It has been three months since the artificial intelligence trend started by chatGPT. Recently, apart from a few occasional news about: xx big factory has released counterfeit products, xx big factory programmers’ gpt actual combat, xx training institution’s gpt teaching news, It seems that few of us who eat melons have taken the initiative to mention it. In my 7 big fan groups and 52 regional promotion groups, sometimes I don't see a word about gpt artificial intelligence for a whole day.
The superficial reason is very simple. Everyone has been tired of gpt, and there are still a few things that can be done for a few months, such as AI painting, intelligent question and answer. And these non-professional mass entertainment functions will lose their popularity sooner or later. People who calm down find that AI's paintings are still the same over and over again, and the aesthetics have long been tired. And smart questions and answers are also hot in three minutes.
If you want a technology to explode for a long time, it is the only way to actually use it in your work. And it is precisely the characteristics of gpt itself that determine its powerlessness in high-end occasions in the actual professional field. Whether it is the automatic code completion that everyone thinks, or the automatic generation of functions, or the intelligent resolution of error reporting and intelligent answering of technical problems, they cannot really solve the problems in the professional field. For most programmers, it is a smarter point. search engine. In addition to letting laymen sigh: Programmers are going to lose their jobs! Other than that, there was no disturbance.
The thing that makes me most sad and sad is this: he is a blogger who talks about economics and finance that I have followed for many years. His theories about the market and finance refresh me every time. I think he is very professional and a master, although the comments below occasionally say that he is just a magic stick fooling laymen. But I still firmly believe in his professionalism and have practiced it for a long time.
After the gpt fire, the blogger in the financial industry suddenly talked about gpt artificial intelligence to write codes. Every day in the live broadcast room, he claimed that programmers were going to be eliminated, and everyone could be a programmer. App and other advertising words. And during the live broadcast, various demonstrations gpt automatically wrote codes, which attracted thousands of viewers in the live broadcast room to applaud, and the comments were overwhelmingly saying that all programmers would lose their jobs. Immediately afterwards, the blogger posted a tutorial on how to use gpt to make an app, and the price was several thousand... Countless laymen rushed to buy it.
When I saw the so-called automatically generated code he demonstrated in the live broadcast room every day, it was nothing more than a calculator adding, subtracting, multiplying and dividing, or adding, deleting, modifying and checking a certain input box. I once again admired his business thinking, but I also chose not to trust this blogger anymore. After all, he looks ugly. To put it seriously, he is bewitching people. I'm also ashamed that I believed him so much.
The above example is just a drop in the ocean. It can be said that 99% of the videos about gpt that we see now are actually purposeful, deceptive, and misleading advertisements. The purpose is to cut leeks. For example, if you see tutorials on AI generating novels and generating beautiful pictures, others tell you that after spending money to learn, you can eliminate painters, eliminate authors, and make money at zero cost. You believe it, and you spend money to learn it. It turned out that the things generated by AI were thrown into the deep sea to be buried like deformed garbage in other people's professional fields. You will naturally regret your previous self-righteousness.
Although it was used by the profiteers above to ruin the reputation of AI. But AI technology is actually the future. Its future strength is inevitable, but in the current era, there is still an unsolvable contradiction, that is, the amount of data.
The intelligence of gpt and many large models depends on the amount of data. The reason why you think the domestic ones are not as good as chatGPT is because there is a gap in the data volume at the plane level, and the models and source codes are almost the same. Then this problem of data volume will be magnified to become a fatal contradiction in our specific company:
If the company chooses a public large-scale model, although the amount of data is huge, the security and professionalism are not enough. All kinds of unreliable and garbage data may cause huge losses to the company. And the public big model can never be improved in depth and professionalism to the extent that the company can directly use it, otherwise the company's technical moat will disappear and everyone can make competing products, then the benefits will disappear and the company will go bankrupt. Therefore, the company will continue to increase the requirements, so that the public large model can never catch up with the required accuracy.
If the company chooses to localize the small model, judging from the company's poor internal data, the trained AI is basically mentally retarded. Money can hire a few interns to be reliable, and it can also solve employment problems and undertake social responsibilities.
So at present, the future development direction should be the following three:
1. The company adopts a combination of public large model and local small model, and the input and analysis of data and the collection of results are solved by the public large model. But the final result chooses this job, which is selected by the localized small model.
2. The company upgrades the local small model, and manually inserts many algorithms and classifications to affect and improve accuracy. For example, like the POT matrix I designed before, although there is little data, it cannot support a complete AI, but after the classification and intervention of the matrix, it is equivalent to a semi-AI, which can also greatly improve the accuracy.
3. (Secrets are being kept in the training course...the secret will be revealed after the protection time has passed)
In summary, you don't need to panic too much, the gpt natural language model has passed the heat....
What should everyone do? Let's learn traditional AI testing techniques.