AI made me unemployed twice

Source: “Economic Observer” (ID: eeo-com-cn), author: Shen Yiran

Image source: Generated by Unbounded AI tool

Introduction

**一|| It takes 3-5 days for a junior painter to draw a commercial illustration, while AI only takes a few seconds. Wenshengtu, Tushengtu) professional package $60/month. **

**Ⅱ|| Large-scale enterprises with a monopoly position in the industry will purchase large-scale industry models for privatization or self-develop large-scale models. The cost is high, the training cycle is long, and the industry model technology itself needs to be improved. Such enterprises will also be interested in AIGC more cautious. There is another category of enterprises that choose to wait and see due to unclear policies. **

**三|| There is a huge knowledge base behind AI, and only with professional knowledge can we be able to mobilize it. **

**四|| The degree of mastery of AI also greatly determines the ability of people to use AI, and those who perceive the opportunity start to learn this skill. **

Because of AI, Li Mei (pseudonym) lost her job twice in a row.

“Unfortunately, remember to come to the office on the 6th of this month to go through the resignation procedures. I wish you a satisfactory job as soon as possible.” At the beginning of May 2023, two months after the introduction of AIGC (Generative Artificial Intelligence), an art concept design company in Shenzhen started a round of layoffs of original painters.

Li Mei is one of them. She graduated from the Academy of Fine Arts of Jiangxi Science and Technology Normal University and has been a painter for 4 years. According to her, 8 of the 10 people in the company’s art original painting team were laid off, and the remaining 2 people were cut in salary. Artists in other project teams were required to learn AI tools and included in the assessment.

After being laid off, Li Mei joined a small-scale entrepreneurial team as a graphic designer. The boss needed 9 cards to design a game app. Each card was paid 1,000 yuan to design an illustration, and the total payment was 9,000 yuan. .

She was fired by her boss before she finished the painting. The other reason was that the project was in a hurry and decided to use AI for drawing.

Since ChatGPT came out, public opinion’s attention to AIGC (generative artificial intelligence) has been cooling down. It seems that this is just a competition among giants, and it is still far away from ordinary people. But in fact, it is gradually being used by Chinese business owners, and it is quietly triggering a round of manpower changes. Some white-collar groups such as painters, new media operations, and sales are losing their jobs or being transferred due to AI.

The impact is first in the content creation industry. Open the recruitment software, and some companies are recruiting AI painters, AI video clips, and AI drawing positions. Posters in subway stations and product advertisements have begun to be generated by AI. In more professional fields such as life sciences and finance, the pace of introduction is also accelerating.

The tense economic environment is forcing business owners to accelerate the introduction of new technologies. AIGC tools represented by ChatGPT and Midjourney are inclusive and easy to use, and can also replace the jobs of some junior white-collar workers, which is like a “life-saving straw” for them. Bosses are trying to find space for AI application in the original process, but they are worried that they will “go too far” when the technology is not perfect and the situation is unclear.

When a seemingly distant cutting-edge technology suddenly “approaches” itself, some people have resistance, and some people accept this new production tool and start to learn (prompt words) skills in order to control this tool. Unknowingly, the replacement of AI has created some jobs, and the bosses and migrant workers have lost their minds and also ushered in opportunities.

AI is approaching to beat workers

“I seem to have been killed by AI.” Li Mei said after experiencing unemployment twice in a row.

It takes 3-5 days for a junior painter to draw a commercial illustration, while AI only takes a few seconds. Figure) Professional package $60/month.

At work, the artist has to communicate and revise with the client for many rounds, and enter a sentence in Midjourney’s dialog box, such as “I want a cat in a spacesuit to eat pie in space”, or disassemble it into "littlecat, in “aerospace suit, eating pies” (kittens, wearing spacesuits, eating pies), “sharp focus” (focus) and other multiple (prompt words), you can get works of the same level.

