【ChatGPT】Will you be the kind of person who is robbed of your job by AI?

---

foreword

Since ChatGPT came out, there have been endless discussions on the Internet. Some people say that AI will replace basic jobs.
A financial practitioner told the "Acorn Business Review" that this is actually a replacement for human experience. Those proud experiences come from the summary of history. After the emergence of AI, "human past experience is no longer It’s important.”
There are also academics who refute that AI has no magic, and it can’t do things that humans can’t do. Take ChatGPT as an example, without human beings splitting the questions in detail, it is impossible to get the correct answer, and the occupation that is easy to be replaced must always do repetitive mechanical labor.
AI also exhibits more exciting capabilities, emerging. A practitioner shared his findings with Acorn Business Review that AI can answer untrained knowledge, as if it has "self-awareness".
But other than that, a more serious problem lies in front of human beings. AI is eliminating small languages ​​and corresponding cultures, such as Bengali and Indian. This is undoubtedly a huge blow to human cultural diversity.
A lot of questions are starting to loom before us:
Will the AI ​​revolution be another scientific revolution? Back in the 18th century, many cultures believed that lightning was a symbol of God's wrath, but Franklin proved that lightning was just an electric wave and invented the lightning rod.
Can co-evolutionary humans do it? Questions followed one after another, astonishment, confusion, excitement, and hidden worries.
Human beings have imagined and explored AI for decades, but it has never been as close as it is at this moment. But throughout human history, this moment is not new.
Human beings domesticate wheat and goats, establish villages, use writing, currency, and even establish empires to explore philosophy and science...not only sailing
to a new world, but also reborn as a new human being.

1. AI replaces "basic work",

From 9500 to 8500 BC, humans shifted from gathering to farming, and domesticated various crops and animals.
Abundant food allows humans to develop the instinct to cooperate on a large scale, forming villages, towns, cities, and kingdoms. This "domestication experience" has been passed down to this day. But with the emergence of AI, human experience seems no longer important.
Zhu Feng, a visa consulting practitioner, told Acorn Business Review that visa consulting is a basic position in the visa consulting industry, that is, to customize visa applications according to customers' needs and circumstances.
Generally speaking, the visa processing rules are a "tree structure", and there will be many subdivisions and rules. The suitable application methods are different according to different conditions. Therefore, according to industry practice, new visa consultants need to study all past visa records before entering the industry.
"After understanding these, we will estimate the probability of the applicant's visa approval, and then recommend relevant plans and communicate preparation materials. This process takes five or six hours." Zhu Feng said.
However, if ChatGPT is applied to the field of visa consultants, it can quickly analyze the probability of customers over-signing, "This almost subverts the work of visa consultants."

The emergence of ChatGPT breaks down the industry barriers established by experience. The financial industry, which analyzes and judges economic trends and avoids and controls investment risks, is already facing AI's "breaking the wall".
Chen Dapeng, an investment analyst at Zijin, told Acorn Business Review that the current mainstream investment strategies in the financial market can be summarized as quantitative analysis and qualitative analysis.
Taking an analyst as an example, there are three main parts of work, collecting extensive and accurate information; using correct processing methods for data; drawing conclusions, practicing and backtesting.
The current ability of ChatGPT is concentrated in the first part of collecting information; after human guidance and teaching, it can also complete the second part of data analysis; but because ChatGPT does not have the ability to judge value, it cannot draw the conclusion of the third part.
"Take the first part as an example. In the past, we wanted to know how the market performed when the Fed raised interest rates in history, or how the prices of assets such as gold, stocks, and bonds changed. We needed to use professional software such as Google, Bloomberg, and Wonder to find out. Search, but now you only need to input natural language in ChatGPT to get good feedback." Chen Dapeng said.
In fact, the "difficulty" of the financial industry is reflected in the collection of information and data analysis.
Chen Dapeng explained that the "threshold" of the financial industry is to monopolize a part of knowledge ownership. For example, only with the help of personal connections and guild organizations can we obtain more data and experience that people in the industry are proud of. But ChatGPT has broken this information barrier; in other words, it has greatly lowered the industry threshold.
Chen Dapeng sighed that after the emergence of AI, "the past experience of human beings is no longer important."

