“At first we thought thisAI)yesinternet"It's an opportunity that doesn't come once in a decade, but the more I think about it, the more I feel that it's an opportunity that doesn't come once in hundreds of years, similar to the industrial revolution that invented electricity." The above paragraph was said by Ma Huateng at Tencent's 2023 shareholders' meeting. Artificial intelligence is the 2023 technology circlemaximumThere is no doubt that AI is a highlight of the industry, but just like the Industrial Revolution, changing the world comes at a price. The more efficient and convenient AIGC (generative AI) has changed the careers of many people in just one year.
Source Note: The image is generated by AI, and the image is authorized by Midjourney
AI replaces real people, starting from the field of game artFirstGun. Earlier in the fall of 2022, two AI painting tools, Midjourney and Stable Diffusion, became popular along with the chatbot ChatGPT. In the past year, on social platforms such as Weibo, Zhihu, and Maimai, many former game artists who were laid off because game manufacturers introduced AI into the art workflow have come forward to share their experiences.
But as recently as the first half of 2022, AI painting was still just a concept that was ridiculed, and painters had no fear that AI would take away their jobs. Therefore, for quite a long time, painters believed that AI painting was the result of a large AI model learning a large number of human paintings, breaking up the massive elements and then splicing them together, and they firmly believed that AI painting was just "stitching together corpses."
However, in less than a year, AI has grown faster than artists could have imagined.up to dateThe Midjourney version even broke the curse that AI cannot draw accurate human hands. At the same time, with the help of different plug-ins, models, and Lora, AI painting tools can almost realize every style you can imagine, and even imitate the styles of masters such as Picasso and Monet.
The most important thing is that the cost of AI painting is extremely low. It does not completely require computing cards specially built for AI, such as NVIDIA Hopper H100 and Ampere A100. Ordinary gaming graphics cards can "run pictures". Even the most productive painters are hit by AI's dimensionality reduction in drawing efficiency. AI can solve quality problems by piling up quantity. It is not a problem for a high-performance graphics card to produce thousands of pictures a day. A major game artist commented on AI painting in this way, "The workload that used to take a week or two now only takes three or four days."
Large companies have introduced AI into early design and later detail depiction, supplemented by manual photo editing, while small and medium-sized companies have used AI instead of outsourcing.maximumThe improvement means increased efficiency and productivity, which means that the development of a large-scale game previously required an art team of 100 people, but now only 20 to 30 people are needed, and the labor cost is greatly reduced. The terrifying thing about AI painting is that it now perfectly balances cost reduction and efficiency improvement.
In addition to the mid- and low-level painters who have been forced out of work by AI, the life of writers is not easy either. In the past 2023, from Germany's Bild to the United StatesmaximumIt is not new for news publishers such as Gannett and CNN to use AI to participate in news writing. Since AI can already draw pictures, it is no problem for writing articles. After all, ChatGPT is famous for its "human" text answers. However, attempts to use AI to replace human editors have almost all failed.
News reports written by AI lack key details, have a lot of repetition, and use strange language styles. This is because under regulatory pressure, AI is constrained by developers to say "correct nonsense", and the arrangement and combination of materials cannot create a style.
In a sense, AI is best at formal official language and clichés, but this will inevitably lead to the content being written being tasteless. In fact, all this is not surprising, because the amount of information conveyed by text is greater than that of images and even more so than that of videos. This is also why AI painting and AI video generation are in full swing, but the most successful use case of AI in the field of text is to generate a corresponding summary based on a piece of text.
If AI is a high threat to painters and a low threat to text creators, then it is a medium threat to programmers. The CEO of Stability AI, the developer of Stable Diffusion, said, "In five years, human programmers will completely disappear."
In fact, in the past few years, the technology industry has been trying to replace programmers, from drag-and-drop website building in the PC Internet era, to low-code in the mobile Internet era, to the current AI writing code.
The work of a programmer is actually very similar to that of a translator, except that translation is usually an interaction between people, while a programmer translates human needs to computers. In simple terms, the work of a programmer is to translate the needs of the real world into source code (Source Code) through computer language, and then convert it into machine code (Native Code) that can be directly processed by the CPU through a compiler, and finally form an executable program, and then maintain the operation of the program in the future.
In this process, AI is currently mainly involved in generating code, which is actually the function of AI programming tools such as GitHub Copilot. Generating code is not the key to a programmer's work, but understanding requirements and implementing them is, and subsequent code maintenance is also very important, but these are currently beyond the ability of AI to solve.
Taking GitHub Copilot as an example, researchers analyzed the effectiveness of GitHub Copilot and found that with its addition, the code churn rate (that is, the situation where the code is reworked, modified, or deleted shortly after being written) has increased significantly.
The researchers collected 150 million lines of code generated by GitHub Copilot and found that GitHub Copilot prefers to add code directly and does not encourage reuse, which means that AI programming tools do not encourage code reuse, but directly rewrite. The problem is that the quality of code directly rewritten by AI is often worrying, otherwise there would not be an increase in code churn, which means that developers dare not directly apply the code generated by GitHub Copilot in the production environment, and must spend time checking the code given by AI and fixing related bugs.
AI will replace humans and cause massive unemployment, which is actually a topic that has been discussed since ChatGPT became popular. But a year has passed, and AI has indeed affected the jobs of some people, but for many people, it is just a false alarm for the time being.