GuShiio.com Wen| Interesting understanding of business, author| Liu Liang,edit| Fun Understanding Business TMT Group
“Artificial intelligence is not as smart as most people hope or fear, and its profound knowledge of the world must be supplemented by humans. rdquo;
This is what an American anthropologist said in the book “Ghost Work”, and netizens also gave a very straightforward translation of this academic description.“Artificial intelligence, there is as much artificial intelligence as there is。& rdquo;
When GPT first became popular, everyone was discussing whether they could keep their jobs. Now, everyone is discussing whether they can do the jobs created by AI.
After three years of the explosion of artificial intelligence, trend changes have taken place from enterprises to the talent market.
Various changes have also brought anxiety and uneasiness to some people in the workplace, especially those in the non-engineering and technical professions, as well as those in liberal arts and business students. While they are trying to squeeze into the AI track, they are worried about whether the popularity of AI will one day be replaced by new things…
01. AI solutions for liberal arts students”
On an ordinary weekday night, Shasha (pseudonym) received a call from a headhunter. Shasha was very surprised because the headhunter served Kimi and the recruitment position was a data tagging specialist.
After graduating from finance, Shasha has been engaged in content work in financial media in Beijing. She really didn’t expect that she could also enter the AI industry with the highest technological content. Moreover, she is also a deep user of Kimi, and her expectations have increased again. At this time, she didn’t know that ten minutes later, the call would be hung up with a completely disappointed ending.
“Ten to eight in the morning, six days a week, big and small weeks, I don’t make as much money as I do now。& rdquo; Regarding the reason for disappointment with this phone call, Sasha simply pointed out that the hourly salary was too low.
As for the content of this job, Shasha concluded:That is, the artificial person who uses artificial people to train intelligence”。
During the training process of generative artificial intelligence, a continuous process of error correction is needed; taking the position where Shasha is communicated, because she is engaged in financial-related work, Shasha is contacted by Kimi’s financial data annotation position. The main job is to generate AI content and use manual professional skills to correct errors.
“For example, when searching Tesla’s 2023 financial data and finding ideal 2023 data, I have to manually tell the AI that it was wrong.” rdquo; Sha Sha gave an example.
Just as Shasha refused the job offer, Ding Ding (pseudonym), who was far away in Fujian, was marking the data. ldquo; I label one thing almost every day in my job.” rdquo; Ding Ding explained that for example, autonomous driving companies need to tell AI what traffic lights are called, and there are all kinds of traffic lights with different shapes that appear at intersections in the world. My job is to mark these pictures with text. rdquo;
Ding Ding said that he works in an Internet outsourcing company, specializing in data annotation;“Highly repetitive, requiring a high degree of concentration, and tiring are the key words he describes this job。
As the saying goes, only how much labor there is how much intelligence there is. Data taggers are a new profession that has developed with the artificial intelligence industry; according to interesting business understanding, in February 2020, data taggers were a type of work for artificial intelligence trainers., was officially included in the National Occupational Classification Catalog, and now more than 10 million people are engaged in this profession.
Source: Weibo screenshot
Generally speaking, basic data annotters do not require too high a degree, and some positions do not require a bachelor’s degree; when training some vertical models, such as those involving law, finance and other fields, relevant professional positions are needed.
Compared with positions such as chip research and development and model building, data annotters have also become a good opportunity for liberal arts students to enter the AI field because of the lower threshold.。& ldquo; Many students around me from the Chinese department have gone to work as data annotators after graduation. rdquo; Xiao Qi (pseudonym), born in 2000, said this.
Also because the threshold is lower, the data labors have become the first to sink in AI positions.
“CCTV has previously reported that in a small county town with a permanent population of only more than 20,000 people in Yonghe County, Linfen City on the Loess Plateau, there are more than 100 data talkers, 70% of whom are Bao Ma. According to interesting business knowledge, some large factories stationed in first-tier and second-tier cities will choose to outsource data labeling to companies in sinking cities in order to save costs; andThe overall market size of data annotation has also increased from 3.63 billion yuan in 2020 to 6.937 billion yuan in 2024。
Source: Guanzhi Haihai Information Network
02. The ever-changing AI
These data talkers who entered the artificial intelligence track also had their own considerations.
“Work is tiring, and I don’t earn as much.& rdquo; Ding Ding said that he felt like a walking Captcha scanner, and he felt a little tired after a long time.” rdquo; Ding Ding, who has been in this outsourcing company for more than two years, is considering leaving this year. If he continues to work like this, he still feels that there will be no long-term development. I want to open a similar outsourcing company myself and pool my own resources.” rdquo;
Shasha is more concerned about job stability. For three consecutive days, at least five headhunters have communicated with me about the same data labeling position, which shows that this need is urgent. The urgent need is because this big model currently lacks training on financial content, but if the training on this kind of content is complete in the future, I will have such concerns as whether I will face the risk of being dismissed.” rdquo;
Their thoughts are not superfluous. After all, we have seen too many stories in the AI industry in the past of the Yangtze River.
Take Nvidia for example. Since the birth of DeepSeek, Nvidia’s share price has suffered a sharp decline; it has dropped from US$3.28 trillion at the end of 2024 to US$2.86 trillion now, setting a historical record of nearly US$600 billion in a single day.
Source: Screenshot of Baidu Stock Market
AI is developing too fast.
