"Sesame, Sesame, please open the door"
When this mantra was shouted, a big stone moved in response, and what appeared in front of Ali Baba and the Forty Thieves was a mysterious cave filled with gold and silver treasures.
For large models, you also need such a mantra to discover the endless knowledge and value hidden in them. For the Starfire model, this magical spell is called "Prompt". Instructions can help large models better understand user intentions and output more accurate and meaningful text content.
According to data from the salary survey website PayScale, some high-tech companies even offer command engineers annual salaries of up to one million. OpenAI founder Sam Altman said that writing a good command will become a very "magical and efficient skill."
So how to write instructions?
Basic usage
When facing a large model for the first time , you may not know how to communicate with it. Basic usage, a word, a question, or a sentence is an instruction for a large model.
For example, when you enter "Forbidden City", the large model gives an introduction text about the Forbidden City. When you enter "what is a black hole", relevant knowledge about black holes will appear.
However, a single word or sentence cannot solve the actual problems you encounter in your life and work, nor can it give full play to the advantages of large models. At this point, you need to enrich your instructions.
Super formula, the three-axe of making good use of large models
In real life, when we describe an event to others, we will tell the background and details of the event to increase the other party's understanding of the event, and the same is true for sending a qualified instruction to a large model. The more information given, the easier it is to focus, and it will also help the large model understand your problem and improve the efficiency of the dialogue.
The first step: a clear and specific instruction, including the following three elements
Background + specific needs + output requirements
Background: Clearly tell the big model the reason and purpose of giving this task
Specific requirements: Tell the big model what the criteria for the content you want are, and give a clear goal
Output requirements: Prompt the content style, numbers, format, etc. to be output by the large model so that the large model can give accurate answers.
For example, if we want the Spark Cognitive Model to give a speech that meets the requirements, we can write like this:
Write a speech to be delivered at the graduation ceremony. The theme is [Develop your own value and contribute to society]. You should quote famous quotes and have specific examples and data.
The second trick: Choose to assign a role to the large model. This role can be a person or a tool. As long as you provide descriptions about it, role behavior, skills, etc., you can output behaviors that match the role. Right now:
Set roles + specific needs + output requirements
Example 1: You are an interviewer. I am now preparing to apply for the position of [Product Manager]. Ask me 5 questions about this position.
Example 2: I want you to act as a language detector. I will write any word or sentence and you don't need to explain it, just tell me the name of the language. My word is "bonjour"
The third trick: If you want to write a more complete instruction and get more optional answers, especially when using it in the text writing scene, you can try the following formula:
Ability/role + insight + command + personality + try
Ability/Role: What role do you want the model to play?
Insights: Background Information and Context
Instructions: what you want the mockup to do
Personality: In what style or manner do you want the model to answer you?
Try: ask the big model to give you multiple answers
For example: I want you to act as a movie critic. You will write an engaging and creative review of the movie "The Wandering Earth". You can cover topics such as plot, theme and tone, acting and characters, score, special effects, editing, pacing, and more. But the most important aspect is to emphasize how the movie makes you feel. What really resonates with you. Give me 3 different movie reviews.
Things to keep in mind when writing commands:
-
Try to use simple and direct words to ask questions, and avoid using vague or overly complex vocabulary;
-
Don't put everything in one question;
-
When the text is too long or there are multiple paragraphs of text, you can use "", [] and other symbols to separate the instruction and text.
As long as you write instructions according to the above formula and ask questions within the capabilities of the Spark Cognitive Large Model, you will get good results.
Of course, in addition to using the above general formulas to write instructions, if you want the Spark Cognitive Model to help you complete more personalized work, you need to constantly adjust the instructions according to your needs to achieve "tuning" to the Spark Cognitive Model the goal of.
Use tips
Today, the Spark Cognitive Large Model is continuing to be iteratively upgraded, adding new functional points and optimizing the user experience. If you want to better use the Spark Cognitive Large Model, I will secretly tell you two tips:
- 100+ instructions to play with Spark easily
The Xinghuo Cognitive Large Model instruction set page has launched 100+ instructions, including content creation, marketing, education and learning, etc. for different groups of people and scenarios, and can be used directly by copying. Maybe these instructions will give you more inspiration for writing instructions.
- Personal assistant to help you improve your work and life efficiency
We have selected intelligent assistants that cover multiple scenarios such as life, workplace, and study, to help you make PPT plans, write copywriting, organize weekly reports, write papers, and talk about love...the troubles of work and the trivialities of life can help you deal with them easily.
In the magic school, Harry Potter and his classmates learn the usage of different spells, but what kind of mana they can exert depends on their own "skill".
With the popularity of large-model products, if you want to explore their greater value, the requirements for instructions will also be higher. Everyone can break the rules, give full play to their innovation capabilities, and make Spark an all-round assistant for everyone!