The content of this article is based on Ibrahim John's "The Art of Asking ChatGPT" (the art of asking questions to ChatGPT).
foreword
Today, ChatGPT is gaining more and more people's favor, through which people can enter questions and get answers. But in addition to simple question and answer, ChatGPT has many hidden ways of asking questions. Do you want to find out? Today, we will introduce to you how to use Prompt technology to extract high-quality answers from the massive data of ChatGPT. Let's unravel the mystery of ChatGPT together!
The following is the directory of this article:
2. How to use Prompt to help ask questions on ChatGPT
2.4 Tips for knowledge integration
1. What is Prompt
Before introducing how to use Prompt to get high-quality answers from ChatGPT, let me briefly explain what Prompt is.
Prompt is a natural language processing model launched by OpenAI in 2015. It is based on the powerful GPT (Generative Pre-training Transformer) processing model. This processing model has the great advantage of being pre-trained with large-scale data and fine-tuned with labeled data to produce high-quality natural language representations. Although the Prompt processing model is developing day by day, researchers still find some problems, for example, a single training cannot fully cover life scenarios, which leads to the inability of the questioner to get the expected answer. Therefore, it has become more and more important to provide appropriate input for the model. In the next section, I will introduce several common questioning models for you.
(The picture comes from the Internet)
2. How to use Prompt to help ask questions on ChatGPT
Recall the way we used to search for questions. For example, if you want to find information about a certain star or a certain company, you usually directly enter "who is xxx star" or "what business does xxx company do" in the search bar, or simpler The method is to directly input "xxx star" and "xxx enterprise" and press the Enter key to find the required information. In Prompt, this search method is called "zero sample, one sample and few sample prompting method". In short, it generates a natural language version from large-scale language model predictions using minimal or no examples. Just like the example above, we don’t know who the star is at all, but the model can output the star’s information in a predictive way for our retrieval reference. This efficient and intelligent search method uses a large amount of data and powerful algorithms to provide us with a more accurate and convenient information acquisition experience. After reading the above examples, do you know something about the intelligent search method of Prompt? Next, the editor will introduce some other commonly used Prompt questioning methods, and let us explore the charm of this prompt language model together.
2.1 Seed Word Prompt
Seed word hinting is a natural language generation technique that takes seed words as input. Users can use specific seed words or phrases to control the output of ChatGPT. The prompt formula of the seed word prompt is a seed word or phrase followed by the command "Please generate text based on the following seed words". Here are some examples and formulas of using seed words to ask questions:
-
Scenario: Text generation, want to generate a story about basketball.
Seed word: "basketball".
Prompt formula: "Please generate text based on the following seed words: basketball"
2. Scenario: I want to write an article with the theme of "College Entrance Examination".
Seed word: "College Entrance Examination".
Prompt formula: "As an author, please use "college entrance examination" as the seed word to generate a short article of 200 characters".
(use seed word hints)
2.2 Dialogue Prompt
A dialogue prompt is a text that simulates a dialogue between two or more entities. It can provide the model with context and a set of role entities, and generate a dialogue between them. This mode is useful for simulating dialogue, generating chatbots, etc. Quests are very useful. Here are a few examples and formulas for dialog prompts:
-
Scenario: dialogue generation, want to generate a dialogue between mobile phone sellers and buyers.
Task: Generate a dialogue between mobile phone sellers and buyers.
Prompt formula: "In the following context, please generate a conversation between a mobile phone seller and a seller".
-
Scene: Story writing, I want to write a story about Pleasant Goat and Bald Qiang meeting in the park.
Task: Create a story about Pleasant Goat and Bald Qiang meeting in the park.
Prompt formula: "In the story of Pleasant Goat and Bald Qiang meeting in the park, generate a dialogue between them".
-
Scenario: Development of a chatbot.
Task: Generate a chatbot that promotes safety education for young children.
Prompt formula: "When a child asks for information on a safety education topic, generate professional and accurate conversations for the chatbot".
