The popularity of large models such as ChatGPT has fueled the new profession of "prompt word engineer". Some people refer to "prompt word engineers" as "the group of people who can spell". It is understood that most of the positions named "Prompt Engineer" or "Prompt Engineer" on major domestic recruitment platforms currently have a monthly salary of 15,000 to 60,000 yuan. Although there is still a lot of debate about whether it is a new career or a temporary job, it is undeniable that AIGC-related careers such as cue word engineers are booming.
It is not new to produce content through Generative Artificial Intelligence (GAI) represented by ChatGPT and Midjourney. However, it is not easy to accurately generate high-level works with the help of GAI. Operators need to split complex tasks Put forward multiple requirements in a language that AI can recognize, so as to obtain more accurate answers. From this, a profession that "specializes in asking AI questions" has also been derived - Prompt Engineer (Prompt Engineer).
Not everyone can have a high-quality conversation with AI, and the prompt word engineer is therefore vividly compared to "the group of people who can spell". Some people think that prompt words are an indispensable part of the popularization of artificial intelligence, and they should keep up with the trend of the times and learn quickly. Some people think that prompt word engineers are a temporary job in the process of training artificial intelligence, and will be in the process of continuous self-improvement of the program. becomes past tense. What kind of occupation is this? How will the future develop? What should we do as ordinary people?
Careers that "chat" with AI
It was a high-paying offer from Scale AI, an artificial intelligence unicorn company, that first made the prompt word engineer popular. In November last year, foreign internet celebrity Riley Goodside (Riley Goodside) joined Scale AI and became "the first prompt word engineer to be recruited" in the industry by virtue of a lot of prompt word skills and experience. It is estimated that the annual salary exceeds million yuan.
The reporter learned that at present, most of the positions named "Prompt Engineer" or "Prompt Engineer" on major domestic recruitment platforms have a monthly salary of 15,000-60,000 yuan, and most job seekers require 1-3 years of experience. Many positions are marked with "experience is not limited". The work content mainly includes designing, managing and optimizing Prompts (prompt words) in related industries for large-scale model training, and cooperating with related business teams to improve the overall product performance.
Unlike traditional programmers, although the prompt word engineer is also a job to deal with machines, the latter sends plain text commands to AI through the natural language that people use every day, and then the AI executes them. Therefore, many people think that this is a job that generates value by skilled chatting, with high income and low threshold.
However, there is still a gap between professional players and ordinary players. The prompt words largely determine the results generated by the model. Even for the same question, inputting different prompt words will result in different results. Tang Xiaolan, an associate professor at the School of Information Engineering of Capital Normal University, pointed out that it is difficult to achieve the final desired complex result through one or two questions. The problem that prompt word engineers need to solve is how to design the problem to achieve the desired result.
In Tang Xiaolan's view, practitioners first need to have certain knowledge in related fields before they can ask questions accurately. "Artificial intelligence is gradually developing in various vertical fields. For example, the lawyer industry requires legal knowledge to accurately describe problems and judge whether the generated results are correct and feasible." At the same time, she believes that practitioners in this profession also need a large number of AIGC (artificial intelligence generated content) experience, know how to optimize the question through multiple rounds of questions to get the expected results. "On this basis, if you can understand the working principle of the large model and improve the quality of questioning from the principle, it will be even more rewarding."
According to the observation of Zhou Xiaokai, a master's student majoring in artificial intelligence at the School of Computer Science of Wuhan University, testing prompt words is not all the work of prompt word engineers. "The large model is in the process of continuous evolution, and the work of the prompt word engineer also involves using the programming language to guide the large language model to produce the output expected by the questioner in a specific task and a specific role. Therefore, this work not only requires understanding natural language, but also Requires an understanding of programming languages."
Take a position of prompt word engineer from a company in the medical field marked "40k-60k·14 salary" on the recruitment platform as an example. The job description requires the position to monitor the performance of the model and the terminal product, continuously optimize the prompt words and Improve product performance; at the same time, it is necessary to cooperate with the medical team to continuously improve product performance through prompt word engineering, improve the medical nature and consistency of products, and provide the best patient experience. The job bonus items also include: familiarity with Python or programming experience, familiarity with doctor's diagnosis and treatment process, etc. According to the reporter's observation, most Prompt-related positions have set certain thresholds in terms of professional knowledge, programming ability, product experience, etc., and are not the "million-dollar annual salary machine chatter" as some people understand.
At present, most Internet technology companies and traditional companies that introduce AI large-scale models have talent needs for prompt word engineers and provide relevant positions. Tang Xiaolan believes that the demand for prompt word engineers in enterprises does exist, and it will be on the rise in a short period of time. "After the emergence of new technology, it takes time to popularize it, so that everyone can gradually become familiar with it. In this process, professional people are needed to do this."
a position? a skill?
Some people also questioned the prompt word engineer as a new profession. Some people believe that from a commercial point of view, products such as ChatGPT and Midjourney have the motivation to further lower the threshold of use so that they can serve more users; at the same time, prompt words are also easy to copy and imitate. Once the entire process is popularized and standardized, this The type of work can easily be replaced.
It can be seen that tech companies are trying to bridge the gap between ordinary users and prompt word engineers. Software such as FusionAI can help users generate more appropriate prompt words; the image generator DALL-E 2 provides users with a very detailed guide for prompt words; the "/describe" command released by Midjourney can reversely deduce from the pictures uploaded by users Prompt words to generate corresponding text descriptions to provide users with creative inspiration.