2. AI has no magic: Humans can’t do it, and neither can it

Chen Dapeng's words are like the suicide note of physicist Yang Dong in the science fiction novel "The Three-Body Problem", "Physics, no longer exists." But in the academic world, there are completely opposite views. Lai
Xiaozheng, an associate professor at the School of Computer Science, South China University of Technology, told Acorn Business Review that AI is not magic, and software engineering is not invalid.
Engaged in the research of chip design and verification, and constantly getting along with AI, he found that this is a process of disenchantment.
At the beginning of ChatGPT, Lai Xiaozheng only asked some broad questions. For example, to issue instructions, a 32-bit CPU needs to be designed, but then he discovered that this code was just "copy-pasted" from ChatGPT, so he began to disassemble the problem (refine the prompt words). For example, specify the required CPU framework and pipeline to make it produce a class diagram for software engineering; then use the class diagram to generate a program framework; finally prompt and ask ChatGPT to supplement the framework details. After such continuous customization of details, the quality of the results generated by ChatGPT began to be satisfactory.

Chen Dapeng also has similar experiences and suggestions. People often ask Chen Dapeng whether ChatGPT can choose stocks, and he answers, "Of course, but the premise is that you have to know how to choose a good stock."
Chen Dapeng believes that ChatGPT can assist in some stock selection strategies. For example, you can enter the historical stock price of a stock in ChatGPT, indicating that the strategy is to buy if it breaks through the highest price in history, and let it backtest whether the strategy is profitable. But you can't just say "choose a profitable stock", ChatGPT can't generate a reasonable answer.
Lai Xiaozheng explained that when you ask ChatGPT questions, the broader and more general the questions, the more random the answers you get. In fact, the process of "asking the right question" has been taught in middle school-problem-solving steps.
Today, Lai Xiaozheng requires students to use ChatGPT throughout the process, from literature review to paper writing, from writing code to participating in projects. And when writing related papers, be sure to write down the steps and AI prompts used like a middle school answer sheet.
Lai Xiaozheng believes that the sooner you embrace AI, the better. The above-mentioned "problem-solving steps" are called prompt engineering. word engineer.
03

Three human fears: being replaced and being surpassed

The birth of new professions means that the old professions are dying.
Since ChatGPT came out, the most discussed topic is which professions and industries will be replaced. It is widely believed that some foundational jobs are at risk.
Liu Cheng (pseudonym), the person in charge of Hefei Yuean Electronics, who is in the chip industry, introduced to Acorn Business Review that after experiments, he found that ChatGPT can already realize the code development of some complex scenarios. , module design, or VIP (Verification IP) design, verification plan extraction, GPT can participate in it.”
For example, ChatGPT can write a set of regression (regression test framework) and collect simulation status scripts; it can complete the writing of AHB (Advanced High Performance Bus) Master module, AXI (Advanced eXtensible Interface) and other module content; can write a VIP of AXI or QSPI (Quad Serial Peripheral Interface); even provide testplan (test plan) based on natural language.
But Liu Cheng also pointed out that ChatGPT currently performs differently in different links. For example, in terms of script testing, ChatGPT has reached the level of a senior engineer; but in the code part, it is only at the level of a junior engineer, "even sometimes it cannot reach it." And that has to do with the nature of the job itself.
Lai Xiaozheng also believes that in the chip industry, the first to be affected by AI is the back-end verification engineer. Because occupations that are easily replaced must always do repetitive mechanical labor. As a back-end verification engineer, the script test is very fixed, mechanized, and precise, and there is not much room for development, so it is easy to be replaced by ChatGPT.
In addition to worrying about being "replaced", people are more afraid of being "overtaken" by AI.
In the so-called "AI evolution" discussion, a term is repeatedly mentioned-emergence. It means that the huge quantitative AI model reaches a certain level, which will cause a leap in existing capabilities, and may even produce unexpected and unexplainable results.
Zhu Feng, a visa consulting practitioner, said that this form of expression is like AI having "self-awareness". For example, in common AI training, no one tells the AI ​​that apples can be juiced. Someone asked how to eat an apple, but it said that it could be juiced. Why AI will answer the knowledge that has not been learned is still inconclusive.
In addition to worry, the emergence, or the "innovative ability" of AI, while worrying people, also brings people more imagination and excitement.
Liu Cheng believes that in the future, artificial intelligence can completely realize "1 to n" innovation through big data. Find potential market demand based on user search data, and launch products that better meet user needs through imitation and splicing of leading products in the market. For example, AI finds that dresses on the e-commerce platform have become the first search, and it can determine that the demand for dresses is the highest. It judges the latest trends and the most popular styles through data, integrates the cultures of different countries and regions, and designs "new clothes".
Similar possibilities also exist in the chip industry. But Liu Cheng also said that at present, in the field of "0 to 1" involving the development of basic science and technology, artificial intelligence cannot achieve real innovation.

