Why can’t AI use prompts well?
Master the art of communicating with AI and improve work efficiency and accuracy.
Core content:
1. The importance and role of prompt words in AI communication
2. The application of gap theory and illusion gap theory in AI communication
3. Practical case analysis of natural language programming and prompt words
Introduction:
There are all kinds of large models on the market, with different IQs and abilities. But in common, if you want to use them well, you must master how to communicate with them - that is, the prompt words.
A previous article mentioned:
Human hallucinations are much greater than large models.
Especially in the workplace, we are constantly dealing with illusions created by the process of information transmission.
"I did not mean that..."
"Why can't you understand?"
"You said you understood!"
"I misunderstood, I thought..."
"You're right, but I don't think so."
The first step to reaching a consensus is to break down everyone’s “I thought”.
Gap theory and hallucinations
The gap theory mainly describes that every customer has an "expected value" for the product before purchasing it, and will generate "actual value" feedback after purchasing and experiencing the product and service.
The gap between "expected value" and "actual value" determines the customer's next purchasing decision.
Some companies attach great importance to collecting customer feedback to improve their products and services.
The situation where "expected value" and "actual value" are completely equal generally occurs when the goods are highly standardized and purchased very rationally.
If we use this theory to judge, how many situations are there in which the "expectation" and "objectivity" of the work content are completely equal? Or how many situations are there in which the work content and output requirements are very standard?
In most cases, we cannot manage "expectations" based on "objective conditions". As long as this matter is not standardized, "objectivity" will be an illusion of personal judgment.
Just as there are a thousand Hamlets in the eyes of a thousand people, it is extremely difficult to reach a consensus on unified standards.
According to the current amount of knowledge in the big model, it has long surpassed the knowledge reserves of each individual, but it is impossible to make it everyone's confidant.
So we need to let it know what "Hamlet" looks like in each of our eyes. And the appearance of "Hamlet" is completely limited by each of our understanding.
No one except Shakespeare can give an accurate definition of Hamlet. When there are no clear requirements and standards for output, illusion is inevitable.
The prompt is to program in natural language
Friends who have watched "Captain America" should remember the character "Winter Soldier Bucky".
After the Winter Soldier fell off a cliff while on a mission with Captain America, he was rescued by Hydra and brainwashed.
When someone read out the command: "желание (desire) ржавчина (rust) семнадцать (seventeen) рассвет (dawn) печь (fire) девять (nine) доброта (kindness) домой (go home) один (one) грузовик (truck) солдат (soldier)", the Winter Soldier began to execute the killing procedure.
This is the "cue word" that controls the brain using natural language programming.
The essence of the prompt word is "command".
Before 1981, the only way for people to interact with computers was the command line interface (CLI).
I believe many of you remember calling out the "cmd" interface in Windows and typing "tasklist" to quickly query all computer processes.
In 1981, Xerox Corporation of the United States invented the "graphical user interface" (GUI), which completely announced a new human-computer interaction paradigm based on graphical interfaces. With the help of a mouse and keyboard, the interaction with computer programs is completed through visual graphics. GUI has laid a solid foundation for the development of today's personal PCs.
In the 1990s, the emergence of the "natural user interface" (NUI) made the interaction with computers not limited to keyboard and mouse. With the development of technology, it has gradually progressed from gestures, eyeballs, and body to the current brain-computer.
With the emergence of large natural language models, the way to drive artificial intelligence has changed from code to text, and "Prompt" has become a way to input commands to control computers.
But it still maintains the basic characteristics of traditional computer language programming: logically combining various instructions and requirements.
In other words, when using prompt words, users must know what they want and be able to understand how the computer thinks in order for the instruction to be executed accurately.
The current scenario of using big models is often that we don’t know what we want before we ask it. Big models are more like a smarter search engine that gives us some ideas to think about.
When we try to make it perform some actions according to its ideas, we will find that it is still different from what we want, because we ourselves have not figured out the ideas and results.
Users only need to make their needs clear
I believe that prompt words are just a transitional solution, and artificial intelligence has just finished the "first half". As intelligent applications go deeper into scenarios, more vertical intelligence will reduce the occurrence of hallucinations.
So there is no need to spend too much energy learning "how to program in natural language". What we need more is to make clear the goals and requirements we want to achieve.
After all, demand is the eternal "command line" of supply.
END
