A new job created by ChatGPT: Prompt Engineer

A new job created by ChatGPT: Prompt Engineer

What is a Prompt Engineer?

It’s about identifying bugs and hidden features of AI so that developers can act on those discoveries.

If you're looking for the hottest job in tech, you can try to learn how to communicate with an AI chatbot, which could land you a job as a Generative AI Prompt Engineer.

Prompt engineers are good at asking questions to AI chatbots running on large language models to get the answers they need. Unlike traditional computer engineers who write code, Prompt engineers test the AI ​​system for any odd behavior or issues by asking the AI ​​a series of logically thoughtful questions.

The so-called prompt engineer, that is, the AI ​​trainer, writes relevant prompts by interacting with AI to help users operate AI to generate more perfect content, and help companies train employees to use these tools. In short, prompt engineers are equivalent to a combination of designers, copywriters and programmers.

The capital feast triggered by ChatGPT is still going on, everyone is talking about the AI ​​technology behind it, everyone is worried about whether they will be replaced by AI, but few people have noticed the new jobs brought by this technology . Known as a Prompt Engineer, the job can pay as much as $335,000 a year, but doesn't require employees to hold a degree in computer engineering. For example, Anthropic, a Google-backed start-up, is hiring a “reminder engineer and librarian” in San Francisco, with a salary of up to $335,000.

 

Why learn Prompt?
Improve the accuracy of generated results: Prompt is an important tool that can be used to guide GPT to generate specific output results. By optimizing the design of Prompt, the accuracy and efficiency of GPT's generated results can be improved, and the error rate of output can be reduced. Ability to control generated content: Prompt allows users to control the content generated by GPT. By changing the content and format of the Prompt, you can control the style and expression generated by GPT. For example, if you want GPT to generate an article, you can set Prompt to let GPT generate an article on a specified topic. Help the model understand the task: Prompt can help the model understand the task better. By designing appropriate prompts, the model can better understand the requirements of the task, thereby generating more accurate and targeted results. For example, for a machine translation task, by setting the prompt, you can instruct the model to translate domain-specific terms and expressions.

 

What are the common usage methods of Prompt?

Basic format
The most basic prompt format is a simple sentence or phrase used to provide a starting topic or direction to the model. For example: "write a science fiction novel", "describe a summer beach", "write a short story", etc. Prompts in this basic format usually guide the model to generate relevant text.

Conversational format
The conversational prompt format can be used to have a dialogue with the model, allowing the model to generate responses based on previous input. For example: "How's the weather in Beijing today?" "...." "What kind of clothes should I wear?" and so on.

Full paragraph format
The full paragraph format is usually used to let the model generate a complete article or paragraph. This format usually provides more detailed background information, such as topics, characters, places, events, and so on. For example: "Write a story about a dog. The main character is a loyal Labrador who lives in a certain city and has an owner named Jack. The story can be centered on the dog, or it can include some other element."

Conditional formatting
The prompt of conditional formatting will provide some conditions, and the model needs to generate text according to these conditions. For example: "If you had a million dollars, what would you do?" This format is often used in interesting scenarios, such as playing a game or generating interesting dialogue.

Answer format
An answer format prompt usually gives some questions and asks the model to answer the questions. This format can be used to generate question answering systems, it can also be used to generate article summaries, and so on. For example: "Who was the first president of the United States?", "What is artificial intelligence?", etc.

Multi-turn dialog format
Prompts in the multi-turn dialog format are often used for more complex conversations that build on previous conversations. For example, "I went to the beach yesterday and had a great time. Did you do anything interesting recently?" This format requires the model to be able to remember previous conversations and carry on.

Format at the beginning of the paragraph
The prompt at the beginning of the paragraph usually provides some sentences or paragraphs at the beginning, allowing the model to continue to generate based on it. For example: "Knights were a very important profession in ancient times. They were often seen as symbols of bravery, justice, and honor. Please write a story about knights." This format is often used to guide models in a specific generated on the topic.

Diversity control format
The diversity control format prompt is usually used to control the diversity of the text generated by the model. For example: "Write a story where one of the keywords is 'cats', but make sure to generate different results each time." This format can be used to generate more interesting and varied text results.

Context generation format
A prompt in context generation format usually provides some text in the context, allowing the model to generate according to the context. For example: "In this story, the main character is a man named Wang Fei. Please continue to write this story, describing some of the challenges and adventures Wang Fei encountered." This format is often used to guide the model in a specific context. generated in .

List-generating formats
Prompts in list-generating formats typically provide a list on which to base the model's text generation. For example: "The following are some topics about travel: 1. Favorite places to travel; 2. Favorite travel companions; 3. Favorite ways to travel. Please write an article about travel based on these topics." This This format is usually used to guide the model to generate a specific list of topics, and generate based on it.

Abstract generation format
A summary generation format prompt typically provides an article and asks the model to generate a summary. For example: "Please generate a summary based on the following article: ..." This format is often used to guide the model to learn to extract key information from a large amount of information and generate a concise summary.

Text modification format
A text modification format prompt usually provides some text for the model to modify. For example: "The following is an incomplete article, please revise and improve it:..." This format is usually used to guide the model to revise and edit the text.

Fill-in-the-
blank format A prompt in fill-in-the-blank format typically provides a piece of text with some spaces left in it for the model to fill in. For example: "The following is a piece of text with spaces, please fill in the correct answer: ________________, sweat drops to the soil." This format is usually used to guide the model to learn to understand the context in the text and fill in the correct answer based on it. s answer.

Agent Dialog Format
Prompts in the agent dialog format typically provide a conversation starter for the model to continue the conversation. For example: "Your current identity is Lu Xun, please talk to me based on this identity", "...", "What was your original intention when you wrote the article "From Baicaoyuan to Sanwei Bookstore"?" This format is often used to guide the model to learn to perform dialogue generation in a specific situation.

Sentiment generation format
The prompt of the emotion generation format usually provides an emotion or sentiment, and the model is generated based on it. For example: "Please write an article describing the emotion of sadness." This format is often used to guide the model to learn to generate under a specific emotion or mood.

Special Note:
In order for GPT to better understand Prompt, the following methods can be used:

1. Provide clear and concise prompts as much as possible, and avoid using overly complex and difficult-to-understand sentences and terms;

2. Include contextual information in the prompt to help GPT better understand tasks and goals;

3. Provide detailed instructions and instructions as much as possible to ensure that GPT can complete the task accurately. This includes using clear language and specific examples.