1
Many programmers have expressed dissatisfaction with the widely discussed "35-year-old crisis" in the Internet industry, and it seems that all programmers have a career shelf life of 35 years old. However, with the rise of AI technology, this radical technological revolution is penetrating all industries in a more brutal and direct way. The core values of programmers are being replaced by automation and smart tools. Programmers are no longer facing a traditional 35-year-old age crisis, but rather an end-of-career crisis.
AI technology is rapidly emerging, iterating at a rate so fast that it is measured in weeks, and almost all of the big model companies have made major breakthroughs in AI programming. Programmers are at risk of profound transformation and obsolescence. But because the transition period where AI technology is replacing programmers is not yet completely over, many programmers are still working in the same jobs that they will be replacing. I like to call this "programmer junk time".
Garbage time is a term used in sports to refer to a game in which the difference between the two teams is too great and the winner has been decided. At this point, the remaining time of the game no longer has a decisive impact on the final result, and the remaining time is called garbage time. The term could not be more appropriate when applied to the vast technological revolution, the raging wheel of history. The inevitability of the times is a law that individuals cannot violate.
2
To understand the deeper logic of programmers being replaced by AI, we have to talk about how AI works.
Behind the ability of AI to exhibit intelligence is the core support of a large-scale language model (LLM). Its training logic is based on massive language data, utilizing GPU clusters that can perform trillions of floating-point operations per second (A100, for example, has a double-precision floating-point capacity of 19.5 TFLOPS), and training the language model through efficient algorithms that tune trillions of parameters. We can understand the model simply and crudely as data plus a parameter structure. While this is simple to say, the cost of training the model is quite high. In the case of GPT-4, for example, it takes tens of thousands of A100 graphics cards worth $10,000 each to train the model, hundreds of days to train it at a time, and tens of thousands of dollars a day in electricity costs alone. Not all intelligence models require such a large investment, but such a massive investment to train superintelligence.
Why can models exhibit intelligence? What is the nature of intelligence?
It is widely recognized in the scientific community that the emergent capabilities of large models shape the intelligence of AI. Emergence is the idea that a large entity consists of many interacting small entities, and that the large entity is able to exhibit properties that none of the small entities that make it up have. With the popularity of ChatGPT, more attention has been paid to the emergence capabilities of large language models. Emergence of a large model refers more to the fact that as the model gets larger, the model at some point suddenly exhibits a capability that it did not have before, i.e., intelligence.
Large models have another surprising property: their accuracy rises in a power law as their size changes, which is known as the scaling law. The larger the model, the faster its performance improves. In other words, the intelligence of the model grows exponentially as you increase the amount of data and computation invested in it.
Nowadays, big AI models are penetrating various fields in the form of a kind of human second brain. In particular, the new model recently released by OpenAI makes our second brain not only a storage and extraction tool for knowledge, but also has the ability to think deeply. It is especially good in the field of programming. Program development is naturally suited to AI scenarios because programs are not dichotomous. Program development is highly structured, and every command and syntactic logic in the code has strict rules. Human language expression is ambiguous, the surface meaning and the deeper meaning may be different. But programming language is not, it does not have any ambiguity, only 0 and 1 black and white.
3
This process of AI programming to replace programmers is very similar to the process of new energy replacing traditional fuel vehicles. With the development of new energy technology, the maturity of the industrial chain, the scale effect makes the marginal cost lower and lower, and the competitive advantage of new energy is getting bigger and bigger. But it is not all of a sudden to replace all the fuel cars, and even the fuel car market will not completely disappear. The fuel car market will gradually begin to differentiate, can be replaced by the civilian car market gradually replaced by new energy vehicles, while fuel cars are beginning to move towards the high-end market, gradually oriented to specific groups, such as car enthusiasts and collectors. You can see manual gearboxes in many supercars, which is a kind of side evidence.