In terms of drawing quality, drawing efficiency, and cost, a fledgling artist is simply incomparable with AI. In terms of workflow, there is a need for repeated communication between the planner and the artist, and now it has become the communication between the planner and AI, and the artist and AI.

Li Mei said that the company first let employees use it by themselves, and later trained all employees to use Midjourney and skills. She is not willing to take the initiative to learn, and many people are still pretending to be calm.

Until it was discovered that the company fed some of the painters’ pictures to AI in order to obtain works based on the styles of these painters, or collected the corrected pictures of the painters and returned them to AI for learning-“picture generation” is the basic function of this type of large model. Feed the graph to the model, and the model feeds back a new graph on this basis through self-learning. This kind of interaction is becoming common. “Its ability is no less than that of human beings, and we are all shocked.” Li Mei said that if you feed some semi-finished products of painters, it is also very good at perfecting details such as light and shadow and lines. Faced with AI, many colleagues are out of balance, but the copyright of the artist’s works belongs to the company, so there is nothing for everyone to say.

Chen Xiyan, CTO of Glance Technology, said that the capabilities of AI are increasingly recognized by the industry. What it does is not trivial work, but delicate and creative things. But the prerequisite for making AI more powerful is the operation behind it. To be more imaginative and creative.

Tian Feng, dean of SenseTime Intelligent Industry Research Institute, told reporters that AI will have an impact on the skills of some content creation positions in the short term. Taking painters as an example, junior painters are mostly engaged in relatively simple basic work and are easily overwhelmed by new productive forces. With the integration of tools, the senior painter who completes the advanced design work is the commander of the project project, and it is difficult to replace talents with compound skills. On the surface, it affects the career path, but in fact it changes the way of getting started. Future painters must learn to draw and use AI. The essence of AI is not to eliminate manpower, but to innovate human production tools. From using pens, using software to using AI, the artist is an evolution of tools.

Li Mei said that it usually takes three years for a painter to advance from elementary to intermediate, and another three years to advance to advanced. When a junior painter loses the channel for advancement, how can he become a senior without practice opportunities? Some people around him have already considered changing careers.

BOSS’S CHOICE

Zhang Yuan (pseudonym) is the manager of a small and medium-sized ToB enterprise. The company is mainly engaged in the business of R&D and sales of industrial software. He plans to do two major things for the company, one is to reduce costs and increase efficiency, and the other is to introduce AIGC. “I planned to use AI to redo every business of the company, but it turned out to be a major adjustment.”

The company has a 4-person team responsible for new media operations. After a month of training, one senior operator was selected to stay, and the remaining three ordinary operators were optimized. The current job is an operator + large model (ChatGPT generates text, Midjourney generates pictures), people are responsible for creating ideas and connecting with customers, and machines are responsible for execution. In Zhang Yuan’s view, the efficiency in practice is not low at all.

First feed the data to the AI large model, and then generate customized content. According to this principle, the large model can also generate customer proposals and bidding documents. So Zhang Yuan aimed at the sales assistant. The job of this position is to help the salesman write plans and bids. After training the salesmen, he has optimized several sales assistants, leaving only one salesman for each business line. He visits customers during the day and writes plans and bids with ChatGPT at night. .

Zhang Yuan’s inner thought is that sales can also be optimized, because ToB company has few customers and stable relationships, and customers can be handed over to the management to connect them. A fresh graduate can be hired to specialize in AI operations, but considering the emotions of employees, let sales stay He said, “The management did not expect so many people to reduce costs and increase efficiency this time.”

Zhang Yuan told reporters that before optimizing personnel, the system requires all employees to learn the prompts of large models, but he found that many people still reject new technologies and don’t like to use them unless they touch their own interests.

As a one-stop video solution provider, Glance Technology has developed a variety of AIGC tools that empower video creators, applied and implemented these tools in various business processes within the company, and conducted prompt word training for all employees.