(dialogue prompt)
2.3 Encouraging hints
Encouraging hints Tips are a kind of technical hints used to encourage ChatGPT to generate thinking. The method used is to add a "let's think about it" sentence in front of the question or topic to be asked. This mode is useful for writing articles and other tasks. Very useful, such as the following example:
Scenario: Want to understand what harm global warming will bring to human beings.
Task; understand the harm of global warming to human beings.
Prompt formula: "Let's think about what harm global warming will do to human beings."
The question in this example will get a more detailed answer than "what harm will global warming bring about".
("Let's think about it" prompt)
2.4 Tips for knowledge integration
Knowledge integration is the use of models to integrate pre-existing knowledge into information or connect different information. This technology helps to generate new knowledge on the basis of the original knowledge system. Here are a few examples and formulas for knowledge integration:
-
Scenario: I want to combine the knowledge of the quadratic equation with the knowledge of chickens and rabbits in the same cage
Task: Combine the knowledge of the quadratic equation with the knowledge of chicken and rabbit in the same cage
Prompt formula: "Integrate existing knowledge of quadratic equations in one variable: chicken and rabbit in the same cage"
-
Scenario: I want to connect the knowledge of the quadratic equation with the knowledge of chickens and rabbits in the same cage.
Task: connect the knowledge of quadratic equations in one variable and the knowledge of chickens and rabbits in the same cage.
Prompt formula: "Connect the following information in a relevant logical and manner: Quadratic equation in one variable. Chicken and rabbit in the same cage."
(knowledge integration tips)
2.5 Self-consistency hints
Self-consistency hints are ways to ensure that what ChatGPT outputs is consistent with the provided input. This method plays a very important role in the verification and verification of data values. The following are some examples and formulas of self-consistency:
-
Scenario: Want to generate a product review consistent with the reporting plugin.
Task: Generate a review about the reporting add-in product.
Prompt formula: "Generate a product review that matches the report plugin product information."
-
Scenario: Check for consistency in a given set of grade information data.
Task: Check the consistency of grade data.
Prompt formula: "Please make sure that the text is self-consistent: "The data shows that the average math score of the second class of the fifth grade is 75 points, but the minimum score is 40 points. "
(self-consistency prompt)
2.6 Sentiment Analysis Tips
As the name implies, the sentiment analysis prompt is a model that can analyze the emotional color or attitude of the text. It can judge whether the sentiment of this piece of information is positive, negative or neutral based on the input information. For example, it can be analyzed in self-consistency. The emotional color of the report plugin product reviews:
Scenario: Analyzing the emotional color of product reviews in the report plug-in
Prompt formula: "Perform sentiment analysis on the following customer reviews [insert review for report plugin product] and classify them as positive, negative, or neutral."
(Sentiment Analysis Tips)
2.7 Training ChatGPT
Training ChatGPT is a delicate work. By constantly asking specific types of questions, we have improved the sensitivity of the model to this type of question in different dialogue scenarios, so that it is more likely to give relevant answers to such questions. . For example, if we want a dialogue that is only for translation, in the first sentence of the dialogue we can say: "You can play the role of a language translator, I will communicate with you in any language, you will automatically detect language and translate it, and self-optimize the answer to continuously improve the quality of the answer based on my instructions." This way ChatGPT can be used like a translator and can be used to create many different types of conversational patterns.
3. Summary:
In addition to the several methods mentioned above, ChatGPT has many broad application fields. For example, it can be used as an editing tool for a table plug-in to help us modify and optimize table data to improve our work efficiency and accuracy. ChatGPT can identify and analyze the context and structure of tables by learning large-scale data sets, and generate corresponding text descriptions. We can enter data and required format requirements, such as dates, numbers, text and other information, and then ChatGPT can automatically generate a form that meets the requirements, improving the accuracy of processing through continuous learning and optimization. In addition, ChatGPT can also be integrated with other databases and plug-ins to expand its functions and applications . In this way, ChatGPT has become a very useful tool that can help us process tabular data, promote the efficiency and accuracy of our work, and also provide a wide range of application prospects that can serve many fields and industries.
References:
Ibrahim John——《The Art of Asking ChatGPT》
Original link: https://qingbaoyuan.vip/6252.html