On February 15, 2024 local time in the United States, OpenAI officially released the Wensheng model SORA, entering the field of generation. It can generate a coherent ** of 60s only according to the prompt words, crushing the "generation length of the industry that is currently only about 4s", and this model has instantly sparked heated discussions in the global science and technology circle with its powerful ability.
The biggest highlight of SORA is that it can automatically generate up to 60 seconds of coherence according to simple prompt words**, imagine, just enter a few keywords or a short description, SORA can quickly transform these ideas into vivid **content, whether it is advertising, promotional videos or teaching**, you can easily get it! This undoubtedly opens a whole new door for content creators, allowing creative wings to fly more freely. This innovation of openai not only represents the latest breakthrough in the field of artificial intelligence technology in the field of generation, but also brings infinite possibilities for the future of content creation. Below I will introduce the prompt words in detail from the following aspects.
1. What is prompt engineering?
Prompt engineering, also known as instruction engineering, is a technique that guides and controls the behavior of AI models by carefully designing and structuring prompts (instructions). It's not just a simple input, it's an advanced strategy and technique designed to allow AI models to better understand and perform complex tasks.
To put it simply, a prompt is an instruction you send to a large model, such as telling a joke, making a snake game in python, writing a love letter to your boyfriend and girlfriend, etc., which seems very simple, but it is very meaningful.
Editing Searching for pictures to learn to prompt engineering, just like learning to use a mouse and keyboard, is a basic skill in the AGI era.
Prompt engineering has a low threshold and a high ceiling, so some people jokingly call prompt a spell.
However, a dedicated prompt engineer will not last long, because in the future work and life, as long as AI is used, everyone must be able to prompt engineering, and the evolution of AI will also make prompt engineering easier and simpler.
2. Typical composition of prompt
The main way of generative AI is dialogue, and the basic modules and ideas of the dialogue system are shown in the figure below
The core idea of the search is summarized in three points:
1.Transform the input natural language dialogue into a structured representation.
2.From a structured representation, build a policy.
3.Convert the strategy into natural language output.
Therefore, prompt will be the basic language of the AI era, and everyone needs and will master prompt engineering. The composition of the prompt words is actually very simple, mainly to be able to explain to the large model his role positioning, what you need him to do, how to do it, and what the requirements for the result are. Here's a typical prompt composition.
For example, the templates provided in the Inspiration Encyclopedia of Zhipu Qingyan are basically composed according to this framework. 
3. Advanced prompt skills
Give examples: You can teach the AI to learn your habits and requirements, and output the content according to the examples you give him, so that the output content can be more stable and more in line with your requirements.
Add vertical knowledge: that is, you first give the AI a piece of material, let him learn the content of the material first, and then you ask him about the expertise in this field, and he will give you the answer you want very accurately.
Add constraints: change the style of tone, tone, etc. For example, you can ask him to omit some modal words or words that are not relevant to the answer and describe the answer directly, or you can output the content in the specified format you request.
Use the chain of thought: The chain of thought is a magical ability that emerges from large models that was discovered by accident (the people of OpenAI didn't expect it to be like this when they were trained). Someone asks a question with let's think step by step, only to find that the AI breaks down the problem into multiple steps and then solves it step by step, making the output more accurate. Let the AI generate more relevant content to form a richer version of the above, thereby increasing the probability that the following will be correct. It is especially effective for complex problems involving computation and logical reasoning.
Let's think about it, isn't it the same with people-to-people communication? First tell him what role he is, and then describe the problem or need to be communicated, and say the desired outcome, so that the communication efficiency is very high. If he is asked to think about it a little longer and disassemble the question step by step, the answer will be more reliable. So, you have to treat AI as a human being.
Ilya Sutskever, chief scientist at OpenAI, said that digital neural networks and biological neural networks of the human brain are mathematically the same. So, let's treat AI as a human being! Look at AI as a human being! Look at AI as a human being! How you get along with people, how you get along with AI.
Understand AI in terms of human seem
Use "When a person looks" to control the AI
Use When People Look to convince users to take AI into perspective.
Fourth, use prompt to optimize prompt
Finally, give a magical spell and let chatgpt write a prompt for you. Just paste it into the chatgpt dialog box, and the original text can be obtained by private message.
prompt to generate prompt