Talking about the process of implementing AIGC tools, Chen Xiyan, CTO of Glance Technology, said, “I have seen the internal stratification of employees. Some people use it smoothly, and some people don’t. This is a change. Those who don’t know how to use AI are doomed. It will be replaced by someone who can use AI.” His vision for evaluating talents is also changing, and he is considering raising the requirements for employees’ work efficiency and average output. He believes that in the future, many business owners will raise their performance standards and evaluate them according to the “human + AI” standard. Of course, he is also hesitating, “Now is the stage of change, not ‘one size fits all’”.

In the cold winter, more companies reduce advertising expenditures, and the content creation industry dominated by animation media generally shrinks. On the one hand, the use of outsourcing companies is reduced, and on the other hand, staff are reduced through performance. Artists are the smallest unit of this industry.

After optimizing a group of personnel, Zhang Yuan is also reflecting on “too big a step”. The dissatisfaction of beating workers is also a kind of stress. The essence of AIGC is not to replace manpower. Perhaps the economic environment has made some business owners radical . AIGC is very inclusive. The general large-scale models on the market only require a very low monthly fee, and some are even open source. They are like a “life-saving straw” for business owners.

Zhang Yuan believes that the normal path should be to train all employees, empower the business, let everyone take on a new role, and redo all products with AI.

Zhangyuan Enterprise was the first to introduce AIGC in the entire chain, and customers and suppliers were very interested in it. What did you do?" Zhang Yuan said.

Zhou Hongyi, the founder of 360 Group, told the Economic Observer that from the perspective of network security, when using large models, enterprises can “take a smaller step at the beginning”, don’t open APIs to it, insist that “people are in the loop”, and let people’s will Play a key role in decision-making.

Zhou Hongyi said that on the whole, enterprises need to take multiple steps from digitalization to intelligence, and it is recommended to adhere to the strategic principles of loose coupling to deep integration and multi-step iteration. It is hoped that enterprises will first enable the “assistant” and “co-pilot” modes when using the large model, so that the large model will remain relatively independent from the existing business system, maintain isolation, and be safer and more controllable for the enterprise.

Career International Business Director Yang Shuo told reporters that different business owners have different attitudes towards the introduction of AIGC, which is related to the business form and industrial status of the company. When the AIGC first appeared, many people were still open-minded, but with the continuous in-depth understanding, some tended to be conservative and cautious. For example, content creation work and creative work require more breadth and divergence, often because their businesses have a higher fault tolerance rate for AIGC; business owners in biogenetics, medical care, and finance are relatively conservative, and the fault tolerance rate for these businesses is very low; especially in some commercial On the key points of the scene, business owners dare not try easily.

Yang Shuo said that in addition, large-scale enterprises with a monopoly position in the industry will purchase large-scale industry models for privatization or self-develop large-scale models. The cost is high, the training period is long, and the industry model technology itself needs to be improved. Such enterprises are also very important to AIGC. will be more cautious. There is another category of enterprises that choose to wait and see due to unclear policies.

Domestic AI

“Join if you can’t beat it.” For two months after being unemployed, Li Mei studied several AIGC tools at home every day, such as ChatGPT, Midjourney, and Stable Diffusion, bought accounts, and learned corresponding prompt words. These AIGC tools from overseas technology companies such as OpenAI have low monthly rental fees. Some functions are free and open, with unlimited usage times. The hardware requirements are not high, and they can run on ordinary graphics cards.

China has also developed large language models such as Wenxin Yiyan, Xunfei Xinghuo, Tongyi Qianwen, and SenseTime’s “Consultation” to benchmark ChatGPT, and SenseTime has also developed “Miaohua” to benchmark overseas Midjourney. “It’s really not difficult to communicate with AI. As long as you know the rules of the prompt words and simply adjust some parameter values, you don’t need to know the code.” AI can directly skip some steps to the finished product, and then refine it, or omit one of the steps that it is not good at, and let AI help you complete it, greatly improving the efficiency of drawing.