Xiaoxue (pseudonym), a self-media person who uses AI tools at work almost every day, has deeply felt the progress of generative AI capabilities in the past two years.” ldquo; I have used everything from ChatGPT, Wenxinyan to Kimi, IMA, and DeepSeek. The functions and usage methods of these AI tools are similar, but there is a big gap in capabilities.” rdquo;
Xiaoxue classifies AI search capabilities into three points based on the needs of a writer: accuracy, logic, and anthropomorphism。& ldquo; When ChatGPT and Wenxin first became popular, the probability of AI talking nonsense was still quite high. For example, asking about the financial situation in 2023 and giving me feedback for 2022, or asking a car company about the model launched by it. Output a completely non-existent model, which is also the so-called unexplainability of a large model, which will inevitably output some absurd answers. rdquo; Xiaoxue recalled, but this year the probability of these outrageous answers has been significantly reduced, and the large model has become increasingly capable of disassembling and exporting long and bilingual texts. rdquo;
Talking about the stickiness of AI products, Xiaoxue said with shame that she can use whoever is easy to use, and there is no stickiness at all. Now I use Kimi and DeepSeek the most, because DS can more flexibly adjust the style of the output language, saving me the process of personifying AI later;Kimi is more rigorous and more logical. rdquo;
As for why it is not highly sticky to an AI product, Xiaoxue believes that the product use process is similar and the learning cost is too low. However, the improvement of capabilities after iteration of different AI products can greatly improve work efficiency, so the cost of changing products is low and the benefits are large.
The fickle nature of users also makes big manufacturers feel very insecure.
According to statistics from interesting business combing,The trend of capital expenditures on AI by global core cloud service giants (meta, Google, Microsoft and Amazon) has continued to grow in the past six quarters, with high before low after high”This reflects that international giants ‘attitude towards AI is to increase investment while also waiting to see changes.
Data source: Dolphin Investment Research
The same is true in China. DeepSeek’s amazing performance has caused all major manufacturers to increase their D content, and it has also sounded a wake-up call for major manufacturers: as the technological paradigm shifts, the industry landscape may be reshaped at any time.
The sea of AI is so blue that whoever swims out first can seize the user’s mind. No one who has been killed in the Internet era does not understand the value of the user’s mind; so when they see the emergence of hot items on the market, there is really no time to spend on self-research, and we can’t help but care about the opportunity cost (potential users) of loss.
03. Can AI add weight to large factories?
In recent years, major Internet companies have been slimming down.
Financial report data shows that Alibaba’s total employee number has dropped from 259,000 at the beginning of 2022 to 194,000 at the end of 2024. In three years, it has decreased by 65,000, basically at a rate of 60 people per day; Baidu will reduce its workforce by about 8800 people in 2024, and its research and development expenses will be cut by 2.1 billion yuan.
Source: Weibo screenshot
Reducing costs and increasing efficiency, divesting non-core businesses, and focusing on core businesses have also been key words in the Internet circle in recent years. The emergence of AI seems to have given big Internet companies a new second curve, and with it comes the demand for talents.
According to interesting business observations, DeepSeek’s popularity has made big model talents popular during the post-holiday spring recruitment season.Recently, Alibaba launched the spring 2026 intern recruitment, opening more than 3000 positions, of which AI-related positions account for nearly 50%; Tencent Yuanbao has also opened a series of social recruitment positions from AI product managers to algorithm engineers; A person familiar with ByteFlying Book also said that Flying Book has been studying the combination of AGI, such as testing that the content of the meeting recorded by the clever plan is directly converted into a text version and adding it to the to-do list based on the key content of the meeting.
“Pulse High Recruitment shows that many DeepSeek executives are also seeking resumes online on the Pulse High Recruitment platform, involving multiple positions such as large model research and development, finance, legal affairs, and interns. Among them, the maximum annual salary for AGI deep learning researcher positions exceeds 1.2 million yuan.
Source: Pulse screenshot
In addition to Internet software applications, the upstream semiconductor manufacturing industry and the downstream ToC consumer industries such as home appliances, communication equipment, and smart hardware have also become the fastest growing track for AI technical talents in recent years in the atmosphere of AI+;“Hunting data shows that the demand for AI technical talents in the home appliance industry increased by 93.75% year-on-year, while communication equipment and intelligent hardware reached 38.61% and 37.22%。
Photo source: Liexian.com
Changes in talent demand correspond to changes in people’s attention in the workplaceBeibei (pseudonym), a marketing manager who used to be engaged in the consumer entertainment industry, has recently been transforming to AI. Traditional AI Four Little Dragons such as Shangtang, newly developed startups such as the Dark Side of the Moon, and consumer companies that have deployed specific intelligence. For example, I have interviewed for pursuing companies. Although the tracks are different, there are some underlying logic in marketing. I am very optimistic about the future development of AI-related industries, and I hope that my career can turn to AI as much as possible.” rdquo;
There are not a few people in the workplace who plan to change careers or learn to use AI tools at work like Beibei.& ldquo; Zhaopin.com ‘s survey of professionals in the first week of the spring recruitment 2025 shows that60% of working people already use AI tools at work, which is significantly higher than 44% surveyed in the same period last year.; Among them, 12.3% of professionals frequently use AI in depth, 29.6% use it occasionally, and 18.1% are learning to use it.
Photo source: Zhaopin Recruitment
Not only professionals, but capital is also looking for potential entrepreneurial talents。An investor in the primary market of hard technology said: We used to invest early and small, but now we are going deep into college teachers and students to find underwater projects and pay attention to talents on the AI track for a long time. If they want to start a business, invest the idea as soon as possible.
“AI should be long-term, the investor said. Unlike the era when the Internet was just emerging, you can make money by building a random website, because nowadays users ‘learning ability is too strong and their requirements for products are higher, and what AI currently shows are all superimposed advantages and have not yet reached the point of creating demand. rdquo;
Whether it is a company or an individual, continuous learning, continuous investment, and long-term thinking of multi-AI is a smart choice.