The same is true for the programming field, which will likewise gradually diverge under the influence of AI programming. Traditional engineering programming work gradually loses its meaning at this stage, becoming more and more mechanized and instrumentalized. The effort and time invested by programmers, compared to the productivity gains brought about by AI, make them feel that the value of their labor is diluted. As a result, the careers of engineered programmers who handle regular development tasks, operations and maintenance, front- and back-end development, and coding of simple business logic are being threatened by AI tools. On the other hand, those few elite programmers are reused, they have deep theoretical foundation and algorithmic ability, their work is more creative and cutting-edge, and they have transcended the scope of traditional engineers and are more like a scientist. This core difference between science thinking and engineering thinking is reflected in the profession is the difference between engineering and science. This is destined to the future of the programmer's work, or the work of the program scientists, only a very small number of people can be competent.
Massachusetts, Harvard, Stanford, Tsinghua, Peking University, Chinese Academy of Sciences, today's researchers in the field of AI have a very high academic background. Their combination with AI can be described as one for ten, one for a hundred. How much value they create, how much "mediocre" value is replaced.
4
However, any change is always characterized by crisis and opportunity. When we cannot see the future, we might as well look back at the past.
Let's review some of the stories in history from a technical and economic perspective.
First, from a technical point of view.
In April 2016, AlphaGo defeated world Go champion Lee Sedol. Since then, humans have never won in the matter of playing Go. If we want to understand the impact of ChatGPT on humans, we might want to see what impact AlphaGo has had on the Go world.
Today, almost all professional Go players rely on Go AI to improve their game. For example, in the opening phase, the AI will boldly choose many moves that are traditionally considered "no-go" moves. After a number of attacks and defenses, the AI will respond to these moves in another place, revealing a surprisingly far-reaching layout. It was not until then that the professional players realized that this was the case!
Some people describe the AI era as being like a train coming from afar. We see it coming from afar, but when it actually arrives, it just passes us by in a flash. Then, we can only watch it get farther and farther away from us, never able to catch up.
In the 19th century, electricity was limited to upper class use. When it was proposed to spread electricity to thousands of homes, there was opposition in Parliament. Some people warned: "If electric plugs were installed in every house, their lethality would be incalculable, just like modern weapons! If a kitten accidentally touches an electric plug, won't it be instantly electrocuted? If an older brother tells his younger brother to touch a plug, won't he be instantly killed? If we universalize electricity throughout London, or even all of England, wouldn't that be like handing out a gun to everyone, including the bad guys and the mentally ill? This is just horrible!""
Consequently, the resolution was not adopted at that time.
However, looking back from today's perspective, electricity has brought enormous value to the entire human community, and almost everyone has benefited from it. It is true that electrocution accidents do occur from time to time, but even so, we will not stop using electricity because of this. What we need to do is to learn how to use electricity properly and safely.
5
Let's look at it again from a capital perspective.
We have to admit that capital is the core force driving the rapid development of technology. The massive investment of capital has made AI the hottest field of the moment. Various startups, research projects, and innovative products have received unprecedented support. Although this will inevitably lead to a bubble in the AI market, this excessive push is precisely to stimulate innovation and filter out competitive companies and products. Frantically throwing money at the market to activate it is exactly how capital plainly acts. In the end, a few companies are bound to stand out and form an oligarchic pattern. These companies are created through full competition, and it doesn't matter who wins in the end.
Does this look like a repeat of the original internet bubble era?
Around 2000, the Internet similarly experienced a bubble and a financial crisis. But the rise of something new is inevitably accompanied by chaos and turbulence. Companies groped their way across the river, eventually figuring out successful business models that shaped today's Internet landscape. And today's tech giants were all created during that Internet bubble era. Baidu, founded in 2000; Tencent, founded in 1998; Alibaba, founded in 1999; Google, founded in 1998; Facebook, founded in 2004; Amazon, founded in 1994; Netflix, founded in 1997. All of these companies arose during the Internet bubble and emerged from that chaos.
The AI technology revolution we are now facing is also accompanied by an economic recession, but this is definitely not the decline of an era, but the beginning of a brand new era. unicorns of the AI era are in the making.
Junk time in a programmer's career (above)
/2024/09/15/
Junk time in a programmer's career (below)
/2024/09/15/