Step Star, and Dark Side of the Moon) have received hundreds of millions of yuan in financing this year.

AI has become a battleground for Internet giants. In the talent war, ByteDance is a relatively aggressive player. According to Maimai data, ByteDance has ranked first in the number of newly released AI positions for the fourth consecutive year since 2021. Among many large companies, ByteDance has the highest recruitment index for big model talents, reaching 17.24, becoming the company with the most newly released big model positions.

In the first half of this year, among the top ten companies recruiting the most AI talents, ByteDance ranked first, Xiaohongshu ranked second, followed by Ant Financial, Meituan, and Huawei. In terms of salary, Huawei offered a monthly salary of 55,000 yuan to big model talents, becoming the largest company with the highest salary for newly-issued big model positions.

"Star employees" from large companies have become the core source of entrepreneurial teams. In February 2023, Wang Huiwen, former co-founder and senior vice president of Meituan, took the lead, followed by executives from major Internet companies such as Baidu, Alibaba, Tencent, ByteDance, JD.com, and NetEase, who successively started large-scale model entrepreneurship.

Among them, Jiang Daxin, former Microsoft global vice president, Wang Yuan, former NetEase vice president and executive director of Hangzhou Research Institute, and Jing Kun, former Baidu vice president, have been exposed to have raised large amounts of funds. In July this year, Yang Hongxia, a technical expert in the development of large language models at ByteDance, and Zhou Chang, a technical backbone of the Alibaba Tongyi large model team, also resigned from large companies and joined the wave of AI entrepreneurship.

However, compared with the rather brutal poaching tactics in the Internet era, the talent competition in the AI ​​field now seems to be undercurrent. "At present, the technical teams of major companies are relatively stable. The core talents are given an annual salary of about 3 million in cash, and the rest are paid in the form of stocks." Ni Yue revealed, "Currently, non-compete agreements are relatively complete, and core talents are protected by both salary and non-compete. Technical bosses often bring their own teams when they change jobs, and the cost of poaching is very high."

In addition to algorithm talents, talents in brand, market, and marketing have also become the next recruitment focus of AI companies. "Many consumer-oriented startups need to further implement their products, so they need to establish a complete promotion team. Compared with algorithm positions with extremely high barriers to entry, these fields have more opportunities." Ni Yue said.

"Today, talents choose large companies based on the various major model technology schools, not just salary." Lin Fan said that talents will first consider whether their judgment on the future mainstream technology route is consistent with the company, and different routes also represent different development directions.OpenAI, Google, and Amazon each have different routes. In addition, whether the company has sufficient computing power and high-quality data reserves will be the priority considerations for talents.

Behind the craze: Is it worth investing heavily in AI? Can it be commercialized?

How to transform technology into commercial returns has always been the sword of Damocles hanging over AI companies.

According to foreign media reports, OpenAI's total operating costs for the whole year are as high as US$8.5 billion, of which nearly US$4 billion is spent on renting Microsoft servers. OpenAI expects total annual revenue to be approximately US$3.5 billion, which means that OpenAI's revenue and expenditure gap may be as high as US$5 billion.

The core investment cost of AI companies is computing power, not manpower. The investment of large-scale model companies in computing power basically exceeds 100 million US dollars. "I am not particularly optimistic about the prospects of large models." Zhao Hong told reporters frankly, "Because each training costs a lot of money, training once costs tens of millions of US dollars. At present, the technology is still in its early stages. In the future, as the technology gradually matures, companies may not be willing to spend such high resources to invest."

The investment community is more cautious about AI companies. Liu Chenghui told reporters that the current investment environment is more severe and the "water temperature" has changed significantly. Investors are more cautious when looking at projects. When examining projects, they not only look at growth space and future value, but also have higher requirements for short-term income.

When it comes to company selection, investors prefer "companies on the table", and the core of "companies on the table" are universities such as Peking University and Tsinghua University, which have deep technical backgrounds. Otherwise, it will be difficult to gain the favor of top investors.

Application layer companies must achieve stable revenue, which means that the company must truly create value for customers. The size of revenue is not the only criterion, but the company must ensure that the revenue is standardized and replicable, which is a big challenge for most companies, because finding application scenarios is still a big problem for large model companies.

"Funding is now showing a trend of head-to-head and centralized development, and it is difficult for small companies to continue to raise money." The CEO of a listed company that has been investing in AI for a long time said, "Capital is beginning to expect returns, competition is becoming increasingly fierce, user scale growth and retention are slowing down, and companies are caught in comprehensive considerations of technology, cost, and market. Most companies have not formed a stable business model and landing scenarios, and the industry reshuffle period may soon come."

"In the past year, the trend of cost reduction has been the most significant in the implementation of AI applications." Zhou Zhifeng, managing partner of Qiming Venture Partners, which has invested in more than 50 AI companies, told The Paper that the cost of calling a large model per million tokens has dropped from $120 (about RMB 800) last year to less than RMB 1 this year, almost a thousand-fold drop. According to estimates, the cost is likely to drop another 1,000 times in the future. AI-related costs will become lower and lower, and the implementation of technology will become easier.

Zhou Zhifeng believes that in the future, many new application areas based on big model technology will emerge, such as AI search, AI games, AI social networking, etc. At the same time, many new content platforms will also appear, such as information, comics, short dramas, role-playing interaction platforms, etc. "In the Internet era, China has bred many technology giants. In the era of generative AI, this phenomenon will surely reappear. China's new generation of great technology companies is already taking shape."

(Zhao Hong and Ni Yue in this article are pseudonyms)