Prompt words applied in various scenarios are also gradually becoming commodities. There are already Krea, PromptBase, PromptHero and other specialized prompt word platforms abroad that sell clearly marked prompt word products, involving GPT, Midjourney, Stable Diffusion, DALL-E and other models, including music, painting, photography and other scenes. Consumers only need to spend less than 10 dollars to buy a system of prompt words to generate high-quality works. On domestic e-commerce platforms, some merchants are also selling prompt word materials for specific models at lower prices, including graphic tutorials, case analysis of prompt word usage, and tens of thousands of keywords.
Some time ago, Microsoft AI Research Institute even launched a new framework called Automatic Prompt Optimization (APO), trying to make AI optimize the prompt words to train itself without manual work, hoping to help users with the least effort. Human intervention and optimization to create better prompt words.
Prompt word Engineer will become a temporary job? Tang Xiaolan believes that it should be judged according to the situation. "The evolution of large-scale model technology must be done in a direction that is more and more suitable for human beings. From this perspective, Prompt will become more and more suitable for human thinking habits in the future. But unless there is technology in the future that can directly transform the human brain Otherwise, we must express our ideas and present them in a form of text. As long as we are still at this stage, Prompt will always exist, and its optimization and exploration will be will always exist."
Previously, Zheng Yizhou, technical product director of Stable Diffusion, once pointed out that prompting technology should be regarded as a skill, not a job. Tang Xiaolan emphasized that whether it is from a professional perspective or not, how to ask questions and how to better communicate with AI will be necessary skills in the future. "This may not be the need of a certain profession, but an ability that every citizen should master in the digital society in the future. Just like everyone uses computers now, being able to use computers is no longer a special ability in itself. Special positions are required to operate."
Make good use of the tools and find your own position
Whether engineers can be regarded as an independent profession is still controversial in the industry. But it is undeniable that AIGC-related careers such as cue word engineers are booming. Perhaps what should be more concerned now is what job seekers should do in such an era.
A few days ago, the "AIGC Employment Trend Big Data Report 2023" released by Liepin Big Data Research Institute shows that AIGC's talent demand is showing a more obvious upward trend than AI. The comparison found that from the first quarter of 2020 to the first quarter of 2021, the job growth of AIGC and AI was comparable; since then, AIGC has continued to grow in general, and the number of new jobs in the first quarter of 2023 is 5.63 times that of the first quarter of 2020; The growth rate of the number of postings has slowed down relatively, which is 1.95 times that of the first quarter of 2020. In the past year, the average annual salary of AIGC's new job recruitment was 401,200 yuan, which was 80,900 yuan higher than the average annual salary of AI recruitment (320,300 yuan) in the same period.
It needs to be admitted that AIGC has improved the quality and efficiency of ordinary people's daily work, such as copywriting, translation, drawing, etc., and some people may face the risk of unemployment. Li Zhifei, the founder and CEO of GoMask, believes that there are two types of occupations that AIGC is most likely to replace. One is occupations that can complete closed-loop work on a computer and have a lot of repetitive work, such as basic art/design; The first category is occupations that are easy to standardize, such as customer service and marketing copywriting that use fixed scripts. Those least likely to be replaced are those that require face-to-face interaction and rely on physical skills, such as tradesmen such as plasterers, electricians and mechanics, and service workers such as hairdressers, cooks, doctors and nurses.
At the same time, new opportunities are emerging. According to Mo Yu, head of Liepin AI technology, more and more small-scale enterprises will be born due to AIGC's technical assistance; and more digital employees will appear in the workplace to undertake a more detailed division of labor and improve the organization The professionalism and stability of the company further promote business exploration and the improvement of iteration efficiency, which makes it possible for small teams or individual operations to perform businesses that previously required larger-scale collaboration.
"At present, there are many talents with computer-related disciplines in the field of AIGC." Li Zhifei pointed out that because AIGC's demand for talents is becoming more and more diversified, in addition to traditional computer science and data science talents, AIGC also needs to have AI model optimization Professionals with knowledge and experience in natural language processing, robot operating systems, etc.
Faced with the employment pressure brought about by AIGC, Qiu Zeyuan, a master student majoring in computer software and theory at Huazhong University of Science and Technology, lamented that on the computer track, the core competitiveness is the technical threshold, but the application of technology in enterprises is changing with each passing day. , learning classic techniques, sometimes it is difficult to get in touch with the latest framework. In his opinion, for ordinary people, it is necessary to keep learning new things at all times.
For the rapid iteration of technology, Qiu Zeyuan feels that there is no need to be too anxious. "If no one knows how to use this new technology, companies need to employ people based on their ability. Through the internship and training projects, they can quickly realize the skills to use the technology. And when they are in school, students will recommend good learning to each other. projects, such as following up cutting-edge research and hotspots through open source communities.”
At the plenary meeting of the 2023 Zhongguancun Forum held in May this year, Robin Li, chairman and CEO of Baidu, boldly predicted that 50% of the world's jobs in 10 years will be prompt word projects, and called on domestic education to strengthen the cultivation of students' ability to ask questions . As an educator in the field of artificial intelligence, Tang Xiaolan suggested that students, no matter they are majors in computer science or not, should be open to new technologies and consciously exercise their ability to master them. "More importantly, we should cultivate interdisciplinary capabilities in this field and other disciplines in a targeted manner, use the tools well, and find our own position in the constant changes."