Four AIs make language return to its ancestors, and small languages ​​and cultures are "endangered"

Humans use the brain to store information, but the brain dies with the death of the human.
In early humans, information mostly disappeared within a century. After the Agricultural Revolution, around 3500 BC, the Sumerians invented numbers and writing to store information. Language is very important to human beings. It is a necessary condition for the human brain to retrieve information efficiently, and it is also the key to the continuation of culture.
However, if AI's large language model is used, countless minority languages ​​will be endangered, and it will affect the existing human culture.
Lai Xiaozheng took programming language as an example. He believes that programming language has already appeared "atavistic phenomenon".
For example, C++ has a history of decades, and engineers prefer to use other concise languages, which allows them to spend less time and energy to write more correct codes, but after ChatGPT came out, this rule was broken, and the amount of code accumulated by C++ Large, using C++, ChatGPT generates the highest quality results.
Also, the large language model is a feedback mechanism. For example, C++ is better than other niche languages, and more people will use C++; the larger the sample size of AI, the more it can satisfy the training, the better the feedback, and the more people will use C++. Form a loop.
When the programming language returns to its ancestors, it is often accompanied by the convergence of global languages. Gradually, some niche subdivided languages ​​​​will slowly disappear. It is worth noting that similar language convergence is also likely to occur in natural language.
Lai Xiaozheng said that as far as ChatGPT is concerned, the English use effect and production quality are higher than those of Chinese, because English materials are the most on the Internet, and other lesser-known languages ​​are also more serious, such as Bengali and Indian.
In Lai Xiaozheng’s words, there are two main reasons for this phenomenon: first, the current AI model has too few samples of minority languages; Some original language details, this can also be confirmed in the official OpenAI.
When OpenAI's official website introduced ChatGPT, it mentioned that many existing ML benchmarks are written in English. In 24 of the 26 languages ​​tested by OpenAI, GPT-4 English language performance outperformed low-resource languages ​​such as Latvian, Welsh, and Swahili.

AI continues to squeeze small languages ​​on the Internet, which is undoubtedly a huge blow to human cultural diversity.

In the future of five human beings, where should education go?