Li Mei’s “domestication” of AI is divided into three steps. “In the beginning, I became an AI retoucher and quality inspector.” Li Mei said that human beings express needs, and AI can give a short, flat and fast result. Its paintings are more like “net celebrity faces” with solid basic skills and lack of personality ", which is why AI can replace a junior painter in a short period of time.

In the second step, Li Mei feeds her work to the AI, and the feedback is a fine work with her style, and people familiar with it will not be able to identify whether the author is a machine or a human. Li Mei believes that at this stage, AI has become her partner.

In the third part, Li Mei created outlines and ideas, which were realized by AI. Once she combined the styles of painters from two different countries and eras to make a portrait of Einstein, whose aesthetic value far exceeds that of the past. In contrast, a person who does not understand art history can only use AI to draw Einstein riding a bicycle, motorcycle, skateboarding, and skiing.

Chen Xiyan said that many people do not have enough imagination, and the basis of imagination is sufficient professional knowledge. There is a huge knowledge base behind AI, and only professional knowledge can mobilize it.

Li Mei said that the state of mind has changed from wait-and-see, disappointed, and hopeless to regaining a balance. From assisting the AI, to working with it, to trying to master it. In the next step, she plans to return to the talent market and apply for the positions of AI painter and painting prompt engineer.

NEW SKILLS, NEW OPPORTUNITIES

People who have just used ChatGPT, Wenxin Yiyan, and Tongyiqianqian often have this feeling: AI cannot give you a perfect result with one click, and there is always a gap between the input command and the feedback obtained, so some people question that AI is “serious”. Nonsense", “artificial mental retardation”, or just used as a search engine. But the popularity of AI has made more people realize: it’s not that AI can’t do it, it’s that you can’t use it.

, Prompt words refer to the design and improvement of prompt word instructions for general and industry AI large models such as ChatGPT to ensure that the output results meet human expectations.

If AI is a new production tool, it is the human skill to domesticate this tool. For large language models such as ChatGPT, the interaction method is mainly based on dialog box input. Therefore, this skill seems to be chatting with AI and constantly asking AI questions. The essence is to translate human language into AI language and bridge the gap between AI language and human language. It is also the “last mile” of human-computer interaction.

For example, human language is a sentence with a complete subject, predicate and object, but it is easier for machines to understand when it is disassembled into “task”, “subject”, “details” and “form”. All commands are predicted, and you also order the prompt words in order of importance.

The degree of mastery of AI also greatly determines the ability of people to use AI, and those who perceive the opportunity start to learn this skill.

Since April, OpenAI engineers represented by Wu Enda have jointly produced ChatGPT prompt word courses, and domestic Baidu and other major companies have also given corresponding public courses. Prompt word engineering is not limited to large language models and graphics models. Large models in some vertical fields also need to be trained or tuned by prompt engineering. For example, in biomedicine and finance, since data has changed from language to professional knowledge, skills The bar is also higher.

Tian Feng said that if a business needs to be transformed with AI, software engineers need to convert human needs into programming language, which requires 2 to 3 steps in the middle. Now AIGC has greatly lowered the threshold for developing and using AI. Humans use natural language instead of computers. Language to describe the task, human-computer interaction only needs one step.

Engineers are emerging overseas. Resume Builder, an overseas organization, said that 29% of companies hope to hire prompt engineers in 2023. Among companies with more than 250 employees, 27% of business leaders said they hope to hire more than 10 prompt engineers. When it comes to compensation for the role, nearly a quarter of business leaders said starting salaries would exceed $200,000. For business leaders at companies with more than 1,000 employees, 17 percent said starting salaries would exceed $300,000.

The recruitment market also has shadows of these people. The reporter found from platforms such as BOSS Zhipin and Liepin that technology companies such as major Internet companies and AI companies, as well as traditional companies such as medical care, video animation, and financial property insurance, have recruited for this position. Compared with the AI-related positions of algorithm engineers and software engineers, the recruitment volume is not large, and the monthly salary ranges from 35k-45k to 8k-10k. A common requirement is to be proficient in using AIGC tools at home and abroad to ensure that AIGC can meet the needs of specific business scenarios. High quality output.