In the past 500 years, human power has grown unprecedentedly. In 1500 AD, the population was about 500 million, and now the global population is 7 billion. This is due to the modern scientific system that dares to admit "ignorance", centers on observation and mathematics, and uses theoretical capabilities to develop new technologies.
In fact, before the scientific revolution, many human cultures did not believe that human beings could progress further, and only needed to abide by the wisdom of their ancestors. For example, in the 18th century, many cultures believed that lightning was a symbol of God's wrath, but Franklin verified that lightning was just an electric wave and invented the lightning rod. The progress of science and technology depends on the investment in education.
Today, the AI ​​revolution, a lesson that human beings must face and learn is to co-evolve with AI. At present, the lesson of co-evolution with AI includes two aspects: "Human Education AI" and "AI Education Human".
According to the "China Artificial Intelligence Talent Development Report (2022)" released by the China Academy of Labor and Social Security Sciences, the number of industrial talents in the AI ​​​​field is about 948,800, of which 68.2% have a bachelor's degree, 9.3% have a master's degree, and 9.3% have a doctor's degree. Education accounted for only 0.1%.
In order to "educate AI", the past three years can be called the outbreak years of new AI majors in colleges and universities. The first batch of colleges and universities in China to offer AI undergraduate majors began to recruit students in 2018. As of 2022, there have been four batches of 440 colleges and universities across the country that have been approved to set up AI majors, accounting for 34.6% of the 1,270 undergraduate colleges and universities.
Taking Zhejiang University as an example, Zhejiang University’s artificial intelligence (Turing class) recruits students from Zhu Kezhen College of Zhejiang University and absorbs the best batch of high school students. The professional compulsory courses include Introduction to Cognitive Neuroscience, Machine Learning, Introduction to Computer Vision, Design Cognitive and Design Intelligence, Artificial Intelligence Chips and Systems, Introduction to Natural Language Processing, etc.
The job market also shows that AI engineers and trainers are in short supply. Many recruitment websites show that the annual salary of undergraduate graduates in the direction of AI is mostly around 200,000 to 300,000 yuan, while the annual salary of graduates with a large amount of experimental experience can reach 40,000 to 500,000 yuan.
In terms of "AI education of human beings", AI tools represented by ChatGPT have shown certain talents. In OpenAI's official test, ChatGPT is like a never-boring and extremely intelligent partner who seems to have infinite patience and professionalism. No matter how naive or tricky the questions asked by students, it can almost always find the answer and never tire of it. explained. At the same time, ChatGPT can also generate personalized test questions that meet the student's level and correct homework.

As long as you master the appropriate questioning method, ChatGPT can make learning unlimited customization.
It is worth noting that it is precisely because of AI's powerful educational auxiliary functions and its tool attributes in the entire industry that it has aroused everyone's thinking about the nature of education. Lai Xiaozheng, associate professor of the School of Computer Science, South China University of Technology, believes that in a broad sense, in the era of AI, our education should abandon the details and focus on cultivating students' logical ability and problem-solving thinking. Because in many fields, the ability to complete specific operations will be fully replaced by AI, and how to "guidance" AI to operate is the job of humans.
Yu Mingfeng, an associate professor at Tongji University’s School of Humanities and deputy director of the Department of Philosophy, also said in an interview with People that if education can actively respond to content-producing artificial intelligence, then there will be many adjustments to the university’s assessment methods and organizational methods .

Summarize

For philosophy education, it is very important for students to get rid of simple knowledge and learn to ask, reflect, and ask more questions. This is also neglected in current education. We should realize that education is not only the training of skills, but also the shaping of personality and the construction of three views-"Who am I? Where do I come from? Where do I want to go?" Throughout the ages, human philosophical speculation and value judgments , spiritual power, etc., have always been shining in the long river of history, and this is the precious humanistic value that AI cannot replace for the time being.
At the same time, Yu Mingfeng also expressed concern that the proficiency of knowledge mastery may be a necessary ladder leading to a reflective ability. It should not be the goal, but it will have the meaning of a ladder. Whether technology is simplifying this ladder or completely removing this necessary ladder is a big question.
After freeing the shackles of excessive knowledge through artificial intelligence, have people's thinking ability, questioning ability, personality shaping, and those concerns about life and society been realized or failed? Can human beings reach the ideal shore? These question marks about the future are the huge questions left for education.