Yang Shuo said that compared with overseas, hint word engineers have not yet formed a large-scale group in China. As the first batch of AIGC engineers in the early days, their occupational outline is not clear enough, and the clear salary range in the talent market has not yet been calculated.

Yang Shuo said, simply put, this is a group of people who understand both prompt words and business. The core value does not lie in prompt word skills, but in fitting the business. A truly competitive prompt engineer must have cross-border capabilities. Yang Shuo further explained that he thinks there are two scenarios for prompting engineers, one is business-driven (90% understand business + 10% curiosity about technology and willingness to learn); the other is technology-driven (90% technical ability + 10% % have a strong willingness and motivation to understand the business).

In some verticals, more specialized capabilities are required.

Baitu Biotech is a life science platform company. Its AI large model “xTrimo” can provide solutions for various proteins, cell functions and disease scenarios according to the parameters and functions given by users, to help universities, pharmaceutical companies Enterprise life science researchers improve R&D efficiency. As a large model developer, the team is also training prompt word skills and recruiting relevant talents.

Song Le, CTO of Baitu Shengke, said that compared with the human-computer dialogue of ChatGPT, the human-computer interaction of life science large models is obviously more difficult. Its prompt words include not only human language, but also protein sequences composed of English and characters. It takes multiple rounds of instructions to allow AI to generate a complete protein.

Song Le said that at present, developers need to use prompt word engineering to train large models, so that the content generated by the machine can be in line with human instructions, thereby lowering the threshold for human-computer interaction. For this reason, the company has also developed a protein generation editor to translate the language on the business side to the machine side as much as possible.

Yang Shuo said that the core proposition that AIGC needs to solve in the current segmented scenarios is how to transform the very structured language system and methodological knowledge in some vertical scenarios into content that can be precipitated by prompt words, such as parameter tuning.

The reminder to engineers here is not just as simple as playing with a few large models. In Song Le’s view, prompt word engineers in this field must understand both life sciences and artificial intelligence algorithms, which is very scarce in the talent market, which is why many employers offer high salaries. At present, the company mainly trains internally, and recruits two types of talents, life sciences and algorithm engineers, to jointly develop and build an interdisciplinary team.

Of course, prompt words are not omnipotent, and “spells” will also fail.

Yang Shuo said that with the introduction of AIGC in various industries, the profession of prompt word engineer may disappear, because when business personnel in various industries can master prompt word engineering, it will become a basic skill. Just like when Li Mei started memorizing some classic prompt words, it was to reproduce the master’s ideas, but she eventually wanted to use her own ideas to drive the AI database.

Tian Feng said that the prompt word is the basic skill of operating AI. Don’t exaggerate its role too much. It is just an interface for human-computer interaction in the AIGC era. With the efforts of developers at home and abroad, the large model will gradually improve the efficiency and efficiency of human-computer interaction. Development efficiency, the function of prompt words will also be weakened. At the same time, with the evolution of industry needs, the efficiency and professionalism of prompt words will become stronger and stronger, which is different from the current popular basic capabilities.

Tian Feng believes that the essence of AI is a new generation of productivity in the process of digital transformation. There will not be too many new occupations, but traditional occupations that master new productivity tools will be born in all walks of life.

Yang Shuo said that many new occupations in the future are not new types of work, but after AIGC has empowered the breadth and breadth of talents, it has opened up the boundaries of the capabilities of many traditional positions, such as the integration of sales and pre-sales, product managers and marketing, journalists and editor fusion etc.

Tian Feng said that most of the current AIGC tools are “point” functions, which help people write an email, a picture, a piece of music, etc., while market users need a complete development environment and a total solution. In the next step, AIGC tools are being coordinated and integrated, developing from point to surface, and gradually scaled, engineered, and automated. A standardized collaboration network protocol and interface standard will be formed among the models. Just like the fully automatic production line of automobiles, human beings usher in the era of network-wide model collaboration.